Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * postmaster.c
4 : : * This program acts as a clearing house for requests to the
5 : : * POSTGRES system. Frontend programs connect to the Postmaster,
6 : : * and postmaster forks a new backend process to handle the
7 : : * connection.
8 : : *
9 : : * The postmaster also manages system-wide operations such as
10 : : * startup and shutdown. The postmaster itself doesn't do those
11 : : * operations, mind you --- it just forks off a subprocess to do them
12 : : * at the right times. It also takes care of resetting the system
13 : : * if a backend crashes.
14 : : *
15 : : * The postmaster process creates the shared memory and semaphore
16 : : * pools during startup, but as a rule does not touch them itself.
17 : : * In particular, it is not a member of the PGPROC array of backends
18 : : * and so it cannot participate in lock-manager operations. Keeping
19 : : * the postmaster away from shared memory operations makes it simpler
20 : : * and more reliable. The postmaster is almost always able to recover
21 : : * from crashes of individual backends by resetting shared memory;
22 : : * if it did much with shared memory then it would be prone to crashing
23 : : * along with the backends.
24 : : *
25 : : * When a request message is received, we now fork() immediately.
26 : : * The child process performs authentication of the request, and
27 : : * then becomes a backend if successful. This allows the auth code
28 : : * to be written in a simple single-threaded style (as opposed to the
29 : : * crufty "poor man's multitasking" code that used to be needed).
30 : : * More importantly, it ensures that blockages in non-multithreaded
31 : : * libraries like SSL or PAM cannot cause denial of service to other
32 : : * clients.
33 : : *
34 : : *
35 : : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
36 : : * Portions Copyright (c) 1994, Regents of the University of California
37 : : *
38 : : *
39 : : * IDENTIFICATION
40 : : * src/backend/postmaster/postmaster.c
41 : : *
42 : : * NOTES
43 : : *
44 : : * Initialization:
45 : : * The Postmaster sets up shared memory data structures
46 : : * for the backends.
47 : : *
48 : : * Synchronization:
49 : : * The Postmaster shares memory with the backends but should avoid
50 : : * touching shared memory, so as not to become stuck if a crashing
51 : : * backend screws up locks or shared memory. Likewise, the Postmaster
52 : : * should never block on messages from frontend clients.
53 : : *
54 : : * Garbage Collection:
55 : : * The Postmaster cleans up after backends if they have an emergency
56 : : * exit and/or core dump.
57 : : *
58 : : * Error Reporting:
59 : : * Use write_stderr() only for reporting "interactive" errors
60 : : * (essentially, bogus arguments on the command line). Once the
61 : : * postmaster is launched, use ereport().
62 : : *
63 : : *-------------------------------------------------------------------------
64 : : */
65 : :
66 : : #include "postgres.h"
67 : :
68 : : #include <unistd.h>
69 : : #include <signal.h>
70 : : #include <time.h>
71 : : #include <sys/wait.h>
72 : : #include <ctype.h>
73 : : #include <sys/stat.h>
74 : : #include <sys/socket.h>
75 : : #include <fcntl.h>
76 : : #include <sys/param.h>
77 : : #include <netdb.h>
78 : : #include <limits.h>
79 : :
80 : : #ifdef USE_BONJOUR
81 : : #include <dns_sd.h>
82 : : #endif
83 : :
84 : : #ifdef USE_SYSTEMD
85 : : #include <systemd/sd-daemon.h>
86 : : #endif
87 : :
88 : : #ifdef HAVE_PTHREAD_IS_THREADED_NP
89 : : #include <pthread.h>
90 : : #endif
91 : :
92 : : #include "access/xlog.h"
93 : : #include "access/xlog_internal.h"
94 : : #include "access/xlogrecovery.h"
95 : : #include "common/file_perm.h"
96 : : #include "common/pg_prng.h"
97 : : #include "lib/ilist.h"
98 : : #include "libpq/libpq.h"
99 : : #include "libpq/pqsignal.h"
100 : : #include "pgstat.h"
101 : : #include "port/pg_bswap.h"
102 : : #include "port/pg_getopt_ctx.h"
103 : : #include "postmaster/autovacuum.h"
104 : : #include "postmaster/bgworker_internals.h"
105 : : #include "postmaster/pgarch.h"
106 : : #include "postmaster/postmaster.h"
107 : : #include "postmaster/syslogger.h"
108 : : #include "postmaster/walsummarizer.h"
109 : : #include "replication/logicallauncher.h"
110 : : #include "replication/slotsync.h"
111 : : #include "replication/walsender.h"
112 : : #include "storage/aio_subsys.h"
113 : : #include "storage/fd.h"
114 : : #include "storage/io_worker.h"
115 : : #include "storage/ipc.h"
116 : : #include "storage/pmsignal.h"
117 : : #include "storage/proc.h"
118 : : #include "storage/shmem_internal.h"
119 : : #include "tcop/backend_startup.h"
120 : : #include "tcop/tcopprot.h"
121 : : #include "utils/datetime.h"
122 : : #include "utils/memutils.h"
123 : : #include "utils/pidfile.h"
124 : : #include "utils/timestamp.h"
125 : : #include "utils/varlena.h"
126 : :
127 : : #ifdef EXEC_BACKEND
128 : : #include "common/file_utils.h"
129 : : #include "storage/pg_shmem.h"
130 : : #endif
131 : :
132 : :
133 : : /*
134 : : * CountChildren and SignalChildren take a bitmask argument to represent
135 : : * BackendTypes to count or signal. Define a separate type and functions to
136 : : * work with the bitmasks, to avoid accidentally passing a plain BackendType
137 : : * in place of a bitmask or vice versa.
138 : : */
139 : : typedef struct
140 : : {
141 : : uint32 mask;
142 : : } BackendTypeMask;
143 : :
144 : : StaticAssertDecl(BACKEND_NUM_TYPES < 32, "too many backend types for uint32");
145 : :
146 : : static const BackendTypeMask BTYPE_MASK_ALL = {(1 << BACKEND_NUM_TYPES) - 1};
147 : : static const BackendTypeMask BTYPE_MASK_NONE = {0};
148 : :
149 : : static inline BackendTypeMask
651 heikki.linnakangas@i 150 :CBC 1725 : btmask(BackendType t)
151 : : {
152 : 1725 : BackendTypeMask mask = {.mask = 1 << t};
153 : :
154 : 1725 : return mask;
155 : : }
156 : :
157 : : static inline BackendTypeMask
594 andres@anarazel.de 158 : 39061 : btmask_add_n(BackendTypeMask mask, int nargs, BackendType *t)
159 : : {
160 [ + + ]: 152881 : for (int i = 0; i < nargs; i++)
161 : 113820 : mask.mask |= 1 << t[i];
651 heikki.linnakangas@i 162 : 39061 : return mask;
163 : : }
164 : :
165 : : #define btmask_add(mask, ...) \
166 : : btmask_add_n(mask, \
167 : : lengthof(((BackendType[]){__VA_ARGS__})), \
168 : : (BackendType[]){__VA_ARGS__} \
169 : : )
170 : :
171 : : static inline BackendTypeMask
172 : 4956 : btmask_del(BackendTypeMask mask, BackendType t)
173 : : {
174 : 4956 : mask.mask &= ~(1 << t);
175 : 4956 : return mask;
176 : : }
177 : :
178 : : static inline BackendTypeMask
594 andres@anarazel.de 179 : 2883 : btmask_all_except_n(int nargs, BackendType *t)
180 : : {
651 heikki.linnakangas@i 181 : 2883 : BackendTypeMask mask = BTYPE_MASK_ALL;
182 : :
594 andres@anarazel.de 183 [ + + ]: 7839 : for (int i = 0; i < nargs; i++)
184 : 4956 : mask = btmask_del(mask, t[i]);
651 heikki.linnakangas@i 185 : 2883 : return mask;
186 : : }
187 : :
188 : : #define btmask_all_except(...) \
189 : : btmask_all_except_n( \
190 : : lengthof(((BackendType[]){__VA_ARGS__})), \
191 : : (BackendType[]){__VA_ARGS__} \
192 : : )
193 : :
194 : : static inline bool
195 : 128550 : btmask_contains(BackendTypeMask mask, BackendType t)
196 : : {
197 : 128550 : return (mask.mask & (1 << t)) != 0;
198 : : }
199 : :
200 : :
201 : : BackgroundWorker *MyBgworkerEntry = NULL;
202 : :
203 : : /* The socket number we are listening for connections on */
204 : : int PostPortNumber = DEF_PGPORT;
205 : :
206 : : /* The directory names for Unix socket(s) */
207 : : char *Unix_socket_directories;
208 : :
209 : : /* The TCP listen address(es) */
210 : : char *ListenAddresses;
211 : :
212 : : /*
213 : : * SuperuserReservedConnections is the number of backends reserved for
214 : : * superuser use, and ReservedConnections is the number of backends reserved
215 : : * for use by roles with privileges of the pg_use_reserved_connections
216 : : * predefined role. These are taken out of the pool of MaxConnections backend
217 : : * slots, so the number of backend slots available for roles that are neither
218 : : * superuser nor have privileges of pg_use_reserved_connections is
219 : : * (MaxConnections - SuperuserReservedConnections - ReservedConnections).
220 : : *
221 : : * If the number of remaining slots is less than or equal to
222 : : * SuperuserReservedConnections, only superusers can make new connections. If
223 : : * the number of remaining slots is greater than SuperuserReservedConnections
224 : : * but less than or equal to
225 : : * (SuperuserReservedConnections + ReservedConnections), only superusers and
226 : : * roles with privileges of pg_use_reserved_connections can make new
227 : : * connections. Note that pre-existing superuser and
228 : : * pg_use_reserved_connections connections don't count against the limits.
229 : : */
230 : : int SuperuserReservedConnections;
231 : : int ReservedConnections;
232 : :
233 : : /* The socket(s) we're listening to. */
234 : : #define MAXLISTEN 64
235 : : static int NumListenSockets = 0;
236 : : static pgsocket *ListenSockets = NULL;
237 : :
238 : : /* still more option variables */
239 : : bool EnableSSL = false;
240 : :
241 : : int PreAuthDelay = 0;
242 : : int AuthenticationTimeout = 60;
243 : :
244 : : bool log_hostname; /* for ps display and logging */
245 : :
246 : : bool enable_bonjour = false;
247 : : char *bonjour_name;
248 : : bool restart_after_crash = true;
249 : : bool remove_temp_files_after_crash = true;
250 : :
251 : : /*
252 : : * When terminating child processes after fatal errors, like a crash of a
253 : : * child process, we normally send SIGQUIT -- and most other comments in this
254 : : * file are written on the assumption that we do -- but developers might
255 : : * prefer to use SIGABRT to collect per-child core dumps.
256 : : */
257 : : bool send_abort_for_crash = false;
258 : : bool send_abort_for_kill = false;
259 : :
260 : : /* special child processes; NULL when not running */
261 : : static PMChild *StartupPMChild = NULL,
262 : : *BgWriterPMChild = NULL,
263 : : *CheckpointerPMChild = NULL,
264 : : *WalWriterPMChild = NULL,
265 : : *WalReceiverPMChild = NULL,
266 : : *WalSummarizerPMChild = NULL,
267 : : *AutoVacLauncherPMChild = NULL,
268 : : *PgArchPMChild = NULL,
269 : : *SysLoggerPMChild = NULL,
270 : : *SlotSyncWorkerPMChild = NULL;
271 : :
272 : : /* Startup process's status */
273 : : typedef enum
274 : : {
275 : : STARTUP_NOT_RUNNING,
276 : : STARTUP_RUNNING,
277 : : STARTUP_SIGNALED, /* we sent it a SIGQUIT or SIGKILL */
278 : : STARTUP_CRASHED,
279 : : } StartupStatusEnum;
280 : :
281 : : static StartupStatusEnum StartupStatus = STARTUP_NOT_RUNNING;
282 : :
283 : : /* Startup/shutdown state */
284 : : #define NoShutdown 0
285 : : #define SmartShutdown 1
286 : : #define FastShutdown 2
287 : : #define ImmediateShutdown 3
288 : :
289 : : static int Shutdown = NoShutdown;
290 : :
291 : : static bool FatalError = false; /* T if recovering from backend crash */
292 : :
293 : : /*
294 : : * We use a simple state machine to control startup, shutdown, and
295 : : * crash recovery (which is rather like shutdown followed by startup).
296 : : *
297 : : * After doing all the postmaster initialization work, we enter PM_STARTUP
298 : : * state and the startup process is launched. The startup process begins by
299 : : * reading the control file and other preliminary initialization steps.
300 : : * In a normal startup, or after crash recovery, the startup process exits
301 : : * with exit code 0 and we switch to PM_RUN state. However, archive recovery
302 : : * is handled specially since it takes much longer and we would like to support
303 : : * hot standby during archive recovery.
304 : : *
305 : : * When the startup process is ready to start archive recovery, it signals the
306 : : * postmaster, and we switch to PM_RECOVERY state. The background writer and
307 : : * checkpointer are already running (as these are launched during PM_STARTUP),
308 : : * and the startup process continues applying WAL. If Hot Standby is enabled,
309 : : * then, after reaching a consistent point in WAL redo, startup process
310 : : * signals us again, and we switch to PM_HOT_STANDBY state and begin accepting
311 : : * connections to perform read-only queries. When archive recovery is
312 : : * finished, the startup process exits with exit code 0 and we switch to
313 : : * PM_RUN state.
314 : : *
315 : : * Normal child backends can only be launched when we are in PM_RUN or
316 : : * PM_HOT_STANDBY state. (connsAllowed can also restrict launching.)
317 : : * In other states we handle connection requests by launching "dead-end"
318 : : * child processes, which will simply send the client an error message and
319 : : * quit. (We track these in the ActiveChildList so that we can know when they
320 : : * are all gone; this is important because they're still connected to shared
321 : : * memory, and would interfere with an attempt to destroy the shmem segment,
322 : : * possibly leading to SHMALL failure when we try to make a new one.)
323 : : * In PM_WAIT_DEAD_END state we are waiting for all the dead-end children
324 : : * to drain out of the system, and therefore stop accepting connection
325 : : * requests at all until the last existing child has quit (which hopefully
326 : : * will not be very long).
327 : : *
328 : : * Notice that this state variable does not distinguish *why* we entered
329 : : * states later than PM_RUN --- Shutdown and FatalError must be consulted
330 : : * to find that out. FatalError is never true in PM_RECOVERY, PM_HOT_STANDBY,
331 : : * or PM_RUN states, nor in PM_WAIT_XLOG_SHUTDOWN states (because we don't
332 : : * enter those states when trying to recover from a crash). It can be true in
333 : : * PM_STARTUP state, because we don't clear it until we've successfully
334 : : * started WAL redo.
335 : : */
336 : : typedef enum
337 : : {
338 : : PM_INIT, /* postmaster starting */
339 : : PM_STARTUP, /* waiting for startup subprocess */
340 : : PM_RECOVERY, /* in archive recovery mode */
341 : : PM_HOT_STANDBY, /* in hot standby mode */
342 : : PM_RUN, /* normal "database is alive" state */
343 : : PM_STOP_BACKENDS, /* need to stop remaining backends */
344 : : PM_WAIT_BACKENDS, /* waiting for live backends to exit */
345 : : PM_WAIT_XLOG_SHUTDOWN, /* waiting for checkpointer to do shutdown
346 : : * ckpt */
347 : : PM_WAIT_XLOG_ARCHIVAL, /* waiting for archiver and walsenders to
348 : : * finish */
349 : : PM_WAIT_IO_WORKERS, /* waiting for io workers to exit */
350 : : PM_WAIT_CHECKPOINTER, /* waiting for checkpointer to shut down */
351 : : PM_WAIT_DEAD_END, /* waiting for dead-end children to exit */
352 : : PM_NO_CHILDREN, /* all important children have exited */
353 : : } PMState;
354 : :
355 : : static PMState pmState = PM_INIT;
356 : :
357 : : /*
358 : : * While performing a "smart shutdown", we restrict new connections but stay
359 : : * in PM_RUN or PM_HOT_STANDBY state until all the client backends are gone.
360 : : * connsAllowed is a sub-state indicator showing the active restriction.
361 : : * It is of no interest unless pmState is PM_RUN or PM_HOT_STANDBY.
362 : : */
363 : : static bool connsAllowed = true;
364 : :
365 : : /* Start time of SIGKILL timeout during immediate shutdown or child crash */
366 : : /* Zero means timeout is not running */
367 : : static time_t AbortStartTime = 0;
368 : :
369 : : /* Length of said timeout */
370 : : #define SIGKILL_CHILDREN_AFTER_SECS 5
371 : :
372 : : static bool ReachedNormalRunning = false; /* T if we've reached PM_RUN */
373 : :
374 : : bool ClientAuthInProgress = false; /* T during new-client
375 : : * authentication */
376 : :
377 : : bool redirection_done = false; /* stderr redirected for syslogger? */
378 : :
379 : : /* received START_AUTOVAC_LAUNCHER signal */
380 : : static bool start_autovac_launcher = false;
381 : :
382 : : /* the launcher needs to be signaled to communicate some condition */
383 : : static bool avlauncher_needs_signal = false;
384 : :
385 : : /* received START_WALRECEIVER signal */
386 : : static bool WalReceiverRequested = false;
387 : :
388 : : /* set when there's a worker that needs to be started up */
389 : : static bool StartWorkerNeeded = true;
390 : : static bool HaveCrashedWorker = false;
391 : :
392 : : /* set when signals arrive */
393 : : static volatile sig_atomic_t pending_pm_pmsignal;
394 : : static volatile sig_atomic_t pending_pm_child_exit;
395 : : static volatile sig_atomic_t pending_pm_reload_request;
396 : : static volatile sig_atomic_t pending_pm_shutdown_request;
397 : : static volatile sig_atomic_t pending_pm_fast_shutdown_request;
398 : : static volatile sig_atomic_t pending_pm_immediate_shutdown_request;
399 : :
400 : : /* event multiplexing object */
401 : : static WaitEventSet *pm_wait_set;
402 : :
403 : : #ifdef USE_SSL
404 : : /* Set when and if SSL has been initialized properly */
405 : : bool LoadedSSL = false;
406 : : #endif
407 : :
408 : : #ifdef USE_BONJOUR
409 : : static DNSServiceRef bonjour_sdref = NULL;
410 : : #endif
411 : :
412 : : /* State for IO worker management. */
413 : : static TimestampTz io_worker_launch_next_time = 0;
414 : : static int io_worker_count = 0;
415 : : static PMChild *io_worker_children[MAX_IO_WORKERS];
416 : :
417 : : /*
418 : : * postmaster.c - function prototypes
419 : : */
420 : : static void CloseServerPorts(int status, Datum arg);
421 : : static void unlink_external_pid_file(int status, Datum arg);
422 : : static void getInstallationPaths(const char *argv0);
423 : : static void checkControlFile(void);
424 : : static void handle_pm_pmsignal_signal(SIGNAL_ARGS);
425 : : static void handle_pm_child_exit_signal(SIGNAL_ARGS);
426 : : static void handle_pm_reload_request_signal(SIGNAL_ARGS);
427 : : static void handle_pm_shutdown_request_signal(SIGNAL_ARGS);
428 : : static void process_pm_pmsignal(void);
429 : : static void process_pm_child_exit(void);
430 : : static void process_pm_reload_request(void);
431 : : static void process_pm_shutdown_request(void);
432 : : static void dummy_handler(SIGNAL_ARGS);
433 : : static void CleanupBackend(PMChild *bp, int exitstatus);
434 : : static void HandleChildCrash(int pid, int exitstatus, const char *procname);
435 : : static void HandleFatalError(QuitSignalReason reason, bool consider_sigabrt);
436 : : static void LogChildExit(int lev, const char *procname,
437 : : int pid, int exitstatus);
438 : : static void PostmasterStateMachine(void);
439 : : static void UpdatePMState(PMState newState);
440 : :
441 : : pg_noreturn static void ExitPostmaster(int status);
442 : : static int ServerLoop(void);
443 : : static int BackendStartup(ClientSocket *client_sock);
444 : : static void report_fork_failure_to_client(ClientSocket *client_sock, int errnum);
445 : : static CAC_state canAcceptConnections(BackendType backend_type);
446 : : static void signal_child(PMChild *pmchild, int signal);
447 : : static bool SignalChildren(int signal, BackendTypeMask targetMask);
448 : : static void TerminateChildren(int signal);
449 : : static int CountChildren(BackendTypeMask targetMask);
450 : : static void LaunchMissingBackgroundProcesses(void);
451 : : static void maybe_start_bgworkers(void);
452 : : static bool maybe_reap_io_worker(int pid);
453 : : static void maybe_start_io_workers(void);
454 : : static TimestampTz maybe_start_io_workers_scheduled_at(void);
455 : : static bool CreateOptsFile(int argc, char *argv[], char *fullprogname);
456 : : static PMChild *StartChildProcess(BackendType type);
457 : : static void StartSysLogger(void);
458 : : static void StartAutovacuumWorker(void);
459 : : static bool StartBackgroundWorker(RegisteredBgWorker *rw);
460 : : static void InitPostmasterDeathWatchHandle(void);
461 : :
462 : : #ifdef WIN32
463 : : #define WNOHANG 0 /* ignored, so any integer value will do */
464 : :
465 : : static pid_t waitpid(pid_t pid, int *exitstatus, int options);
466 : : static void WINAPI pgwin32_deadchild_callback(PVOID lpParameter, BOOLEAN TimerOrWaitFired);
467 : :
468 : : static HANDLE win32ChildQueue;
469 : :
470 : : typedef struct
471 : : {
472 : : HANDLE waitHandle;
473 : : HANDLE procHandle;
474 : : DWORD procId;
475 : : } win32_deadchild_waitinfo;
476 : : #endif /* WIN32 */
477 : :
478 : : /* Macros to check exit status of a child process */
479 : : #define EXIT_STATUS_0(st) ((st) == 0)
480 : : #define EXIT_STATUS_1(st) (WIFEXITED(st) && WEXITSTATUS(st) == 1)
481 : : #define EXIT_STATUS_3(st) (WIFEXITED(st) && WEXITSTATUS(st) == 3)
482 : :
483 : : #ifndef WIN32
484 : : /*
485 : : * File descriptors for pipe used to monitor if postmaster is alive.
486 : : * First is POSTMASTER_FD_WATCH, second is POSTMASTER_FD_OWN.
487 : : */
488 : : int postmaster_alive_fds[2] = {-1, -1};
489 : : #else
490 : : /* Process handle of postmaster used for the same purpose on Windows */
491 : : HANDLE PostmasterHandle;
492 : : #endif
493 : :
494 : : /*
495 : : * Postmaster main entry point
496 : : */
497 : : void
11006 scrappy@hub.org 498 : 1012 : PostmasterMain(int argc, char *argv[])
499 : : {
500 : : pg_getopt_ctx optctx;
501 : : int opt;
502 : : int status;
7993 tgl@sss.pgh.pa.us 503 : 1012 : char *userDoption = NULL;
5705 504 : 1012 : bool listen_addr_saved = false;
4583 peter_e@gmx.net 505 : 1012 : char *output_config_variable = NULL;
506 : :
2869 tmunro@postgresql.or 507 : 1012 : InitProcessGlobals();
508 : :
509 : 1012 : PostmasterPid = MyProcPid;
510 : :
8492 tgl@sss.pgh.pa.us 511 : 1012 : IsPostmasterEnvironment = true;
512 : :
513 : : /*
514 : : * Start our win32 signal implementation
515 : : */
516 : : #ifdef WIN32
517 : : pgwin32_signal_initialize();
518 : : #endif
519 : :
520 : : /*
521 : : * We should not be creating any files or directories before we check the
522 : : * data directory (see checkDataDir()), but just in case set the umask to
523 : : * the most restrictive (owner-only) permissions.
524 : : *
525 : : * checkDataDir() will reset the umask based on the data directory
526 : : * permissions.
527 : : */
3064 sfrost@snowman.net 528 : 1012 : umask(PG_MODE_MASK_OWNER);
529 : :
530 : : /*
531 : : * By default, palloc() requests in the postmaster will be allocated in
532 : : * the PostmasterContext, which is space that can be recycled by backends.
533 : : * Allocated data that needs to be available to backends should be
534 : : * allocated in TopMemoryContext.
535 : : */
9556 tgl@sss.pgh.pa.us 536 : 1012 : PostmasterContext = AllocSetContextCreate(TopMemoryContext,
537 : : "Postmaster",
538 : : ALLOCSET_DEFAULT_SIZES);
539 : 1012 : MemoryContextSwitchTo(PostmasterContext);
540 : :
541 : : /* Initialize paths to installation files */
6326 542 : 1012 : getInstallationPaths(argv[0]);
543 : :
544 : : /*
545 : : * Set up signal handlers for the postmaster process.
546 : : *
547 : : * CAUTION: when changing this list, check for side-effects on the signal
548 : : * handling setup of child processes. See tcop/postgres.c,
549 : : * bootstrap/bootstrap.c, postmaster/bgwriter.c, postmaster/walwriter.c,
550 : : * postmaster/autovacuum.c, postmaster/pgarch.c, postmaster/syslogger.c,
551 : : * postmaster/bgworker.c and postmaster/checkpointer.c.
552 : : */
4527 553 : 1012 : pqinitmask();
1301 tmunro@postgresql.or 554 : 1012 : sigprocmask(SIG_SETMASK, &BlockSig, NULL);
555 : :
1323 556 : 1012 : pqsignal(SIGHUP, handle_pm_reload_request_signal);
557 : 1012 : pqsignal(SIGINT, handle_pm_shutdown_request_signal);
558 : 1012 : pqsignal(SIGQUIT, handle_pm_shutdown_request_signal);
559 : 1012 : pqsignal(SIGTERM, handle_pm_shutdown_request_signal);
135 andrew@dunslane.net 560 : 1012 : pqsignal(SIGALRM, PG_SIG_IGN); /* ignored */
561 : 1012 : pqsignal(SIGPIPE, PG_SIG_IGN); /* ignored */
1323 tmunro@postgresql.or 562 : 1012 : pqsignal(SIGUSR1, handle_pm_pmsignal_signal);
563 : 1012 : pqsignal(SIGUSR2, dummy_handler); /* unused, reserve for children */
564 : 1012 : pqsignal(SIGCHLD, handle_pm_child_exit_signal);
565 : :
566 : : /* This may configure SIGURG, depending on platform. */
539 heikki.linnakangas@i 567 : 1012 : InitializeWaitEventSupport();
1323 tmunro@postgresql.or 568 : 1012 : InitProcessLocalLatch();
569 : :
570 : : /*
571 : : * No other place in Postgres should touch SIGTTIN/SIGTTOU handling. We
572 : : * ignore those signals in a postmaster environment, so that there is no
573 : : * risk of a child process freezing up due to writing to stderr. But for
574 : : * a standalone backend, their default handling is reasonable. Hence, all
575 : : * child processes should just allow the inherited settings to stand.
576 : : */
577 : : #ifdef SIGTTIN
135 andrew@dunslane.net 578 : 1012 : pqsignal(SIGTTIN, PG_SIG_IGN); /* ignored */
579 : : #endif
580 : : #ifdef SIGTTOU
581 : 1012 : pqsignal(SIGTTOU, PG_SIG_IGN); /* ignored */
582 : : #endif
583 : :
584 : : /* ignore SIGXFSZ, so that ulimit violations work like disk full */
585 : : #ifdef SIGXFSZ
586 : 1012 : pqsignal(SIGXFSZ, PG_SIG_IGN); /* ignored */
587 : : #endif
588 : :
589 : : /* Begin accepting signals. */
1301 tmunro@postgresql.or 590 : 1012 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
591 : :
592 : : /*
593 : : * Options setup
594 : : */
8868 tgl@sss.pgh.pa.us 595 : 1012 : InitializeGUCOptions();
596 : :
597 : : /*
598 : : * Parse command-line options. CAUTION: keep this in sync with
599 : : * tcop/postgres.c (the option sets should not conflict) and with the
600 : : * common help() function in main/main.c.
601 : : */
150 heikki.linnakangas@i 602 : 1012 : pg_getopt_start(&optctx, argc, argv, "B:bC:c:D:d:EeFf:h:ijk:lN:OPp:r:S:sTt:W:-:");
603 : 1012 : optctx.opterr = 1;
604 [ + + ]: 3635 : while ((opt = pg_getopt_next(&optctx)) != -1)
605 : : {
10581 bruce@momjian.us 606 [ - + + + : 2627 : switch (opt)
+ + - - -
+ - - - -
+ - - - -
+ - - - -
- - - ]
607 : : {
10580 bruce@momjian.us 608 :UBC 0 : case 'B':
150 heikki.linnakangas@i 609 : 0 : SetConfigOption("shared_buffers", optctx.optarg, PGC_POSTMASTER, PGC_S_ARGV);
10580 bruce@momjian.us 610 : 0 : break;
611 : :
5603 bruce@momjian.us 612 :CBC 57 : case 'b':
613 : : /* Undocumented flag used for binary upgrades */
614 : 57 : IsBinaryUpgrade = true;
615 : 57 : break;
616 : :
5439 617 : 5 : case 'C':
150 heikki.linnakangas@i 618 : 5 : output_config_variable = strdup(optctx.optarg);
5439 bruce@momjian.us 619 : 5 : break;
620 : :
1354 peter@eisentraut.org 621 : 820 : case '-':
622 : :
623 : : /*
624 : : * Error if the user misplaced a special must-be-first option
625 : : * for dispatching to a subprogram. parse_dispatch_option()
626 : : * returns DISPATCH_POSTMASTER if it doesn't find a match, so
627 : : * error for anything else.
628 : : */
150 heikki.linnakangas@i 629 [ - + ]: 820 : if (parse_dispatch_option(optctx.optarg) != DISPATCH_POSTMASTER)
631 nathan@postgresql.or 630 [ # # ]:UBC 0 : ereport(ERROR,
631 : : (errcode(ERRCODE_SYNTAX_ERROR),
632 : : errmsg("--%s must be first argument", optctx.optarg)));
633 : :
634 : : pg_fallthrough;
635 : : case 'c':
636 : : {
637 : : char *name,
638 : : *value;
639 : :
150 heikki.linnakangas@i 640 :CBC 1282 : ParseLongOption(optctx.optarg, &name, &value);
1354 peter@eisentraut.org 641 [ + + ]: 1282 : if (!value)
642 : : {
643 [ + - ]: 1 : if (opt == '-')
644 [ + - ]: 1 : ereport(ERROR,
645 : : (errcode(ERRCODE_SYNTAX_ERROR),
646 : : errmsg("--%s requires a value",
647 : : optctx.optarg)));
648 : : else
1354 peter@eisentraut.org 649 [ # # ]:UBC 0 : ereport(ERROR,
650 : : (errcode(ERRCODE_SYNTAX_ERROR),
651 : : errmsg("-c %s requires a value",
652 : : optctx.optarg)));
653 : : }
654 : :
1354 peter@eisentraut.org 655 :CBC 1281 : SetConfigOption(name, value, PGC_POSTMASTER, PGC_S_ARGV);
656 : 1278 : pfree(name);
657 : 1278 : pfree(value);
658 : 1278 : break;
659 : : }
660 : :
10580 bruce@momjian.us 661 : 1008 : case 'D':
150 heikki.linnakangas@i 662 : 1008 : userDoption = strdup(optctx.optarg);
10580 bruce@momjian.us 663 : 1008 : break;
664 : :
10580 bruce@momjian.us 665 :UBC 0 : case 'd':
150 heikki.linnakangas@i 666 : 0 : set_debug_options(atoi(optctx.optarg), PGC_POSTMASTER, PGC_S_ARGV);
7956 tgl@sss.pgh.pa.us 667 : 0 : break;
668 : :
7539 peter_e@gmx.net 669 : 0 : case 'E':
670 : 0 : SetConfigOption("log_statement", "all", PGC_POSTMASTER, PGC_S_ARGV);
671 : 0 : break;
672 : :
673 : 0 : case 'e':
674 : 0 : SetConfigOption("datestyle", "euro", PGC_POSTMASTER, PGC_S_ARGV);
675 : 0 : break;
676 : :
9582 bruce@momjian.us 677 :CBC 103 : case 'F':
8951 peter_e@gmx.net 678 : 103 : SetConfigOption("fsync", "false", PGC_POSTMASTER, PGC_S_ARGV);
9582 bruce@momjian.us 679 : 103 : break;
680 : :
7539 peter_e@gmx.net 681 :UBC 0 : case 'f':
150 heikki.linnakangas@i 682 [ # # ]: 0 : if (!set_plan_disabling_options(optctx.optarg, PGC_POSTMASTER, PGC_S_ARGV))
683 : : {
7539 peter_e@gmx.net 684 : 0 : write_stderr("%s: invalid argument for option -f: \"%s\"\n",
685 : : progname, optctx.optarg);
686 : 0 : ExitPostmaster(1);
687 : : }
688 : 0 : break;
689 : :
9418 bruce@momjian.us 690 : 0 : case 'h':
150 heikki.linnakangas@i 691 : 0 : SetConfigOption("listen_addresses", optctx.optarg, PGC_POSTMASTER, PGC_S_ARGV);
9418 bruce@momjian.us 692 : 0 : break;
693 : :
10517 694 : 0 : case 'i':
8192 tgl@sss.pgh.pa.us 695 : 0 : SetConfigOption("listen_addresses", "*", PGC_POSTMASTER, PGC_S_ARGV);
9820 bruce@momjian.us 696 : 0 : break;
697 : :
7539 peter_e@gmx.net 698 : 0 : case 'j':
699 : : /* only used by interactive backend */
700 : 0 : break;
701 : :
9418 bruce@momjian.us 702 :CBC 103 : case 'k':
150 heikki.linnakangas@i 703 : 103 : SetConfigOption("unix_socket_directories", optctx.optarg, PGC_POSTMASTER, PGC_S_ARGV);
9418 bruce@momjian.us 704 : 103 : break;
705 : :
9820 bruce@momjian.us 706 :UBC 0 : case 'l':
8951 peter_e@gmx.net 707 : 0 : SetConfigOption("ssl", "true", PGC_POSTMASTER, PGC_S_ARGV);
10520 bruce@momjian.us 708 : 0 : break;
709 : :
10051 tgl@sss.pgh.pa.us 710 : 0 : case 'N':
150 heikki.linnakangas@i 711 : 0 : SetConfigOption("max_connections", optctx.optarg, PGC_POSTMASTER, PGC_S_ARGV);
10051 tgl@sss.pgh.pa.us 712 : 0 : break;
713 : :
7539 peter_e@gmx.net 714 : 0 : case 'O':
715 : 0 : SetConfigOption("allow_system_table_mods", "true", PGC_POSTMASTER, PGC_S_ARGV);
716 : 0 : break;
717 : :
718 : 0 : case 'P':
719 : 0 : SetConfigOption("ignore_system_indexes", "true", PGC_POSTMASTER, PGC_S_ARGV);
720 : 0 : break;
721 : :
10580 bruce@momjian.us 722 :CBC 69 : case 'p':
150 heikki.linnakangas@i 723 : 69 : SetConfigOption("port", optctx.optarg, PGC_POSTMASTER, PGC_S_ARGV);
10580 bruce@momjian.us 724 : 69 : break;
725 : :
7539 peter_e@gmx.net 726 :UBC 0 : case 'r':
727 : : /* only used by single-user backend */
728 : 0 : break;
729 : :
730 : 0 : case 'S':
150 heikki.linnakangas@i 731 : 0 : SetConfigOption("work_mem", optctx.optarg, PGC_POSTMASTER, PGC_S_ARGV);
10580 bruce@momjian.us 732 : 0 : break;
733 : :
734 : 0 : case 's':
7175 tgl@sss.pgh.pa.us 735 : 0 : SetConfigOption("log_statement_stats", "true", PGC_POSTMASTER, PGC_S_ARGV);
7539 peter_e@gmx.net 736 : 0 : break;
737 : :
738 : 0 : case 'T':
739 : :
740 : : /*
741 : : * This option used to be defined as sending SIGSTOP after a
742 : : * backend crash, but sending SIGABRT seems more useful.
743 : : */
1375 tgl@sss.pgh.pa.us 744 : 0 : SetConfigOption("send_abort_for_crash", "true", PGC_POSTMASTER, PGC_S_ARGV);
10580 bruce@momjian.us 745 : 0 : break;
746 : :
7539 peter_e@gmx.net 747 : 0 : case 't':
748 : : {
150 heikki.linnakangas@i 749 : 0 : const char *tmp = get_stats_option_name(optctx.optarg);
750 : :
7267 bruce@momjian.us 751 [ # # ]: 0 : if (tmp)
752 : : {
753 : 0 : SetConfigOption(tmp, "true", PGC_POSTMASTER, PGC_S_ARGV);
754 : : }
755 : : else
756 : : {
757 : 0 : write_stderr("%s: invalid argument for option -t: \"%s\"\n",
758 : : progname, optctx.optarg);
759 : 0 : ExitPostmaster(1);
760 : : }
761 : 0 : break;
762 : : }
763 : :
7539 peter_e@gmx.net 764 : 0 : case 'W':
150 heikki.linnakangas@i 765 : 0 : SetConfigOption("post_auth_delay", optctx.optarg, PGC_POSTMASTER, PGC_S_ARGV);
7539 peter_e@gmx.net 766 : 0 : break;
767 : :
10580 bruce@momjian.us 768 : 0 : default:
8072 769 : 0 : write_stderr("Try \"%s --help\" for more information.\n",
770 : : progname);
9402 tgl@sss.pgh.pa.us 771 : 0 : ExitPostmaster(1);
772 : : }
773 : : }
774 : :
775 : : /*
776 : : * Postmaster accepts no non-option switch arguments.
777 : : */
150 heikki.linnakangas@i 778 [ - + ]:CBC 1008 : if (optctx.optind < argc)
779 : : {
8072 bruce@momjian.us 780 :UBC 0 : write_stderr("%s: invalid argument: \"%s\"\n",
150 heikki.linnakangas@i 781 : 0 : progname, argv[optctx.optind]);
8072 bruce@momjian.us 782 : 0 : write_stderr("Try \"%s --help\" for more information.\n",
783 : : progname);
8951 peter_e@gmx.net 784 : 0 : ExitPostmaster(1);
785 : : }
786 : :
787 : : /*
788 : : * Locate the proper configuration files and data directory, and read
789 : : * postgresql.conf for the first time.
790 : : */
7993 tgl@sss.pgh.pa.us 791 [ - + ]:CBC 1008 : if (!SelectConfigFiles(userDoption, progname))
7993 tgl@sss.pgh.pa.us 792 :UBC 0 : ExitPostmaster(2);
793 : :
5439 bruce@momjian.us 794 [ + + ]:CBC 1008 : if (output_config_variable != NULL)
795 : : {
796 : : /*
797 : : * If this is a runtime-computed GUC, it hasn't yet been initialized,
798 : : * and the present value is not useful. However, this is a convenient
799 : : * place to print the value for most GUCs because it is safe to run
800 : : * postmaster startup to this point even if the server is already
801 : : * running. For the handful of runtime-computed GUCs that we cannot
802 : : * provide meaningful values for yet, we wait until later in
803 : : * postmaster startup to print the value. We won't be able to use -C
804 : : * on running servers for those GUCs, but using this option now would
805 : : * lead to incorrect results for them.
806 : : */
1806 michael@paquier.xyz 807 : 2 : int flags = GetConfigOptionFlags(output_config_variable, true);
808 : :
809 [ + + ]: 2 : if ((flags & GUC_RUNTIME_COMPUTED) == 0)
810 : : {
811 : : /*
812 : : * "-C guc" was specified, so print GUC's value and exit. No
813 : : * extra permission check is needed because the user is reading
814 : : * inside the data dir.
815 : : */
816 : 1 : const char *config_val = GetConfigOption(output_config_variable,
817 : : false, false);
818 : :
819 [ + - ]: 1 : puts(config_val ? config_val : "");
820 : 1 : ExitPostmaster(0);
821 : : }
822 : :
823 : : /*
824 : : * A runtime-computed GUC will be printed later on. As we initialize
825 : : * a server startup sequence, silence any log messages that may show
826 : : * up in the output generated. FATAL and more severe messages are
827 : : * useful to show, even if one would only expect at least PANIC. LOG
828 : : * entries are hidden.
829 : : */
1541 tgl@sss.pgh.pa.us 830 : 1 : SetConfigOption("log_min_messages", "FATAL", PGC_SUSET,
831 : : PGC_S_OVERRIDE);
832 : : }
833 : :
834 : : /* Verify that DataDir looks reasonable */
7993 835 : 1007 : checkDataDir();
836 : :
837 : : /* Check that pg_control exists */
3064 sfrost@snowman.net 838 : 1007 : checkControlFile();
839 : :
840 : : /* And switch working directory into it */
7724 tgl@sss.pgh.pa.us 841 : 1007 : ChangeToDataDir();
842 : :
843 : : /*
844 : : * Check for invalid combinations of GUC settings.
845 : : */
1315 rhaas@postgresql.org 846 [ - + ]: 1007 : if (SuperuserReservedConnections + ReservedConnections >= MaxConnections)
847 : : {
832 peter@eisentraut.org 848 :UBC 0 : write_stderr("%s: \"superuser_reserved_connections\" (%d) plus \"reserved_connections\" (%d) must be less than \"max_connections\" (%d)\n",
849 : : progname,
850 : : SuperuserReservedConnections, ReservedConnections,
851 : : MaxConnections);
5130 magnus@hagander.net 852 : 0 : ExitPostmaster(1);
853 : : }
4122 heikki.linnakangas@i 854 [ + + - + ]:CBC 1007 : if (XLogArchiveMode > ARCHIVE_MODE_OFF && wal_level == WAL_LEVEL_MINIMAL)
5965 heikki.linnakangas@i 855 [ # # ]:UBC 0 : ereport(ERROR,
856 : : (errmsg("WAL archival cannot be enabled when \"wal_level\" is \"minimal\"")));
5965 heikki.linnakangas@i 857 [ + + - + ]:CBC 1007 : if (max_wal_senders > 0 && wal_level == WAL_LEVEL_MINIMAL)
5965 heikki.linnakangas@i 858 [ # # ]:UBC 0 : ereport(ERROR,
859 : : (errmsg("WAL streaming (\"max_wal_senders\" > 0) requires \"wal_level\" to be \"replica\" or \"logical\"")));
981 rhaas@postgresql.org 860 [ + + - + ]:CBC 1007 : if (summarize_wal && wal_level == WAL_LEVEL_MINIMAL)
981 rhaas@postgresql.org 861 [ # # ]:UBC 0 : ereport(ERROR,
862 : : (errmsg("WAL cannot be summarized when \"wal_level\" is \"minimal\"")));
247 msawada@postgresql.o 863 [ + + - + ]:CBC 1007 : if (sync_replication_slots && wal_level == WAL_LEVEL_MINIMAL)
371 fujii@postgresql.org 864 [ # # ]:UBC 0 : ereport(ERROR,
865 : : (errmsg("replication slot synchronization (\"sync_replication_slots\" = on) requires \"wal_level\" to be \"replica\" or \"logical\"")));
866 : :
867 : : /*
868 : : * Other one-time internal sanity checks can go here, if they are fast.
869 : : * (Put any slow processing further down, after postmaster.pid creation.)
870 : : */
8624 tgl@sss.pgh.pa.us 871 [ - + ]:CBC 1007 : if (!CheckDateTokenTables())
872 : : {
8072 bruce@momjian.us 873 :UBC 0 : write_stderr("%s: invalid datetoken tables, please fix\n", progname);
8624 tgl@sss.pgh.pa.us 874 : 0 : ExitPostmaster(1);
875 : : }
876 : :
877 : : /* For debugging: display postmaster environment */
638 andres@anarazel.de 878 [ + + ]:CBC 1007 : if (message_level_is_interesting(DEBUG3))
879 : : {
880 : : #if !defined(WIN32)
881 : : extern char **environ;
882 : : #endif
883 : : char **p;
884 : : StringInfoData si;
885 : :
886 : 7 : initStringInfo(&si);
887 : :
888 : 7 : appendStringInfoString(&si, "initial environment dump:");
9419 tgl@sss.pgh.pa.us 889 [ + + ]: 282 : for (p = environ; *p; ++p)
638 andres@anarazel.de 890 : 275 : appendStringInfo(&si, "\n%s", *p);
891 : :
892 [ + - ]: 7 : ereport(DEBUG3, errmsg_internal("%s", si.data));
893 : 7 : pfree(si.data);
894 : : }
895 : :
896 : : /*
897 : : * Create lockfile for data directory.
898 : : *
899 : : * We want to do this before we try to grab the input sockets, because the
900 : : * data directory interlock is more reliable than the socket-file
901 : : * interlock (thanks to whoever decided to put socket files in /tmp :-().
902 : : * For the same reason, it's best to grab the TCP socket(s) before the
903 : : * Unix socket(s).
904 : : *
905 : : * Also note that this internally sets up the on_proc_exit function that
906 : : * is responsible for removing both data directory and socket lockfiles;
907 : : * so it must happen before opening sockets so that at exit, the socket
908 : : * lockfiles go away after CloseServerPorts runs.
909 : : */
7724 tgl@sss.pgh.pa.us 910 : 1007 : CreateDataDirLockFile(true);
911 : :
912 : : /*
913 : : * Read the control file (for error checking and config info).
914 : : *
915 : : * Since we verify the control file's CRC, this has a useful side effect
916 : : * on machines where we need a run-time test for CRC support instructions.
917 : : * The postmaster will do the test once at startup, and then its child
918 : : * processes will inherit the correct function pointer and not need to
919 : : * repeat the test.
920 : : */
3266 andres@anarazel.de 921 : 1006 : LocalProcessControlFile(false);
922 : :
923 : : /*
924 : : * Register the apply launcher. It's probably a good idea to call this
925 : : * before any modules had a chance to take the background worker slots.
926 : : */
3507 peter_e@gmx.net 927 : 1006 : ApplyLauncherRegister();
928 : :
929 : : /*
930 : : * Register the shared memory needs of all core subsystems.
931 : : */
143 heikki.linnakangas@i 932 : 1006 : RegisterBuiltinShmemCallbacks();
933 : :
934 : : /*
935 : : * process any libraries that should be preloaded at postmaster start
936 : : */
7317 tgl@sss.pgh.pa.us 937 : 1006 : process_shared_preload_libraries();
938 : :
939 : : /*
940 : : * Initialize SSL library, if specified.
941 : : */
942 : : #ifdef USE_SSL
2346 andrew@dunslane.net 943 [ + + ]: 1006 : if (EnableSSL)
944 : : {
945 : 48 : (void) secure_initialize(true);
946 : 35 : LoadedSSL = true;
947 : : }
948 : : #endif
949 : :
950 : : /*
951 : : * Now that loadable modules have had their chance to alter any GUCs,
952 : : * calculate MaxBackends and initialize the machinery to track child
953 : : * processes.
954 : : */
4985 alvherre@alvh.no-ip. 955 : 993 : InitializeMaxBackends();
651 heikki.linnakangas@i 956 : 993 : InitPostmasterChildSlots();
957 : :
958 : : /*
959 : : * Calculate the size of the PGPROC fast-path lock arrays.
960 : : */
705 tomas.vondra@postgre 961 : 993 : InitializeFastPathLocks();
962 : :
963 : : /*
964 : : * Also call any legacy shmem request hooks that might've been installed
965 : : * by preloaded libraries.
966 : : *
967 : : * Note: this must be done before ShmemCallRequestCallbacks(), because the
968 : : * hooks may request LWLocks with RequestNamedLWLockTranche(), which in
969 : : * turn affects the size of the LWLock array calculated in lwlock.c.
970 : : */
1567 rhaas@postgresql.org 971 : 993 : process_shmem_requests();
972 : :
973 : : /*
974 : : * Ask all subsystems, including preloaded libraries, to register their
975 : : * shared memory needs.
976 : : */
143 heikki.linnakangas@i 977 : 993 : ShmemCallRequestCallbacks();
978 : :
979 : : /*
980 : : * Now that loadable modules have had their chance to request additional
981 : : * shared memory, determine the value of any runtime-computed GUCs that
982 : : * depend on the amount of shared memory required.
983 : : */
1814 michael@paquier.xyz 984 : 993 : InitializeShmemGUCs();
985 : :
986 : : /*
987 : : * Now that modules have been loaded, we can process any custom resource
988 : : * managers specified in the wal_consistency_checking GUC.
989 : : */
1604 jdavis@postgresql.or 990 : 993 : InitializeWalConsistencyChecking();
991 : :
992 : : /*
993 : : * If -C was specified with a runtime-computed GUC, we held off printing
994 : : * the value earlier, as the GUC was not yet initialized. We handle -C
995 : : * for most GUCs before we lock the data directory so that the option may
996 : : * be used on a running server. However, a handful of GUCs are runtime-
997 : : * computed and do not have meaningful values until after locking the data
998 : : * directory, and we cannot safely calculate their values earlier on a
999 : : * running server. At this point, such GUCs should be properly
1000 : : * initialized, and we haven't yet set up shared memory, so this is a good
1001 : : * time to handle the -C option for these special GUCs.
1002 : : */
1806 michael@paquier.xyz 1003 [ + + ]: 993 : if (output_config_variable != NULL)
1004 : : {
1005 : 1 : const char *config_val = GetConfigOption(output_config_variable,
1006 : : false, false);
1007 : :
1008 [ + - ]: 1 : puts(config_val ? config_val : "");
1009 : 1 : ExitPostmaster(0);
1010 : : }
1011 : :
1012 : : /*
1013 : : * Set up shared memory and semaphores.
1014 : : *
1015 : : * Note: if using SysV shmem and/or semas, each postmaster startup will
1016 : : * normally choose the same IPC keys. This helps ensure that we will
1017 : : * clean up dead IPC objects if the postmaster crashes and is restarted.
1018 : : */
1503 tgl@sss.pgh.pa.us 1019 : 992 : CreateSharedMemoryAndSemaphores();
1020 : :
1021 : : /*
1022 : : * Estimate number of openable files. This must happen after setting up
1023 : : * semaphores, because on some platforms semaphores count as open files.
1024 : : */
2542 1025 : 991 : set_max_safe_fds();
1026 : :
1027 : : /*
1028 : : * Initialize pipe (or process handle on Windows) that allows children to
1029 : : * wake up from sleep on postmaster death.
1030 : : */
1031 : 991 : InitPostmasterDeathWatchHandle();
1032 : :
1033 : : #ifdef WIN32
1034 : :
1035 : : /*
1036 : : * Initialize I/O completion port used to deliver list of dead children.
1037 : : */
1038 : : win32ChildQueue = CreateIoCompletionPort(INVALID_HANDLE_VALUE, NULL, 0, 1);
1039 : : if (win32ChildQueue == NULL)
1040 : : ereport(FATAL,
1041 : : (errmsg("could not create I/O completion port for child queue")));
1042 : : #endif
1043 : :
1044 : : #ifdef EXEC_BACKEND
1045 : : /* Write out nondefault GUC settings for child processes to use */
1046 : : write_nondefault_variables(PGC_POSTMASTER);
1047 : :
1048 : : /*
1049 : : * Clean out the temp directory used to transmit parameters to child
1050 : : * processes (see internal_forkexec). We must do this before launching
1051 : : * any child processes, else we have a race condition: we could remove a
1052 : : * parameter file before the child can read it. It should be safe to do
1053 : : * so now, because we verified earlier that there are no conflicting
1054 : : * Postgres processes in this data directory.
1055 : : */
1056 : : RemovePgTempFilesInDir(PG_TEMP_FILES_DIR, true, false);
1057 : : #endif
1058 : :
1059 : : /*
1060 : : * Forcibly remove the files signaling a standby promotion request.
1061 : : * Otherwise, the existence of those files triggers a promotion too early,
1062 : : * whether a user wants that or not.
1063 : : *
1064 : : * This removal of files is usually unnecessary because they can exist
1065 : : * only during a few moments during a standby promotion. However there is
1066 : : * a race condition: if pg_ctl promote is executed and creates the files
1067 : : * during a promotion, the files can stay around even after the server is
1068 : : * brought up to be the primary. Then, if a new standby starts by using
1069 : : * the backup taken from the new primary, the files can exist at server
1070 : : * startup and must be removed in order to avoid an unexpected promotion.
1071 : : *
1072 : : * Note that promotion signal files need to be removed before the startup
1073 : : * process is invoked. Because, after that, they can be used by
1074 : : * postmaster's SIGUSR1 signal handler.
1075 : : */
1076 : 991 : RemovePromoteSignalFiles();
1077 : :
1078 : : /* Do the same for logrotate signal file */
1079 : 991 : RemoveLogrotateSignalFiles();
1080 : :
1081 : : /* Remove any outdated file holding the current log filenames. */
1082 [ + - - + ]: 991 : if (unlink(LOG_METAINFO_DATAFILE) < 0 && errno != ENOENT)
2542 tgl@sss.pgh.pa.us 1083 [ # # ]:UBC 0 : ereport(LOG,
1084 : : (errcode_for_file_access(),
1085 : : errmsg("could not remove file \"%s\": %m",
1086 : : LOG_METAINFO_DATAFILE)));
1087 : :
1088 : : /*
1089 : : * If enabled, start up syslogger collection subprocess
1090 : : */
651 heikki.linnakangas@i 1091 [ + + ]:CBC 991 : if (Logging_collector)
1092 : 1 : StartSysLogger();
1093 : :
1094 : : /*
1095 : : * Reset whereToSendOutput from DestDebug (its starting state) to
1096 : : * DestNone. This stops ereport from sending log messages to stderr unless
1097 : : * Log_destination permits. We don't do this until the postmaster is
1098 : : * fully launched, since startup failures may as well be reported to
1099 : : * stderr.
1100 : : *
1101 : : * If we are in fact disabling logging to stderr, first emit a log message
1102 : : * saying so, to provide a breadcrumb trail for users who may not remember
1103 : : * that their logging is configured to go somewhere else.
1104 : : */
2542 tgl@sss.pgh.pa.us 1105 [ - + ]: 991 : if (!(Log_destination & LOG_DESTINATION_STDERR))
2542 tgl@sss.pgh.pa.us 1106 [ # # ]:UBC 0 : ereport(LOG,
1107 : : (errmsg("ending log output to stderr"),
1108 : : errhint("Future log output will go to log destination \"%s\".",
1109 : : Log_destination_string)));
1110 : :
2542 tgl@sss.pgh.pa.us 1111 :CBC 991 : whereToSendOutput = DestNone;
1112 : :
1113 : : /*
1114 : : * Report server startup in log. While we could emit this much earlier,
1115 : : * it seems best to do so after starting the log collector, if we intend
1116 : : * to use one.
1117 : : */
2766 peter@eisentraut.org 1118 [ + - ]: 991 : ereport(LOG,
1119 : : (errmsg("starting %s", PG_VERSION_STR)));
1120 : :
1121 : : /*
1122 : : * Establish input sockets.
1123 : : *
1124 : : * First set up an on_proc_exit function that's charged with closing the
1125 : : * sockets again at postmaster shutdown.
1126 : : */
10 michael@paquier.xyz 1127 :GNC 991 : ListenSockets = palloc_array(pgsocket, MAXLISTEN);
1057 heikki.linnakangas@i 1128 :CBC 991 : on_proc_exit(CloseServerPorts, 0);
1129 : :
8192 tgl@sss.pgh.pa.us 1130 [ + - ]: 991 : if (ListenAddresses)
1131 : : {
1132 : : char *rawstring;
1133 : : List *elemlist;
1134 : : ListCell *l;
7728 peter_e@gmx.net 1135 : 991 : int success = 0;
1136 : :
1137 : : /* Need a modifiable copy of ListenAddresses */
8054 tgl@sss.pgh.pa.us 1138 : 991 : rawstring = pstrdup(ListenAddresses);
1139 : :
1140 : : /* Parse string into list of hostnames */
2479 1141 [ - + ]: 991 : if (!SplitGUCList(rawstring, ',', &elemlist))
1142 : : {
1143 : : /* syntax error in list */
8054 tgl@sss.pgh.pa.us 1144 [ # # ]:UBC 0 : ereport(FATAL,
1145 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1146 : : errmsg("invalid list syntax in parameter \"%s\"",
1147 : : "listen_addresses")));
1148 : : }
1149 : :
8054 tgl@sss.pgh.pa.us 1150 [ + + + + :CBC 1027 : foreach(l, elemlist)
+ + ]
1151 : : {
8033 bruce@momjian.us 1152 : 36 : char *curhost = (char *) lfirst(l);
1153 : :
8127 1154 [ - + ]: 36 : if (strcmp(curhost, "*") == 0)
898 heikki.linnakangas@i 1155 :UBC 0 : status = ListenServerPort(AF_UNSPEC, NULL,
8192 tgl@sss.pgh.pa.us 1156 : 0 : (unsigned short) PostPortNumber,
1157 : : NULL,
1158 : : ListenSockets,
1159 : : &NumListenSockets,
1160 : : MAXLISTEN);
1161 : : else
898 heikki.linnakangas@i 1162 :CBC 36 : status = ListenServerPort(AF_UNSPEC, curhost,
8436 tgl@sss.pgh.pa.us 1163 : 36 : (unsigned short) PostPortNumber,
1164 : : NULL,
1165 : : ListenSockets,
1166 : : &NumListenSockets,
1167 : : MAXLISTEN);
1168 : :
7728 peter_e@gmx.net 1169 [ + - ]: 36 : if (status == STATUS_OK)
1170 : : {
1171 : 36 : success++;
1172 : : /* record the first successful host addr in lockfile */
5705 tgl@sss.pgh.pa.us 1173 [ + - ]: 36 : if (!listen_addr_saved)
1174 : : {
1175 : 36 : AddToDataDirLockFile(LOCK_FILE_LINE_LISTEN_ADDR, curhost);
1176 : 36 : listen_addr_saved = true;
1177 : : }
1178 : : }
1179 : : else
8192 tgl@sss.pgh.pa.us 1180 [ # # ]:UBC 0 : ereport(WARNING,
1181 : : (errmsg("could not create listen socket for \"%s\"",
1182 : : curhost)));
1183 : : }
1184 : :
5130 tgl@sss.pgh.pa.us 1185 [ + + - + ]:CBC 991 : if (!success && elemlist != NIL)
7728 peter_e@gmx.net 1186 [ # # ]:UBC 0 : ereport(FATAL,
1187 : : (errmsg("could not create any TCP/IP sockets")));
1188 : :
8054 tgl@sss.pgh.pa.us 1189 :CBC 991 : list_free(elemlist);
1190 : 991 : pfree(rawstring);
1191 : : }
1192 : :
1193 : : #ifdef USE_BONJOUR
1194 : : /* Register for Bonjour only if we opened TCP socket(s) */
1195 : : if (enable_bonjour && NumListenSockets > 0)
1196 : : {
1197 : : DNSServiceErrorType err;
1198 : :
1199 : : /*
1200 : : * We pass 0 for interface_index, which will result in registering on
1201 : : * all "applicable" interfaces. It's not entirely clear from the
1202 : : * DNS-SD docs whether this would be appropriate if we have bound to
1203 : : * just a subset of the available network interfaces.
1204 : : */
1205 : : err = DNSServiceRegister(&bonjour_sdref,
1206 : : 0,
1207 : : 0,
1208 : : bonjour_name,
1209 : : "_postgresql._tcp.",
1210 : : NULL,
1211 : : NULL,
1212 : : pg_hton16(PostPortNumber),
1213 : : 0,
1214 : : NULL,
1215 : : NULL,
1216 : : NULL);
1217 : : if (err != kDNSServiceErr_NoError)
1218 : : ereport(LOG,
1219 : : (errmsg("DNSServiceRegister() failed: error code %ld",
1220 : : (long) err)));
1221 : :
1222 : : /*
1223 : : * We don't bother to read the mDNS daemon's reply, and we expect that
1224 : : * it will automatically terminate our registration when the socket is
1225 : : * closed at postmaster termination. So there's nothing more to be
1226 : : * done here. However, the bonjour_sdref is kept around so that
1227 : : * forked children can close their copies of the socket.
1228 : : */
1229 : : }
1230 : : #endif
1231 : :
5130 1232 [ + - ]: 991 : if (Unix_socket_directories)
1233 : : {
1234 : : char *rawstring;
1235 : : List *elemlist;
1236 : : ListCell *l;
1237 : 991 : int success = 0;
1238 : :
1239 : : /* Need a modifiable copy of Unix_socket_directories */
1240 : 991 : rawstring = pstrdup(Unix_socket_directories);
1241 : :
1242 : : /* Parse string into list of directories */
1243 [ - + ]: 991 : if (!SplitDirectoriesString(rawstring, ',', &elemlist))
1244 : : {
1245 : : /* syntax error in list */
5130 tgl@sss.pgh.pa.us 1246 [ # # ]:UBC 0 : ereport(FATAL,
1247 : : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
1248 : : errmsg("invalid list syntax in parameter \"%s\"",
1249 : : "unix_socket_directories")));
1250 : : }
1251 : :
5130 tgl@sss.pgh.pa.us 1252 [ + - + + :CBC 1982 : foreach(l, elemlist)
+ + ]
1253 : : {
1254 : 991 : char *socketdir = (char *) lfirst(l);
1255 : :
898 heikki.linnakangas@i 1256 : 991 : status = ListenServerPort(AF_UNIX, NULL,
5130 tgl@sss.pgh.pa.us 1257 : 991 : (unsigned short) PostPortNumber,
1258 : : socketdir,
1259 : : ListenSockets,
1260 : : &NumListenSockets,
1261 : : MAXLISTEN);
1262 : :
1263 [ + - ]: 991 : if (status == STATUS_OK)
1264 : : {
1265 : 991 : success++;
1266 : : /* record the first successful Unix socket in lockfile */
1267 [ + - ]: 991 : if (success == 1)
1268 : 991 : AddToDataDirLockFile(LOCK_FILE_LINE_SOCKET_DIR, socketdir);
1269 : : }
1270 : : else
5130 tgl@sss.pgh.pa.us 1271 [ # # ]:UBC 0 : ereport(WARNING,
1272 : : (errmsg("could not create Unix-domain socket in directory \"%s\"",
1273 : : socketdir)));
1274 : : }
1275 : :
5130 tgl@sss.pgh.pa.us 1276 [ - + - - ]:CBC 991 : if (!success && elemlist != NIL)
5130 tgl@sss.pgh.pa.us 1277 [ # # ]:UBC 0 : ereport(FATAL,
1278 : : (errmsg("could not create any Unix-domain sockets")));
1279 : :
5130 tgl@sss.pgh.pa.us 1280 :CBC 991 : list_free_deep(elemlist);
1281 : 991 : pfree(rawstring);
1282 : : }
1283 : :
1284 : : /*
1285 : : * check that we have some socket to listen on
1286 : : */
1057 heikki.linnakangas@i 1287 [ - + ]: 991 : if (NumListenSockets == 0)
8192 tgl@sss.pgh.pa.us 1288 [ # # ]:UBC 0 : ereport(FATAL,
1289 : : (errmsg("no socket created for listening")));
1290 : :
1291 : : /*
1292 : : * If no valid TCP ports, write an empty line for listen address,
1293 : : * indicating the Unix socket must be used. Note that this line is not
1294 : : * added to the lock file until there is a socket backing it.
1295 : : */
5705 tgl@sss.pgh.pa.us 1296 [ + + ]:CBC 991 : if (!listen_addr_saved)
1297 : 955 : AddToDataDirLockFile(LOCK_FILE_LINE_LISTEN_ADDR, "");
1298 : :
1299 : : /*
1300 : : * Record postmaster options. We delay this till now to avoid recording
1301 : : * bogus options (eg, unusable port number).
1302 : : */
8141 bruce@momjian.us 1303 [ - + ]: 991 : if (!CreateOptsFile(argc, argv, my_exec_path))
9402 tgl@sss.pgh.pa.us 1304 :UBC 0 : ExitPostmaster(1);
1305 : :
1306 : : /*
1307 : : * Write the external PID file if requested
1308 : : */
7992 tgl@sss.pgh.pa.us 1309 [ - + ]:CBC 991 : if (external_pid_file)
1310 : : {
7992 tgl@sss.pgh.pa.us 1311 :UBC 0 : FILE *fpidfile = fopen(external_pid_file, "w");
1312 : :
7993 1313 [ # # ]: 0 : if (fpidfile)
1314 : : {
1315 : 0 : fprintf(fpidfile, "%d\n", MyProcPid);
1316 : 0 : fclose(fpidfile);
1317 : :
1318 : : /* Make PID file world readable */
5548 peter_e@gmx.net 1319 [ # # ]: 0 : if (chmod(external_pid_file, S_IRUSR | S_IWUSR | S_IRGRP | S_IROTH) != 0)
898 michael@paquier.xyz 1320 : 0 : write_stderr("%s: could not change permissions of external PID file \"%s\": %m\n",
1321 : : progname, external_pid_file);
1322 : : }
1323 : : else
1324 : 0 : write_stderr("%s: could not write external PID file \"%s\": %m\n",
1325 : : progname, external_pid_file);
1326 : :
5120 peter_e@gmx.net 1327 : 0 : on_proc_exit(unlink_external_pid_file, 0);
1328 : : }
1329 : :
1330 : : /*
1331 : : * Remove old temporary files. At this point there can be no other
1332 : : * Postgres processes running in this directory, so this should be safe.
1333 : : */
4244 andres@anarazel.de 1334 :CBC 991 : RemovePgTempFiles();
1335 : :
1336 : : /*
1337 : : * Initialize the autovacuum subsystem (again, no process start yet)
1338 : : */
7338 tgl@sss.pgh.pa.us 1339 : 991 : autovac_init();
1340 : :
1341 : : /*
1342 : : * Load configuration files for client authentication.
1343 : : */
6212 1344 [ - + ]: 991 : if (!load_hba())
1345 : : {
1346 : : /*
1347 : : * It makes no sense to continue if we fail to load the HBA file,
1348 : : * since there is no way to connect to the database in this case.
1349 : : */
6212 tgl@sss.pgh.pa.us 1350 [ # # ]:UBC 0 : ereport(FATAL,
1351 : : /* translator: %s is a configuration file */
1352 : : (errmsg("could not load %s", HbaFileName)));
1353 : : }
5088 heikki.linnakangas@i 1354 :CBC 991 : if (!load_ident())
1355 : : {
1356 : : /*
1357 : : * We can start up without the IDENT file, although it means that you
1358 : : * cannot log in using any of the authentication methods that need a
1359 : : * user name mapping. load_ident() already logged the details of error
1360 : : * to the log.
1361 : : */
1362 : : }
1363 : :
1364 : : #ifdef HAVE_PTHREAD_IS_THREADED_NP
1365 : :
1366 : : /*
1367 : : * On macOS, libintl replaces setlocale() with a version that calls
1368 : : * CFLocaleCopyCurrent() when its second argument is "" and every relevant
1369 : : * environment variable is unset or empty. CFLocaleCopyCurrent() makes
1370 : : * the process multithreaded. The postmaster calls sigprocmask() and
1371 : : * calls fork() without an immediate exec(), both of which have undefined
1372 : : * behavior in a multithreaded program. A multithreaded postmaster is the
1373 : : * normal case on Windows, which offers neither fork() nor sigprocmask().
1374 : : * Currently, macOS is the only platform having pthread_is_threaded_np(),
1375 : : * so we need not worry whether this HINT is appropriate elsewhere.
1376 : : */
1377 : : if (pthread_is_threaded_np() != 0)
1378 : : ereport(FATAL,
1379 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
1380 : : errmsg("postmaster became multithreaded during startup"),
1381 : : errhint("Set the LC_ALL environment variable to a valid locale.")));
1382 : : #endif
1383 : :
1384 : : /*
1385 : : * Remember postmaster startup time
1386 : : */
7729 tgl@sss.pgh.pa.us 1387 : 991 : PgStartTime = GetCurrentTimestamp();
1388 : :
1389 : : /*
1390 : : * Report postmaster status in the postmaster.pid file, to allow pg_ctl to
1391 : : * see what's happening.
1392 : : */
3347 1393 : 991 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STARTING);
1394 : :
527 andres@anarazel.de 1395 : 991 : UpdatePMState(PM_STARTUP);
1396 : :
1397 : : /* Make sure we can perform I/O while starting up. */
141 tmunro@postgresql.or 1398 : 991 : maybe_start_io_workers();
1399 : :
1400 : : /* Start bgwriter and checkpointer so they can help with recovery */
651 heikki.linnakangas@i 1401 [ + - ]: 991 : if (CheckpointerPMChild == NULL)
1402 : 991 : CheckpointerPMChild = StartChildProcess(B_CHECKPOINTER);
1403 [ + - ]: 991 : if (BgWriterPMChild == NULL)
1404 : 991 : BgWriterPMChild = StartChildProcess(B_BG_WRITER);
1405 : :
1406 : : /*
1407 : : * We're ready to rock and roll...
1408 : : */
1409 : 991 : StartupPMChild = StartChildProcess(B_STARTUP);
1410 [ - + ]: 991 : Assert(StartupPMChild != NULL);
4067 tgl@sss.pgh.pa.us 1411 : 991 : StartupStatus = STARTUP_RUNNING;
1412 : :
1413 : : /* Some workers may be scheduled to start now */
3410 1414 : 991 : maybe_start_bgworkers();
1415 : :
10581 bruce@momjian.us 1416 : 991 : status = ServerLoop();
1417 : :
1418 : : /*
1419 : : * ServerLoop probably shouldn't ever return, but if it does, close down.
1420 : : */
10581 bruce@momjian.us 1421 :UBC 0 : ExitPostmaster(status != STATUS_OK);
1422 : :
1423 : : abort(); /* not reached */
1424 : : }
1425 : :
1426 : :
1427 : : /*
1428 : : * on_proc_exit callback to close server's listen sockets
1429 : : */
1430 : : static void
4043 tgl@sss.pgh.pa.us 1431 :CBC 991 : CloseServerPorts(int status, Datum arg)
1432 : : {
1433 : : int i;
1434 : :
1435 : : /*
1436 : : * First, explicitly close all the socket FDs. We used to just let this
1437 : : * happen implicitly at postmaster exit, but it's better to close them
1438 : : * before we remove the postmaster.pid lockfile; otherwise there's a race
1439 : : * condition if a new postmaster wants to re-use the TCP port number.
1440 : : */
1057 heikki.linnakangas@i 1441 [ + + ]: 2018 : for (i = 0; i < NumListenSockets; i++)
1442 : : {
898 1443 [ - + ]: 1027 : if (closesocket(ListenSockets[i]) != 0)
898 heikki.linnakangas@i 1444 [ # # ]:UBC 0 : elog(LOG, "could not close listen socket: %m");
1445 : : }
1057 heikki.linnakangas@i 1446 :CBC 991 : NumListenSockets = 0;
1447 : :
1448 : : /*
1449 : : * Next, remove any filesystem entries for Unix sockets. To avoid race
1450 : : * conditions against incoming postmasters, this must happen after closing
1451 : : * the sockets and before removing lock files.
1452 : : */
4043 tgl@sss.pgh.pa.us 1453 : 991 : RemoveSocketFiles();
1454 : :
1455 : : /*
1456 : : * We don't do anything about socket lock files here; those will be
1457 : : * removed in a later on_proc_exit callback.
1458 : : */
1459 : 991 : }
1460 : :
1461 : : /*
1462 : : * on_proc_exit callback to delete external_pid_file
1463 : : */
1464 : : static void
5120 peter_e@gmx.net 1465 :UBC 0 : unlink_external_pid_file(int status, Datum arg)
1466 : : {
1467 [ # # ]: 0 : if (external_pid_file)
1468 : 0 : unlink(external_pid_file);
1469 : 0 : }
1470 : :
1471 : :
1472 : : /*
1473 : : * Compute and check the directory paths to files that are part of the
1474 : : * installation (as deduced from the postgres executable's own location)
1475 : : */
1476 : : static void
6326 tgl@sss.pgh.pa.us 1477 :CBC 1012 : getInstallationPaths(const char *argv0)
1478 : : {
1479 : : DIR *pdir;
1480 : :
1481 : : /* Locate the postgres executable itself */
1482 [ - + ]: 1012 : if (find_my_exec(argv0, my_exec_path) < 0)
2092 peter@eisentraut.org 1483 [ # # ]:UBC 0 : ereport(FATAL,
1484 : : (errmsg("%s: could not locate my own executable path", argv0)));
1485 : :
1486 : : #ifdef EXEC_BACKEND
1487 : : /* Locate executable backend before we change working directory */
1488 : : if (find_other_exec(argv0, "postgres", PG_BACKEND_VERSIONSTR,
1489 : : postgres_exec_path) < 0)
1490 : : ereport(FATAL,
1491 : : (errmsg("%s: could not locate matching postgres executable",
1492 : : argv0)));
1493 : : #endif
1494 : :
1495 : : /*
1496 : : * Locate the pkglib directory --- this has to be set early in case we try
1497 : : * to load any modules from it in response to postgresql.conf entries.
1498 : : */
6326 tgl@sss.pgh.pa.us 1499 :CBC 1012 : get_pkglib_path(my_exec_path, pkglib_path);
1500 : :
1501 : : /*
1502 : : * Verify that there's a readable directory there; otherwise the Postgres
1503 : : * installation is incomplete or corrupt. (A typical cause of this
1504 : : * failure is that the postgres executable has been moved or hardlinked to
1505 : : * some directory that's not a sibling of the installation lib/
1506 : : * directory.)
1507 : : */
1508 : 1012 : pdir = AllocateDir(pkglib_path);
1509 [ - + ]: 1012 : if (pdir == NULL)
6326 tgl@sss.pgh.pa.us 1510 [ # # ]:UBC 0 : ereport(ERROR,
1511 : : (errcode_for_file_access(),
1512 : : errmsg("could not open directory \"%s\": %m",
1513 : : pkglib_path),
1514 : : errhint("This may indicate an incomplete PostgreSQL installation, or that the file \"%s\" has been moved away from its proper location.",
1515 : : my_exec_path)));
6326 tgl@sss.pgh.pa.us 1516 :CBC 1012 : FreeDir(pdir);
1517 : :
1518 : : /*
1519 : : * It's not worth checking the share/ directory. If the lib/ directory is
1520 : : * there, then share/ probably is too.
1521 : : */
1522 : 1012 : }
1523 : :
1524 : : /*
1525 : : * Check that pg_control exists in the correct location in the data directory.
1526 : : *
1527 : : * No attempt is made to validate the contents of pg_control here. This is
1528 : : * just a sanity check to see if we are looking at a real data directory.
1529 : : */
1530 : : static void
3064 sfrost@snowman.net 1531 : 1007 : checkControlFile(void)
1532 : : {
1533 : : char path[MAXPGPATH];
1534 : : FILE *fp;
1535 : :
507 fujii@postgresql.org 1536 : 1007 : snprintf(path, sizeof(path), "%s/%s", DataDir, XLOG_CONTROL_FILE);
1537 : :
8126 tgl@sss.pgh.pa.us 1538 : 1007 : fp = AllocateFile(path, PG_BINARY_R);
1539 [ - + ]: 1007 : if (fp == NULL)
1540 : : {
8072 bruce@momjian.us 1541 :UBC 0 : write_stderr("%s: could not find the database system\n"
1542 : : "Expected to find it in the directory \"%s\",\n"
1543 : : "but could not open file \"%s\": %m\n",
1544 : : progname, DataDir, path);
8126 tgl@sss.pgh.pa.us 1545 : 0 : ExitPostmaster(2);
1546 : : }
8126 tgl@sss.pgh.pa.us 1547 :CBC 1007 : FreeFile(fp);
1548 : 1007 : }
1549 : :
1550 : : /*
1551 : : * Determine how long should we let ServerLoop sleep, in milliseconds.
1552 : : *
1553 : : * In normal conditions we wait at most one minute, to ensure that the other
1554 : : * background tasks handled by ServerLoop get done even when no requests are
1555 : : * arriving. However, if there are background workers waiting to be started,
1556 : : * we don't actually sleep so that they are quickly serviced. Other exception
1557 : : * cases are as shown in the code.
1558 : : */
1559 : : static int
1323 tmunro@postgresql.or 1560 : 45724 : DetermineSleepTime(void)
1561 : : {
1562 : : TimestampTz next_wakeup;
1563 : :
1564 : : /*
1565 : : * If in ImmediateShutdown with a SIGKILL timeout, ignore everything else
1566 : : * and wait for that.
1567 : : *
1568 : : * XXX Shouldn't this also test FatalError?
1569 : : */
141 1570 [ + + ]: 45724 : if (Shutdown >= ImmediateShutdown)
1571 : : {
4087 tgl@sss.pgh.pa.us 1572 [ + - ]: 1495 : if (AbortStartTime != 0)
1573 : : {
309 1574 : 1495 : time_t curtime = time(NULL);
1575 : : int seconds;
1576 : :
1577 : : /*
1578 : : * time left to abort; clamp to 0 if it already expired, or if
1579 : : * time goes backwards
1580 : : */
1581 [ + - ]: 1495 : if (curtime < AbortStartTime ||
1582 [ - + ]: 1495 : curtime - AbortStartTime >= SIGKILL_CHILDREN_AFTER_SECS)
309 tgl@sss.pgh.pa.us 1583 :UBC 0 : seconds = 0;
1584 : : else
309 tgl@sss.pgh.pa.us 1585 :CBC 1495 : seconds = SIGKILL_CHILDREN_AFTER_SECS -
1586 : : (curtime - AbortStartTime);
1587 : :
1588 : 1495 : return seconds * 1000;
1589 : : }
1590 : : }
1591 : :
1592 : : /* Time of next maybe_start_io_workers() call, or 0 for none. */
141 tmunro@postgresql.or 1593 : 44229 : next_wakeup = maybe_start_io_workers_scheduled_at();
1594 : :
1595 : : /* Ignore bgworkers during shutdown. */
1596 [ - + - - ]: 44229 : if (StartWorkerNeeded && Shutdown == NoShutdown)
1323 tmunro@postgresql.or 1597 :UBC 0 : return 0;
1598 : :
141 tmunro@postgresql.or 1599 [ + + - + ]:CBC 44229 : if (HaveCrashedWorker && Shutdown == NoShutdown)
1600 : : {
1601 : : dlist_mutable_iter iter;
1602 : :
1603 : : /*
1604 : : * When there are crashed bgworkers, we sleep just long enough that
1605 : : * they are restarted when they request to be. Scan the list to
1606 : : * determine the minimum of all wakeup times according to most recent
1607 : : * crash time and requested restart interval.
1608 : : */
748 heikki.linnakangas@i 1609 [ # # # # ]:UBC 0 : dlist_foreach_modify(iter, &BackgroundWorkerList)
1610 : : {
1611 : : RegisteredBgWorker *rw;
1612 : : TimestampTz this_wakeup;
1613 : :
1614 : 0 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
1615 : :
5012 alvherre@alvh.no-ip. 1616 [ # # ]: 0 : if (rw->rw_crashed_at == 0)
1617 : 0 : continue;
1618 : :
4696 rhaas@postgresql.org 1619 [ # # ]: 0 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART
1620 [ # # ]: 0 : || rw->rw_terminate)
1621 : : {
748 heikki.linnakangas@i 1622 : 0 : ForgetBackgroundWorker(rw);
5012 alvherre@alvh.no-ip. 1623 : 0 : continue;
1624 : : }
1625 : :
1626 : 0 : this_wakeup = TimestampTzPlusMilliseconds(rw->rw_crashed_at,
1627 : : 1000L * rw->rw_worker.bgw_restart_time);
1628 [ # # # # ]: 0 : if (next_wakeup == 0 || this_wakeup < next_wakeup)
1629 : 0 : next_wakeup = this_wakeup;
1630 : : }
1631 : : }
1632 : :
5012 alvherre@alvh.no-ip. 1633 [ + + ]:CBC 44229 : if (next_wakeup != 0)
1634 : : {
1635 : : int ms;
1636 : :
1637 : : /* result of TimestampDifferenceMilliseconds is in [0, INT_MAX] */
1309 tgl@sss.pgh.pa.us 1638 : 34 : ms = (int) TimestampDifferenceMilliseconds(GetCurrentTimestamp(),
1639 : : next_wakeup);
1640 : 34 : return Min(60 * 1000, ms);
1641 : : }
1642 : :
1323 tmunro@postgresql.or 1643 : 44195 : return 60 * 1000;
1644 : : }
1645 : :
1646 : : /*
1647 : : * Activate or deactivate notifications of server socket events. Since we
1648 : : * don't currently have a way to remove events from an existing WaitEventSet,
1649 : : * we'll just destroy and recreate the whole thing. This is called during
1650 : : * shutdown so we can wait for backends to exit without accepting new
1651 : : * connections, and during crash reinitialization when we need to start
1652 : : * listening for new connections again. The WaitEventSet will be freed in fork
1653 : : * children by ClosePostmasterPorts().
1654 : : */
1655 : : static void
1656 : 1992 : ConfigurePostmasterWaitSet(bool accept_connections)
1657 : : {
1658 [ + + ]: 1992 : if (pm_wait_set)
1659 : 1001 : FreeWaitEventSet(pm_wait_set);
1660 : 1992 : pm_wait_set = NULL;
1661 : :
1008 heikki.linnakangas@i 1662 : 3984 : pm_wait_set = CreateWaitEventSet(NULL,
1057 1663 : 1992 : accept_connections ? (1 + NumListenSockets) : 1);
1323 tmunro@postgresql.or 1664 : 1992 : AddWaitEventToSet(pm_wait_set, WL_LATCH_SET, PGINVALID_SOCKET, MyLatch,
1665 : : NULL);
1666 : :
1667 [ + + ]: 1992 : if (accept_connections)
1668 : : {
1057 heikki.linnakangas@i 1669 [ + + ]: 2028 : for (int i = 0; i < NumListenSockets; i++)
1670 : 1032 : AddWaitEventToSet(pm_wait_set, WL_SOCKET_ACCEPT, ListenSockets[i],
1671 : : NULL, NULL);
1672 : : }
1323 tmunro@postgresql.or 1673 : 1992 : }
1674 : :
1675 : : /*
1676 : : * Main idle loop of postmaster
1677 : : */
1678 : : static int
10919 scrappy@hub.org 1679 : 991 : ServerLoop(void)
1680 : : {
1681 : : time_t last_lockfile_recheck_time,
1682 : : last_touch_time;
1683 : : WaitEvent events[MAXLISTEN];
1684 : : int nevents;
1685 : :
1323 tmunro@postgresql.or 1686 : 991 : ConfigurePostmasterWaitSet(true);
3978 tgl@sss.pgh.pa.us 1687 : 991 : last_lockfile_recheck_time = last_touch_time = time(NULL);
1688 : :
1689 : : for (;;)
10581 bruce@momjian.us 1690 : 44733 : {
1691 : : time_t now;
1692 : :
1323 tmunro@postgresql.or 1693 : 45724 : nevents = WaitEventSetWait(pm_wait_set,
1694 : 45724 : DetermineSleepTime(),
1695 : : events,
1696 : : lengthof(events),
1697 : : 0 /* postmaster posts no wait_events */ );
1698 : :
1699 : : /*
1700 : : * Latch set by signal handler, or new connection pending on any of
1701 : : * our sockets? If the latter, fork a child process to deal with it.
1702 : : */
1703 [ + + ]: 90451 : for (int i = 0; i < nevents; i++)
1704 : : {
1705 [ + + ]: 45718 : if (events[i].events & WL_LATCH_SET)
1706 : 30103 : ResetLatch(MyLatch);
1707 : :
1708 : : /*
1709 : : * The following requests are handled unconditionally, even if we
1710 : : * didn't see WL_LATCH_SET. This gives high priority to shutdown
1711 : : * and reload requests where the latch happens to appear later in
1712 : : * events[] or will be reported by a later call to
1713 : : * WaitEventSetWait().
1714 : : */
1310 1715 [ + + ]: 45718 : if (pending_pm_shutdown_request)
1716 : 980 : process_pm_shutdown_request();
1717 [ + + ]: 45718 : if (pending_pm_reload_request)
1718 : 172 : process_pm_reload_request();
1719 [ + + ]: 45718 : if (pending_pm_child_exit)
1720 : 24328 : process_pm_child_exit();
1721 [ + + ]: 44727 : if (pending_pm_pmsignal)
1722 : 4684 : process_pm_pmsignal();
1723 : :
1724 [ + + ]: 44727 : if (events[i].events & WL_SOCKET_ACCEPT)
1725 : : {
1726 : : ClientSocket s;
1727 : :
898 heikki.linnakangas@i 1728 [ + - ]: 15615 : if (AcceptConnection(events[i].fd, &s) == STATUS_OK)
1729 : 15615 : BackendStartup(&s);
1730 : :
1731 : : /* We no longer need the open socket in this process */
1732 [ + - ]: 15615 : if (s.sock != PGINVALID_SOCKET)
1733 : : {
1734 [ - + ]: 15615 : if (closesocket(s.sock) != 0)
898 heikki.linnakangas@i 1735 [ # # ]:UBC 0 : elog(LOG, "could not close client socket: %m");
1736 : : }
1737 : : }
1738 : : }
1739 : :
1740 : : /*
1741 : : * If we need to launch any background processes after changing state
1742 : : * or because some exited, do so now.
1743 : : */
745 heikki.linnakangas@i 1744 :CBC 44733 : LaunchMissingBackgroundProcesses();
1745 : :
1746 : : /* If we need to signal the autovacuum launcher, do so now */
6212 alvherre@alvh.no-ip. 1747 [ - + ]: 44733 : if (avlauncher_needs_signal)
1748 : : {
6212 alvherre@alvh.no-ip. 1749 :UBC 0 : avlauncher_needs_signal = false;
651 heikki.linnakangas@i 1750 [ # # ]: 0 : if (AutoVacLauncherPMChild != NULL)
594 andres@anarazel.de 1751 : 0 : signal_child(AutoVacLauncherPMChild, SIGUSR2);
1752 : : }
1753 : :
1754 : : #ifdef HAVE_PTHREAD_IS_THREADED_NP
1755 : :
1756 : : /*
1757 : : * With assertions enabled, check regularly for appearance of
1758 : : * additional threads. All builds check at start and exit.
1759 : : */
1760 : : Assert(pthread_is_threaded_np() == 0);
1761 : : #endif
1762 : :
1763 : : /*
1764 : : * Lastly, check to see if it's time to do some things that we don't
1765 : : * want to do every single time through the loop, because they're a
1766 : : * bit expensive. Note that there's up to a minute of slop in when
1767 : : * these tasks will be performed, since DetermineSleepTime() will let
1768 : : * us sleep at most that long; except for SIGKILL timeout which has
1769 : : * special-case logic there.
1770 : : */
3975 tgl@sss.pgh.pa.us 1771 :CBC 44733 : now = time(NULL);
1772 : :
1773 : : /*
1774 : : * If we already sent SIGQUIT to children and they are slow to shut
1775 : : * down, it's time to send them SIGKILL (or SIGABRT if requested).
1776 : : * This doesn't happen normally, but under certain conditions backends
1777 : : * can get stuck while shutting down. This is a last measure to get
1778 : : * them unwedged.
1779 : : *
1780 : : * Note we also do this during recovery from a process crash.
1781 : : */
1375 1782 [ + + + + ]: 44733 : if ((Shutdown >= ImmediateShutdown || FatalError) &&
4087 1783 [ + + ]: 1566 : AbortStartTime != 0 &&
1784 [ - + ]: 1557 : (now - AbortStartTime) >= SIGKILL_CHILDREN_AFTER_SECS)
1785 : : {
1786 : : /* We were gentle with them before. Not anymore */
2176 tgl@sss.pgh.pa.us 1787 [ # # # # ]:UBC 0 : ereport(LOG,
1788 : : /* translator: %s is SIGKILL or SIGABRT */
1789 : : (errmsg("issuing %s to recalcitrant children",
1790 : : send_abort_for_kill ? "SIGABRT" : "SIGKILL")));
1375 1791 [ # # ]: 0 : TerminateChildren(send_abort_for_kill ? SIGABRT : SIGKILL);
1792 : : /* reset flag so we don't SIGKILL again */
4709 alvherre@alvh.no-ip. 1793 : 0 : AbortStartTime = 0;
1794 : : }
1795 : :
1796 : : /*
1797 : : * Once a minute, verify that postmaster.pid hasn't been removed or
1798 : : * overwritten. If it has, we force a shutdown. This avoids having
1799 : : * postmasters and child processes hanging around after their database
1800 : : * is gone, and maybe causing problems if a new database cluster is
1801 : : * created in the same place. It also provides some protection
1802 : : * against a DBA foolishly removing postmaster.pid and manually
1803 : : * starting a new postmaster. Data corruption is likely to ensue from
1804 : : * that anyway, but we can minimize the damage by aborting ASAP.
1805 : : */
3978 tgl@sss.pgh.pa.us 1806 [ + + ]:CBC 44733 : if (now - last_lockfile_recheck_time >= 1 * SECS_PER_MINUTE)
1807 : : {
1808 [ - + ]: 42 : if (!RecheckDataDirLockFile())
1809 : : {
3978 tgl@sss.pgh.pa.us 1810 [ # # ]:UBC 0 : ereport(LOG,
1811 : : (errmsg("performing immediate shutdown because data directory lock file is invalid")));
1812 : 0 : kill(MyProcPid, SIGQUIT);
1813 : : }
3978 tgl@sss.pgh.pa.us 1814 :CBC 42 : last_lockfile_recheck_time = now;
1815 : : }
1816 : :
1817 : : /*
1818 : : * Touch Unix socket and lock files every 58 minutes, to ensure that
1819 : : * they are not removed by overzealous /tmp-cleaning tasks. We assume
1820 : : * no one runs cleaners with cutoff times of less than an hour ...
1821 : : */
1822 [ - + ]: 44733 : if (now - last_touch_time >= 58 * SECS_PER_MINUTE)
1823 : : {
3978 tgl@sss.pgh.pa.us 1824 :UBC 0 : TouchSocketFiles();
1825 : 0 : TouchSocketLockFiles();
1826 : 0 : last_touch_time = now;
1827 : : }
1828 : : }
1829 : : }
1830 : :
1831 : : /*
1832 : : * canAcceptConnections --- check to see if database state allows connections
1833 : : * of the specified type. backend_type can be B_BACKEND or B_AUTOVAC_WORKER.
1834 : : * (Note that we don't yet know whether a normal B_BACKEND connection might
1835 : : * turn into a walsender.)
1836 : : */
1837 : : static CAC_state
651 heikki.linnakangas@i 1838 :CBC 15771 : canAcceptConnections(BackendType backend_type)
1839 : : {
5765 tgl@sss.pgh.pa.us 1840 : 15771 : CAC_state result = CAC_OK;
1841 : :
651 heikki.linnakangas@i 1842 [ + + - + ]: 15771 : Assert(backend_type == B_BACKEND || backend_type == B_AUTOVAC_WORKER);
1843 : :
1844 : : /*
1845 : : * Can't start backends when in startup/shutdown/inconsistent recovery
1846 : : * state. We treat autovac workers the same as user backends for this
1847 : : * purpose.
1848 : : */
1849 [ + + + + ]: 15771 : if (pmState != PM_RUN && pmState != PM_HOT_STANDBY)
1850 : : {
2204 tgl@sss.pgh.pa.us 1851 [ + + ]: 260 : if (Shutdown > NoShutdown)
6860 bruce@momjian.us 1852 : 55 : return CAC_SHUTDOWN; /* shutdown is pending */
1981 fujii@postgresql.org 1853 [ + + + + ]: 205 : else if (!FatalError && pmState == PM_STARTUP)
5618 bruce@momjian.us 1854 : 194 : return CAC_STARTUP; /* normal startup */
1981 fujii@postgresql.org 1855 [ + + + - ]: 11 : else if (!FatalError && pmState == PM_RECOVERY)
512 1856 : 7 : return CAC_NOTHOTSTANDBY; /* not yet ready for hot standby */
1857 : : else
5765 tgl@sss.pgh.pa.us 1858 : 4 : return CAC_RECOVERY; /* else must be crash recovery */
1859 : : }
1860 : :
1861 : : /*
1862 : : * "Smart shutdown" restrictions are applied only to normal connections,
1863 : : * not to autovac workers.
1864 : : */
651 heikki.linnakangas@i 1865 [ - + - - ]: 15511 : if (!connsAllowed && backend_type == B_BACKEND)
1604 sfrost@snowman.net 1866 :UBC 0 : return CAC_SHUTDOWN; /* shutdown is pending */
1867 : :
5765 tgl@sss.pgh.pa.us 1868 :CBC 15511 : return result;
1869 : : }
1870 : :
1871 : : /*
1872 : : * ClosePostmasterPorts -- close all the postmaster's open sockets
1873 : : *
1874 : : * This is called during child process startup to release file descriptors
1875 : : * that are not needed by that child process. The postmaster still has
1876 : : * them open, of course.
1877 : : *
1878 : : * Note: we pass am_syslogger as a boolean because we don't want to set
1879 : : * the global variable yet when this is called.
1880 : : */
1881 : : void
8057 1882 : 23502 : ClosePostmasterPorts(bool am_syslogger)
1883 : : {
1884 : : /* Release resources held by the postmaster's WaitEventSet. */
1323 tmunro@postgresql.or 1885 [ + + ]: 23502 : if (pm_wait_set)
1886 : : {
1887 : 20073 : FreeWaitEventSetAfterFork(pm_wait_set);
1888 : 20073 : pm_wait_set = NULL;
1889 : : }
1890 : :
1891 : : #ifndef WIN32
1892 : :
1893 : : /*
1894 : : * Close the write end of postmaster death watch pipe. It's important to
1895 : : * do this as early as possible, so that if postmaster dies, others won't
1896 : : * think that it's still running because we're holding the pipe open.
1897 : : */
2609 peter@eisentraut.org 1898 [ - + ]: 23502 : if (close(postmaster_alive_fds[POSTMASTER_FD_OWN]) != 0)
5529 heikki.linnakangas@i 1899 [ # # ]:UBC 0 : ereport(FATAL,
1900 : : (errcode_for_file_access(),
1901 : : errmsg_internal("could not close postmaster death monitoring pipe in child process: %m")));
5529 heikki.linnakangas@i 1902 :CBC 23502 : postmaster_alive_fds[POSTMASTER_FD_OWN] = -1;
1903 : : /* Notify fd.c that we released one pipe FD. */
2376 tgl@sss.pgh.pa.us 1904 : 23502 : ReleaseExternalFD();
1905 : : #endif
1906 : :
1907 : : /*
1908 : : * Close the postmaster's listen sockets. These aren't tracked by fd.c,
1909 : : * so we don't call ReleaseExternalFD() here.
1910 : : *
1911 : : * The listen sockets are marked as FD_CLOEXEC, so this isn't needed in
1912 : : * EXEC_BACKEND mode.
1913 : : */
1914 : : #ifndef EXEC_BACKEND
1056 heikki.linnakangas@i 1915 [ + + ]: 23502 : if (ListenSockets)
1916 : : {
1917 [ + + ]: 47428 : for (int i = 0; i < NumListenSockets; i++)
1918 : : {
898 1919 [ - + ]: 23927 : if (closesocket(ListenSockets[i]) != 0)
898 heikki.linnakangas@i 1920 [ # # ]:UBC 0 : elog(LOG, "could not close listen socket: %m");
1921 : : }
1056 heikki.linnakangas@i 1922 :CBC 23501 : pfree(ListenSockets);
1923 : : }
1057 1924 : 23502 : NumListenSockets = 0;
1925 : 23502 : ListenSockets = NULL;
1926 : : #endif
1927 : :
1928 : : /*
1929 : : * If using syslogger, close the read side of the pipe. We don't bother
1930 : : * tracking this in fd.c, either.
1931 : : */
8057 tgl@sss.pgh.pa.us 1932 [ + + ]: 23502 : if (!am_syslogger)
1933 : : {
1934 : : #ifndef WIN32
1935 [ + + ]: 23501 : if (syslogPipe[0] >= 0)
1936 : 16 : close(syslogPipe[0]);
1937 : 23501 : syslogPipe[0] = -1;
1938 : : #else
1939 : : if (syslogPipe[0])
1940 : : CloseHandle(syslogPipe[0]);
1941 : : syslogPipe[0] = 0;
1942 : : #endif
1943 : : }
1944 : :
1945 : : #ifdef USE_BONJOUR
1946 : : /* If using Bonjour, close the connection to the mDNS daemon */
1947 : : if (bonjour_sdref)
1948 : : close(DNSServiceRefSockFD(bonjour_sdref));
1949 : : #endif
9331 1950 : 23502 : }
1951 : :
1952 : :
1953 : : /*
1954 : : * InitProcessGlobals -- set MyStartTime[stamp], random seeds
1955 : : *
1956 : : * Called early in the postmaster and every backend.
1957 : : */
1958 : : void
2869 tmunro@postgresql.or 1959 : 24781 : InitProcessGlobals(void)
1960 : : {
1961 : 24781 : MyStartTimestamp = GetCurrentTimestamp();
1962 : 24781 : MyStartTime = timestamptz_to_time_t(MyStartTimestamp);
1963 : :
1964 : : /* initialize timing infrastructure (required for INSTR_* calls) */
142 andres@anarazel.de 1965 : 24781 : pg_initialize_timing();
1966 : :
1967 : : /*
1968 : : * Set a different global seed in every process. We want something
1969 : : * unpredictable, so if possible, use high-quality random bits for the
1970 : : * seed. Otherwise, fall back to a seed based on timestamp and PID.
1971 : : */
1733 tgl@sss.pgh.pa.us 1972 [ + - - + ]: 24781 : if (unlikely(!pg_prng_strong_seed(&pg_global_prng_state)))
1973 : : {
1974 : : uint64 rseed;
1975 : :
1976 : : /*
1977 : : * Since PIDs and timestamps tend to change more frequently in their
1978 : : * least significant bits, shift the timestamp left to allow a larger
1979 : : * total number of seeds in a given time period. Since that would
1980 : : * leave only 20 bits of the timestamp that cycle every ~1 second,
1981 : : * also mix in some higher bits.
1982 : : */
2798 tgl@sss.pgh.pa.us 1983 :UBC 0 : rseed = ((uint64) MyProcPid) ^
2842 tmunro@postgresql.or 1984 : 0 : ((uint64) MyStartTimestamp << 12) ^
2798 tgl@sss.pgh.pa.us 1985 : 0 : ((uint64) MyStartTimestamp >> 20);
1986 : :
1733 1987 : 0 : pg_prng_seed(&pg_global_prng_state, rseed);
1988 : : }
1989 : :
1990 : : /*
1991 : : * Also make sure that we've set a good seed for random(3). Use of that
1992 : : * is deprecated in core Postgres, but extensions might use it.
1993 : : */
1994 : : #ifndef WIN32
1733 tgl@sss.pgh.pa.us 1995 :CBC 24781 : srandom(pg_prng_uint32(&pg_global_prng_state));
1996 : : #endif
2869 tmunro@postgresql.or 1997 : 24781 : }
1998 : :
1999 : : /*
2000 : : * Child processes use SIGUSR1 to notify us of 'pmsignals'. pg_ctl uses
2001 : : * SIGUSR1 to ask postmaster to check for logrotate and promote files.
2002 : : */
2003 : : static void
1323 2004 : 4732 : handle_pm_pmsignal_signal(SIGNAL_ARGS)
2005 : : {
2006 : 4732 : pending_pm_pmsignal = true;
2007 : 4732 : SetLatch(MyLatch);
2008 : 4732 : }
2009 : :
2010 : : /*
2011 : : * pg_ctl uses SIGHUP to request a reload of the configuration files.
2012 : : */
2013 : : static void
2014 : 172 : handle_pm_reload_request_signal(SIGNAL_ARGS)
2015 : : {
2016 : 172 : pending_pm_reload_request = true;
2017 : 172 : SetLatch(MyLatch);
2018 : 172 : }
2019 : :
2020 : : /*
2021 : : * Re-read config files, and tell children to do same.
2022 : : */
2023 : : static void
2024 : 172 : process_pm_reload_request(void)
2025 : : {
2026 : 172 : pending_pm_reload_request = false;
2027 : :
2028 [ + + ]: 172 : ereport(DEBUG2,
2029 : : (errmsg_internal("postmaster received reload request signal")));
2030 : :
9205 tgl@sss.pgh.pa.us 2031 [ + - ]: 172 : if (Shutdown <= SmartShutdown)
2032 : : {
8437 2033 [ + - ]: 172 : ereport(LOG,
2034 : : (errmsg("received SIGHUP, reloading configuration files")));
9062 2035 : 172 : ProcessConfigFile(PGC_SIGHUP);
651 heikki.linnakangas@i 2036 : 172 : SignalChildren(SIGHUP, btmask_all_except(B_DEAD_END_BACKEND));
2037 : :
2038 : : /* Reload authentication config files too */
6555 magnus@hagander.net 2039 [ - + ]: 172 : if (!load_hba())
3524 tgl@sss.pgh.pa.us 2040 [ # # ]:UBC 0 : ereport(LOG,
2041 : : /* translator: %s is a configuration file */
2042 : : (errmsg("%s was not reloaded", HbaFileName)));
2043 : :
5088 heikki.linnakangas@i 2044 [ - + ]:CBC 172 : if (!load_ident())
3524 tgl@sss.pgh.pa.us 2045 [ # # ]:UBC 0 : ereport(LOG,
2046 : : (errmsg("%s was not reloaded", IdentFileName)));
2047 : :
2048 : : #ifdef USE_SSL
2049 : : /* Reload SSL configuration as well */
3524 tgl@sss.pgh.pa.us 2050 [ + + ]:CBC 172 : if (EnableSSL)
2051 : : {
2052 [ + + ]: 13 : if (secure_initialize(false) == 0)
2053 : 11 : LoadedSSL = true;
2054 : : else
2055 [ + - ]: 2 : ereport(LOG,
2056 : : (errmsg("SSL configuration was not reloaded")));
2057 : : }
2058 : : else
2059 : : {
2060 : 159 : secure_destroy();
2061 : 159 : LoadedSSL = false;
2062 : : }
2063 : : #endif
2064 : :
2065 : : #ifdef EXEC_BACKEND
2066 : : /* Update the starting-point file for future children */
2067 : : write_nondefault_variables(PGC_SIGHUP);
2068 : : #endif
2069 : : }
1323 tmunro@postgresql.or 2070 : 172 : }
2071 : :
2072 : : /*
2073 : : * pg_ctl uses SIGTERM, SIGINT and SIGQUIT to request different types of
2074 : : * shutdown.
2075 : : */
2076 : : static void
2077 : 980 : handle_pm_shutdown_request_signal(SIGNAL_ARGS)
2078 : : {
2079 [ + + + - ]: 980 : switch (postgres_signal_arg)
2080 : : {
2081 : 49 : case SIGTERM:
2082 : : /* smart is implied if the other two flags aren't set */
2083 : 49 : pending_pm_shutdown_request = true;
2084 : 49 : break;
2085 : 573 : case SIGINT:
2086 : 573 : pending_pm_fast_shutdown_request = true;
2087 : 573 : pending_pm_shutdown_request = true;
2088 : 573 : break;
2089 : 358 : case SIGQUIT:
2090 : 358 : pending_pm_immediate_shutdown_request = true;
2091 : 358 : pending_pm_shutdown_request = true;
2092 : 358 : break;
2093 : : }
2094 : 980 : SetLatch(MyLatch);
9584 peter_e@gmx.net 2095 : 980 : }
2096 : :
2097 : : /*
2098 : : * Process shutdown request.
2099 : : */
2100 : : static void
1323 tmunro@postgresql.or 2101 : 980 : process_pm_shutdown_request(void)
2102 : : {
2103 : : int mode;
2104 : :
8416 tgl@sss.pgh.pa.us 2105 [ + + ]: 980 : ereport(DEBUG2,
2106 : : (errmsg_internal("postmaster received shutdown request signal")));
2107 : :
1323 tmunro@postgresql.or 2108 : 980 : pending_pm_shutdown_request = false;
2109 : :
2110 : : /*
2111 : : * If more than one shutdown request signal arrived since the last server
2112 : : * loop, take the one that is the most immediate. That matches the
2113 : : * priority that would apply if we processed them one by one in any order.
2114 : : */
2115 [ + + ]: 980 : if (pending_pm_immediate_shutdown_request)
2116 : : {
2117 : 358 : pending_pm_immediate_shutdown_request = false;
2118 : 358 : pending_pm_fast_shutdown_request = false;
2119 : 358 : mode = ImmediateShutdown;
2120 : : }
2121 [ + + ]: 622 : else if (pending_pm_fast_shutdown_request)
2122 : : {
2123 : 573 : pending_pm_fast_shutdown_request = false;
2124 : 573 : mode = FastShutdown;
2125 : : }
2126 : : else
2127 : 49 : mode = SmartShutdown;
2128 : :
2129 [ + + + - ]: 980 : switch (mode)
2130 : : {
2131 : 49 : case SmartShutdown:
2132 : :
2133 : : /*
2134 : : * Smart Shutdown:
2135 : : *
2136 : : * Wait for children to end their work, then shut down.
2137 : : */
9822 vadim4o@yahoo.com 2138 [ - + ]: 49 : if (Shutdown >= SmartShutdown)
9062 tgl@sss.pgh.pa.us 2139 :UBC 0 : break;
9822 vadim4o@yahoo.com 2140 :CBC 49 : Shutdown = SmartShutdown;
8437 tgl@sss.pgh.pa.us 2141 [ + - ]: 49 : ereport(LOG,
2142 : : (errmsg("received smart shutdown request")));
2143 : :
2144 : : /* Report status */
3347 2145 : 49 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STOPPING);
2146 : : #ifdef USE_SYSTEMD
2147 : : sd_notify(0, "STOPPING=1");
2148 : : #endif
2149 : :
2150 : : /*
2151 : : * If we reached normal running, we go straight to waiting for
2152 : : * client backends to exit. If already in PM_STOP_BACKENDS or a
2153 : : * later state, do not change it.
2154 : : */
1604 sfrost@snowman.net 2155 [ - + - - ]: 49 : if (pmState == PM_RUN || pmState == PM_HOT_STANDBY)
2156 : 49 : connsAllowed = false;
2204 tgl@sss.pgh.pa.us 2157 [ # # # # ]:UBC 0 : else if (pmState == PM_STARTUP || pmState == PM_RECOVERY)
2158 : : {
2159 : : /* There should be no clients, so proceed to stop children */
594 andres@anarazel.de 2160 : 0 : UpdatePMState(PM_STOP_BACKENDS);
2161 : : }
2162 : :
2163 : : /*
2164 : : * Now wait for online backup mode to end and backends to exit. If
2165 : : * that is already the case, PostmasterStateMachine will take the
2166 : : * next step.
2167 : : */
6958 tgl@sss.pgh.pa.us 2168 :CBC 49 : PostmasterStateMachine();
9062 2169 : 49 : break;
2170 : :
1323 tmunro@postgresql.or 2171 : 573 : case FastShutdown:
2172 : :
2173 : : /*
2174 : : * Fast Shutdown:
2175 : : *
2176 : : * Abort all children with SIGTERM (rollback active transactions
2177 : : * and exit) and shut down when they are gone.
2178 : : */
9822 vadim4o@yahoo.com 2179 [ - + ]: 573 : if (Shutdown >= FastShutdown)
9062 tgl@sss.pgh.pa.us 2180 :UBC 0 : break;
8233 tgl@sss.pgh.pa.us 2181 :CBC 573 : Shutdown = FastShutdown;
8437 2182 [ + - ]: 573 : ereport(LOG,
2183 : : (errmsg("received fast shutdown request")));
2184 : :
2185 : : /* Report status */
3347 2186 : 573 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STOPPING);
2187 : : #ifdef USE_SYSTEMD
2188 : : sd_notify(0, "STOPPING=1");
2189 : : #endif
2190 : :
2920 michael@paquier.xyz 2191 [ + - - + ]: 573 : if (pmState == PM_STARTUP || pmState == PM_RECOVERY)
2192 : : {
2193 : : /* Just shut down background processes silently */
594 andres@anarazel.de 2194 :UBC 0 : UpdatePMState(PM_STOP_BACKENDS);
2195 : : }
5908 rhaas@postgresql.org 2196 [ + + ]:CBC 573 : else if (pmState == PM_RUN ||
2197 [ + - ]: 70 : pmState == PM_HOT_STANDBY)
2198 : : {
2199 : : /* Report that we're about to zap live client sessions */
6958 tgl@sss.pgh.pa.us 2200 [ + - ]: 573 : ereport(LOG,
2201 : : (errmsg("aborting any active transactions")));
594 andres@anarazel.de 2202 : 573 : UpdatePMState(PM_STOP_BACKENDS);
2203 : : }
2204 : :
2205 : : /*
2206 : : * PostmasterStateMachine will issue any necessary signals, or
2207 : : * take the next step if no child processes need to be killed.
2208 : : */
6958 tgl@sss.pgh.pa.us 2209 : 573 : PostmasterStateMachine();
9062 2210 : 573 : break;
2211 : :
1323 tmunro@postgresql.or 2212 : 358 : case ImmediateShutdown:
2213 : :
2214 : : /*
2215 : : * Immediate Shutdown:
2216 : : *
2217 : : * abort all children with SIGQUIT, wait for them to exit,
2218 : : * terminate remaining ones with SIGKILL, then exit without
2219 : : * attempt to properly shut down the data base system.
2220 : : */
4808 alvherre@alvh.no-ip. 2221 [ - + ]: 358 : if (Shutdown >= ImmediateShutdown)
4808 alvherre@alvh.no-ip. 2222 :UBC 0 : break;
4808 alvherre@alvh.no-ip. 2223 :CBC 358 : Shutdown = ImmediateShutdown;
8437 tgl@sss.pgh.pa.us 2224 [ + - ]: 358 : ereport(LOG,
2225 : : (errmsg("received immediate shutdown request")));
2226 : :
2227 : : /* Report status */
3347 2228 : 358 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STOPPING);
2229 : : #ifdef USE_SYSTEMD
2230 : : sd_notify(0, "STOPPING=1");
2231 : : #endif
2232 : :
2233 : : /* tell children to shut down ASAP */
2234 : : /* (note we don't apply send_abort_for_crash here) */
2072 2235 : 358 : SetQuitSignalReason(PMQUIT_FOR_STOP);
4808 alvherre@alvh.no-ip. 2236 : 358 : TerminateChildren(SIGQUIT);
594 andres@anarazel.de 2237 : 358 : UpdatePMState(PM_WAIT_BACKENDS);
2238 : :
2239 : : /* set stopwatch for them to die */
4808 alvherre@alvh.no-ip. 2240 : 358 : AbortStartTime = time(NULL);
2241 : :
2242 : : /*
2243 : : * Now wait for backends to exit. If there are none,
2244 : : * PostmasterStateMachine will take the next step.
2245 : : */
2246 : 358 : PostmasterStateMachine();
9822 vadim4o@yahoo.com 2247 : 358 : break;
2248 : : }
1323 tmunro@postgresql.or 2249 : 980 : }
2250 : :
2251 : : static void
2252 : 24387 : handle_pm_child_exit_signal(SIGNAL_ARGS)
2253 : : {
2254 : 24387 : pending_pm_child_exit = true;
2255 : 24387 : SetLatch(MyLatch);
11006 scrappy@hub.org 2256 : 24387 : }
2257 : :
2258 : : /*
2259 : : * Cleanup after a child process dies.
2260 : : */
2261 : : static void
1323 tmunro@postgresql.or 2262 : 24328 : process_pm_child_exit(void)
2263 : : {
2264 : : int pid; /* process id of dead child process */
2265 : : int exitstatus; /* its exit status */
2266 : :
2267 : 24328 : pending_pm_child_exit = false;
2268 : :
8416 tgl@sss.pgh.pa.us 2269 [ - + ]: 24328 : ereport(DEBUG4,
2270 : : (errmsg_internal("reaping dead processes")));
2271 : :
5166 2272 [ + + ]: 50627 : while ((pid = waitpid(-1, &exitstatus, WNOHANG)) > 0)
2273 : : {
2274 : : PMChild *pmchild;
2275 : :
2276 : : /*
2277 : : * Check if this child was a startup process.
2278 : : */
651 heikki.linnakangas@i 2279 [ + + + + ]: 26299 : if (StartupPMChild && pid == StartupPMChild->pid)
2280 : : {
2281 : 996 : ReleasePostmasterChildSlot(StartupPMChild);
2282 : 996 : StartupPMChild = NULL;
2283 : :
2284 : : /*
2285 : : * Startup process exited in response to a shutdown request (or it
2286 : : * completed normally regardless of the shutdown request).
2287 : : */
5027 tgl@sss.pgh.pa.us 2288 [ + + ]: 996 : if (Shutdown > NoShutdown &&
2289 [ + + + - : 122 : (EXIT_STATUS_0(exitstatus) || EXIT_STATUS_1(exitstatus)))
+ + ]
2290 : : {
4067 2291 : 70 : StartupStatus = STARTUP_NOT_RUNNING;
594 andres@anarazel.de 2292 : 70 : UpdatePMState(PM_WAIT_BACKENDS);
2293 : : /* PostmasterStateMachine logic does the rest */
4293 simon@2ndQuadrant.co 2294 : 70 : continue;
2295 : : }
2296 : :
2297 [ + - - + ]: 926 : if (EXIT_STATUS_3(exitstatus))
2298 : : {
4293 simon@2ndQuadrant.co 2299 [ # # ]:UBC 0 : ereport(LOG,
2300 : : (errmsg("shutdown at recovery target")));
4067 tgl@sss.pgh.pa.us 2301 : 0 : StartupStatus = STARTUP_NOT_RUNNING;
2204 2302 : 0 : Shutdown = Max(Shutdown, SmartShutdown);
4293 simon@2ndQuadrant.co 2303 : 0 : TerminateChildren(SIGTERM);
594 andres@anarazel.de 2304 : 0 : UpdatePMState(PM_WAIT_BACKENDS);
2305 : : /* PostmasterStateMachine logic does the rest */
5027 tgl@sss.pgh.pa.us 2306 : 0 : continue;
2307 : : }
2308 : :
2309 : : /*
2310 : : * Any unexpected exit (including FATAL exit) of the startup
2311 : : * process is catastrophic, so kill other children, and set
2312 : : * StartupStatus so we don't try to reinitialize after they're
2313 : : * gone. Exception: if StartupStatus is STARTUP_SIGNALED, then we
2314 : : * previously sent the startup process a SIGQUIT; so that's
2315 : : * probably the reason it died, and we do want to try to restart
2316 : : * in that case.
2317 : : *
2318 : : * This stanza also handles the case where we sent a SIGQUIT
2319 : : * during PM_STARTUP due to some dead-end child crashing: in that
2320 : : * situation, if the startup process dies on the SIGQUIT, we need
2321 : : * to transition to PM_WAIT_BACKENDS state which will allow
2322 : : * PostmasterStateMachine to restart the startup process. (On the
2323 : : * other hand, the startup process might complete normally, if we
2324 : : * were too late with the SIGQUIT. In that case we'll fall
2325 : : * through and commence normal operations.)
2326 : : */
6399 heikki.linnakangas@i 2327 [ + + ]:CBC 926 : if (!EXIT_STATUS_0(exitstatus))
2328 : : {
4067 tgl@sss.pgh.pa.us 2329 [ + + ]: 63 : if (StartupStatus == STARTUP_SIGNALED)
2330 : : {
2331 : 52 : StartupStatus = STARTUP_NOT_RUNNING;
2558 2332 [ - + ]: 52 : if (pmState == PM_STARTUP)
594 andres@anarazel.de 2333 :UBC 0 : UpdatePMState(PM_WAIT_BACKENDS);
2334 : : }
2335 : : else
4067 tgl@sss.pgh.pa.us 2336 :CBC 11 : StartupStatus = STARTUP_CRASHED;
2337 : :
2338 : : /*
2339 : : * If FatalError is already set, we are reinitializing after a
2340 : : * previous crash, and HandleChildCrash() would do nothing,
2341 : : * leaving the state machine stuck at PM_STARTUP. Give up,
2342 : : * signal the remaining children and head for PM_NO_CHILDREN,
2343 : : * where STARTUP_CRASHED makes us exit.
2344 : : */
7 michael@paquier.xyz 2345 [ + + - + ]: 63 : if (StartupStatus == STARTUP_CRASHED &&
7 michael@paquier.xyz 2346 [ # # ]:UBC 0 : FatalError && Shutdown != ImmediateShutdown)
2347 : : {
2348 : 0 : LogChildExit(LOG, _("startup process"), pid, exitstatus);
2349 [ # # ]: 0 : ereport(LOG,
2350 : : (errmsg("aborting startup due to startup process failure")));
2351 : 0 : HandleFatalError(PMQUIT_FOR_CRASH, true);
2352 : : }
2353 : : else
7 michael@paquier.xyz 2354 :CBC 63 : HandleChildCrash(pid, exitstatus,
2355 : 63 : _("startup process"));
6958 tgl@sss.pgh.pa.us 2356 : 63 : continue;
2357 : : }
2358 : :
2359 : : /*
2360 : : * Startup succeeded, commence normal operations
2361 : : */
4067 2362 : 863 : StartupStatus = STARTUP_NOT_RUNNING;
6394 heikki.linnakangas@i 2363 : 863 : FatalError = false;
2558 tgl@sss.pgh.pa.us 2364 : 863 : AbortStartTime = 0;
5937 rhaas@postgresql.org 2365 : 863 : ReachedNormalRunning = true;
594 andres@anarazel.de 2366 : 863 : UpdatePMState(PM_RUN);
1604 sfrost@snowman.net 2367 : 863 : connsAllowed = true;
2368 : :
2369 : : /*
2370 : : * At the next iteration of the postmaster's main loop, we will
2371 : : * crank up the background tasks like the autovacuum launcher and
2372 : : * background workers that were not started earlier already.
2373 : : */
745 heikki.linnakangas@i 2374 : 863 : StartWorkerNeeded = true;
2375 : :
2376 : : /* at this point we are really open for business */
6394 2377 [ + - ]: 863 : ereport(LOG,
2378 : : (errmsg("database system is ready to accept connections")));
2379 : :
2380 : : /* Report status */
3347 tgl@sss.pgh.pa.us 2381 : 863 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_READY);
2382 : : #ifdef USE_SYSTEMD
2383 : : sd_notify(0, "READY=1");
2384 : : #endif
2385 : :
6392 heikki.linnakangas@i 2386 : 863 : continue;
2387 : : }
2388 : :
2389 : : /*
2390 : : * Was it the bgwriter? Normal exit can be ignored; we'll start a new
2391 : : * one at the next iteration of the postmaster's main loop, if
2392 : : * necessary. Any other exit condition is treated as a crash.
2393 : : */
651 2394 [ + + + + ]: 25303 : if (BgWriterPMChild && pid == BgWriterPMChild->pid)
2395 : : {
2396 : 996 : ReleasePostmasterChildSlot(BgWriterPMChild);
2397 : 996 : BgWriterPMChild = NULL;
5413 simon@2ndQuadrant.co 2398 [ + + ]: 996 : if (!EXIT_STATUS_0(exitstatus))
2399 : 374 : HandleChildCrash(pid, exitstatus,
2400 : 374 : _("background writer process"));
2401 : 996 : continue;
2402 : : }
2403 : :
2404 : : /*
2405 : : * Was it the checkpointer?
2406 : : */
651 heikki.linnakangas@i 2407 [ + + + + ]: 24307 : if (CheckpointerPMChild && pid == CheckpointerPMChild->pid)
2408 : : {
2409 : 996 : ReleasePostmasterChildSlot(CheckpointerPMChild);
2410 : 996 : CheckpointerPMChild = NULL;
579 andres@anarazel.de 2411 [ + + + - ]: 996 : if (EXIT_STATUS_0(exitstatus) && pmState == PM_WAIT_CHECKPOINTER)
9822 vadim4o@yahoo.com 2412 : 622 : {
2413 : : /*
2414 : : * OK, we saw normal exit of the checkpointer after it's been
2415 : : * told to shut down. We know checkpointer wrote a shutdown
2416 : : * checkpoint, otherwise we'd still be in
2417 : : * PM_WAIT_XLOG_SHUTDOWN state.
2418 : : *
2419 : : * At this point only dead-end children and logger should be
2420 : : * left.
2421 : : */
579 andres@anarazel.de 2422 : 622 : UpdatePMState(PM_WAIT_DEAD_END);
2423 : 622 : ConfigurePostmasterWaitSet(false);
2424 : 622 : SignalChildren(SIGTERM, btmask_all_except(B_LOGGER));
2425 : : }
2426 : : else
2427 : : {
2428 : : /*
2429 : : * Any unexpected exit of the checkpointer (including FATAL
2430 : : * exit) is treated as a crash.
2431 : : */
6958 tgl@sss.pgh.pa.us 2432 : 374 : HandleChildCrash(pid, exitstatus,
5413 simon@2ndQuadrant.co 2433 : 374 : _("checkpointer process"));
2434 : : }
2435 : :
8125 tgl@sss.pgh.pa.us 2436 : 996 : continue;
2437 : : }
2438 : :
2439 : : /*
2440 : : * Was it the wal writer? Normal exit can be ignored; we'll start a
2441 : : * new one at the next iteration of the postmaster's main loop, if
2442 : : * necessary. Any other exit condition is treated as a crash.
2443 : : */
651 heikki.linnakangas@i 2444 [ + + + + ]: 23311 : if (WalWriterPMChild && pid == WalWriterPMChild->pid)
2445 : : {
2446 : 863 : ReleasePostmasterChildSlot(WalWriterPMChild);
2447 : 863 : WalWriterPMChild = NULL;
6974 tgl@sss.pgh.pa.us 2448 [ + + ]: 863 : if (!EXIT_STATUS_0(exitstatus))
2449 : 311 : HandleChildCrash(pid, exitstatus,
6867 peter_e@gmx.net 2450 : 311 : _("WAL writer process"));
6974 tgl@sss.pgh.pa.us 2451 : 863 : continue;
2452 : : }
2453 : :
2454 : : /*
2455 : : * Was it the wal receiver? If exit status is zero (normal) or one
2456 : : * (FATAL exit), we assume everything is all right just like normal
2457 : : * backends. (If we need a new wal receiver, we'll start one at the
2458 : : * next iteration of the postmaster's main loop.)
2459 : : */
651 heikki.linnakangas@i 2460 [ + + + + ]: 22448 : if (WalReceiverPMChild && pid == WalReceiverPMChild->pid)
2461 : : {
2462 : 291 : ReleasePostmasterChildSlot(WalReceiverPMChild);
2463 : 291 : WalReceiverPMChild = NULL;
6068 2464 [ + - + - : 291 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
+ + ]
2465 : 21 : HandleChildCrash(pid, exitstatus,
2466 : 21 : _("WAL receiver process"));
2467 : 291 : continue;
2468 : : }
2469 : :
2470 : : /*
2471 : : * Was it the wal summarizer? Normal exit can be ignored; we'll start
2472 : : * a new one at the next iteration of the postmaster's main loop, if
2473 : : * necessary. Any other exit condition is treated as a crash.
2474 : : */
651 2475 [ + + + + ]: 22157 : if (WalSummarizerPMChild && pid == WalSummarizerPMChild->pid)
2476 : : {
2477 : 24 : ReleasePostmasterChildSlot(WalSummarizerPMChild);
2478 : 24 : WalSummarizerPMChild = NULL;
981 rhaas@postgresql.org 2479 [ + + ]: 24 : if (!EXIT_STATUS_0(exitstatus))
2480 : 19 : HandleChildCrash(pid, exitstatus,
2481 : 19 : _("WAL summarizer process"));
2482 : 24 : continue;
2483 : : }
2484 : :
2485 : : /*
2486 : : * Was it the autovacuum launcher? Normal exit can be ignored; we'll
2487 : : * start a new one at the next iteration of the postmaster's main
2488 : : * loop, if necessary. Any other exit condition is treated as a
2489 : : * crash.
2490 : : */
651 heikki.linnakangas@i 2491 [ + + + + ]: 22133 : if (AutoVacLauncherPMChild && pid == AutoVacLauncherPMChild->pid)
2492 : : {
2493 : 714 : ReleasePostmasterChildSlot(AutoVacLauncherPMChild);
2494 : 714 : AutoVacLauncherPMChild = NULL;
7133 alvherre@alvh.no-ip. 2495 [ + + ]: 714 : if (!EXIT_STATUS_0(exitstatus))
7714 tgl@sss.pgh.pa.us 2496 : 260 : HandleChildCrash(pid, exitstatus,
7133 alvherre@alvh.no-ip. 2497 : 260 : _("autovacuum launcher process"));
7714 tgl@sss.pgh.pa.us 2498 : 714 : continue;
2499 : : }
2500 : :
2501 : : /*
2502 : : * Was it the archiver? If exit status is zero (normal) or one (FATAL
2503 : : * exit), we assume everything is all right just like normal backends
2504 : : * and just try to start a new one on the next cycle of the
2505 : : * postmaster's main loop, to retry archiving remaining files.
2506 : : */
651 heikki.linnakangas@i 2507 [ + + + + ]: 21419 : if (PgArchPMChild && pid == PgArchPMChild->pid)
2508 : : {
2509 : 63 : ReleasePostmasterChildSlot(PgArchPMChild);
2510 : 63 : PgArchPMChild = NULL;
1991 fujii@postgresql.org 2511 [ + + + - : 63 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
+ - ]
2512 : 44 : HandleChildCrash(pid, exitstatus,
2513 : 44 : _("archiver process"));
8074 tgl@sss.pgh.pa.us 2514 : 63 : continue;
2515 : : }
2516 : :
2517 : : /* Was it the system logger? If so, try to start a new one */
651 heikki.linnakangas@i 2518 [ + + - + ]: 21356 : if (SysLoggerPMChild && pid == SysLoggerPMChild->pid)
2519 : : {
651 heikki.linnakangas@i 2520 :UBC 0 : ReleasePostmasterChildSlot(SysLoggerPMChild);
2521 : 0 : SysLoggerPMChild = NULL;
2522 : :
2523 : : /* for safety's sake, launch new logger *first* */
2524 [ # # ]: 0 : if (Logging_collector)
2525 : 0 : StartSysLogger();
2526 : :
7219 tgl@sss.pgh.pa.us 2527 [ # # ]: 0 : if (!EXIT_STATUS_0(exitstatus))
7856 bruce@momjian.us 2528 : 0 : LogChildExit(LOG, _("system logger process"),
2529 : : pid, exitstatus);
8057 tgl@sss.pgh.pa.us 2530 : 0 : continue;
2531 : : }
2532 : :
2533 : : /*
2534 : : * Was it the slot sync worker? Normal exit or FATAL exit can be
2535 : : * ignored (FATAL can be caused by libpqwalreceiver on receiving
2536 : : * shutdown request by the startup process during promotion); we'll
2537 : : * start a new one at the next iteration of the postmaster's main
2538 : : * loop, if necessary. Any other exit condition is treated as a crash.
2539 : : */
651 heikki.linnakangas@i 2540 [ + + + + ]:CBC 21356 : if (SlotSyncWorkerPMChild && pid == SlotSyncWorkerPMChild->pid)
2541 : : {
2542 : 6 : ReleasePostmasterChildSlot(SlotSyncWorkerPMChild);
2543 : 6 : SlotSyncWorkerPMChild = NULL;
917 akapila@postgresql.o 2544 [ + + + - : 6 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
- + ]
917 akapila@postgresql.o 2545 :UBC 0 : HandleChildCrash(pid, exitstatus,
2546 : 0 : _("slot sync worker process"));
917 akapila@postgresql.o 2547 :CBC 6 : continue;
2548 : : }
2549 : :
2550 : : /* Was it an IO worker? */
527 andres@anarazel.de 2551 [ + + ]: 21350 : if (maybe_reap_io_worker(pid))
2552 : : {
2553 [ + + + - : 2039 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
+ + ]
2554 : 751 : HandleChildCrash(pid, exitstatus, _("io worker"));
2555 : :
2556 : : /*
2557 : : * A worker that exited with an error might have brought the pool
2558 : : * size below io_min_workers, or allowed the queue to grow to the
2559 : : * point where another worker called for growth.
2560 : : *
2561 : : * In the common case that a worker timed out due to idleness, no
2562 : : * replacement needs to be started. maybe_start_io_workers() will
2563 : : * figure that out.
2564 : : */
141 tmunro@postgresql.or 2565 : 2039 : maybe_start_io_workers();
2566 : :
527 andres@anarazel.de 2567 : 2039 : continue;
2568 : : }
2569 : :
2570 : : /*
2571 : : * Was it a backend or a background worker?
2572 : : */
651 heikki.linnakangas@i 2573 : 19311 : pmchild = FindPostmasterChildByPid(pid);
2574 [ + - ]: 19311 : if (pmchild)
2575 : : {
2576 : 19311 : CleanupBackend(pmchild, exitstatus);
2577 : : }
2578 : :
2579 : : /*
2580 : : * We don't know anything about this child process. That's highly
2581 : : * unexpected, as we do track all the child processes that we fork.
2582 : : */
2583 : : else
2584 : : {
747 heikki.linnakangas@i 2585 [ # # # # :UBC 0 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
# # ]
2586 : 0 : HandleChildCrash(pid, exitstatus, _("untracked child process"));
2587 : : else
2588 : 0 : LogChildExit(LOG, _("untracked child process"), pid, exitstatus);
2589 : : }
2590 : : } /* loop over pending child-death reports */
2591 : :
2592 : : /*
2593 : : * After cleaning out the SIGCHLD queue, see if we have any state changes
2594 : : * or actions to make.
2595 : : */
6958 tgl@sss.pgh.pa.us 2596 :CBC 24328 : PostmasterStateMachine();
11006 scrappy@hub.org 2597 : 23337 : }
2598 : :
2599 : : /*
2600 : : * CleanupBackend -- cleanup after terminated backend or background worker.
2601 : : *
2602 : : * Remove all local state associated with the child process and release its
2603 : : * PMChild slot.
2604 : : */
2605 : : static void
651 heikki.linnakangas@i 2606 : 19311 : CleanupBackend(PMChild *bp,
2607 : : int exitstatus) /* child's exit status. */
2608 : : {
2609 : : char namebuf[MAXPGPATH];
2610 : : const char *procname;
747 2611 : 19311 : bool crashed = false;
2612 : 19311 : bool logged = false;
2613 : : pid_t bp_pid;
2614 : : bool bp_bgworker_notify;
2615 : : BackendType bp_bkend_type;
2616 : : RegisteredBgWorker *rw;
2617 : :
2618 : : /* Construct a process name for the log message */
651 2619 [ + + ]: 19311 : if (bp->bkend_type == B_BG_WORKER)
2620 : : {
3283 peter_e@gmx.net 2621 : 3540 : snprintf(namebuf, MAXPGPATH, _("background worker \"%s\""),
747 heikki.linnakangas@i 2622 : 3540 : bp->rw->rw_worker.bgw_type);
2623 : 3540 : procname = namebuf;
2624 : : }
2625 : : else
651 2626 : 15771 : procname = _(GetBackendTypeDesc(bp->bkend_type));
2627 : :
2628 : : /*
2629 : : * If a backend dies in an ugly way then we must signal all other backends
2630 : : * to quickdie. If exit status is zero (normal) or one (FATAL exit), we
2631 : : * assume everything is all right and proceed to remove the backend from
2632 : : * the active child list.
2633 : : */
747 2634 [ + + + + : 19311 : if (!EXIT_STATUS_0(exitstatus) && !EXIT_STATUS_1(exitstatus))
+ + ]
2635 : 714 : crashed = true;
2636 : :
2637 : : #ifdef WIN32
2638 : :
2639 : : /*
2640 : : * On win32, also treat ERROR_WAIT_NO_CHILDREN (128) as nonfatal case,
2641 : : * since that sometimes happens under load when the process fails to start
2642 : : * properly (long before it starts using shared memory). Microsoft reports
2643 : : * it is related to mutex failure:
2644 : : * http://archives.postgresql.org/pgsql-hackers/2010-09/msg00790.php
2645 : : */
2646 : : if (exitstatus == ERROR_WAIT_NO_CHILDREN)
2647 : : {
2648 : : LogChildExit(LOG, procname, bp->pid, exitstatus);
2649 : : logged = true;
2650 : : crashed = false;
2651 : : }
2652 : : #endif
2653 : :
2654 : : /*
2655 : : * Release the PMChild entry.
2656 : : *
2657 : : * If the process attached to shared memory, this also checks that it
2658 : : * detached cleanly.
2659 : : */
651 2660 : 19311 : bp_pid = bp->pid;
2661 : 19311 : bp_bgworker_notify = bp->bgworker_notify;
2662 : 19311 : bp_bkend_type = bp->bkend_type;
2663 : 19311 : rw = bp->rw;
2664 [ + + ]: 19311 : if (!ReleasePostmasterChildSlot(bp))
2665 : : {
2666 : : /*
2667 : : * Uh-oh, the child failed to clean itself up. Treat as a crash after
2668 : : * all.
2669 : : */
2670 : 401 : crashed = true;
2671 : : }
2672 : 19311 : bp = NULL;
2673 : :
2674 : : /*
2675 : : * In a crash case, exit immediately without resetting background worker
2676 : : * state. However, if restart_after_crash is enabled, the background
2677 : : * worker state (e.g., rw_pid) still needs be reset so the worker can
2678 : : * restart after crash recovery. This reset is handled in
2679 : : * ResetBackgroundWorkerCrashTimes(), not here.
2680 : : */
747 2681 [ + + ]: 19311 : if (crashed)
2682 : : {
651 2683 : 714 : HandleChildCrash(bp_pid, exitstatus, procname);
8125 tgl@sss.pgh.pa.us 2684 : 714 : return;
2685 : : }
2686 : :
2687 : : /*
2688 : : * This backend may have been slated to receive SIGUSR1 when some
2689 : : * background worker started or stopped. Cancel those notifications, as
2690 : : * we don't want to signal PIDs that are not PostgreSQL backends. This
2691 : : * gets skipped in the (probably very common) case where the backend has
2692 : : * never requested any such notifications.
2693 : : */
651 heikki.linnakangas@i 2694 [ + + ]: 18597 : if (bp_bgworker_notify)
2695 : 363 : BackgroundWorkerStopNotifications(bp_pid);
2696 : :
2697 : : /*
2698 : : * If it was an autovacuum worker, wake up the launcher so that it can
2699 : : * immediately launch a new worker or rebalance to cost limit setting of
2700 : : * the remaining workers.
2701 : : */
224 2702 [ + + + + ]: 18597 : if (bp_bkend_type == B_AUTOVAC_WORKER && AutoVacLauncherPMChild != NULL)
2703 : 155 : signal_child(AutoVacLauncherPMChild, SIGUSR2);
2704 : :
2705 : : /*
2706 : : * If it was a background worker, also update its RegisteredBgWorker
2707 : : * entry.
2708 : : */
651 2709 [ + + ]: 18597 : if (bp_bkend_type == B_BG_WORKER)
2710 : : {
747 2711 [ + + ]: 3202 : if (!EXIT_STATUS_0(exitstatus))
2712 : : {
2713 : : /* Record timestamp, so we know when to restart the worker. */
2714 : 799 : rw->rw_crashed_at = GetCurrentTimestamp();
2715 : : }
2716 : : else
2717 : : {
2718 : : /* Zero exit status means terminate */
2719 : 2403 : rw->rw_crashed_at = 0;
2720 : 2403 : rw->rw_terminate = true;
2721 : : }
2722 : :
2723 : 3202 : rw->rw_pid = 0;
2724 : 3202 : ReportBackgroundWorkerExit(rw); /* report child death */
2725 : :
2726 [ + - ]: 3202 : if (!logged)
2727 : : {
2728 [ + + ]: 3202 : LogChildExit(EXIT_STATUS_0(exitstatus) ? DEBUG1 : LOG,
2729 : : procname, bp_pid, exitstatus);
2730 : 3202 : logged = true;
2731 : : }
2732 : :
2733 : : /* have it be restarted */
2734 : 3202 : HaveCrashedWorker = true;
2735 : : }
2736 : :
2737 [ + + ]: 18597 : if (!logged)
651 2738 : 15395 : LogChildExit(DEBUG2, procname, bp_pid, exitstatus);
2739 : : }
2740 : :
2741 : : /*
2742 : : * Transition into FatalError state, in response to something bad having
2743 : : * happened. Commonly the caller will have logged the reason for entering
2744 : : * FatalError state.
2745 : : *
2746 : : * This should only be called when not already in ImmediateShutdown state.
2747 : : */
2748 : : static void
580 andres@anarazel.de 2749 : 16 : HandleFatalError(QuitSignalReason reason, bool consider_sigabrt)
2750 : : {
2751 : : int sigtosend;
2752 : :
2753 [ - + ]: 16 : Assert(Shutdown != ImmediateShutdown);
2754 : :
2755 : 16 : SetQuitSignalReason(reason);
2756 : :
2757 [ + - - + ]: 16 : if (consider_sigabrt && send_abort_for_crash)
580 andres@anarazel.de 2758 :UBC 0 : sigtosend = SIGABRT;
2759 : : else
580 andres@anarazel.de 2760 :CBC 16 : sigtosend = SIGQUIT;
2761 : :
2762 : : /*
2763 : : * Signal all other child processes to exit.
2764 : : *
2765 : : * We could exclude dead-end children here, but at least when sending
2766 : : * SIGABRT it seems better to include them.
2767 : : */
2768 : 16 : TerminateChildren(sigtosend);
2769 : :
2770 : 16 : FatalError = true;
2771 : :
2772 : : /*
2773 : : * Choose the appropriate new state to react to the fatal error. Unless we
2774 : : * were already in the process of shutting down, we go through
2775 : : * PM_WAIT_BACKENDS. For errors during the shutdown sequence, we directly
2776 : : * switch to PM_WAIT_DEAD_END.
2777 : : */
2778 [ - + - - : 16 : switch (pmState)
- - ]
2779 : : {
580 andres@anarazel.de 2780 :UBC 0 : case PM_INIT:
2781 : : /* shouldn't have any children */
2782 : 0 : Assert(false);
2783 : : break;
2784 : :
2785 : : /* wait for children to die */
128 michael@paquier.xyz 2786 :CBC 16 : case PM_STARTUP:
2787 : : case PM_RECOVERY:
2788 : : case PM_HOT_STANDBY:
2789 : : case PM_RUN:
2790 : : case PM_STOP_BACKENDS:
580 andres@anarazel.de 2791 : 16 : UpdatePMState(PM_WAIT_BACKENDS);
2792 : 16 : break;
2793 : :
580 andres@anarazel.de 2794 :UBC 0 : case PM_WAIT_BACKENDS:
2795 : : /* there might be more backends to wait for */
2796 : 0 : break;
2797 : :
2798 : 0 : case PM_WAIT_XLOG_SHUTDOWN:
2799 : : case PM_WAIT_XLOG_ARCHIVAL:
2800 : : case PM_WAIT_CHECKPOINTER:
2801 : : case PM_WAIT_IO_WORKERS:
2802 : :
2803 : : /*
2804 : : * NB: Similar code exists in PostmasterStateMachine()'s handling
2805 : : * of FatalError in PM_STOP_BACKENDS/PM_WAIT_BACKENDS states.
2806 : : */
2807 : 0 : ConfigurePostmasterWaitSet(false);
2808 : 0 : UpdatePMState(PM_WAIT_DEAD_END);
2809 : 0 : break;
2810 : :
2811 : 0 : case PM_WAIT_DEAD_END:
2812 : : case PM_NO_CHILDREN:
2813 : 0 : break;
2814 : : }
2815 : :
2816 : : /*
2817 : : * .. and if this doesn't happen quickly enough, now the clock is ticking
2818 : : * for us to kill them without mercy.
2819 : : */
4808 alvherre@alvh.no-ip. 2820 [ + - ]:CBC 16 : if (AbortStartTime == 0)
2821 : 16 : AbortStartTime = time(NULL);
11006 scrappy@hub.org 2822 : 16 : }
2823 : :
2824 : : /*
2825 : : * HandleChildCrash -- cleanup after failed backend, bgwriter, checkpointer,
2826 : : * walwriter, autovacuum, archiver, slot sync worker, or background worker.
2827 : : *
2828 : : * The objectives here are to clean up our local state about the child
2829 : : * process, and to signal all other remaining children to quickdie.
2830 : : *
2831 : : * The caller has already released its PMChild slot.
2832 : : */
2833 : : static void
580 andres@anarazel.de 2834 : 2931 : HandleChildCrash(int pid, int exitstatus, const char *procname)
2835 : : {
2836 : : /*
2837 : : * We only log messages and send signals if this is the first process
2838 : : * crash and we're not doing an immediate shutdown; otherwise, we're only
2839 : : * here to update postmaster's idea of live processes. If we have already
2840 : : * signaled children, nonzero exit status is to be expected, so don't
2841 : : * clutter log.
2842 : : */
2843 [ + + + + ]: 2931 : if (FatalError || Shutdown == ImmediateShutdown)
2844 : 2915 : return;
2845 : :
2846 : 16 : LogChildExit(LOG, procname, pid, exitstatus);
2847 [ + - ]: 16 : ereport(LOG,
2848 : : (errmsg("terminating any other active server processes")));
2849 : :
2850 : : /*
2851 : : * Switch into error state. The crashed process has already been removed
2852 : : * from ActiveChildList.
2853 : : */
2854 : 16 : HandleFatalError(PMQUIT_FOR_CRASH, true);
2855 : : }
2856 : :
2857 : : /*
2858 : : * Log the death of a child process.
2859 : : */
2860 : : static void
8942 tgl@sss.pgh.pa.us 2861 : 18613 : LogChildExit(int lev, const char *procname, int pid, int exitstatus)
2862 : : {
2863 : : /*
2864 : : * size of activity_buffer is arbitrary, but set equal to default
2865 : : * track_activity_query_size
2866 : : */
2867 : : char activity_buffer[1024];
5424 2868 : 18613 : const char *activity = NULL;
2869 : :
2870 [ + + ]: 18613 : if (!EXIT_STATUS_0(exitstatus))
2871 : 1219 : activity = pgstat_get_crashed_backend_activity(pid,
2872 : : activity_buffer,
2873 : : sizeof(activity_buffer));
2874 : :
9060 2875 [ + + ]: 18613 : if (WIFEXITED(exitstatus))
8437 2876 [ + + + + ]: 18609 : ereport(lev,
2877 : :
2878 : : /*------
2879 : : translator: %s is a noun phrase describing a child process, such as
2880 : : "server process" */
2881 : : (errmsg("%s (PID %d) exited with exit code %d",
2882 : : procname, pid, WEXITSTATUS(exitstatus)),
2883 : : activity ? errdetail("Failed process was running: %s", activity) : 0));
9060 2884 [ + - ]: 4 : else if (WIFSIGNALED(exitstatus))
2885 : : {
2886 : : #if defined(WIN32)
2887 : : ereport(lev,
2888 : :
2889 : : /*------
2890 : : translator: %s is a noun phrase describing a child process, such as
2891 : : "server process" */
2892 : : (errmsg("%s (PID %d) was terminated by exception 0x%X",
2893 : : procname, pid, WTERMSIG(exitstatus)),
2894 : : errhint("See C include file \"ntstatus.h\" for a description of the hexadecimal value."),
2895 : : activity ? errdetail("Failed process was running: %s", activity) : 0));
2896 : : #else
3354 2897 [ + - + - ]: 4 : ereport(lev,
2898 : :
2899 : : /*------
2900 : : translator: %s is a noun phrase describing a child process, such as
2901 : : "server process" */
2902 : : (errmsg("%s (PID %d) was terminated by signal %d: %s",
2903 : : procname, pid, WTERMSIG(exitstatus),
2904 : : pg_strsignal(WTERMSIG(exitstatus))),
2905 : : activity ? errdetail("Failed process was running: %s", activity) : 0));
2906 : : #endif
2907 : : }
2908 : : else
8437 tgl@sss.pgh.pa.us 2909 [ # # # # ]:UBC 0 : ereport(lev,
2910 : :
2911 : : /*------
2912 : : translator: %s is a noun phrase describing a child process, such as
2913 : : "server process" */
2914 : : (errmsg("%s (PID %d) exited with unrecognized status %d",
2915 : : procname, pid, exitstatus),
2916 : : activity ? errdetail("Failed process was running: %s", activity) : 0));
9060 tgl@sss.pgh.pa.us 2917 :CBC 18613 : }
2918 : :
2919 : : /*
2920 : : * Advance the postmaster's state machine and take actions as appropriate
2921 : : *
2922 : : * This is common code for process_pm_shutdown_request(),
2923 : : * process_pm_child_exit() and process_pm_pmsignal(), which process the signals
2924 : : * that might mean we need to change state.
2925 : : */
2926 : : static void
6958 2927 : 27316 : PostmasterStateMachine(void)
2928 : : {
2929 : : /* If we're doing a smart shutdown, try to advance that state. */
2204 2930 [ + + + + ]: 27316 : if (pmState == PM_RUN || pmState == PM_HOT_STANDBY)
2931 : : {
1604 sfrost@snowman.net 2932 [ + + ]: 19436 : if (!connsAllowed)
2933 : : {
2934 : : /*
2935 : : * This state ends when we have no normal client backends running.
2936 : : * Then we're ready to stop other children.
2937 : : */
651 heikki.linnakangas@i 2938 [ + + ]: 107 : if (CountChildren(btmask(B_BACKEND)) == 0)
594 andres@anarazel.de 2939 : 49 : UpdatePMState(PM_STOP_BACKENDS);
2940 : : }
2941 : : }
2942 : :
2943 : : /*
2944 : : * In the PM_WAIT_BACKENDS state, wait for all the regular backends and
2945 : : * processes like autovacuum and background workers that are comparable to
2946 : : * backends to exit.
2947 : : *
2948 : : * PM_STOP_BACKENDS is a transient state that means the same as
2949 : : * PM_WAIT_BACKENDS, but we signal the processes first, before waiting for
2950 : : * them. Treating it as a distinct pmState allows us to share this code
2951 : : * across multiple shutdown code paths.
2952 : : */
651 heikki.linnakangas@i 2953 [ + + + + ]: 27316 : if (pmState == PM_STOP_BACKENDS || pmState == PM_WAIT_BACKENDS)
2954 : : {
2955 : 5303 : BackendTypeMask targetMask = BTYPE_MASK_NONE;
2956 : :
2957 : : /*
2958 : : * PM_WAIT_BACKENDS state ends when we have no regular backends, no
2959 : : * autovac launcher or workers, and no bgworkers (including
2960 : : * unconnected ones).
2961 : : */
594 andres@anarazel.de 2962 : 5303 : targetMask = btmask_add(targetMask,
2963 : : B_BACKEND,
2964 : : B_AUTOVAC_LAUNCHER,
2965 : : B_AUTOVAC_WORKER,
2966 : : B_BG_WORKER);
2967 : :
2968 : : /*
2969 : : * No walwriter, bgwriter, slot sync worker, or WAL summarizer either.
2970 : : */
2971 : 5303 : targetMask = btmask_add(targetMask,
2972 : : B_WAL_WRITER,
2973 : : B_BG_WRITER,
2974 : : B_SLOTSYNC_WORKER,
2975 : : B_WAL_SUMMARIZER);
2976 : :
2977 : : /* If we're in recovery, also stop startup and walreceiver procs */
2978 : 5303 : targetMask = btmask_add(targetMask,
2979 : : B_STARTUP,
2980 : : B_WAL_RECEIVER);
2981 : :
2982 : : /*
2983 : : * If we are doing crash recovery or an immediate shutdown then we
2984 : : * expect archiver, checkpointer, io workers and walsender to exit as
2985 : : * well, otherwise not.
2986 : : */
651 heikki.linnakangas@i 2987 [ + + + + ]: 5303 : if (FatalError || Shutdown >= ImmediateShutdown)
580 andres@anarazel.de 2988 : 1940 : targetMask = btmask_add(targetMask,
2989 : : B_CHECKPOINTER,
2990 : : B_ARCHIVER,
2991 : : B_IO_WORKER,
2992 : : B_WAL_SENDER);
2993 : :
2994 : : /*
2995 : : * Normally archiver, checkpointer, IO workers and walsenders will
2996 : : * continue running; they will be terminated later after writing the
2997 : : * checkpoint record. We also let dead-end children to keep running
2998 : : * for now. The syslogger process exits last.
2999 : : *
3000 : : * This assertion checks that we have covered all backend types,
3001 : : * either by including them in targetMask, or by noting here that they
3002 : : * are allowed to continue running.
3003 : : */
3004 : : #ifdef USE_ASSERT_CHECKING
3005 : : {
651 heikki.linnakangas@i 3006 : 5303 : BackendTypeMask remainMask = BTYPE_MASK_NONE;
3007 : :
594 andres@anarazel.de 3008 : 5303 : remainMask = btmask_add(remainMask,
3009 : : B_DEAD_END_BACKEND,
3010 : : B_LOGGER);
3011 : :
3012 : : /*
3013 : : * Archiver, checkpointer, IO workers, and walsender may or may
3014 : : * not be in targetMask already.
3015 : : */
580 3016 : 5303 : remainMask = btmask_add(remainMask,
3017 : : B_ARCHIVER,
3018 : : B_CHECKPOINTER,
3019 : : B_IO_WORKER,
3020 : : B_WAL_SENDER);
3021 : :
3022 : : /* these are not real postmaster children */
594 3023 : 5303 : remainMask = btmask_add(remainMask,
3024 : : B_INVALID,
3025 : : B_STANDALONE_BACKEND);
3026 : :
3027 : : /* also add data checksums processes */
146 dgustafsson@postgres 3028 : 5303 : remainMask = btmask_add(remainMask,
3029 : : B_DATACHECKSUMSWORKER_LAUNCHER,
3030 : : B_DATACHECKSUMSWORKER_WORKER);
3031 : :
3032 : : /* All types should be included in targetMask or remainMask */
651 heikki.linnakangas@i 3033 [ - + ]: 5303 : Assert((remainMask.mask | targetMask.mask) == BTYPE_MASK_ALL.mask);
3034 : : }
3035 : : #endif
3036 : :
3037 : : /* If we had not yet signaled the processes to exit, do so now */
3038 [ + + ]: 5303 : if (pmState == PM_STOP_BACKENDS)
3039 : : {
3040 : : /*
3041 : : * Forget any pending requests for background workers, since we're
3042 : : * no longer willing to launch any new workers. (If additional
3043 : : * requests arrive, BackgroundWorkerStateChange will reject them.)
3044 : : */
3045 : 622 : ForgetUnstartedBackgroundWorkers();
3046 : :
3047 : 622 : SignalChildren(SIGTERM, targetMask);
3048 : :
594 andres@anarazel.de 3049 : 622 : UpdatePMState(PM_WAIT_BACKENDS);
3050 : : }
3051 : :
3052 : : /* Are any of the target processes still running? */
651 heikki.linnakangas@i 3053 [ + + ]: 5303 : if (CountChildren(targetMask) == 0)
3054 : : {
4808 alvherre@alvh.no-ip. 3055 [ + + + + ]: 996 : if (Shutdown >= ImmediateShutdown || FatalError)
3056 : : {
3057 : : /*
3058 : : * Stop any dead-end children and stop creating new ones.
3059 : : *
3060 : : * NB: Similar code exists in HandleFatalError(), when the
3061 : : * error happens in pmState > PM_WAIT_BACKENDS.
3062 : : */
594 andres@anarazel.de 3063 : 374 : UpdatePMState(PM_WAIT_DEAD_END);
651 heikki.linnakangas@i 3064 : 374 : ConfigurePostmasterWaitSet(false);
3065 : 374 : SignalChildren(SIGQUIT, btmask(B_DEAD_END_BACKEND));
3066 : :
3067 : : /*
3068 : : * We already SIGQUIT'd auxiliary processes (other than
3069 : : * logger), if any, when we started immediate shutdown or
3070 : : * entered FatalError state.
3071 : : */
3072 : : }
3073 : : else
3074 : : {
3075 : : /*
3076 : : * If we get here, we are proceeding with normal shutdown. All
3077 : : * the regular children are gone, and it's time to tell the
3078 : : * checkpointer to do a shutdown checkpoint.
3079 : : */
6958 tgl@sss.pgh.pa.us 3080 [ - + ]: 622 : Assert(Shutdown > NoShutdown);
3081 : : /* Start the checkpointer if not running */
651 heikki.linnakangas@i 3082 [ - + ]: 622 : if (CheckpointerPMChild == NULL)
651 heikki.linnakangas@i 3083 :UBC 0 : CheckpointerPMChild = StartChildProcess(B_CHECKPOINTER);
3084 : : /* And tell it to write the shutdown checkpoint */
651 heikki.linnakangas@i 3085 [ + - ]:CBC 622 : if (CheckpointerPMChild != NULL)
3086 : : {
579 andres@anarazel.de 3087 : 622 : signal_child(CheckpointerPMChild, SIGINT);
594 3088 : 622 : UpdatePMState(PM_WAIT_XLOG_SHUTDOWN);
3089 : : }
3090 : : else
3091 : : {
3092 : : /*
3093 : : * If we failed to fork a checkpointer, just shut down.
3094 : : * Any required cleanup will happen at next restart. We
3095 : : * set FatalError so that an "abnormal shutdown" message
3096 : : * gets logged when we exit.
3097 : : *
3098 : : * We don't consult send_abort_for_crash here, as it's
3099 : : * unlikely that dumping cores would illuminate the reason
3100 : : * for checkpointer fork failure.
3101 : : *
3102 : : * XXX: It may be worth to introduce a different PMQUIT
3103 : : * value that signals that the cluster is in a bad state,
3104 : : * without a process having crashed. But right now this
3105 : : * path is very unlikely to be reached, so it isn't
3106 : : * obviously worthwhile adding a distinct error message in
3107 : : * quickdie().
3108 : : */
580 andres@anarazel.de 3109 :UBC 0 : HandleFatalError(PMQUIT_FOR_CRASH, false);
3110 : : }
3111 : : }
3112 : : }
3113 : : }
3114 : :
3115 : : /*
3116 : : * The state transition from PM_WAIT_XLOG_SHUTDOWN to
3117 : : * PM_WAIT_XLOG_ARCHIVAL is in process_pm_pmsignal(), in response to
3118 : : * PMSIGNAL_XLOG_IS_SHUTDOWN.
3119 : : */
3120 : :
594 andres@anarazel.de 3121 [ + + ]:CBC 27316 : if (pmState == PM_WAIT_XLOG_ARCHIVAL)
3122 : : {
3123 : : /*
3124 : : * PM_WAIT_XLOG_ARCHIVAL state ends when there are no children other
3125 : : * than checkpointer, io workers and dead-end children left. There
3126 : : * shouldn't be any regular backends left by now anyway; what we're
3127 : : * really waiting for is for walsenders and archiver to exit.
3128 : : */
527 3129 [ + + ]: 691 : if (CountChildren(btmask_all_except(B_CHECKPOINTER, B_IO_WORKER,
3130 : : B_LOGGER, B_DEAD_END_BACKEND)) == 0)
3131 : : {
3132 : 622 : UpdatePMState(PM_WAIT_IO_WORKERS);
3133 : 622 : SignalChildren(SIGUSR2, btmask(B_IO_WORKER));
3134 : : }
3135 : : }
3136 : :
3137 [ + + ]: 27316 : if (pmState == PM_WAIT_IO_WORKERS)
3138 : : {
3139 : : /*
3140 : : * PM_WAIT_IO_WORKERS state ends when there's only checkpointer and
3141 : : * dead-end children left.
3142 : : */
3143 [ + + ]: 1828 : if (io_worker_count == 0)
3144 : : {
579 3145 : 622 : UpdatePMState(PM_WAIT_CHECKPOINTER);
3146 : :
3147 : : /*
3148 : : * Now that the processes mentioned above are gone, tell
3149 : : * checkpointer to shut down too. That allows checkpointer to
3150 : : * perform some last bits of cleanup without other processes
3151 : : * interfering.
3152 : : */
3153 [ + - ]: 622 : if (CheckpointerPMChild != NULL)
3154 : 622 : signal_child(CheckpointerPMChild, SIGUSR2);
3155 : : }
3156 : : }
3157 : :
3158 : : /*
3159 : : * The state transition from PM_WAIT_CHECKPOINTER to PM_WAIT_DEAD_END is
3160 : : * in process_pm_child_exit().
3161 : : */
3162 : :
6958 tgl@sss.pgh.pa.us 3163 [ + + ]: 27316 : if (pmState == PM_WAIT_DEAD_END)
3164 : : {
3165 : : /*
3166 : : * PM_WAIT_DEAD_END state ends when all other children are gone except
3167 : : * for the logger. During normal shutdown, all that remains are
3168 : : * dead-end backends, but in FatalError processing we jump straight
3169 : : * here with more processes remaining. Note that they have already
3170 : : * been sent appropriate shutdown signals, either during a normal
3171 : : * state transition leading up to PM_WAIT_DEAD_END, or during
3172 : : * FatalError processing.
3173 : : *
3174 : : * The reason we wait is to protect against a new postmaster starting
3175 : : * conflicting subprocesses; this isn't an ironclad protection, but it
3176 : : * at least helps in the shutdown-and-immediately-restart scenario.
3177 : : */
651 heikki.linnakangas@i 3178 [ + + ]: 1024 : if (CountChildren(btmask_all_except(B_LOGGER)) == 0)
3179 : : {
3180 : : /* These other guys should be dead already */
3181 [ - + ]: 996 : Assert(StartupPMChild == NULL);
3182 [ - + ]: 996 : Assert(WalReceiverPMChild == NULL);
3183 [ - + ]: 996 : Assert(WalSummarizerPMChild == NULL);
3184 [ - + ]: 996 : Assert(BgWriterPMChild == NULL);
3185 [ - + ]: 996 : Assert(CheckpointerPMChild == NULL);
3186 [ - + ]: 996 : Assert(WalWriterPMChild == NULL);
3187 [ - + ]: 996 : Assert(AutoVacLauncherPMChild == NULL);
3188 [ - + ]: 996 : Assert(SlotSyncWorkerPMChild == NULL);
3189 : : /* syslogger is not considered here */
594 andres@anarazel.de 3190 : 996 : UpdatePMState(PM_NO_CHILDREN);
3191 : : }
3192 : : }
3193 : :
3194 : : /*
3195 : : * If we've been told to shut down, we exit as soon as there are no
3196 : : * remaining children. If there was a crash, cleanup will occur at the
3197 : : * next startup. (Before PostgreSQL 8.3, we tried to recover from the
3198 : : * crash before exiting, but that seems unwise if we are quitting because
3199 : : * we got SIGTERM from init --- there may well not be time for recovery
3200 : : * before init decides to SIGKILL us.)
3201 : : *
3202 : : * Note that the syslogger continues to run. It will exit when it sees
3203 : : * EOF on its input pipe, which happens when there are no more upstream
3204 : : * processes.
3205 : : */
6958 tgl@sss.pgh.pa.us 3206 [ + + + + ]: 27316 : if (Shutdown > NoShutdown && pmState == PM_NO_CHILDREN)
3207 : : {
3208 [ - + ]: 980 : if (FatalError)
3209 : : {
6958 tgl@sss.pgh.pa.us 3210 [ # # ]:UBC 0 : ereport(LOG, (errmsg("abnormal database system shutdown")));
3211 : 0 : ExitPostmaster(1);
3212 : : }
3213 : : else
3214 : : {
3215 : : /*
3216 : : * Normal exit from the postmaster is here. We don't need to log
3217 : : * anything here, since the UnlinkLockFiles proc_exit callback
3218 : : * will do so, and that should be the last user-visible action.
3219 : : */
6958 tgl@sss.pgh.pa.us 3220 :CBC 980 : ExitPostmaster(0);
3221 : : }
3222 : : }
3223 : :
3224 : : /*
3225 : : * If the startup process failed, or the user does not want an automatic
3226 : : * restart after backend crashes, wait for all non-syslogger children to
3227 : : * exit, and then exit postmaster. We don't try to reinitialize when the
3228 : : * startup process fails, because more than likely it will just fail again
3229 : : * and we will keep trying forever.
3230 : : */
1922 3231 [ + + ]: 26336 : if (pmState == PM_NO_CHILDREN)
3232 : : {
3233 [ + + ]: 16 : if (StartupStatus == STARTUP_CRASHED)
3234 : : {
3235 [ + - ]: 11 : ereport(LOG,
3236 : : (errmsg("shutting down due to startup process failure")));
3237 : 11 : ExitPostmaster(1);
3238 : : }
3239 [ - + ]: 5 : if (!restart_after_crash)
3240 : : {
1922 tgl@sss.pgh.pa.us 3241 [ # # ]:UBC 0 : ereport(LOG,
3242 : : (errmsg("shutting down because \"restart_after_crash\" is off")));
3243 : 0 : ExitPostmaster(1);
3244 : : }
3245 : : }
3246 : :
3247 : : /*
3248 : : * If we need to recover from a crash, wait for all non-syslogger children
3249 : : * to exit, then reset shmem and start the startup process.
3250 : : */
6958 tgl@sss.pgh.pa.us 3251 [ + + + + ]:CBC 26325 : if (FatalError && pmState == PM_NO_CHILDREN)
3252 : : {
3253 [ + - ]: 5 : ereport(LOG,
3254 : : (errmsg("all server processes terminated; reinitializing")));
3255 : :
3256 : : /* remove leftover temporary files after a crash */
1988 tomas.vondra@postgre 3257 [ + + ]: 5 : if (remove_temp_files_after_crash)
3258 : 4 : RemovePgTempFiles();
3259 : :
3260 : : /* allow background workers to immediately restart */
4495 rhaas@postgresql.org 3261 : 5 : ResetBackgroundWorkerCrashTimes();
3262 : :
6444 tgl@sss.pgh.pa.us 3263 : 5 : shmem_exit(1);
3264 : :
3265 : : /* re-read control file into local memory */
3266 andres@anarazel.de 3266 : 5 : LocalProcessControlFile(true);
3267 : :
3268 : : /*
3269 : : * Re-initialize shared memory and semaphores. Note: We don't call
3270 : : * RegisterBuiltinShmemCallbacks(), we keep the old registrations. In
3271 : : * order to re-register structs in extensions, we'd need to reload
3272 : : * shared preload libraries, and we don't want to do that.
3273 : : */
143 heikki.linnakangas@i 3274 : 5 : ResetShmemAllocator();
3275 : 5 : ShmemCallRequestCallbacks();
1503 tgl@sss.pgh.pa.us 3276 : 5 : CreateSharedMemoryAndSemaphores();
3277 : :
527 andres@anarazel.de 3278 : 5 : UpdatePMState(PM_STARTUP);
3279 : :
3280 : : /* Make sure we can perform I/O while starting up. */
141 tmunro@postgresql.or 3281 : 5 : maybe_start_io_workers();
3282 : :
651 heikki.linnakangas@i 3283 : 5 : StartupPMChild = StartChildProcess(B_STARTUP);
3284 [ - + ]: 5 : Assert(StartupPMChild != NULL);
4067 tgl@sss.pgh.pa.us 3285 : 5 : StartupStatus = STARTUP_RUNNING;
3286 : : /* crash recovery started, reset SIGKILL flag */
4709 alvherre@alvh.no-ip. 3287 : 5 : AbortStartTime = 0;
3288 : :
3289 : : /* start accepting server socket connection events again */
1323 tmunro@postgresql.or 3290 : 5 : ConfigurePostmasterWaitSet(true);
3291 : : }
6958 tgl@sss.pgh.pa.us 3292 : 26325 : }
3293 : :
3294 : : static const char *
594 andres@anarazel.de 3295 : 1308 : pmstate_name(PMState state)
3296 : : {
3297 : : #define PM_TOSTR_CASE(sym) case sym: return #sym
3298 [ + + + + : 1308 : switch (state)
+ + + + +
+ + + +
- ]
3299 : : {
3300 : 71 : PM_TOSTR_CASE(PM_INIT);
3301 : 142 : PM_TOSTR_CASE(PM_STARTUP);
3302 : 30 : PM_TOSTR_CASE(PM_RECOVERY);
3303 : 28 : PM_TOSTR_CASE(PM_HOT_STANDBY);
3304 : 127 : PM_TOSTR_CASE(PM_RUN);
3305 : 106 : PM_TOSTR_CASE(PM_STOP_BACKENDS);
3306 : 158 : PM_TOSTR_CASE(PM_WAIT_BACKENDS);
3307 : 106 : PM_TOSTR_CASE(PM_WAIT_XLOG_SHUTDOWN);
3308 : 106 : PM_TOSTR_CASE(PM_WAIT_XLOG_ARCHIVAL);
527 3309 : 106 : PM_TOSTR_CASE(PM_WAIT_IO_WORKERS);
594 3310 : 148 : PM_TOSTR_CASE(PM_WAIT_DEAD_END);
579 3311 : 106 : PM_TOSTR_CASE(PM_WAIT_CHECKPOINTER);
594 3312 : 74 : PM_TOSTR_CASE(PM_NO_CHILDREN);
3313 : : }
3314 : : #undef PM_TOSTR_CASE
3315 : :
594 andres@anarazel.de 3316 :UBC 0 : pg_unreachable();
3317 : : return ""; /* silence compiler */
3318 : : }
3319 : :
3320 : : /*
3321 : : * Simple wrapper for updating pmState. The main reason to have this wrapper
3322 : : * is that it makes it easy to log all state transitions.
3323 : : */
3324 : : static void
594 andres@anarazel.de 3325 :CBC 8489 : UpdatePMState(PMState newState)
3326 : : {
3327 [ + + ]: 8489 : elog(DEBUG1, "updating PMState from %s to %s",
3328 : : pmstate_name(pmState), pmstate_name(newState));
3329 : 8489 : pmState = newState;
3330 : 8489 : }
3331 : :
3332 : : /*
3333 : : * Launch background processes after state change, or relaunch after an
3334 : : * existing process has exited.
3335 : : *
3336 : : * Check the current pmState and the status of any background processes. If
3337 : : * there are any background processes missing that should be running in the
3338 : : * current state, but are not, launch them.
3339 : : */
3340 : : static void
745 heikki.linnakangas@i 3341 : 44733 : LaunchMissingBackgroundProcesses(void)
3342 : : {
3343 : : /* Syslogger is active in all states */
651 3344 [ + + - + ]: 44733 : if (SysLoggerPMChild == NULL && Logging_collector)
651 heikki.linnakangas@i 3345 :UBC 0 : StartSysLogger();
3346 : :
3347 : : /*
3348 : : * The number of configured workers might have changed, or a prior start
3349 : : * of a worker might have failed. Check if we need to start/stop any
3350 : : * workers.
3351 : : *
3352 : : * A config file change will always lead to this function being called, so
3353 : : * we always will process the config change in a timely manner.
3354 : : */
141 tmunro@postgresql.or 3355 :CBC 44733 : maybe_start_io_workers();
3356 : :
3357 : : /*
3358 : : * The checkpointer and the background writer are active from the start,
3359 : : * until shutdown is initiated.
3360 : : *
3361 : : * (If the checkpointer is not running when we enter the
3362 : : * PM_WAIT_XLOG_SHUTDOWN state, it is launched one more time to perform
3363 : : * the shutdown checkpoint. That's done in PostmasterStateMachine(), not
3364 : : * here.)
3365 : : */
745 heikki.linnakangas@i 3366 [ + + + + ]: 44733 : if (pmState == PM_RUN || pmState == PM_RECOVERY ||
3367 [ + + + + ]: 9858 : pmState == PM_HOT_STANDBY || pmState == PM_STARTUP)
3368 : : {
651 3369 [ + + ]: 37817 : if (CheckpointerPMChild == NULL)
3370 : 5 : CheckpointerPMChild = StartChildProcess(B_CHECKPOINTER);
3371 [ + + ]: 37817 : if (BgWriterPMChild == NULL)
3372 : 5 : BgWriterPMChild = StartChildProcess(B_BG_WRITER);
3373 : : }
3374 : :
3375 : : /*
3376 : : * WAL writer is needed only in normal operation (else we cannot be
3377 : : * writing any new WAL).
3378 : : */
3379 [ + + + + ]: 44733 : if (WalWriterPMChild == NULL && pmState == PM_RUN)
3380 : 863 : WalWriterPMChild = StartChildProcess(B_WAL_WRITER);
3381 : :
3382 : : /*
3383 : : * We don't want autovacuum to run in binary upgrade mode because
3384 : : * autovacuum might update relfrozenxid for empty tables before the
3385 : : * physical files are put in place.
3386 : : */
3387 [ + + + + : 55761 : if (!IsBinaryUpgrade && AutoVacLauncherPMChild == NULL &&
+ + ]
745 3388 [ - + ]: 15747 : (AutoVacuumingActive() || start_autovac_launcher) &&
3389 [ + + ]: 6309 : pmState == PM_RUN)
3390 : : {
651 3391 : 714 : AutoVacLauncherPMChild = StartChildProcess(B_AUTOVAC_LAUNCHER);
3392 [ + - ]: 714 : if (AutoVacLauncherPMChild != NULL)
745 3393 : 714 : start_autovac_launcher = false; /* signal processed */
3394 : : }
3395 : :
3396 : : /*
3397 : : * If WAL archiving is enabled always, we are allowed to start archiver
3398 : : * even during recovery.
3399 : : */
651 3400 [ + + ]: 44733 : if (PgArchPMChild == NULL &&
745 3401 [ + + - + : 43501 : ((XLogArchivingActive() && pmState == PM_RUN) ||
+ + + + ]
3402 [ + + - + : 43501 : (XLogArchivingAlways() && (pmState == PM_RECOVERY || pmState == PM_HOT_STANDBY))) &&
+ + + - -
+ + - ]
3403 : 58 : PgArchCanRestart())
651 3404 : 58 : PgArchPMChild = StartChildProcess(B_ARCHIVER);
3405 : :
3406 : : /*
3407 : : * If we need to start a slot sync worker, try to do that now
3408 : : *
3409 : : * We allow to start the slot sync worker when we are on a hot standby,
3410 : : * fast or immediate shutdown is not in progress, slot sync parameters are
3411 : : * configured correctly, and it is the first time of worker's launch, or
3412 : : * enough time has passed since the worker was launched last.
3413 : : */
3414 [ + + + + ]: 44733 : if (SlotSyncWorkerPMChild == NULL && pmState == PM_HOT_STANDBY &&
745 3415 [ + - + + : 2668 : Shutdown <= SmartShutdown && sync_replication_slots &&
+ + ]
3416 [ + + ]: 31 : ValidateSlotSyncParams(LOG) && SlotSyncWorkerCanRestart())
651 3417 : 6 : SlotSyncWorkerPMChild = StartChildProcess(B_SLOTSYNC_WORKER);
3418 : :
3419 : : /*
3420 : : * If we need to start a WAL receiver, try to do that now
3421 : : *
3422 : : * Note: if a walreceiver process is already running, it might seem that
3423 : : * we should clear WalReceiverRequested. However, there's a race
3424 : : * condition if the walreceiver terminates and the startup process
3425 : : * immediately requests a new one: it's quite possible to get the signal
3426 : : * for the request before reaping the dead walreceiver process. Better to
3427 : : * risk launching an extra walreceiver than to miss launching one we need.
3428 : : * (The walreceiver code has logic to recognize that it should go away if
3429 : : * not needed.)
3430 : : */
745 3431 [ + + ]: 44733 : if (WalReceiverRequested)
3432 : : {
651 3433 [ + + ]: 436 : if (WalReceiverPMChild == NULL &&
745 3434 [ + + + - ]: 315 : (pmState == PM_STARTUP || pmState == PM_RECOVERY ||
3435 [ + + ]: 314 : pmState == PM_HOT_STANDBY) &&
3436 [ + - ]: 291 : Shutdown <= SmartShutdown)
3437 : : {
651 3438 : 291 : WalReceiverPMChild = StartChildProcess(B_WAL_RECEIVER);
268 peter@eisentraut.org 3439 [ + - ]: 291 : if (WalReceiverPMChild != NULL)
745 heikki.linnakangas@i 3440 : 291 : WalReceiverRequested = false;
3441 : : /* else leave the flag set, so we'll try again later */
3442 : : }
3443 : : }
3444 : :
3445 : : /* If we need to start a WAL summarizer, try to do that now */
651 3446 [ + + + + ]: 44733 : if (summarize_wal && WalSummarizerPMChild == NULL &&
745 3447 [ + + + + ]: 105 : (pmState == PM_RUN || pmState == PM_HOT_STANDBY) &&
3448 [ + - ]: 24 : Shutdown <= SmartShutdown)
651 3449 : 24 : WalSummarizerPMChild = StartChildProcess(B_WAL_SUMMARIZER);
3450 : :
3451 : : /* Get other worker processes running, if needed */
745 3452 [ + + + + ]: 44733 : if (StartWorkerNeeded || HaveCrashedWorker)
3453 : 8236 : maybe_start_bgworkers();
3454 : 44733 : }
3455 : :
3456 : : /*
3457 : : * Return string representation of signal.
3458 : : *
3459 : : * Because this is only implemented for signals we already rely on in this
3460 : : * file we don't need to deal with unimplemented or same-numeric-value signals
3461 : : * (as we'd e.g. have to for EWOULDBLOCK / EAGAIN).
3462 : : */
3463 : : static const char *
594 andres@anarazel.de 3464 : 63 : pm_signame(int signal)
3465 : : {
3466 : : #define PM_TOSTR_CASE(sym) case sym: return #sym
3467 [ - - - + : 63 : switch (signal)
- - + - +
- ]
3468 : : {
594 andres@anarazel.de 3469 :UBC 0 : PM_TOSTR_CASE(SIGABRT);
3470 : 0 : PM_TOSTR_CASE(SIGCHLD);
3471 : 0 : PM_TOSTR_CASE(SIGHUP);
594 andres@anarazel.de 3472 :CBC 7 : PM_TOSTR_CASE(SIGINT);
594 andres@anarazel.de 3473 :UBC 0 : PM_TOSTR_CASE(SIGKILL);
3474 : 0 : PM_TOSTR_CASE(SIGQUIT);
594 andres@anarazel.de 3475 :CBC 41 : PM_TOSTR_CASE(SIGTERM);
594 andres@anarazel.de 3476 :UBC 0 : PM_TOSTR_CASE(SIGUSR1);
594 andres@anarazel.de 3477 :CBC 15 : PM_TOSTR_CASE(SIGUSR2);
594 andres@anarazel.de 3478 :UBC 0 : default:
3479 : : /* all signals sent by postmaster should be listed here */
3480 : 0 : Assert(false);
3481 : : return "(unknown)";
3482 : : }
3483 : : #undef PM_TOSTR_CASE
3484 : :
3485 : : return ""; /* silence compiler */
3486 : : }
3487 : :
3488 : : /*
3489 : : * Send a signal to a postmaster child process
3490 : : *
3491 : : * On systems that have setsid(), each child process sets itself up as a
3492 : : * process group leader. For signals that are generally interpreted in the
3493 : : * appropriate fashion, we signal the entire process group not just the
3494 : : * direct child process. This allows us to, for example, SIGQUIT a blocked
3495 : : * archive_recovery script, or SIGINT a script being run by a backend via
3496 : : * system().
3497 : : *
3498 : : * There is a race condition for recently-forked children: they might not
3499 : : * have executed setsid() yet. So we signal the child directly as well as
3500 : : * the group. We assume such a child will handle the signal before trying
3501 : : * to spawn any grandchild processes. We also assume that signaling the
3502 : : * child twice will not cause any problems.
3503 : : */
3504 : : static void
651 heikki.linnakangas@i 3505 :CBC 10209 : signal_child(PMChild *pmchild, int signal)
3506 : : {
3507 : 10209 : pid_t pid = pmchild->pid;
3508 : :
594 andres@anarazel.de 3509 [ + + ]: 10209 : ereport(DEBUG3,
3510 : : (errmsg_internal("sending signal %d/%s to %s process with pid %d",
3511 : : signal, pm_signame(signal),
3512 : : GetBackendTypeDesc(pmchild->bkend_type),
3513 : : (int) pmchild->pid)));
3514 : :
7219 tgl@sss.pgh.pa.us 3515 [ - + ]: 10209 : if (kill(pid, signal) < 0)
7219 tgl@sss.pgh.pa.us 3516 [ # # ]:UBC 0 : elog(DEBUG3, "kill(%ld,%d) failed: %m", (long) pid, signal);
3517 : : #ifdef HAVE_SETSID
7219 tgl@sss.pgh.pa.us 3518 [ + + ]:CBC 10209 : switch (signal)
3519 : : {
3520 : 6517 : case SIGINT:
3521 : : case SIGTERM:
3522 : : case SIGQUIT:
3523 : : case SIGKILL:
3524 : : case SIGABRT:
3525 [ + + ]: 6517 : if (kill(-pid, signal) < 0)
3526 [ - + ]: 10 : elog(DEBUG3, "kill(%ld,%d) failed: %m", (long) (-pid), signal);
3527 : 6517 : break;
3528 : 3692 : default:
3529 : 3692 : break;
3530 : : }
3531 : : #endif
3532 : 10209 : }
3533 : :
3534 : : /*
3535 : : * Send a signal to the targeted children.
3536 : : */
3537 : : static bool
651 heikki.linnakangas@i 3538 : 3408 : SignalChildren(int signal, BackendTypeMask targetMask)
3539 : : {
3540 : : dlist_iter iter;
6068 3541 : 3408 : bool signaled = false;
3542 : :
651 3543 [ + - + + ]: 16692 : dlist_foreach(iter, &ActiveChildList)
3544 : : {
3545 : 13284 : PMChild *bp = dlist_container(PMChild, elem, iter.cur);
3546 : :
3547 : : /*
3548 : : * If we need to distinguish between B_BACKEND and B_WAL_SENDER, check
3549 : : * if any B_BACKEND backends have recently announced that they are
3550 : : * actually WAL senders.
3551 : : */
3552 [ + + ]: 13284 : if (btmask_contains(targetMask, B_WAL_SENDER) != btmask_contains(targetMask, B_BACKEND) &&
3553 [ + + ]: 6885 : bp->bkend_type == B_BACKEND)
3554 : : {
3555 [ + + ]: 673 : if (IsPostmasterChildWalSender(bp->child_slot))
3556 : 51 : bp->bkend_type = B_WAL_SENDER;
3557 : : }
3558 : :
3559 [ + + ]: 13284 : if (!btmask_contains(targetMask, bp->bkend_type))
3560 : 4549 : continue;
3561 : :
3562 : 8735 : signal_child(bp, signal);
6068 3563 : 8735 : signaled = true;
3564 : : }
3565 : 3408 : return signaled;
3566 : : }
3567 : :
3568 : : /*
3569 : : * Send a termination signal to children. This considers all of our children
3570 : : * processes, except syslogger.
3571 : : */
3572 : : static void
4808 alvherre@alvh.no-ip. 3573 : 374 : TerminateChildren(int signal)
3574 : : {
651 heikki.linnakangas@i 3575 : 374 : SignalChildren(signal, btmask_all_except(B_LOGGER));
3576 [ + + ]: 374 : if (StartupPMChild != NULL)
3577 : : {
1375 tgl@sss.pgh.pa.us 3578 [ - + - - : 52 : if (signal == SIGQUIT || signal == SIGKILL || signal == SIGABRT)
- - ]
4067 3579 : 52 : StartupStatus = STARTUP_SIGNALED;
3580 : : }
4808 alvherre@alvh.no-ip. 3581 : 374 : }
3582 : :
3583 : : /*
3584 : : * BackendStartup -- start backend process
3585 : : *
3586 : : * returns: STATUS_ERROR if the fork failed, STATUS_OK otherwise.
3587 : : *
3588 : : * Note: if you change this code, also consider StartAutovacuumWorker and
3589 : : * StartBackgroundWorker.
3590 : : */
3591 : : static int
898 heikki.linnakangas@i 3592 : 15615 : BackendStartup(ClientSocket *client_sock)
3593 : : {
651 3594 : 15615 : PMChild *bn = NULL;
3595 : : pid_t pid;
3596 : : BackendStartupData startup_data;
3597 : : CAC_state cac;
3598 : :
3599 : : /*
3600 : : * Capture time that Postmaster got a socket from accept (for logging
3601 : : * connection establishment and setup total duration).
3602 : : */
533 melanieplageman@gmai 3603 : 15615 : startup_data.socket_created = GetCurrentTimestamp();
3604 : :
3605 : : /*
3606 : : * Allocate and assign the child slot. Note we must do this before
3607 : : * forking, so that we can handle failures (out of memory or child-process
3608 : : * slots) cleanly.
3609 : : */
651 heikki.linnakangas@i 3610 : 15615 : cac = canAcceptConnections(B_BACKEND);
3611 [ + + ]: 15615 : if (cac == CAC_OK)
3612 : : {
3613 : : /* Can change later to B_WAL_SENDER */
3614 : 15355 : bn = AssignPostmasterChildSlot(B_BACKEND);
3615 [ + + ]: 15355 : if (!bn)
3616 : : {
3617 : : /*
3618 : : * Too many regular child processes; launch a dead-end child
3619 : : * process instead.
3620 : : */
3621 : 28 : cac = CAC_TOOMANY;
3622 : : }
3623 : : }
9076 tgl@sss.pgh.pa.us 3624 [ + + ]: 15615 : if (!bn)
3625 : : {
651 heikki.linnakangas@i 3626 : 288 : bn = AllocDeadEndChild();
3627 [ - + ]: 288 : if (!bn)
3628 : : {
651 heikki.linnakangas@i 3629 [ # # ]:UBC 0 : ereport(LOG,
3630 : : (errcode(ERRCODE_OUT_OF_MEMORY),
3631 : : errmsg("out of memory")));
3632 : 0 : return STATUS_ERROR;
3633 : : }
3634 : : }
3635 : :
3636 : : /* Pass down canAcceptConnections state */
651 heikki.linnakangas@i 3637 :CBC 15615 : startup_data.canAcceptConnections = cac;
747 3638 : 15615 : bn->rw = NULL;
3639 : :
3640 : : /* Hasn't asked to be notified about any bgworkers yet */
4747 rhaas@postgresql.org 3641 : 15615 : bn->bgworker_notify = false;
3642 : :
651 heikki.linnakangas@i 3643 : 15615 : pid = postmaster_child_launch(bn->bkend_type, bn->child_slot,
3644 : : &startup_data, sizeof(startup_data),
3645 : : client_sock);
10581 bruce@momjian.us 3646 [ - + ]: 15615 : if (pid < 0)
3647 : : {
3648 : : /* in parent, fork failed */
8999 tgl@sss.pgh.pa.us 3649 :UBC 0 : int save_errno = errno;
3650 : :
651 heikki.linnakangas@i 3651 : 0 : (void) ReleasePostmasterChildSlot(bn);
8437 tgl@sss.pgh.pa.us 3652 : 0 : errno = save_errno;
3653 [ # # ]: 0 : ereport(LOG,
3654 : : (errmsg("could not fork new process for connection: %m")));
898 heikki.linnakangas@i 3655 : 0 : report_fork_failure_to_client(client_sock, save_errno);
10222 bruce@momjian.us 3656 : 0 : return STATUS_ERROR;
3657 : : }
3658 : :
3659 : : /* in parent, successful fork */
8416 tgl@sss.pgh.pa.us 3660 [ + + ]:CBC 15615 : ereport(DEBUG2,
3661 : : (errmsg_internal("forked new %s, pid=%d socket=%d",
3662 : : GetBackendTypeDesc(bn->bkend_type),
3663 : : (int) pid, (int) client_sock->sock)));
3664 : :
3665 : : /*
3666 : : * Everything's been successful, it's safe to add this backend to our list
3667 : : * of backends.
3668 : : */
10581 bruce@momjian.us 3669 : 15615 : bn->pid = pid;
10222 3670 : 15615 : return STATUS_OK;
3671 : : }
3672 : :
3673 : : /*
3674 : : * Try to report backend fork() failure to client before we close the
3675 : : * connection. Since we do not care to risk blocking the postmaster on
3676 : : * this connection, we set the connection to non-blocking and try only once.
3677 : : *
3678 : : * This is grungy special-purpose code; we cannot use backend libpq since
3679 : : * it's not up and running.
3680 : : */
3681 : : static void
898 heikki.linnakangas@i 3682 :UBC 0 : report_fork_failure_to_client(ClientSocket *client_sock, int errnum)
3683 : : {
3684 : : char buffer[1000];
3685 : : int rc;
3686 : :
3687 : : /* Format the error message packet (always V2 protocol) */
8999 tgl@sss.pgh.pa.us 3688 : 0 : snprintf(buffer, sizeof(buffer), "E%s%s\n",
3689 : : _("could not fork new process for connection: "),
3690 : : strerror(errnum));
3691 : :
3692 : : /* Set port to non-blocking. Don't do send() if this fails */
898 heikki.linnakangas@i 3693 [ # # ]: 0 : if (!pg_set_noblock(client_sock->sock))
8999 tgl@sss.pgh.pa.us 3694 : 0 : return;
3695 : :
3696 : : /* We'll retry after EINTR, but ignore all other failures */
3697 : : do
3698 : : {
898 heikki.linnakangas@i 3699 : 0 : rc = send(client_sock->sock, buffer, strlen(buffer) + 1, 0);
7347 tgl@sss.pgh.pa.us 3700 [ # # # # ]: 0 : } while (rc < 0 && errno == EINTR);
3701 : : }
3702 : :
3703 : : /*
3704 : : * ExitPostmaster -- cleanup
3705 : : *
3706 : : * Do NOT call exit() directly --- always go through here!
3707 : : */
3708 : : static void
11006 scrappy@hub.org 3709 :CBC 993 : ExitPostmaster(int status)
3710 : : {
3711 : : #ifdef HAVE_PTHREAD_IS_THREADED_NP
3712 : :
3713 : : /*
3714 : : * There is no known cause for a postmaster to become multithreaded after
3715 : : * startup. However, we might reach here via an error exit before
3716 : : * reaching the test in PostmasterMain, so provide the same hint as there.
3717 : : * This message uses LOG level, because an unclean shutdown at this point
3718 : : * would usually not look much different from a clean shutdown.
3719 : : */
3720 : : if (pthread_is_threaded_np() != 0)
3721 : : ereport(LOG,
3722 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3723 : : errmsg("postmaster became multithreaded"),
3724 : : errhint("Set the LC_ALL environment variable to a valid locale.")));
3725 : : #endif
3726 : :
3727 : : /* should cleanup shared memory and kill all backends */
3728 : :
3729 : : /*
3730 : : * Not sure of the semantics here. When the Postmaster dies, should the
3731 : : * backends all be killed? probably not.
3732 : : *
3733 : : * MUST -- vadim 05-10-1999
3734 : : */
3735 : :
10288 bruce@momjian.us 3736 : 993 : proc_exit(status);
3737 : : }
3738 : :
3739 : : /*
3740 : : * Handle pmsignal conditions representing requests from backends,
3741 : : * and check for promote and logrotate requests from pg_ctl.
3742 : : */
3743 : : static void
1323 tmunro@postgresql.or 3744 : 4684 : process_pm_pmsignal(void)
3745 : : {
579 andres@anarazel.de 3746 : 4684 : bool request_state_update = false;
3747 : :
1323 tmunro@postgresql.or 3748 : 4684 : pending_pm_pmsignal = false;
3749 : :
3750 [ + + ]: 4684 : ereport(DEBUG2,
3751 : : (errmsg_internal("postmaster received pmsignal signal")));
3752 : :
3753 : : /*
3754 : : * RECOVERY_STARTED and BEGIN_HOT_STANDBY signals are ignored in
3755 : : * unexpected states. If the startup process quickly starts up, completes
3756 : : * recovery, exits, we might process the death of the startup process
3757 : : * first. We don't want to go back to recovery in that case.
3758 : : */
6394 heikki.linnakangas@i 3759 [ + + ]: 4684 : if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_STARTED) &&
5027 tgl@sss.pgh.pa.us 3760 [ + - + - ]: 282 : pmState == PM_STARTUP && Shutdown == NoShutdown)
3761 : : {
3762 : : /* WAL redo has started. We're out of reinitialization. */
6394 heikki.linnakangas@i 3763 : 282 : FatalError = false;
2558 tgl@sss.pgh.pa.us 3764 : 282 : AbortStartTime = 0;
512 fujii@postgresql.org 3765 : 282 : reachedConsistency = false;
3766 : :
3767 : : /*
3768 : : * Start the archiver if we're responsible for (re-)archiving received
3769 : : * files.
3770 : : */
651 heikki.linnakangas@i 3771 [ - + ]: 282 : Assert(PgArchPMChild == NULL);
4094 fujii@postgresql.org 3772 [ + + - + : 282 : if (XLogArchivingAlways())
+ + ]
651 heikki.linnakangas@i 3773 : 5 : PgArchPMChild = StartChildProcess(B_ARCHIVER);
3774 : :
3775 : : /*
3776 : : * If we aren't planning to enter hot standby mode later, treat
3777 : : * RECOVERY_STARTED as meaning we're out of startup, and report status
3778 : : * accordingly.
3779 : : */
3936 peter_e@gmx.net 3780 [ + + ]: 282 : if (!EnableHotStandby)
3781 : : {
3347 tgl@sss.pgh.pa.us 3782 : 2 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_STANDBY);
3783 : : #ifdef USE_SYSTEMD
3784 : : sd_notify(0, "READY=1");
3785 : : #endif
3786 : : }
3787 : :
594 andres@anarazel.de 3788 : 282 : UpdatePMState(PM_RECOVERY);
3789 : : }
3790 : :
512 fujii@postgresql.org 3791 [ + + ]: 4684 : if (CheckPostmasterSignal(PMSIGNAL_RECOVERY_CONSISTENT) &&
5027 tgl@sss.pgh.pa.us 3792 [ + - + - ]: 190 : pmState == PM_RECOVERY && Shutdown == NoShutdown)
3793 : : {
512 fujii@postgresql.org 3794 : 190 : reachedConsistency = true;
3795 : : }
3796 : :
3797 [ + + ]: 4684 : if (CheckPostmasterSignal(PMSIGNAL_BEGIN_HOT_STANDBY) &&
3798 [ + - + - ]: 180 : (pmState == PM_RECOVERY && Shutdown == NoShutdown))
3799 : : {
6095 simon@2ndQuadrant.co 3800 [ + - ]: 180 : ereport(LOG,
3801 : : (errmsg("database system is ready to accept read-only connections")));
3802 : :
3803 : : /* Report status */
3347 tgl@sss.pgh.pa.us 3804 : 180 : AddToDataDirLockFile(LOCK_FILE_LINE_PM_STATUS, PM_STATUS_READY);
3805 : : #ifdef USE_SYSTEMD
3806 : : sd_notify(0, "READY=1");
3807 : : #endif
3808 : :
594 andres@anarazel.de 3809 : 180 : UpdatePMState(PM_HOT_STANDBY);
1604 sfrost@snowman.net 3810 : 180 : connsAllowed = true;
3811 : :
3812 : : /* Some workers may be scheduled to start now */
4345 rhaas@postgresql.org 3813 : 180 : StartWorkerNeeded = true;
3814 : : }
3815 : :
3816 : : /* Process IO worker start requests. */
141 tmunro@postgresql.or 3817 : 4684 : if (CheckPostmasterSignal(PMSIGNAL_IO_WORKER_GROW))
3818 : : {
3819 : : /*
3820 : : * No local flag, as the state is exposed through pgaio_worker_*()
3821 : : * functions. This signal is received on potentially actionable level
3822 : : * changes, so that maybe_start_io_workers() will run.
3823 : : */
3824 : : }
3825 : :
3826 : : /* Process background worker state changes. */
2072 tgl@sss.pgh.pa.us 3827 [ + + ]: 4684 : if (CheckPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE))
3828 : : {
3829 : : /* Accept new worker requests only if not stopping. */
3830 : 1849 : BackgroundWorkerStateChange(pmState < PM_STOP_BACKENDS);
3831 : 1849 : StartWorkerNeeded = true;
3832 : : }
3833 : :
3834 : : /* Tell syslogger to rotate logfile if requested */
651 heikki.linnakangas@i 3835 [ + + ]: 4684 : if (SysLoggerPMChild != NULL)
3836 : : {
2917 akorotkov@postgresql 3837 [ + + ]: 2 : if (CheckLogrotateSignal())
3838 : : {
651 heikki.linnakangas@i 3839 : 1 : signal_child(SysLoggerPMChild, SIGUSR1);
2917 akorotkov@postgresql 3840 : 1 : RemoveLogrotateSignalFiles();
3841 : : }
3842 [ - + ]: 1 : else if (CheckPostmasterSignal(PMSIGNAL_ROTATE_LOGFILE))
3843 : : {
651 heikki.linnakangas@i 3844 :UBC 0 : signal_child(SysLoggerPMChild, SIGUSR1);
3845 : : }
3846 : : }
3847 : :
5027 tgl@sss.pgh.pa.us 3848 [ - + ]:CBC 4684 : if (CheckPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER) &&
2204 tgl@sss.pgh.pa.us 3849 [ # # # # ]:UBC 0 : Shutdown <= SmartShutdown && pmState < PM_STOP_BACKENDS)
3850 : : {
3851 : : /*
3852 : : * Start one iteration of the autovacuum daemon, even if autovacuuming
3853 : : * is nominally not enabled. This is so we can have an active defense
3854 : : * against transaction ID wraparound. We set a flag for the main loop
3855 : : * to do it rather than trying to do it here --- this is because the
3856 : : * autovac process itself may send the signal, and we want to handle
3857 : : * that by launching another iteration as soon as the current one
3858 : : * completes.
3859 : : */
7133 alvherre@alvh.no-ip. 3860 : 0 : start_autovac_launcher = true;
3861 : : }
3862 : :
5027 tgl@sss.pgh.pa.us 3863 [ + + ]:CBC 4684 : if (CheckPostmasterSignal(PMSIGNAL_START_AUTOVAC_WORKER) &&
2204 3864 [ + - + - ]: 156 : Shutdown <= SmartShutdown && pmState < PM_STOP_BACKENDS)
3865 : : {
3866 : : /* The autovacuum launcher wants us to start a worker process. */
7133 alvherre@alvh.no-ip. 3867 : 156 : StartAutovacuumWorker();
3868 : : }
3869 : :
3349 tgl@sss.pgh.pa.us 3870 [ + + ]: 4684 : if (CheckPostmasterSignal(PMSIGNAL_START_WALRECEIVER))
3871 : : {
3872 : : /* Startup Process wants us to start the walreceiver process. */
3873 : 295 : WalReceiverRequested = true;
3874 : : }
3875 : :
579 andres@anarazel.de 3876 [ + + ]: 4684 : if (CheckPostmasterSignal(PMSIGNAL_XLOG_IS_SHUTDOWN))
3877 : : {
3878 : : /* Checkpointer completed the shutdown checkpoint */
3879 [ + - ]: 622 : if (pmState == PM_WAIT_XLOG_SHUTDOWN)
3880 : : {
3881 : : /*
3882 : : * If we have an archiver subprocess, tell it to do a last archive
3883 : : * cycle and quit. Likewise, if we have walsender processes, tell
3884 : : * them to send any remaining WAL and quit.
3885 : : */
3886 [ - + ]: 622 : Assert(Shutdown > NoShutdown);
3887 : :
3888 : : /* Waken archiver for the last time */
3889 [ + + ]: 622 : if (PgArchPMChild != NULL)
3890 : 19 : signal_child(PgArchPMChild, SIGUSR2);
3891 : :
3892 : : /*
3893 : : * Waken walsenders for the last time. No regular backends should
3894 : : * be around anymore.
3895 : : */
3896 : 622 : SignalChildren(SIGUSR2, btmask(B_WAL_SENDER));
3897 : :
3898 : 622 : UpdatePMState(PM_WAIT_XLOG_ARCHIVAL);
3899 : : }
579 andres@anarazel.de 3900 [ # # # # ]:UBC 0 : else if (!FatalError && Shutdown != ImmediateShutdown)
3901 : : {
3902 : : /*
3903 : : * Checkpointer only ought to perform the shutdown checkpoint
3904 : : * during shutdown. If somehow checkpointer did so in another
3905 : : * situation, we have no choice but to crash-restart.
3906 : : *
3907 : : * It's possible however that we get PMSIGNAL_XLOG_IS_SHUTDOWN
3908 : : * outside of PM_WAIT_XLOG_SHUTDOWN if an orderly shutdown was
3909 : : * "interrupted" by a crash or an immediate shutdown.
3910 : : */
3911 [ # # ]: 0 : ereport(LOG,
3912 : : (errmsg("WAL was shut down unexpectedly")));
3913 : :
3914 : : /*
3915 : : * Doesn't seem likely to help to take send_abort_for_crash into
3916 : : * account here.
3917 : : */
3918 : 0 : HandleFatalError(PMQUIT_FOR_CRASH, false);
3919 : : }
3920 : :
3921 : : /*
3922 : : * Need to run PostmasterStateMachine() to check if we already can go
3923 : : * to the next state.
3924 : : */
579 andres@anarazel.de 3925 :CBC 622 : request_state_update = true;
3926 : : }
3927 : :
3928 : : /*
3929 : : * Try to advance postmaster's state machine, if a child requests it.
3930 : : */
3931 [ + + ]: 4684 : if (CheckPostmasterSignal(PMSIGNAL_ADVANCE_STATE_MACHINE))
3932 : : {
3933 : 1386 : request_state_update = true;
3934 : : }
3935 : :
3936 : : /*
3937 : : * Be careful about the order of this action relative to this function's
3938 : : * other actions. Generally, this should be after other actions, in case
3939 : : * they have effects PostmasterStateMachine would need to know about.
3940 : : * However, we should do it before the CheckPromoteSignal step, which
3941 : : * cannot have any (immediate) effect on the state machine, but does
3942 : : * depend on what state we're in now.
3943 : : */
3944 [ + + ]: 4684 : if (request_state_update)
3945 : : {
5625 rhaas@postgresql.org 3946 : 2008 : PostmasterStateMachine();
3947 : : }
3948 : :
651 heikki.linnakangas@i 3949 [ + + ]: 4684 : if (StartupPMChild != NULL &&
5672 rhaas@postgresql.org 3950 [ + + + + ]: 719 : (pmState == PM_STARTUP || pmState == PM_RECOVERY ||
2204 tgl@sss.pgh.pa.us 3951 [ + + + + ]: 1227 : pmState == PM_HOT_STANDBY) &&
2682 3952 : 717 : CheckPromoteSignal())
3953 : : {
3954 : : /*
3955 : : * Tell startup process to finish recovery.
3956 : : *
3957 : : * Leave the promote signal file in place and let the Startup process
3958 : : * do the unlink.
3959 : : */
651 heikki.linnakangas@i 3960 : 55 : signal_child(StartupPMChild, SIGUSR2);
3961 : : }
9298 tgl@sss.pgh.pa.us 3962 : 4684 : }
3963 : :
3964 : : /*
3965 : : * Dummy signal handler
3966 : : *
3967 : : * We use this for signals that we don't actually use in the postmaster,
3968 : : * but we do use in backends. If we were to PG_SIG_IGN such signals in the
3969 : : * postmaster, then a newly started backend might drop a signal that arrives
3970 : : * before it's able to reconfigure its signal processing. (See notes in
3971 : : * tcop/postgres.c.)
3972 : : */
3973 : : static void
9062 tgl@sss.pgh.pa.us 3974 :UBC 0 : dummy_handler(SIGNAL_ARGS)
3975 : : {
3976 : 0 : }
3977 : :
3978 : : /*
3979 : : * Count up number of child processes of specified types.
3980 : : */
3981 : : static int
651 heikki.linnakangas@i 3982 :CBC 7125 : CountChildren(BackendTypeMask targetMask)
3983 : : {
3984 : : dlist_iter iter;
10071 tgl@sss.pgh.pa.us 3985 : 7125 : int cnt = 0;
3986 : :
651 heikki.linnakangas@i 3987 [ + - + + ]: 36691 : dlist_foreach(iter, &ActiveChildList)
3988 : : {
3989 : 29566 : PMChild *bp = dlist_container(PMChild, elem, iter.cur);
3990 : :
3991 : : /*
3992 : : * If we need to distinguish between B_BACKEND and B_WAL_SENDER, check
3993 : : * if any B_BACKEND backends have recently announced that they are
3994 : : * actually WAL senders.
3995 : : */
3996 [ + + ]: 29566 : if (btmask_contains(targetMask, B_WAL_SENDER) != btmask_contains(targetMask, B_BACKEND) &&
3997 [ + + ]: 19955 : bp->bkend_type == B_BACKEND)
3998 : : {
3999 [ - + ]: 1099 : if (IsPostmasterChildWalSender(bp->child_slot))
651 heikki.linnakangas@i 4000 :UBC 0 : bp->bkend_type = B_WAL_SENDER;
4001 : : }
4002 : :
651 heikki.linnakangas@i 4003 [ + + ]:CBC 29566 : if (!btmask_contains(targetMask, bp->bkend_type))
4004 : 13473 : continue;
4005 : :
4006 [ - + ]: 16093 : ereport(DEBUG4,
4007 : : (errmsg_internal("%s process %d is still running",
4008 : : GetBackendTypeDesc(bp->bkend_type), (int) bp->pid)));
4009 : :
6068 4010 : 16093 : cnt++;
4011 : : }
10071 tgl@sss.pgh.pa.us 4012 : 7125 : return cnt;
4013 : : }
4014 : :
4015 : :
4016 : : /*
4017 : : * StartChildProcess -- start an auxiliary process for the postmaster
4018 : : *
4019 : : * "type" determines what kind of child will be started. All child types
4020 : : * initially go to AuxiliaryProcessMain, which will handle common setup.
4021 : : *
4022 : : * Return value of StartChildProcess is subprocess' PMChild entry, or NULL on
4023 : : * failure.
4024 : : */
4025 : : static PMChild *
906 heikki.linnakangas@i 4026 : 7144 : StartChildProcess(BackendType type)
4027 : : {
4028 : : PMChild *pmchild;
4029 : : pid_t pid;
4030 : :
651 4031 : 7144 : pmchild = AssignPostmasterChildSlot(type);
4032 [ - + ]: 7144 : if (!pmchild)
4033 : : {
651 heikki.linnakangas@i 4034 [ # # ]:UBC 0 : if (type == B_AUTOVAC_WORKER)
4035 [ # # ]: 0 : ereport(LOG,
4036 : : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
4037 : : errmsg("no slot available for new autovacuum worker process")));
4038 : : else
4039 : : {
4040 : : /* shouldn't happen because we allocate enough slots */
4041 [ # # ]: 0 : elog(LOG, "no postmaster child slot available for aux process");
4042 : : }
4043 : 0 : return NULL;
4044 : : }
4045 : :
651 heikki.linnakangas@i 4046 :CBC 7144 : pid = postmaster_child_launch(type, pmchild->child_slot, NULL, 0, NULL);
9822 vadim4o@yahoo.com 4047 [ - + ]: 7144 : if (pid < 0)
4048 : : {
4049 : : /* in parent, fork failed */
651 heikki.linnakangas@i 4050 :UBC 0 : ReleasePostmasterChildSlot(pmchild);
892 4051 [ # # ]: 0 : ereport(LOG,
4052 : : (errmsg("could not fork \"%s\" process: %m", PostmasterChildName(type))));
4053 : :
4054 : : /*
4055 : : * fork failure is fatal during startup, but there's no need to choke
4056 : : * immediately if starting other child types fails.
4057 : : */
906 4058 [ # # ]: 0 : if (type == B_STARTUP)
8125 tgl@sss.pgh.pa.us 4059 : 0 : ExitPostmaster(1);
651 heikki.linnakangas@i 4060 : 0 : return NULL;
4061 : : }
4062 : :
4063 : : /* in parent, successful fork */
651 heikki.linnakangas@i 4064 :CBC 7144 : pmchild->pid = pid;
4065 : 7144 : return pmchild;
4066 : : }
4067 : :
4068 : : /*
4069 : : * StartSysLogger -- start the syslogger process
4070 : : */
4071 : : void
4072 : 1 : StartSysLogger(void)
4073 : : {
4074 [ - + ]: 1 : Assert(SysLoggerPMChild == NULL);
4075 : :
4076 : 1 : SysLoggerPMChild = AssignPostmasterChildSlot(B_LOGGER);
4077 [ - + ]: 1 : if (!SysLoggerPMChild)
651 heikki.linnakangas@i 4078 [ # # ]:UBC 0 : elog(PANIC, "no postmaster child slot available for syslogger");
651 heikki.linnakangas@i 4079 :CBC 1 : SysLoggerPMChild->pid = SysLogger_Start(SysLoggerPMChild->child_slot);
4080 [ - + ]: 1 : if (SysLoggerPMChild->pid == 0)
4081 : : {
651 heikki.linnakangas@i 4082 :UBC 0 : ReleasePostmasterChildSlot(SysLoggerPMChild);
4083 : 0 : SysLoggerPMChild = NULL;
4084 : : }
9822 vadim4o@yahoo.com 4085 :CBC 1 : }
4086 : :
4087 : : /*
4088 : : * StartAutovacuumWorker
4089 : : * Start an autovac worker process.
4090 : : *
4091 : : * This function is here because it enters the resulting PID into the
4092 : : * postmaster's private backends list.
4093 : : *
4094 : : * NB -- this code very roughly matches BackendStartup.
4095 : : */
4096 : : static void
7133 alvherre@alvh.no-ip. 4097 : 156 : StartAutovacuumWorker(void)
4098 : : {
4099 : : PMChild *bn;
4100 : :
4101 : : /*
4102 : : * If not in condition to run a process, don't try, but handle it like a
4103 : : * fork failure. This does not normally happen, since the signal is only
4104 : : * supposed to be sent by autovacuum launcher when it's OK to do it, but
4105 : : * we have to check to avoid race-condition problems during DB state
4106 : : * changes.
4107 : : */
651 heikki.linnakangas@i 4108 [ + - ]: 156 : if (canAcceptConnections(B_AUTOVAC_WORKER) == CAC_OK)
4109 : : {
4110 : 156 : bn = StartChildProcess(B_AUTOVAC_WORKER);
6958 tgl@sss.pgh.pa.us 4111 [ + - ]: 156 : if (bn)
4112 : : {
3600 heikki.linnakangas@i 4113 : 156 : bn->bgworker_notify = false;
747 4114 : 156 : bn->rw = NULL;
651 4115 : 156 : return;
4116 : : }
4117 : : else
4118 : : {
4119 : : /*
4120 : : * fork failed, fall through to report -- actual error message was
4121 : : * logged by StartChildProcess
4122 : : */
4123 : : }
4124 : : }
4125 : :
4126 : : /*
4127 : : * Report the failure to the launcher, if it's running. (If it's not, we
4128 : : * might not even be connected to shared memory, so don't try to call
4129 : : * AutoVacWorkerFailed.) Note that we also need to signal it so that it
4130 : : * responds to the condition, but we don't do that here, instead waiting
4131 : : * for ServerLoop to do it. This way we avoid a ping-pong signaling in
4132 : : * quick succession between the autovac launcher and postmaster in case
4133 : : * things get ugly.
4134 : : */
651 heikki.linnakangas@i 4135 [ # # ]:UBC 0 : if (AutoVacLauncherPMChild != NULL)
4136 : : {
6958 tgl@sss.pgh.pa.us 4137 : 0 : AutoVacWorkerFailed();
6212 alvherre@alvh.no-ip. 4138 : 0 : avlauncher_needs_signal = true;
4139 : : }
4140 : : }
4141 : :
4142 : :
4143 : : /*
4144 : : * Create the opts file
4145 : : */
4146 : : static bool
8141 bruce@momjian.us 4147 :CBC 991 : CreateOptsFile(int argc, char *argv[], char *fullprogname)
4148 : : {
4149 : : FILE *fp;
4150 : : int i;
4151 : :
4152 : : #define OPTS_FILE "postmaster.opts"
4153 : :
7724 tgl@sss.pgh.pa.us 4154 [ - + ]: 991 : if ((fp = fopen(OPTS_FILE, "w")) == NULL)
4155 : : {
2092 peter@eisentraut.org 4156 [ # # ]:UBC 0 : ereport(LOG,
4157 : : (errcode_for_file_access(),
4158 : : errmsg("could not create file \"%s\": %m", OPTS_FILE)));
9545 peter_e@gmx.net 4159 : 0 : return false;
4160 : : }
4161 : :
9545 peter_e@gmx.net 4162 :CBC 991 : fprintf(fp, "%s", fullprogname);
4163 [ + + ]: 5197 : for (i = 1; i < argc; i++)
6636 bruce@momjian.us 4164 : 4206 : fprintf(fp, " \"%s\"", argv[i]);
9545 peter_e@gmx.net 4165 : 991 : fputs("\n", fp);
4166 : :
8249 tgl@sss.pgh.pa.us 4167 [ - + ]: 991 : if (fclose(fp))
4168 : : {
2092 peter@eisentraut.org 4169 [ # # ]:UBC 0 : ereport(LOG,
4170 : : (errcode_for_file_access(),
4171 : : errmsg("could not write file \"%s\": %m", OPTS_FILE)));
9545 peter_e@gmx.net 4172 : 0 : return false;
4173 : : }
4174 : :
9545 peter_e@gmx.net 4175 :CBC 991 : return true;
4176 : : }
4177 : :
4178 : :
4179 : : /*
4180 : : * Start a new bgworker.
4181 : : * Starting time conditions must have been checked already.
4182 : : *
4183 : : * Returns true on success, false on failure.
4184 : : * In either case, update the RegisteredBgWorker's state appropriately.
4185 : : *
4186 : : * NB -- this code very roughly matches BackendStartup.
4187 : : */
4188 : : static bool
651 heikki.linnakangas@i 4189 : 3540 : StartBackgroundWorker(RegisteredBgWorker *rw)
4190 : : {
4191 : : PMChild *bn;
4192 : : pid_t worker_pid;
4193 : :
3412 tgl@sss.pgh.pa.us 4194 [ - + ]: 3540 : Assert(rw->rw_pid == 0);
4195 : :
4196 : : /*
4197 : : * Allocate and assign the child slot. Note we must do this before
4198 : : * forking, so that we can handle failures (out of memory or child-process
4199 : : * slots) cleanly.
4200 : : *
4201 : : * Treat failure as though the worker had crashed. That way, the
4202 : : * postmaster will wait a bit before attempting to start it again; if we
4203 : : * tried again right away, most likely we'd find ourselves hitting the
4204 : : * same resource-exhaustion condition.
4205 : : */
651 heikki.linnakangas@i 4206 : 3540 : bn = AssignPostmasterChildSlot(B_BG_WORKER);
756 4207 [ - + ]: 3540 : if (bn == NULL)
4208 : : {
651 heikki.linnakangas@i 4209 [ # # ]:UBC 0 : ereport(LOG,
4210 : : (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
4211 : : errmsg("no slot available for new background worker process")));
3412 tgl@sss.pgh.pa.us 4212 : 0 : rw->rw_crashed_at = GetCurrentTimestamp();
4213 : 0 : return false;
4214 : : }
747 heikki.linnakangas@i 4215 :CBC 3540 : bn->rw = rw;
651 4216 : 3540 : bn->bkend_type = B_BG_WORKER;
4217 : 3540 : bn->bgworker_notify = false;
4218 : :
4080 rhaas@postgresql.org 4219 [ + + ]: 3540 : ereport(DEBUG1,
4220 : : (errmsg_internal("starting background worker process \"%s\"",
4221 : : rw->rw_worker.bgw_name)));
4222 : :
651 heikki.linnakangas@i 4223 : 3540 : worker_pid = postmaster_child_launch(B_BG_WORKER, bn->child_slot,
552 peter@eisentraut.org 4224 : 3540 : &rw->rw_worker, sizeof(BackgroundWorker), NULL);
892 heikki.linnakangas@i 4225 [ - + ]: 3540 : if (worker_pid == -1)
4226 : : {
4227 : : /* in postmaster, fork failed ... */
892 heikki.linnakangas@i 4228 [ # # ]:UBC 0 : ereport(LOG,
4229 : : (errmsg("could not fork background worker process: %m")));
4230 : : /* undo what AssignPostmasterChildSlot did */
651 4231 : 0 : ReleasePostmasterChildSlot(bn);
4232 : :
4233 : : /* mark entry as crashed, so we'll try again later */
892 4234 : 0 : rw->rw_crashed_at = GetCurrentTimestamp();
4235 : 0 : return false;
4236 : : }
4237 : :
4238 : : /* in postmaster, fork successful ... */
892 heikki.linnakangas@i 4239 :CBC 3540 : rw->rw_pid = worker_pid;
747 4240 : 3540 : bn->pid = rw->rw_pid;
892 4241 : 3540 : ReportBackgroundWorkerPID(rw);
4242 : 3540 : return true;
4243 : : }
4244 : :
4245 : : /*
4246 : : * Does the current postmaster state require starting a worker with the
4247 : : * specified start_time?
4248 : : */
4249 : : static bool
5012 alvherre@alvh.no-ip. 4250 : 4653 : bgworker_should_start_now(BgWorkerStartTime start_time)
4251 : : {
4252 [ - + + + : 4653 : switch (pmState)
- ]
4253 : : {
5012 alvherre@alvh.no-ip. 4254 :UBC 0 : case PM_NO_CHILDREN:
4255 : : case PM_WAIT_CHECKPOINTER:
4256 : : case PM_WAIT_DEAD_END:
4257 : : case PM_WAIT_XLOG_ARCHIVAL:
4258 : : case PM_WAIT_XLOG_SHUTDOWN:
4259 : : case PM_WAIT_IO_WORKERS:
4260 : : case PM_WAIT_BACKENDS:
4261 : : case PM_STOP_BACKENDS:
4262 : 0 : break;
4263 : :
5012 alvherre@alvh.no-ip. 4264 :CBC 3540 : case PM_RUN:
4265 [ + + ]: 3540 : if (start_time == BgWorkerStart_RecoveryFinished)
4266 : 1519 : return true;
4267 : : pg_fallthrough;
4268 : :
4269 : : case PM_HOT_STANDBY:
4270 [ + + ]: 2197 : if (start_time == BgWorkerStart_ConsistentState)
4271 : 2021 : return true;
4272 : : pg_fallthrough;
4273 : :
4274 : : case PM_RECOVERY:
4275 : : case PM_STARTUP:
4276 : : case PM_INIT:
4277 [ - + ]: 1113 : if (start_time == BgWorkerStart_PostmasterStart)
5012 alvherre@alvh.no-ip. 4278 :UBC 0 : return true;
4279 : : }
4280 : :
5012 alvherre@alvh.no-ip. 4281 :CBC 1113 : return false;
4282 : : }
4283 : :
4284 : : /*
4285 : : * If the time is right, start background worker(s).
4286 : : *
4287 : : * As a side effect, the bgworker control variables are set or reset
4288 : : * depending on whether more workers may need to be started.
4289 : : *
4290 : : * We limit the number of workers started per call, to avoid consuming the
4291 : : * postmaster's attention for too long when many such requests are pending.
4292 : : * As long as StartWorkerNeeded is true, ServerLoop will not block and will
4293 : : * call this function again after dealing with any other issues.
4294 : : */
4295 : : static void
3410 tgl@sss.pgh.pa.us 4296 : 9227 : maybe_start_bgworkers(void)
4297 : : {
4298 : : #define MAX_BGWORKERS_TO_LAUNCH 100
4299 : 9227 : int num_launched = 0;
5012 alvherre@alvh.no-ip. 4300 : 9227 : TimestampTz now = 0;
4301 : : dlist_mutable_iter iter;
4302 : :
4303 : : /*
4304 : : * During crash recovery, we have no need to be called until the state
4305 : : * transition out of recovery.
4306 : : */
4307 [ - + ]: 9227 : if (FatalError)
4308 : : {
5012 alvherre@alvh.no-ip. 4309 :UBC 0 : StartWorkerNeeded = false;
4310 : 0 : HaveCrashedWorker = false;
3412 tgl@sss.pgh.pa.us 4311 : 0 : return;
4312 : : }
4313 : :
4314 : : /* Don't need to be called again unless we find a reason for it below */
3412 tgl@sss.pgh.pa.us 4315 :CBC 9227 : StartWorkerNeeded = false;
5012 alvherre@alvh.no-ip. 4316 : 9227 : HaveCrashedWorker = false;
4317 : :
748 heikki.linnakangas@i 4318 [ + - + + ]: 25154 : dlist_foreach_modify(iter, &BackgroundWorkerList)
4319 : : {
4320 : : RegisteredBgWorker *rw;
4321 : :
4322 : 15927 : rw = dlist_container(RegisteredBgWorker, rw_lnode, iter.cur);
4323 : :
4324 : : /* ignore if already running */
5012 alvherre@alvh.no-ip. 4325 [ + + ]: 15927 : if (rw->rw_pid != 0)
4326 : 8406 : continue;
4327 : :
4328 : : /* if marked for death, clean up and remove from list */
4696 rhaas@postgresql.org 4329 [ - + ]: 7521 : if (rw->rw_terminate)
4330 : : {
748 heikki.linnakangas@i 4331 :UBC 0 : ForgetBackgroundWorker(rw);
4696 rhaas@postgresql.org 4332 : 0 : continue;
4333 : : }
4334 : :
4335 : : /*
4336 : : * If this worker has crashed previously, maybe it needs to be
4337 : : * restarted (unless on registration it specified it doesn't want to
4338 : : * be restarted at all). Check how long ago did a crash last happen.
4339 : : * If the last crash is too recent, don't start it right away; let it
4340 : : * be restarted once enough time has passed.
4341 : : */
5012 alvherre@alvh.no-ip. 4342 [ + + ]:CBC 7521 : if (rw->rw_crashed_at != 0)
4343 : : {
4344 [ - + ]: 2868 : if (rw->rw_worker.bgw_restart_time == BGW_NEVER_RESTART)
4790 rhaas@postgresql.org 4345 :UBC 0 : {
4346 : : int notify_pid;
4347 : :
3186 4348 : 0 : notify_pid = rw->rw_worker.bgw_notify_pid;
4349 : :
748 heikki.linnakangas@i 4350 : 0 : ForgetBackgroundWorker(rw);
4351 : :
4352 : : /* Report worker is gone now. */
3186 rhaas@postgresql.org 4353 [ # # ]: 0 : if (notify_pid != 0)
4354 : 0 : kill(notify_pid, SIGUSR1);
4355 : :
5012 alvherre@alvh.no-ip. 4356 : 0 : continue;
4357 : : }
4358 : :
4359 : : /* read system time only when needed */
5012 alvherre@alvh.no-ip. 4360 [ + - ]:CBC 2868 : if (now == 0)
4361 : 2868 : now = GetCurrentTimestamp();
4362 : :
4363 [ + - ]: 2868 : if (!TimestampDifferenceExceeds(rw->rw_crashed_at, now,
3354 tgl@sss.pgh.pa.us 4364 : 2868 : rw->rw_worker.bgw_restart_time * 1000))
4365 : : {
4366 : : /* Set flag to remember that we have workers to start later */
5012 alvherre@alvh.no-ip. 4367 : 2868 : HaveCrashedWorker = true;
4368 : 2868 : continue;
4369 : : }
4370 : : }
4371 : :
4372 [ + + ]: 4653 : if (bgworker_should_start_now(rw->rw_worker.bgw_start_time))
4373 : : {
4374 : : /* reset crash time before trying to start worker */
4375 : 3540 : rw->rw_crashed_at = 0;
4376 : :
4377 : : /*
4378 : : * Try to start the worker.
4379 : : *
4380 : : * On failure, give up processing workers for now, but set
4381 : : * StartWorkerNeeded so we'll come back here on the next iteration
4382 : : * of ServerLoop to try again. (We don't want to wait, because
4383 : : * there might be additional ready-to-run workers.) We could set
4384 : : * HaveCrashedWorker as well, since this worker is now marked
4385 : : * crashed, but there's no need because the next run of this
4386 : : * function will do that.
4387 : : */
651 heikki.linnakangas@i 4388 [ - + ]: 3540 : if (!StartBackgroundWorker(rw))
4389 : : {
3412 tgl@sss.pgh.pa.us 4390 :UBC 0 : StartWorkerNeeded = true;
4013 rhaas@postgresql.org 4391 : 0 : return;
4392 : : }
4393 : :
4394 : : /*
4395 : : * If we've launched as many workers as allowed, quit, but have
4396 : : * ServerLoop call us again to look for additional ready-to-run
4397 : : * workers. There might not be any, but we'll find out the next
4398 : : * time we run.
4399 : : */
3410 tgl@sss.pgh.pa.us 4400 [ - + ]:CBC 3540 : if (++num_launched >= MAX_BGWORKERS_TO_LAUNCH)
4401 : : {
3410 tgl@sss.pgh.pa.us 4402 :UBC 0 : StartWorkerNeeded = true;
4403 : 0 : return;
4404 : : }
4405 : : }
4406 : : }
4407 : : }
4408 : :
4409 : : static bool
527 andres@anarazel.de 4410 :CBC 21350 : maybe_reap_io_worker(int pid)
4411 : : {
411 tmunro@postgresql.or 4412 [ + + ]: 640835 : for (int i = 0; i < MAX_IO_WORKERS; ++i)
4413 : : {
4414 [ + + ]: 621524 : if (io_worker_children[i] &&
4415 [ + + ]: 41783 : io_worker_children[i]->pid == pid)
4416 : : {
4417 : 2039 : ReleasePostmasterChildSlot(io_worker_children[i]);
4418 : :
527 andres@anarazel.de 4419 : 2039 : --io_worker_count;
411 tmunro@postgresql.or 4420 : 2039 : io_worker_children[i] = NULL;
527 andres@anarazel.de 4421 : 2039 : return true;
4422 : : }
4423 : : }
4424 : 19311 : return false;
4425 : : }
4426 : :
4427 : : /*
4428 : : * Returns the next time at which maybe_start_io_workers() would start one or
4429 : : * more I/O workers. Any time in the past means ASAP, and 0 means no worker
4430 : : * is currently scheduled.
4431 : : *
4432 : : * This is called by DetermineSleepTime() and also maybe_start_io_workers()
4433 : : * itself, to make sure that they agree.
4434 : : */
4435 : : static TimestampTz
141 tmunro@postgresql.or 4436 : 94036 : maybe_start_io_workers_scheduled_at(void)
4437 : : {
527 andres@anarazel.de 4438 [ + + ]: 94036 : if (!pgaio_workers_enabled())
141 tmunro@postgresql.or 4439 : 232 : return 0;
4440 : :
4441 : : /*
4442 : : * If we're in final shutting down state, then we're just waiting for all
4443 : : * processes to exit.
4444 : : */
527 andres@anarazel.de 4445 [ + + ]: 93804 : if (pmState >= PM_WAIT_IO_WORKERS)
141 tmunro@postgresql.or 4446 : 5148 : return 0;
4447 : :
4448 : : /* Don't start new workers during an immediate shutdown either. */
527 andres@anarazel.de 4449 [ + + ]: 88656 : if (Shutdown >= ImmediateShutdown)
141 tmunro@postgresql.or 4450 : 2213 : return 0;
4451 : :
4452 : : /*
4453 : : * Don't start new workers if we're in the shutdown phase of a crash
4454 : : * restart. But we *do* need to start if we're already starting up again.
4455 : : */
527 andres@anarazel.de 4456 [ + + + + ]: 86443 : if (FatalError && pmState >= PM_STOP_BACKENDS)
141 tmunro@postgresql.or 4457 : 156 : return 0;
4458 : :
4459 : : /*
4460 : : * Don't start a worker if we're at or above the maximum. (Excess workers
4461 : : * exit when the GUC is lowered, but the count can be temporarily too high
4462 : : * until they are reaped.)
4463 : : */
4464 [ + + ]: 86287 : if (io_worker_count >= io_max_workers)
4465 : 37 : return 0;
4466 : :
4467 : : /* If we're under the minimum, start a worker as soon as possible. */
4468 [ + + ]: 86250 : if (io_worker_count < io_min_workers)
4469 : 2022 : return TIMESTAMP_MINUS_INFINITY; /* start worker ASAP */
4470 : :
4471 : : /* Only proceed if a "grow" signal has been received from a worker. */
4472 [ + + ]: 84228 : if (!pgaio_worker_pm_test_grow_signal_sent())
4473 : 84106 : return 0;
4474 : :
4475 : : /*
4476 : : * maybe_start_io_workers() should start a new I/O worker after this time,
4477 : : * or as soon as possible if is already in the past.
4478 : : */
4479 : 122 : return io_worker_launch_next_time;
4480 : : }
4481 : :
4482 : : /*
4483 : : * Start I/O workers if required. Used at startup, to respond to change of
4484 : : * the io_min_workers GUC, when asked to start a new one due to submission
4485 : : * queue backlog, and after workers terminate in response to errors (by
4486 : : * starting "replacement" workers).
4487 : : */
4488 : : static void
4489 : 47768 : maybe_start_io_workers(void)
4490 : : {
4491 : : TimestampTz scheduled_at;
4492 : :
4493 [ + + ]: 49807 : while ((scheduled_at = maybe_start_io_workers_scheduled_at()) != 0)
4494 : : {
4495 : 2110 : TimestampTz now = GetCurrentTimestamp();
4496 : : PMChild *child;
4497 : : int i;
4498 : :
4499 [ - + ]: 2110 : Assert(pmState < PM_WAIT_IO_WORKERS);
4500 : :
4501 : : /* Still waiting for the scheduled time? */
4502 [ + + ]: 2110 : if (scheduled_at > now)
4503 : 34 : break;
4504 : :
4505 : : /*
4506 : : * Compute next launch time relative to the previous value, so that
4507 : : * time spent on the postmaster's other duties don't result in an
4508 : : * inaccurate launch interval.
4509 : : */
4510 : 2076 : io_worker_launch_next_time =
4511 : 2076 : TimestampTzPlusMilliseconds(io_worker_launch_next_time,
4512 : : io_worker_launch_interval);
4513 : :
4514 : : /*
4515 : : * If that's already in the past, the interval is either impossibly
4516 : : * short or we received no requests for new workers for a period.
4517 : : * Compute a new future time relative to now instead.
4518 : : */
4519 [ + + ]: 2076 : if (io_worker_launch_next_time <= now)
4520 : 1062 : io_worker_launch_next_time =
4521 : 1062 : TimestampTzPlusMilliseconds(now, io_worker_launch_interval);
4522 : :
4523 : : /*
4524 : : * Check if a grow signal has been received, but the grow request has
4525 : : * been canceled since then because work ran out. We've still
4526 : : * advanced the next launch time, to suppress repeat signals from
4527 : : * workers until then.
4528 : : */
4529 [ + + + + ]: 2076 : if (io_worker_count >= io_min_workers && !pgaio_worker_pm_test_grow())
4530 : : {
4531 : 37 : pgaio_worker_pm_clear_grow_signal_sent();
4532 : 37 : break;
4533 : : }
4534 : :
4535 : : /* find unused entry in io_worker_children array */
411 4536 [ + - ]: 3572 : for (i = 0; i < MAX_IO_WORKERS; ++i)
4537 : : {
4538 [ + + ]: 3572 : if (io_worker_children[i] == NULL)
527 andres@anarazel.de 4539 : 2039 : break;
4540 : : }
411 tmunro@postgresql.or 4541 [ - + ]: 2039 : if (i == MAX_IO_WORKERS)
411 tmunro@postgresql.or 4542 [ # # ]:UBC 0 : elog(ERROR, "could not find a free IO worker slot");
4543 : :
4544 : : /* Try to launch one. */
527 andres@anarazel.de 4545 :CBC 2039 : child = StartChildProcess(B_IO_WORKER);
4546 [ + - ]: 2039 : if (child != NULL)
4547 : : {
411 tmunro@postgresql.or 4548 : 2039 : io_worker_children[i] = child;
527 andres@anarazel.de 4549 : 2039 : ++io_worker_count;
4550 : : }
4551 : : else
4552 : : {
4553 : : /*
4554 : : * Fork failure: we'll try again after the launch interval
4555 : : * expires, or be called again without delay if we don't yet have
4556 : : * io_min_workers. Don't loop here though, the postmaster has
4557 : : * other duties.
4558 : : */
141 tmunro@postgresql.or 4559 :UBC 0 : break;
4560 : : }
4561 : : }
4562 : :
4563 : : /*
4564 : : * Workers decide when to shut down by themselves, according to the
4565 : : * io_max_workers and io_worker_idle_timeout GUCs.
4566 : : */
527 andres@anarazel.de 4567 :CBC 47768 : }
4568 : :
4569 : :
4570 : : /*
4571 : : * When a backend asks to be notified about worker state changes, we
4572 : : * set a flag in its backend entry. The background worker machinery needs
4573 : : * to know when such backends exit.
4574 : : */
4575 : : bool
4747 rhaas@postgresql.org 4576 : 2728 : PostmasterMarkPIDForWorkerNotify(int pid)
4577 : : {
4578 : : dlist_iter iter;
4579 : : PMChild *bp;
4580 : :
651 heikki.linnakangas@i 4581 [ + - + - ]: 7191 : dlist_foreach(iter, &ActiveChildList)
4582 : : {
4583 : 7191 : bp = dlist_container(PMChild, elem, iter.cur);
4747 rhaas@postgresql.org 4584 [ + + ]: 7191 : if (bp->pid == pid)
4585 : : {
4586 : 2728 : bp->bgworker_notify = true;
4587 : 2728 : return true;
4588 : : }
4589 : : }
4747 rhaas@postgresql.org 4590 :UBC 0 : return false;
4591 : : }
4592 : :
4593 : : #ifdef WIN32
4594 : :
4595 : : /*
4596 : : * Subset implementation of waitpid() for Windows. We assume pid is -1
4597 : : * (that is, check all child processes) and options is WNOHANG (don't wait).
4598 : : */
4599 : : static pid_t
4600 : : waitpid(pid_t pid, int *exitstatus, int options)
4601 : : {
4602 : : win32_deadchild_waitinfo *childinfo;
4603 : : DWORD exitcode;
4604 : : DWORD dwd;
4605 : : ULONG_PTR key;
4606 : : OVERLAPPED *ovl;
4607 : :
4608 : : /* Try to consume one win32_deadchild_waitinfo from the queue. */
4609 : : if (!GetQueuedCompletionStatus(win32ChildQueue, &dwd, &key, &ovl, 0))
4610 : : {
4611 : : errno = EAGAIN;
4612 : : return -1;
4613 : : }
4614 : :
4615 : : childinfo = (win32_deadchild_waitinfo *) key;
4616 : : pid = childinfo->procId;
4617 : :
4618 : : /*
4619 : : * Remove handle from wait - required even though it's set to wait only
4620 : : * once
4621 : : */
4622 : : UnregisterWaitEx(childinfo->waitHandle, NULL);
4623 : :
4624 : : if (!GetExitCodeProcess(childinfo->procHandle, &exitcode))
4625 : : {
4626 : : /*
4627 : : * Should never happen. Inform user and set a fixed exitcode.
4628 : : */
4629 : : write_stderr("could not read exit code for process\n");
4630 : : exitcode = 255;
4631 : : }
4632 : : *exitstatus = exitcode;
4633 : :
4634 : : /*
4635 : : * Close the process handle. Only after this point can the PID can be
4636 : : * recycled by the kernel.
4637 : : */
4638 : : CloseHandle(childinfo->procHandle);
4639 : :
4640 : : /*
4641 : : * Free struct that was allocated before the call to
4642 : : * RegisterWaitForSingleObject()
4643 : : */
4644 : : pfree(childinfo);
4645 : :
4646 : : return pid;
4647 : : }
4648 : :
4649 : : /*
4650 : : * Note! Code below executes on a thread pool! All operations must
4651 : : * be thread safe! Note that elog() and friends must *not* be used.
4652 : : */
4653 : : static void WINAPI
4654 : : pgwin32_deadchild_callback(PVOID lpParameter, BOOLEAN TimerOrWaitFired)
4655 : : {
4656 : : /* Should never happen, since we use INFINITE as timeout value. */
4657 : : if (TimerOrWaitFired)
4658 : : return;
4659 : :
4660 : : /*
4661 : : * Post the win32_deadchild_waitinfo object for waitpid() to deal with. If
4662 : : * that fails, we leak the object, but we also leak a whole process and
4663 : : * get into an unrecoverable state, so there's not much point in worrying
4664 : : * about that. We'd like to panic, but we can't use that infrastructure
4665 : : * from this thread.
4666 : : */
4667 : : if (!PostQueuedCompletionStatus(win32ChildQueue,
4668 : : 0,
4669 : : (ULONG_PTR) lpParameter,
4670 : : NULL))
4671 : : write_stderr("could not post child completion status\n");
4672 : :
4673 : : /* Queue SIGCHLD signal. */
4674 : : pg_queue_signal(SIGCHLD);
4675 : : }
4676 : :
4677 : : /*
4678 : : * Queue a waiter to signal when this child dies. The wait will be handled
4679 : : * automatically by an operating system thread pool. The memory and the
4680 : : * process handle will be freed by a later call to waitpid().
4681 : : */
4682 : : void
4683 : : pgwin32_register_deadchild_callback(HANDLE procHandle, DWORD procId)
4684 : : {
4685 : : win32_deadchild_waitinfo *childinfo;
4686 : :
4687 : : childinfo = palloc_object(win32_deadchild_waitinfo);
4688 : : childinfo->procHandle = procHandle;
4689 : : childinfo->procId = procId;
4690 : :
4691 : : if (!RegisterWaitForSingleObject(&childinfo->waitHandle,
4692 : : procHandle,
4693 : : pgwin32_deadchild_callback,
4694 : : childinfo,
4695 : : INFINITE,
4696 : : WT_EXECUTEONLYONCE | WT_EXECUTEINWAITTHREAD))
4697 : : ereport(FATAL,
4698 : : (errmsg_internal("could not register process for wait: error code %lu",
4699 : : GetLastError())));
4700 : : }
4701 : :
4702 : : #endif /* WIN32 */
4703 : :
4704 : : /*
4705 : : * Initialize one and only handle for monitoring postmaster death.
4706 : : *
4707 : : * Called once in the postmaster, so that child processes can subsequently
4708 : : * monitor if their parent is dead.
4709 : : */
4710 : : static void
5529 heikki.linnakangas@i 4711 :CBC 991 : InitPostmasterDeathWatchHandle(void)
4712 : : {
4713 : : #ifndef WIN32
4714 : :
4715 : : /*
4716 : : * Create a pipe. Postmaster holds the write end of the pipe open
4717 : : * (POSTMASTER_FD_OWN), and children hold the read end. Children can pass
4718 : : * the read file descriptor to select() to wake up in case postmaster
4719 : : * dies, or check for postmaster death with a (read() == 0). Children must
4720 : : * close the write end as soon as possible after forking, because EOF
4721 : : * won't be signaled in the read end until all processes have closed the
4722 : : * write fd. That is taken care of in ClosePostmasterPorts().
4723 : : */
4724 [ - + ]: 991 : Assert(MyProcPid == PostmasterPid);
3415 tgl@sss.pgh.pa.us 4725 [ - + ]: 991 : if (pipe(postmaster_alive_fds) < 0)
5529 heikki.linnakangas@i 4726 [ # # ]:UBC 0 : ereport(FATAL,
4727 : : (errcode_for_file_access(),
4728 : : errmsg_internal("could not create pipe to monitor postmaster death: %m")));
4729 : :
4730 : : /* Notify fd.c that we've eaten two FDs for the pipe. */
2376 tgl@sss.pgh.pa.us 4731 :CBC 991 : ReserveExternalFD();
4732 : 991 : ReserveExternalFD();
4733 : :
4734 : : /*
4735 : : * Set O_NONBLOCK to allow testing for the fd's presence with a read()
4736 : : * call.
4737 : : */
3415 4738 [ - + ]: 991 : if (fcntl(postmaster_alive_fds[POSTMASTER_FD_WATCH], F_SETFL, O_NONBLOCK) == -1)
5529 heikki.linnakangas@i 4739 [ # # ]:UBC 0 : ereport(FATAL,
4740 : : (errcode_for_socket_access(),
4741 : : errmsg_internal("could not set postmaster death monitoring pipe to nonblocking mode: %m")));
4742 : : #else
4743 : :
4744 : : /*
4745 : : * On Windows, we use a process handle for the same purpose.
4746 : : */
4747 : : if (DuplicateHandle(GetCurrentProcess(),
4748 : : GetCurrentProcess(),
4749 : : GetCurrentProcess(),
4750 : : &PostmasterHandle,
4751 : : 0,
4752 : : TRUE,
4753 : : DUPLICATE_SAME_ACCESS) == 0)
4754 : : ereport(FATAL,
4755 : : (errmsg_internal("could not duplicate postmaster handle: error code %lu",
4756 : : GetLastError())));
4757 : : #endif /* WIN32 */
5529 heikki.linnakangas@i 4758 :CBC 991 : }
|
