Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pg_createsubscriber.c
4 : * Create a new logical replica from a standby server
5 : *
6 : * Copyright (C) 2024, PostgreSQL Global Development Group
7 : *
8 : * IDENTIFICATION
9 : * src/bin/pg_basebackup/pg_createsubscriber.c
10 : *
11 : *-------------------------------------------------------------------------
12 : */
13 :
14 : #include "postgres_fe.h"
15 :
16 : #include <sys/time.h>
17 : #include <sys/wait.h>
18 : #include <time.h>
19 :
20 : #include "catalog/pg_authid_d.h"
21 : #include "common/connect.h"
22 : #include "common/controldata_utils.h"
23 : #include "common/file_perm.h"
24 : #include "common/logging.h"
25 : #include "common/pg_prng.h"
26 : #include "common/restricted_token.h"
27 : #include "fe_utils/recovery_gen.h"
28 : #include "fe_utils/simple_list.h"
29 : #include "getopt_long.h"
30 :
31 : #define DEFAULT_SUB_PORT "50432"
32 :
33 : /* Command-line options */
34 : struct CreateSubscriberOptions
35 : {
36 : char *config_file; /* configuration file */
37 : char *pub_conninfo_str; /* publisher connection string */
38 : char *socket_dir; /* directory for Unix-domain socket, if any */
39 : char *sub_port; /* subscriber port number */
40 : const char *sub_username; /* subscriber username */
41 : SimpleStringList database_names; /* list of database names */
42 : SimpleStringList pub_names; /* list of publication names */
43 : SimpleStringList sub_names; /* list of subscription names */
44 : SimpleStringList replslot_names; /* list of replication slot names */
45 : int recovery_timeout; /* stop recovery after this time */
46 : };
47 :
48 : struct LogicalRepInfo
49 : {
50 : Oid oid; /* database OID */
51 : char *dbname; /* database name */
52 : char *pubconninfo; /* publisher connection string */
53 : char *subconninfo; /* subscriber connection string */
54 : char *pubname; /* publication name */
55 : char *subname; /* subscription name */
56 : char *replslotname; /* replication slot name */
57 :
58 : bool made_replslot; /* replication slot was created */
59 : bool made_publication; /* publication was created */
60 : };
61 :
62 : static void cleanup_objects_atexit(void);
63 : static void usage();
64 : static char *get_base_conninfo(const char *conninfo, char **dbname);
65 : static char *get_sub_conninfo(const struct CreateSubscriberOptions *opt);
66 : static char *get_exec_path(const char *argv0, const char *progname);
67 : static void check_data_directory(const char *datadir);
68 : static char *concat_conninfo_dbname(const char *conninfo, const char *dbname);
69 : static struct LogicalRepInfo *store_pub_sub_info(const struct CreateSubscriberOptions *opt,
70 : const char *pub_base_conninfo,
71 : const char *sub_base_conninfo);
72 : static PGconn *connect_database(const char *conninfo, bool exit_on_error);
73 : static void disconnect_database(PGconn *conn, bool exit_on_error);
74 : static uint64 get_primary_sysid(const char *conninfo);
75 : static uint64 get_standby_sysid(const char *datadir);
76 : static void modify_subscriber_sysid(const struct CreateSubscriberOptions *opt);
77 : static bool server_is_in_recovery(PGconn *conn);
78 : static char *generate_object_name(PGconn *conn);
79 : static void check_publisher(const struct LogicalRepInfo *dbinfo);
80 : static char *setup_publisher(struct LogicalRepInfo *dbinfo);
81 : static void check_subscriber(const struct LogicalRepInfo *dbinfo);
82 : static void setup_subscriber(struct LogicalRepInfo *dbinfo,
83 : const char *consistent_lsn);
84 : static void setup_recovery(const struct LogicalRepInfo *dbinfo, const char *datadir,
85 : const char *lsn);
86 : static void drop_primary_replication_slot(struct LogicalRepInfo *dbinfo,
87 : const char *slotname);
88 : static char *create_logical_replication_slot(PGconn *conn,
89 : struct LogicalRepInfo *dbinfo);
90 : static void drop_replication_slot(PGconn *conn, struct LogicalRepInfo *dbinfo,
91 : const char *slot_name);
92 : static void pg_ctl_status(const char *pg_ctl_cmd, int rc);
93 : static void start_standby_server(const struct CreateSubscriberOptions *opt,
94 : bool restricted_access);
95 : static void stop_standby_server(const char *datadir);
96 : static void wait_for_end_recovery(const char *conninfo,
97 : const struct CreateSubscriberOptions *opt);
98 : static void create_publication(PGconn *conn, struct LogicalRepInfo *dbinfo);
99 : static void drop_publication(PGconn *conn, struct LogicalRepInfo *dbinfo);
100 : static void create_subscription(PGconn *conn, const struct LogicalRepInfo *dbinfo);
101 : static void set_replication_progress(PGconn *conn, const struct LogicalRepInfo *dbinfo,
102 : const char *lsn);
103 : static void enable_subscription(PGconn *conn, const struct LogicalRepInfo *dbinfo);
104 :
105 : #define USEC_PER_SEC 1000000
106 : #define WAIT_INTERVAL 1 /* 1 second */
107 :
108 : static const char *progname;
109 :
110 : static char *primary_slot_name = NULL;
111 : static bool dry_run = false;
112 :
113 : static bool success = false;
114 :
115 : static struct LogicalRepInfo *dbinfo;
116 : static int num_dbs = 0; /* number of specified databases */
117 : static int num_pubs = 0; /* number of specified publications */
118 : static int num_subs = 0; /* number of specified subscriptions */
119 : static int num_replslots = 0; /* number of specified replication slots */
120 :
121 : static pg_prng_state prng_state;
122 :
123 : static char *pg_ctl_path = NULL;
124 : static char *pg_resetwal_path = NULL;
125 :
126 : /* standby / subscriber data directory */
127 : static char *subscriber_dir = NULL;
128 :
129 : static bool recovery_ended = false;
130 : static bool standby_running = false;
131 :
132 : enum WaitPMResult
133 : {
134 : POSTMASTER_READY,
135 : POSTMASTER_STILL_STARTING
136 : };
137 :
138 :
139 : /*
140 : * Cleanup objects that were created by pg_createsubscriber if there is an
141 : * error.
142 : *
143 : * Publications and replication slots are created on primary. Depending on the
144 : * step it failed, it should remove the already created objects if it is
145 : * possible (sometimes it won't work due to a connection issue).
146 : * There is no cleanup on the target server. The steps on the target server are
147 : * executed *after* promotion, hence, at this point, a failure means recreate
148 : * the physical replica and start again.
149 : */
150 : static void
151 18 : cleanup_objects_atexit(void)
152 : {
153 18 : if (success)
154 6 : return;
155 :
156 : /*
157 : * If the server is promoted, there is no way to use the current setup
158 : * again. Warn the user that a new replication setup should be done before
159 : * trying again.
160 : */
161 12 : if (recovery_ended)
162 : {
163 0 : pg_log_warning("failed after the end of recovery");
164 0 : pg_log_warning_hint("The target server cannot be used as a physical replica anymore. "
165 : "You must recreate the physical replica before continuing.");
166 : }
167 :
168 36 : for (int i = 0; i < num_dbs; i++)
169 : {
170 24 : if (dbinfo[i].made_publication || dbinfo[i].made_replslot)
171 : {
172 : PGconn *conn;
173 :
174 0 : conn = connect_database(dbinfo[i].pubconninfo, false);
175 0 : if (conn != NULL)
176 : {
177 0 : if (dbinfo[i].made_publication)
178 0 : drop_publication(conn, &dbinfo[i]);
179 0 : if (dbinfo[i].made_replslot)
180 0 : drop_replication_slot(conn, &dbinfo[i], dbinfo[i].replslotname);
181 0 : disconnect_database(conn, false);
182 : }
183 : else
184 : {
185 : /*
186 : * If a connection could not be established, inform the user
187 : * that some objects were left on primary and should be
188 : * removed before trying again.
189 : */
190 0 : if (dbinfo[i].made_publication)
191 : {
192 0 : pg_log_warning("publication \"%s\" in database \"%s\" on primary might be left behind",
193 : dbinfo[i].pubname, dbinfo[i].dbname);
194 0 : pg_log_warning_hint("Consider dropping this publication before trying again.");
195 : }
196 0 : if (dbinfo[i].made_replslot)
197 : {
198 0 : pg_log_warning("replication slot \"%s\" in database \"%s\" on primary might be left behind",
199 : dbinfo[i].replslotname, dbinfo[i].dbname);
200 0 : pg_log_warning_hint("Drop this replication slot soon to avoid retention of WAL files.");
201 : }
202 : }
203 : }
204 : }
205 :
206 12 : if (standby_running)
207 8 : stop_standby_server(subscriber_dir);
208 : }
209 :
210 : static void
211 2 : usage(void)
212 : {
213 2 : printf(_("%s creates a new logical replica from a standby server.\n\n"),
214 : progname);
215 2 : printf(_("Usage:\n"));
216 2 : printf(_(" %s [OPTION]...\n"), progname);
217 2 : printf(_("\nOptions:\n"));
218 2 : printf(_(" -d, --database=DBNAME database to create a subscription\n"));
219 2 : printf(_(" -D, --pgdata=DATADIR location for the subscriber data directory\n"));
220 2 : printf(_(" -n, --dry-run dry run, just show what would be done\n"));
221 2 : printf(_(" -p, --subscriber-port=PORT subscriber port number (default %s)\n"), DEFAULT_SUB_PORT);
222 2 : printf(_(" -P, --publisher-server=CONNSTR publisher connection string\n"));
223 2 : printf(_(" -s, --socket-directory=DIR socket directory to use (default current directory)\n"));
224 2 : printf(_(" -t, --recovery-timeout=SECS seconds to wait for recovery to end\n"));
225 2 : printf(_(" -U, --subscriber-username=NAME subscriber username\n"));
226 2 : printf(_(" -v, --verbose output verbose messages\n"));
227 2 : printf(_(" --config-file=FILENAME use specified main server configuration\n"
228 : " file when running target cluster\n"));
229 2 : printf(_(" --publication=NAME publication name\n"));
230 2 : printf(_(" --replication-slot=NAME replication slot name\n"));
231 2 : printf(_(" --subscription=NAME subscription name\n"));
232 2 : printf(_(" -V, --version output version information, then exit\n"));
233 2 : printf(_(" -?, --help show this help, then exit\n"));
234 2 : printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT);
235 2 : printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL);
236 2 : }
237 :
238 : /*
239 : * Validate a connection string. Returns a base connection string that is a
240 : * connection string without a database name.
241 : *
242 : * Since we might process multiple databases, each database name will be
243 : * appended to this base connection string to provide a final connection
244 : * string. If the second argument (dbname) is not null, returns dbname if the
245 : * provided connection string contains it. If option --database is not
246 : * provided, uses dbname as the only database to setup the logical replica.
247 : *
248 : * It is the caller's responsibility to free the returned connection string and
249 : * dbname.
250 : */
251 : static char *
252 26 : get_base_conninfo(const char *conninfo, char **dbname)
253 : {
254 : PQExpBuffer buf;
255 : PQconninfoOption *conn_opts;
256 : PQconninfoOption *conn_opt;
257 26 : char *errmsg = NULL;
258 : char *ret;
259 : int i;
260 :
261 26 : conn_opts = PQconninfoParse(conninfo, &errmsg);
262 26 : if (conn_opts == NULL)
263 : {
264 0 : pg_log_error("could not parse connection string: %s", errmsg);
265 0 : PQfreemem(errmsg);
266 0 : return NULL;
267 : }
268 :
269 26 : buf = createPQExpBuffer();
270 26 : i = 0;
271 1092 : for (conn_opt = conn_opts; conn_opt->keyword != NULL; conn_opt++)
272 : {
273 1066 : if (strcmp(conn_opt->keyword, "dbname") == 0 && conn_opt->val != NULL)
274 : {
275 18 : if (dbname)
276 18 : *dbname = pg_strdup(conn_opt->val);
277 18 : continue;
278 : }
279 :
280 1048 : if (conn_opt->val != NULL && conn_opt->val[0] != '\0')
281 : {
282 44 : if (i > 0)
283 18 : appendPQExpBufferChar(buf, ' ');
284 44 : appendPQExpBuffer(buf, "%s=%s", conn_opt->keyword, conn_opt->val);
285 44 : i++;
286 : }
287 : }
288 :
289 26 : ret = pg_strdup(buf->data);
290 :
291 26 : destroyPQExpBuffer(buf);
292 26 : PQconninfoFree(conn_opts);
293 :
294 26 : return ret;
295 : }
296 :
297 : /*
298 : * Build a subscriber connection string. Only a few parameters are supported
299 : * since it starts a server with restricted access.
300 : */
301 : static char *
302 26 : get_sub_conninfo(const struct CreateSubscriberOptions *opt)
303 : {
304 26 : PQExpBuffer buf = createPQExpBuffer();
305 : char *ret;
306 :
307 26 : appendPQExpBuffer(buf, "port=%s", opt->sub_port);
308 : #if !defined(WIN32)
309 26 : appendPQExpBuffer(buf, " host=%s", opt->socket_dir);
310 : #endif
311 26 : if (opt->sub_username != NULL)
312 0 : appendPQExpBuffer(buf, " user=%s", opt->sub_username);
313 26 : appendPQExpBuffer(buf, " fallback_application_name=%s", progname);
314 :
315 26 : ret = pg_strdup(buf->data);
316 :
317 26 : destroyPQExpBuffer(buf);
318 :
319 26 : return ret;
320 : }
321 :
322 : /*
323 : * Verify if a PostgreSQL binary (progname) is available in the same directory as
324 : * pg_createsubscriber and it has the same version. It returns the absolute
325 : * path of the progname.
326 : */
327 : static char *
328 36 : get_exec_path(const char *argv0, const char *progname)
329 : {
330 : char *versionstr;
331 : char *exec_path;
332 : int ret;
333 :
334 36 : versionstr = psprintf("%s (PostgreSQL) %s\n", progname, PG_VERSION);
335 36 : exec_path = pg_malloc(MAXPGPATH);
336 36 : ret = find_other_exec(argv0, progname, versionstr, exec_path);
337 :
338 36 : if (ret < 0)
339 : {
340 : char full_path[MAXPGPATH];
341 :
342 0 : if (find_my_exec(argv0, full_path) < 0)
343 0 : strlcpy(full_path, progname, sizeof(full_path));
344 :
345 0 : if (ret == -1)
346 0 : pg_fatal("program \"%s\" is needed by %s but was not found in the same directory as \"%s\"",
347 : progname, "pg_createsubscriber", full_path);
348 : else
349 0 : pg_fatal("program \"%s\" was found by \"%s\" but was not the same version as %s",
350 : progname, full_path, "pg_createsubscriber");
351 : }
352 :
353 36 : pg_log_debug("%s path is: %s", progname, exec_path);
354 :
355 36 : return exec_path;
356 : }
357 :
358 : /*
359 : * Is it a cluster directory? These are preliminary checks. It is far from
360 : * making an accurate check. If it is not a clone from the publisher, it will
361 : * eventually fail in a future step.
362 : */
363 : static void
364 18 : check_data_directory(const char *datadir)
365 : {
366 : struct stat statbuf;
367 : char versionfile[MAXPGPATH];
368 :
369 18 : pg_log_info("checking if directory \"%s\" is a cluster data directory",
370 : datadir);
371 :
372 18 : if (stat(datadir, &statbuf) != 0)
373 : {
374 0 : if (errno == ENOENT)
375 0 : pg_fatal("data directory \"%s\" does not exist", datadir);
376 : else
377 0 : pg_fatal("could not access directory \"%s\": %m", datadir);
378 : }
379 :
380 18 : snprintf(versionfile, MAXPGPATH, "%s/PG_VERSION", datadir);
381 18 : if (stat(versionfile, &statbuf) != 0 && errno == ENOENT)
382 : {
383 0 : pg_fatal("directory \"%s\" is not a database cluster directory",
384 : datadir);
385 : }
386 18 : }
387 :
388 : /*
389 : * Append database name into a base connection string.
390 : *
391 : * dbname is the only parameter that changes so it is not included in the base
392 : * connection string. This function concatenates dbname to build a "real"
393 : * connection string.
394 : */
395 : static char *
396 68 : concat_conninfo_dbname(const char *conninfo, const char *dbname)
397 : {
398 68 : PQExpBuffer buf = createPQExpBuffer();
399 : char *ret;
400 :
401 : Assert(conninfo != NULL);
402 :
403 68 : appendPQExpBufferStr(buf, conninfo);
404 68 : appendPQExpBuffer(buf, " dbname=%s", dbname);
405 :
406 68 : ret = pg_strdup(buf->data);
407 68 : destroyPQExpBuffer(buf);
408 :
409 68 : return ret;
410 : }
411 :
412 : /*
413 : * Store publication and subscription information.
414 : *
415 : * If publication, replication slot and subscription names were specified,
416 : * store it here. Otherwise, a generated name will be assigned to the object in
417 : * setup_publisher().
418 : */
419 : static struct LogicalRepInfo *
420 18 : store_pub_sub_info(const struct CreateSubscriberOptions *opt,
421 : const char *pub_base_conninfo,
422 : const char *sub_base_conninfo)
423 : {
424 : struct LogicalRepInfo *dbinfo;
425 18 : SimpleStringListCell *pubcell = NULL;
426 18 : SimpleStringListCell *subcell = NULL;
427 18 : SimpleStringListCell *replslotcell = NULL;
428 18 : int i = 0;
429 :
430 18 : dbinfo = pg_malloc_array(struct LogicalRepInfo, num_dbs);
431 :
432 18 : if (num_pubs > 0)
433 4 : pubcell = opt->pub_names.head;
434 18 : if (num_subs > 0)
435 2 : subcell = opt->sub_names.head;
436 18 : if (num_replslots > 0)
437 4 : replslotcell = opt->replslot_names.head;
438 :
439 52 : for (SimpleStringListCell *cell = opt->database_names.head; cell; cell = cell->next)
440 : {
441 : char *conninfo;
442 :
443 : /* Fill publisher attributes */
444 34 : conninfo = concat_conninfo_dbname(pub_base_conninfo, cell->val);
445 34 : dbinfo[i].pubconninfo = conninfo;
446 34 : dbinfo[i].dbname = cell->val;
447 34 : if (num_pubs > 0)
448 8 : dbinfo[i].pubname = pubcell->val;
449 : else
450 26 : dbinfo[i].pubname = NULL;
451 34 : if (num_replslots > 0)
452 6 : dbinfo[i].replslotname = replslotcell->val;
453 : else
454 28 : dbinfo[i].replslotname = NULL;
455 34 : dbinfo[i].made_replslot = false;
456 34 : dbinfo[i].made_publication = false;
457 : /* Fill subscriber attributes */
458 34 : conninfo = concat_conninfo_dbname(sub_base_conninfo, cell->val);
459 34 : dbinfo[i].subconninfo = conninfo;
460 34 : if (num_subs > 0)
461 4 : dbinfo[i].subname = subcell->val;
462 : else
463 30 : dbinfo[i].subname = NULL;
464 : /* Other fields will be filled later */
465 :
466 34 : pg_log_debug("publisher(%d): publication: %s ; replication slot: %s ; connection string: %s", i,
467 : dbinfo[i].pubname ? dbinfo[i].pubname : "(auto)",
468 : dbinfo[i].replslotname ? dbinfo[i].replslotname : "(auto)",
469 : dbinfo[i].pubconninfo);
470 34 : pg_log_debug("subscriber(%d): subscription: %s ; connection string: %s", i,
471 : dbinfo[i].subname ? dbinfo[i].subname : "(auto)",
472 : dbinfo[i].subconninfo);
473 :
474 34 : if (num_pubs > 0)
475 8 : pubcell = pubcell->next;
476 34 : if (num_subs > 0)
477 4 : subcell = subcell->next;
478 34 : if (num_replslots > 0)
479 6 : replslotcell = replslotcell->next;
480 :
481 34 : i++;
482 : }
483 :
484 18 : return dbinfo;
485 : }
486 :
487 : /*
488 : * Open a new connection. If exit_on_error is true, it has an undesired
489 : * condition and it should exit immediately.
490 : */
491 : static PGconn *
492 80 : connect_database(const char *conninfo, bool exit_on_error)
493 : {
494 : PGconn *conn;
495 : PGresult *res;
496 :
497 80 : conn = PQconnectdb(conninfo);
498 80 : if (PQstatus(conn) != CONNECTION_OK)
499 : {
500 0 : pg_log_error("connection to database failed: %s",
501 : PQerrorMessage(conn));
502 0 : PQfinish(conn);
503 :
504 0 : if (exit_on_error)
505 0 : exit(1);
506 0 : return NULL;
507 : }
508 :
509 : /* Secure search_path */
510 80 : res = PQexec(conn, ALWAYS_SECURE_SEARCH_PATH_SQL);
511 80 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
512 : {
513 0 : pg_log_error("could not clear search_path: %s",
514 : PQresultErrorMessage(res));
515 0 : PQclear(res);
516 0 : PQfinish(conn);
517 :
518 0 : if (exit_on_error)
519 0 : exit(1);
520 0 : return NULL;
521 : }
522 80 : PQclear(res);
523 :
524 80 : return conn;
525 : }
526 :
527 : /*
528 : * Close the connection. If exit_on_error is true, it has an undesired
529 : * condition and it should exit immediately.
530 : */
531 : static void
532 80 : disconnect_database(PGconn *conn, bool exit_on_error)
533 : {
534 : Assert(conn != NULL);
535 :
536 80 : PQfinish(conn);
537 :
538 80 : if (exit_on_error)
539 4 : exit(1);
540 76 : }
541 :
542 : /*
543 : * Obtain the system identifier using the provided connection. It will be used
544 : * to compare if a data directory is a clone of another one.
545 : */
546 : static uint64
547 18 : get_primary_sysid(const char *conninfo)
548 : {
549 : PGconn *conn;
550 : PGresult *res;
551 : uint64 sysid;
552 :
553 18 : pg_log_info("getting system identifier from publisher");
554 :
555 18 : conn = connect_database(conninfo, true);
556 :
557 18 : res = PQexec(conn, "SELECT system_identifier FROM pg_catalog.pg_control_system()");
558 18 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
559 : {
560 0 : pg_log_error("could not get system identifier: %s",
561 : PQresultErrorMessage(res));
562 0 : disconnect_database(conn, true);
563 : }
564 18 : if (PQntuples(res) != 1)
565 : {
566 0 : pg_log_error("could not get system identifier: got %d rows, expected %d row",
567 : PQntuples(res), 1);
568 0 : disconnect_database(conn, true);
569 : }
570 :
571 18 : sysid = strtou64(PQgetvalue(res, 0, 0), NULL, 10);
572 :
573 18 : pg_log_info("system identifier is %llu on publisher",
574 : (unsigned long long) sysid);
575 :
576 18 : PQclear(res);
577 18 : disconnect_database(conn, false);
578 :
579 18 : return sysid;
580 : }
581 :
582 : /*
583 : * Obtain the system identifier from control file. It will be used to compare
584 : * if a data directory is a clone of another one. This routine is used locally
585 : * and avoids a connection.
586 : */
587 : static uint64
588 18 : get_standby_sysid(const char *datadir)
589 : {
590 : ControlFileData *cf;
591 : bool crc_ok;
592 : uint64 sysid;
593 :
594 18 : pg_log_info("getting system identifier from subscriber");
595 :
596 18 : cf = get_controlfile(datadir, &crc_ok);
597 18 : if (!crc_ok)
598 0 : pg_fatal("control file appears to be corrupt");
599 :
600 18 : sysid = cf->system_identifier;
601 :
602 18 : pg_log_info("system identifier is %llu on subscriber",
603 : (unsigned long long) sysid);
604 :
605 18 : pg_free(cf);
606 :
607 18 : return sysid;
608 : }
609 :
610 : /*
611 : * Modify the system identifier. Since a standby server preserves the system
612 : * identifier, it makes sense to change it to avoid situations in which WAL
613 : * files from one of the systems might be used in the other one.
614 : */
615 : static void
616 6 : modify_subscriber_sysid(const struct CreateSubscriberOptions *opt)
617 : {
618 : ControlFileData *cf;
619 : bool crc_ok;
620 : struct timeval tv;
621 :
622 : char *cmd_str;
623 :
624 6 : pg_log_info("modifying system identifier of subscriber");
625 :
626 6 : cf = get_controlfile(subscriber_dir, &crc_ok);
627 6 : if (!crc_ok)
628 0 : pg_fatal("control file appears to be corrupt");
629 :
630 : /*
631 : * Select a new system identifier.
632 : *
633 : * XXX this code was extracted from BootStrapXLOG().
634 : */
635 6 : gettimeofday(&tv, NULL);
636 6 : cf->system_identifier = ((uint64) tv.tv_sec) << 32;
637 6 : cf->system_identifier |= ((uint64) tv.tv_usec) << 12;
638 6 : cf->system_identifier |= getpid() & 0xFFF;
639 :
640 6 : if (!dry_run)
641 2 : update_controlfile(subscriber_dir, cf, true);
642 :
643 6 : pg_log_info("system identifier is %llu on subscriber",
644 : (unsigned long long) cf->system_identifier);
645 :
646 6 : pg_log_info("running pg_resetwal on the subscriber");
647 :
648 6 : cmd_str = psprintf("\"%s\" -D \"%s\" > \"%s\"", pg_resetwal_path,
649 : subscriber_dir, DEVNULL);
650 :
651 6 : pg_log_debug("pg_resetwal command is: %s", cmd_str);
652 :
653 6 : if (!dry_run)
654 : {
655 2 : int rc = system(cmd_str);
656 :
657 2 : if (rc == 0)
658 2 : pg_log_info("subscriber successfully changed the system identifier");
659 : else
660 0 : pg_fatal("subscriber failed to change system identifier: exit code: %d", rc);
661 : }
662 :
663 6 : pg_free(cf);
664 6 : }
665 :
666 : /*
667 : * Generate an object name using a prefix, database oid and a random integer.
668 : * It is used in case the user does not specify an object name (publication,
669 : * subscription, replication slot).
670 : */
671 : static char *
672 10 : generate_object_name(PGconn *conn)
673 : {
674 : PGresult *res;
675 : Oid oid;
676 : uint32 rand;
677 : char *objname;
678 :
679 10 : res = PQexec(conn,
680 : "SELECT oid FROM pg_catalog.pg_database "
681 : "WHERE datname = pg_catalog.current_database()");
682 10 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
683 : {
684 0 : pg_log_error("could not obtain database OID: %s",
685 : PQresultErrorMessage(res));
686 0 : disconnect_database(conn, true);
687 : }
688 :
689 10 : if (PQntuples(res) != 1)
690 : {
691 0 : pg_log_error("could not obtain database OID: got %d rows, expected %d row",
692 : PQntuples(res), 1);
693 0 : disconnect_database(conn, true);
694 : }
695 :
696 : /* Database OID */
697 10 : oid = strtoul(PQgetvalue(res, 0, 0), NULL, 10);
698 :
699 10 : PQclear(res);
700 :
701 : /* Random unsigned integer */
702 10 : rand = pg_prng_uint32(&prng_state);
703 :
704 : /*
705 : * Build the object name. The name must not exceed NAMEDATALEN - 1. This
706 : * current schema uses a maximum of 40 characters (20 + 10 + 1 + 8 +
707 : * '\0').
708 : */
709 10 : objname = psprintf("pg_createsubscriber_%u_%x", oid, rand);
710 :
711 10 : return objname;
712 : }
713 :
714 : /*
715 : * Create the publications and replication slots in preparation for logical
716 : * replication. Returns the LSN from latest replication slot. It will be the
717 : * replication start point that is used to adjust the subscriptions (see
718 : * set_replication_progress).
719 : */
720 : static char *
721 6 : setup_publisher(struct LogicalRepInfo *dbinfo)
722 : {
723 6 : char *lsn = NULL;
724 :
725 6 : pg_prng_seed(&prng_state, (uint64) (getpid() ^ time(NULL)));
726 :
727 16 : for (int i = 0; i < num_dbs; i++)
728 : {
729 : PGconn *conn;
730 10 : char *genname = NULL;
731 :
732 10 : conn = connect_database(dbinfo[i].pubconninfo, true);
733 :
734 : /*
735 : * If an object name was not specified as command-line options, assign
736 : * a generated object name. The replication slot has a different rule.
737 : * The subscription name is assigned to the replication slot name if
738 : * no replication slot is specified. It follows the same rule as
739 : * CREATE SUBSCRIPTION.
740 : */
741 10 : if (num_pubs == 0 || num_subs == 0 || num_replslots == 0)
742 10 : genname = generate_object_name(conn);
743 10 : if (num_pubs == 0)
744 2 : dbinfo[i].pubname = pg_strdup(genname);
745 10 : if (num_subs == 0)
746 6 : dbinfo[i].subname = pg_strdup(genname);
747 10 : if (num_replslots == 0)
748 4 : dbinfo[i].replslotname = pg_strdup(dbinfo[i].subname);
749 :
750 : /*
751 : * Create publication on publisher. This step should be executed
752 : * *before* promoting the subscriber to avoid any transactions between
753 : * consistent LSN and the new publication rows (such transactions
754 : * wouldn't see the new publication rows resulting in an error).
755 : */
756 10 : create_publication(conn, &dbinfo[i]);
757 :
758 : /* Create replication slot on publisher */
759 10 : if (lsn)
760 2 : pg_free(lsn);
761 10 : lsn = create_logical_replication_slot(conn, &dbinfo[i]);
762 10 : if (lsn != NULL || dry_run)
763 10 : pg_log_info("create replication slot \"%s\" on publisher",
764 : dbinfo[i].replslotname);
765 : else
766 0 : exit(1);
767 :
768 10 : disconnect_database(conn, false);
769 : }
770 :
771 6 : return lsn;
772 : }
773 :
774 : /*
775 : * Is recovery still in progress?
776 : */
777 : static bool
778 48 : server_is_in_recovery(PGconn *conn)
779 : {
780 : PGresult *res;
781 : int ret;
782 :
783 48 : res = PQexec(conn, "SELECT pg_catalog.pg_is_in_recovery()");
784 :
785 48 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
786 : {
787 0 : pg_log_error("could not obtain recovery progress: %s",
788 : PQresultErrorMessage(res));
789 0 : disconnect_database(conn, true);
790 : }
791 :
792 :
793 48 : ret = strcmp("t", PQgetvalue(res, 0, 0));
794 :
795 48 : PQclear(res);
796 :
797 48 : return ret == 0;
798 : }
799 :
800 : /*
801 : * Is the primary server ready for logical replication?
802 : *
803 : * XXX Does it not allow a synchronous replica?
804 : */
805 : static void
806 10 : check_publisher(const struct LogicalRepInfo *dbinfo)
807 : {
808 : PGconn *conn;
809 : PGresult *res;
810 10 : bool failed = false;
811 :
812 : char *wal_level;
813 : int max_repslots;
814 : int cur_repslots;
815 : int max_walsenders;
816 : int cur_walsenders;
817 :
818 10 : pg_log_info("checking settings on publisher");
819 :
820 10 : conn = connect_database(dbinfo[0].pubconninfo, true);
821 :
822 : /*
823 : * If the primary server is in recovery (i.e. cascading replication),
824 : * objects (publication) cannot be created because it is read only.
825 : */
826 10 : if (server_is_in_recovery(conn))
827 : {
828 2 : pg_log_error("primary server cannot be in recovery");
829 2 : disconnect_database(conn, true);
830 : }
831 :
832 : /*------------------------------------------------------------------------
833 : * Logical replication requires a few parameters to be set on publisher.
834 : * Since these parameters are not a requirement for physical replication,
835 : * we should check it to make sure it won't fail.
836 : *
837 : * - wal_level = logical
838 : * - max_replication_slots >= current + number of dbs to be converted
839 : * - max_wal_senders >= current + number of dbs to be converted
840 : * -----------------------------------------------------------------------
841 : */
842 8 : res = PQexec(conn,
843 : "WITH wl AS "
844 : "(SELECT setting AS wallevel FROM pg_catalog.pg_settings "
845 : "WHERE name = 'wal_level'), "
846 : "total_mrs AS "
847 : "(SELECT setting AS tmrs FROM pg_catalog.pg_settings "
848 : "WHERE name = 'max_replication_slots'), "
849 : "cur_mrs AS "
850 : "(SELECT count(*) AS cmrs "
851 : "FROM pg_catalog.pg_replication_slots), "
852 : "total_mws AS "
853 : "(SELECT setting AS tmws FROM pg_catalog.pg_settings "
854 : "WHERE name = 'max_wal_senders'), "
855 : "cur_mws AS "
856 : "(SELECT count(*) AS cmws FROM pg_catalog.pg_stat_activity "
857 : "WHERE backend_type = 'walsender') "
858 : "SELECT wallevel, tmrs, cmrs, tmws, cmws "
859 : "FROM wl, total_mrs, cur_mrs, total_mws, cur_mws");
860 :
861 8 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
862 : {
863 0 : pg_log_error("could not obtain publisher settings: %s",
864 : PQresultErrorMessage(res));
865 0 : disconnect_database(conn, true);
866 : }
867 :
868 8 : wal_level = pg_strdup(PQgetvalue(res, 0, 0));
869 8 : max_repslots = atoi(PQgetvalue(res, 0, 1));
870 8 : cur_repslots = atoi(PQgetvalue(res, 0, 2));
871 8 : max_walsenders = atoi(PQgetvalue(res, 0, 3));
872 8 : cur_walsenders = atoi(PQgetvalue(res, 0, 4));
873 :
874 8 : PQclear(res);
875 :
876 8 : pg_log_debug("publisher: wal_level: %s", wal_level);
877 8 : pg_log_debug("publisher: max_replication_slots: %d", max_repslots);
878 8 : pg_log_debug("publisher: current replication slots: %d", cur_repslots);
879 8 : pg_log_debug("publisher: max_wal_senders: %d", max_walsenders);
880 8 : pg_log_debug("publisher: current wal senders: %d", cur_walsenders);
881 :
882 : /*
883 : * If standby sets primary_slot_name, check if this replication slot is in
884 : * use on primary for WAL retention purposes. This replication slot has no
885 : * use after the transformation, hence, it will be removed at the end of
886 : * this process.
887 : */
888 8 : if (primary_slot_name)
889 : {
890 8 : PQExpBuffer str = createPQExpBuffer();
891 8 : char *psn_esc = PQescapeLiteral(conn, primary_slot_name, strlen(primary_slot_name));
892 :
893 8 : appendPQExpBuffer(str,
894 : "SELECT 1 FROM pg_catalog.pg_replication_slots "
895 : "WHERE active AND slot_name = %s",
896 : psn_esc);
897 :
898 8 : pg_free(psn_esc);
899 :
900 8 : pg_log_debug("command is: %s", str->data);
901 :
902 8 : res = PQexec(conn, str->data);
903 8 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
904 : {
905 0 : pg_log_error("could not obtain replication slot information: %s",
906 : PQresultErrorMessage(res));
907 0 : disconnect_database(conn, true);
908 : }
909 :
910 8 : if (PQntuples(res) != 1)
911 : {
912 0 : pg_log_error("could not obtain replication slot information: got %d rows, expected %d row",
913 : PQntuples(res), 1);
914 0 : disconnect_database(conn, true);
915 : }
916 : else
917 8 : pg_log_info("primary has replication slot \"%s\"",
918 : primary_slot_name);
919 :
920 8 : PQclear(res);
921 : }
922 :
923 8 : disconnect_database(conn, false);
924 :
925 8 : if (strcmp(wal_level, "logical") != 0)
926 : {
927 2 : pg_log_error("publisher requires wal_level >= \"logical\"");
928 2 : failed = true;
929 : }
930 :
931 8 : if (max_repslots - cur_repslots < num_dbs)
932 : {
933 2 : pg_log_error("publisher requires %d replication slots, but only %d remain",
934 : num_dbs, max_repslots - cur_repslots);
935 2 : pg_log_error_hint("Consider increasing max_replication_slots to at least %d.",
936 : cur_repslots + num_dbs);
937 2 : failed = true;
938 : }
939 :
940 8 : if (max_walsenders - cur_walsenders < num_dbs)
941 : {
942 2 : pg_log_error("publisher requires %d wal sender processes, but only %d remain",
943 : num_dbs, max_walsenders - cur_walsenders);
944 2 : pg_log_error_hint("Consider increasing max_wal_senders to at least %d.",
945 : cur_walsenders + num_dbs);
946 2 : failed = true;
947 : }
948 :
949 8 : pg_free(wal_level);
950 :
951 8 : if (failed)
952 2 : exit(1);
953 6 : }
954 :
955 : /*
956 : * Is the standby server ready for logical replication?
957 : *
958 : * XXX Does it not allow a time-delayed replica?
959 : *
960 : * XXX In a cascaded replication scenario (P -> S -> C), if the target server
961 : * is S, it cannot detect there is a replica (server C) because server S starts
962 : * accepting only local connections and server C cannot connect to it. Hence,
963 : * there is not a reliable way to provide a suitable error saying the server C
964 : * will be broken at the end of this process (due to pg_resetwal).
965 : */
966 : static void
967 14 : check_subscriber(const struct LogicalRepInfo *dbinfo)
968 : {
969 : PGconn *conn;
970 : PGresult *res;
971 14 : bool failed = false;
972 :
973 : int max_lrworkers;
974 : int max_repslots;
975 : int max_wprocs;
976 :
977 14 : pg_log_info("checking settings on subscriber");
978 :
979 14 : conn = connect_database(dbinfo[0].subconninfo, true);
980 :
981 : /* The target server must be a standby */
982 14 : if (!server_is_in_recovery(conn))
983 : {
984 2 : pg_log_error("target server must be a standby");
985 2 : disconnect_database(conn, true);
986 : }
987 :
988 : /*------------------------------------------------------------------------
989 : * Logical replication requires a few parameters to be set on subscriber.
990 : * Since these parameters are not a requirement for physical replication,
991 : * we should check it to make sure it won't fail.
992 : *
993 : * - max_replication_slots >= number of dbs to be converted
994 : * - max_logical_replication_workers >= number of dbs to be converted
995 : * - max_worker_processes >= 1 + number of dbs to be converted
996 : *------------------------------------------------------------------------
997 : */
998 12 : res = PQexec(conn,
999 : "SELECT setting FROM pg_catalog.pg_settings WHERE name IN ("
1000 : "'max_logical_replication_workers', "
1001 : "'max_replication_slots', "
1002 : "'max_worker_processes', "
1003 : "'primary_slot_name') "
1004 : "ORDER BY name");
1005 :
1006 12 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1007 : {
1008 0 : pg_log_error("could not obtain subscriber settings: %s",
1009 : PQresultErrorMessage(res));
1010 0 : disconnect_database(conn, true);
1011 : }
1012 :
1013 12 : max_lrworkers = atoi(PQgetvalue(res, 0, 0));
1014 12 : max_repslots = atoi(PQgetvalue(res, 1, 0));
1015 12 : max_wprocs = atoi(PQgetvalue(res, 2, 0));
1016 12 : if (strcmp(PQgetvalue(res, 3, 0), "") != 0)
1017 10 : primary_slot_name = pg_strdup(PQgetvalue(res, 3, 0));
1018 :
1019 12 : pg_log_debug("subscriber: max_logical_replication_workers: %d",
1020 : max_lrworkers);
1021 12 : pg_log_debug("subscriber: max_replication_slots: %d", max_repslots);
1022 12 : pg_log_debug("subscriber: max_worker_processes: %d", max_wprocs);
1023 12 : if (primary_slot_name)
1024 10 : pg_log_debug("subscriber: primary_slot_name: %s", primary_slot_name);
1025 :
1026 12 : PQclear(res);
1027 :
1028 12 : disconnect_database(conn, false);
1029 :
1030 12 : if (max_repslots < num_dbs)
1031 : {
1032 2 : pg_log_error("subscriber requires %d replication slots, but only %d remain",
1033 : num_dbs, max_repslots);
1034 2 : pg_log_error_hint("Consider increasing max_replication_slots to at least %d.",
1035 : num_dbs);
1036 2 : failed = true;
1037 : }
1038 :
1039 12 : if (max_lrworkers < num_dbs)
1040 : {
1041 2 : pg_log_error("subscriber requires %d logical replication workers, but only %d remain",
1042 : num_dbs, max_lrworkers);
1043 2 : pg_log_error_hint("Consider increasing max_logical_replication_workers to at least %d.",
1044 : num_dbs);
1045 2 : failed = true;
1046 : }
1047 :
1048 12 : if (max_wprocs < num_dbs + 1)
1049 : {
1050 2 : pg_log_error("subscriber requires %d worker processes, but only %d remain",
1051 : num_dbs + 1, max_wprocs);
1052 2 : pg_log_error_hint("Consider increasing max_worker_processes to at least %d.",
1053 : num_dbs + 1);
1054 2 : failed = true;
1055 : }
1056 :
1057 12 : if (failed)
1058 2 : exit(1);
1059 10 : }
1060 :
1061 : /*
1062 : * Create the subscriptions, adjust the initial location for logical
1063 : * replication and enable the subscriptions. That's the last step for logical
1064 : * replication setup.
1065 : */
1066 : static void
1067 6 : setup_subscriber(struct LogicalRepInfo *dbinfo, const char *consistent_lsn)
1068 : {
1069 16 : for (int i = 0; i < num_dbs; i++)
1070 : {
1071 : PGconn *conn;
1072 :
1073 : /* Connect to subscriber. */
1074 10 : conn = connect_database(dbinfo[i].subconninfo, true);
1075 :
1076 : /*
1077 : * Since the publication was created before the consistent LSN, it is
1078 : * available on the subscriber when the physical replica is promoted.
1079 : * Remove publications from the subscriber because it has no use.
1080 : */
1081 10 : drop_publication(conn, &dbinfo[i]);
1082 :
1083 10 : create_subscription(conn, &dbinfo[i]);
1084 :
1085 : /* Set the replication progress to the correct LSN */
1086 10 : set_replication_progress(conn, &dbinfo[i], consistent_lsn);
1087 :
1088 : /* Enable subscription */
1089 10 : enable_subscription(conn, &dbinfo[i]);
1090 :
1091 10 : disconnect_database(conn, false);
1092 : }
1093 6 : }
1094 :
1095 : /*
1096 : * Write the required recovery parameters.
1097 : */
1098 : static void
1099 6 : setup_recovery(const struct LogicalRepInfo *dbinfo, const char *datadir, const char *lsn)
1100 : {
1101 : PGconn *conn;
1102 : PQExpBuffer recoveryconfcontents;
1103 :
1104 : /*
1105 : * Despite of the recovery parameters will be written to the subscriber,
1106 : * use a publisher connection. The primary_conninfo is generated using the
1107 : * connection settings.
1108 : */
1109 6 : conn = connect_database(dbinfo[0].pubconninfo, true);
1110 :
1111 : /*
1112 : * Write recovery parameters.
1113 : *
1114 : * The subscriber is not running yet. In dry run mode, the recovery
1115 : * parameters *won't* be written. An invalid LSN is used for printing
1116 : * purposes. Additional recovery parameters are added here. It avoids
1117 : * unexpected behavior such as end of recovery as soon as a consistent
1118 : * state is reached (recovery_target) and failure due to multiple recovery
1119 : * targets (name, time, xid, LSN).
1120 : */
1121 6 : recoveryconfcontents = GenerateRecoveryConfig(conn, NULL, NULL);
1122 6 : appendPQExpBuffer(recoveryconfcontents, "recovery_target = ''\n");
1123 6 : appendPQExpBuffer(recoveryconfcontents,
1124 : "recovery_target_timeline = 'latest'\n");
1125 6 : appendPQExpBuffer(recoveryconfcontents,
1126 : "recovery_target_inclusive = true\n");
1127 6 : appendPQExpBuffer(recoveryconfcontents,
1128 : "recovery_target_action = promote\n");
1129 6 : appendPQExpBuffer(recoveryconfcontents, "recovery_target_name = ''\n");
1130 6 : appendPQExpBuffer(recoveryconfcontents, "recovery_target_time = ''\n");
1131 6 : appendPQExpBuffer(recoveryconfcontents, "recovery_target_xid = ''\n");
1132 :
1133 6 : if (dry_run)
1134 : {
1135 4 : appendPQExpBuffer(recoveryconfcontents, "# dry run mode");
1136 4 : appendPQExpBuffer(recoveryconfcontents,
1137 : "recovery_target_lsn = '%X/%X'\n",
1138 4 : LSN_FORMAT_ARGS((XLogRecPtr) InvalidXLogRecPtr));
1139 : }
1140 : else
1141 : {
1142 2 : appendPQExpBuffer(recoveryconfcontents, "recovery_target_lsn = '%s'\n",
1143 : lsn);
1144 2 : WriteRecoveryConfig(conn, datadir, recoveryconfcontents);
1145 : }
1146 6 : disconnect_database(conn, false);
1147 :
1148 6 : pg_log_debug("recovery parameters:\n%s", recoveryconfcontents->data);
1149 6 : }
1150 :
1151 : /*
1152 : * Drop physical replication slot on primary if the standby was using it. After
1153 : * the transformation, it has no use.
1154 : *
1155 : * XXX we might not fail here. Instead, we provide a warning so the user
1156 : * eventually drops this replication slot later.
1157 : */
1158 : static void
1159 6 : drop_primary_replication_slot(struct LogicalRepInfo *dbinfo, const char *slotname)
1160 : {
1161 : PGconn *conn;
1162 :
1163 : /* Replication slot does not exist, do nothing */
1164 6 : if (!primary_slot_name)
1165 0 : return;
1166 :
1167 6 : conn = connect_database(dbinfo[0].pubconninfo, false);
1168 6 : if (conn != NULL)
1169 : {
1170 6 : drop_replication_slot(conn, &dbinfo[0], slotname);
1171 6 : disconnect_database(conn, false);
1172 : }
1173 : else
1174 : {
1175 0 : pg_log_warning("could not drop replication slot \"%s\" on primary",
1176 : slotname);
1177 0 : pg_log_warning_hint("Drop this replication slot soon to avoid retention of WAL files.");
1178 : }
1179 : }
1180 :
1181 : /*
1182 : * Create a logical replication slot and returns a LSN.
1183 : *
1184 : * CreateReplicationSlot() is not used because it does not provide the one-row
1185 : * result set that contains the LSN.
1186 : */
1187 : static char *
1188 10 : create_logical_replication_slot(PGconn *conn, struct LogicalRepInfo *dbinfo)
1189 : {
1190 10 : PQExpBuffer str = createPQExpBuffer();
1191 10 : PGresult *res = NULL;
1192 10 : const char *slot_name = dbinfo->replslotname;
1193 : char *slot_name_esc;
1194 10 : char *lsn = NULL;
1195 :
1196 : Assert(conn != NULL);
1197 :
1198 10 : pg_log_info("creating the replication slot \"%s\" on database \"%s\"",
1199 : slot_name, dbinfo->dbname);
1200 :
1201 10 : slot_name_esc = PQescapeLiteral(conn, slot_name, strlen(slot_name));
1202 :
1203 10 : appendPQExpBuffer(str,
1204 : "SELECT lsn FROM pg_catalog.pg_create_logical_replication_slot(%s, 'pgoutput', false, false, false)",
1205 : slot_name_esc);
1206 :
1207 10 : pg_free(slot_name_esc);
1208 :
1209 10 : pg_log_debug("command is: %s", str->data);
1210 :
1211 10 : if (!dry_run)
1212 : {
1213 4 : res = PQexec(conn, str->data);
1214 4 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1215 : {
1216 0 : pg_log_error("could not create replication slot \"%s\" on database \"%s\": %s",
1217 : slot_name, dbinfo->dbname,
1218 : PQresultErrorMessage(res));
1219 0 : PQclear(res);
1220 0 : destroyPQExpBuffer(str);
1221 0 : return NULL;
1222 : }
1223 :
1224 4 : lsn = pg_strdup(PQgetvalue(res, 0, 0));
1225 4 : PQclear(res);
1226 : }
1227 :
1228 : /* For cleanup purposes */
1229 10 : dbinfo->made_replslot = true;
1230 :
1231 10 : destroyPQExpBuffer(str);
1232 :
1233 10 : return lsn;
1234 : }
1235 :
1236 : static void
1237 6 : drop_replication_slot(PGconn *conn, struct LogicalRepInfo *dbinfo,
1238 : const char *slot_name)
1239 : {
1240 6 : PQExpBuffer str = createPQExpBuffer();
1241 : char *slot_name_esc;
1242 : PGresult *res;
1243 :
1244 : Assert(conn != NULL);
1245 :
1246 6 : pg_log_info("dropping the replication slot \"%s\" on database \"%s\"",
1247 : slot_name, dbinfo->dbname);
1248 :
1249 6 : slot_name_esc = PQescapeLiteral(conn, slot_name, strlen(slot_name));
1250 :
1251 6 : appendPQExpBuffer(str, "SELECT pg_catalog.pg_drop_replication_slot(%s)", slot_name_esc);
1252 :
1253 6 : pg_free(slot_name_esc);
1254 :
1255 6 : pg_log_debug("command is: %s", str->data);
1256 :
1257 6 : if (!dry_run)
1258 : {
1259 2 : res = PQexec(conn, str->data);
1260 2 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1261 : {
1262 0 : pg_log_error("could not drop replication slot \"%s\" on database \"%s\": %s",
1263 : slot_name, dbinfo->dbname, PQresultErrorMessage(res));
1264 0 : dbinfo->made_replslot = false; /* don't try again. */
1265 : }
1266 :
1267 2 : PQclear(res);
1268 : }
1269 :
1270 6 : destroyPQExpBuffer(str);
1271 6 : }
1272 :
1273 : /*
1274 : * Reports a suitable message if pg_ctl fails.
1275 : */
1276 : static void
1277 40 : pg_ctl_status(const char *pg_ctl_cmd, int rc)
1278 : {
1279 40 : if (rc != 0)
1280 : {
1281 0 : if (WIFEXITED(rc))
1282 : {
1283 0 : pg_log_error("pg_ctl failed with exit code %d", WEXITSTATUS(rc));
1284 : }
1285 0 : else if (WIFSIGNALED(rc))
1286 : {
1287 : #if defined(WIN32)
1288 : pg_log_error("pg_ctl was terminated by exception 0x%X",
1289 : WTERMSIG(rc));
1290 : pg_log_error_detail("See C include file \"ntstatus.h\" for a description of the hexadecimal value.");
1291 : #else
1292 0 : pg_log_error("pg_ctl was terminated by signal %d: %s",
1293 : WTERMSIG(rc), pg_strsignal(WTERMSIG(rc)));
1294 : #endif
1295 : }
1296 : else
1297 : {
1298 0 : pg_log_error("pg_ctl exited with unrecognized status %d", rc);
1299 : }
1300 :
1301 0 : pg_log_error_detail("The failed command was: %s", pg_ctl_cmd);
1302 0 : exit(1);
1303 : }
1304 40 : }
1305 :
1306 : static void
1307 20 : start_standby_server(const struct CreateSubscriberOptions *opt, bool restricted_access)
1308 : {
1309 20 : PQExpBuffer pg_ctl_cmd = createPQExpBuffer();
1310 : int rc;
1311 :
1312 20 : appendPQExpBuffer(pg_ctl_cmd, "\"%s\" start -D \"%s\" -s",
1313 : pg_ctl_path, subscriber_dir);
1314 20 : if (restricted_access)
1315 : {
1316 20 : appendPQExpBuffer(pg_ctl_cmd, " -o \"-p %s\"", opt->sub_port);
1317 : #if !defined(WIN32)
1318 :
1319 : /*
1320 : * An empty listen_addresses list means the server does not listen on
1321 : * any IP interfaces; only Unix-domain sockets can be used to connect
1322 : * to the server. Prevent external connections to minimize the chance
1323 : * of failure.
1324 : */
1325 20 : appendPQExpBufferStr(pg_ctl_cmd, " -o \"-c listen_addresses='' -c unix_socket_permissions=0700");
1326 20 : if (opt->socket_dir)
1327 20 : appendPQExpBuffer(pg_ctl_cmd, " -c unix_socket_directories='%s'",
1328 : opt->socket_dir);
1329 20 : appendPQExpBufferChar(pg_ctl_cmd, '"');
1330 : #endif
1331 : }
1332 20 : if (opt->config_file != NULL)
1333 0 : appendPQExpBuffer(pg_ctl_cmd, " -o \"-c config_file=%s\"",
1334 : opt->config_file);
1335 20 : pg_log_debug("pg_ctl command is: %s", pg_ctl_cmd->data);
1336 20 : rc = system(pg_ctl_cmd->data);
1337 20 : pg_ctl_status(pg_ctl_cmd->data, rc);
1338 20 : standby_running = true;
1339 20 : destroyPQExpBuffer(pg_ctl_cmd);
1340 20 : pg_log_info("server was started");
1341 20 : }
1342 :
1343 : static void
1344 20 : stop_standby_server(const char *datadir)
1345 : {
1346 : char *pg_ctl_cmd;
1347 : int rc;
1348 :
1349 20 : pg_ctl_cmd = psprintf("\"%s\" stop -D \"%s\" -s", pg_ctl_path,
1350 : datadir);
1351 20 : pg_log_debug("pg_ctl command is: %s", pg_ctl_cmd);
1352 20 : rc = system(pg_ctl_cmd);
1353 20 : pg_ctl_status(pg_ctl_cmd, rc);
1354 20 : standby_running = false;
1355 20 : pg_log_info("server was stopped");
1356 20 : }
1357 :
1358 : /*
1359 : * Returns after the server finishes the recovery process.
1360 : *
1361 : * If recovery_timeout option is set, terminate abnormally without finishing
1362 : * the recovery process. By default, it waits forever.
1363 : */
1364 : static void
1365 6 : wait_for_end_recovery(const char *conninfo, const struct CreateSubscriberOptions *opt)
1366 : {
1367 : PGconn *conn;
1368 6 : int status = POSTMASTER_STILL_STARTING;
1369 6 : int timer = 0;
1370 6 : int count = 0; /* number of consecutive connection attempts */
1371 :
1372 : #define NUM_CONN_ATTEMPTS 10
1373 :
1374 6 : pg_log_info("waiting for the target server to reach the consistent state");
1375 :
1376 6 : conn = connect_database(conninfo, true);
1377 :
1378 : for (;;)
1379 18 : {
1380 : PGresult *res;
1381 24 : bool in_recovery = server_is_in_recovery(conn);
1382 :
1383 : /*
1384 : * Does the recovery process finish? In dry run mode, there is no
1385 : * recovery mode. Bail out as the recovery process has ended.
1386 : */
1387 24 : if (!in_recovery || dry_run)
1388 : {
1389 6 : status = POSTMASTER_READY;
1390 6 : recovery_ended = true;
1391 6 : break;
1392 : }
1393 :
1394 : /*
1395 : * If it is still in recovery, make sure the target server is
1396 : * connected to the primary so it can receive the required WAL to
1397 : * finish the recovery process. If it is disconnected try
1398 : * NUM_CONN_ATTEMPTS in a row and bail out if not succeed.
1399 : */
1400 18 : res = PQexec(conn,
1401 : "SELECT 1 FROM pg_catalog.pg_stat_wal_receiver");
1402 18 : if (PQntuples(res) == 0)
1403 : {
1404 0 : if (++count > NUM_CONN_ATTEMPTS)
1405 : {
1406 0 : stop_standby_server(subscriber_dir);
1407 0 : pg_log_error("standby server disconnected from the primary");
1408 0 : break;
1409 : }
1410 : }
1411 : else
1412 18 : count = 0; /* reset counter if it connects again */
1413 :
1414 18 : PQclear(res);
1415 :
1416 : /* Bail out after recovery_timeout seconds if this option is set */
1417 18 : if (opt->recovery_timeout > 0 && timer >= opt->recovery_timeout)
1418 : {
1419 0 : stop_standby_server(subscriber_dir);
1420 0 : pg_log_error("recovery timed out");
1421 0 : disconnect_database(conn, true);
1422 : }
1423 :
1424 : /* Keep waiting */
1425 18 : pg_usleep(WAIT_INTERVAL * USEC_PER_SEC);
1426 :
1427 18 : timer += WAIT_INTERVAL;
1428 : }
1429 :
1430 6 : disconnect_database(conn, false);
1431 :
1432 6 : if (status == POSTMASTER_STILL_STARTING)
1433 0 : pg_fatal("server did not end recovery");
1434 :
1435 6 : pg_log_info("target server reached the consistent state");
1436 6 : pg_log_info_hint("If pg_createsubscriber fails after this point, you must recreate the physical replica before continuing.");
1437 6 : }
1438 :
1439 : /*
1440 : * Create a publication that includes all tables in the database.
1441 : */
1442 : static void
1443 10 : create_publication(PGconn *conn, struct LogicalRepInfo *dbinfo)
1444 : {
1445 10 : PQExpBuffer str = createPQExpBuffer();
1446 : PGresult *res;
1447 : char *ipubname_esc;
1448 : char *spubname_esc;
1449 :
1450 : Assert(conn != NULL);
1451 :
1452 10 : ipubname_esc = PQescapeIdentifier(conn, dbinfo->pubname, strlen(dbinfo->pubname));
1453 10 : spubname_esc = PQescapeLiteral(conn, dbinfo->pubname, strlen(dbinfo->pubname));
1454 :
1455 : /* Check if the publication already exists */
1456 10 : appendPQExpBuffer(str,
1457 : "SELECT 1 FROM pg_catalog.pg_publication "
1458 : "WHERE pubname = %s",
1459 : spubname_esc);
1460 10 : res = PQexec(conn, str->data);
1461 10 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1462 : {
1463 0 : pg_log_error("could not obtain publication information: %s",
1464 : PQresultErrorMessage(res));
1465 0 : disconnect_database(conn, true);
1466 : }
1467 :
1468 10 : if (PQntuples(res) == 1)
1469 : {
1470 : /*
1471 : * Unfortunately, if it reaches this code path, it will always fail
1472 : * (unless you decide to change the existing publication name). That's
1473 : * bad but it is very unlikely that the user will choose a name with
1474 : * pg_createsubscriber_ prefix followed by the exact database oid and
1475 : * a random number.
1476 : */
1477 0 : pg_log_error("publication \"%s\" already exists", dbinfo->pubname);
1478 0 : pg_log_error_hint("Consider renaming this publication before continuing.");
1479 0 : disconnect_database(conn, true);
1480 : }
1481 :
1482 10 : PQclear(res);
1483 10 : resetPQExpBuffer(str);
1484 :
1485 10 : pg_log_info("creating publication \"%s\" on database \"%s\"",
1486 : dbinfo->pubname, dbinfo->dbname);
1487 :
1488 10 : appendPQExpBuffer(str, "CREATE PUBLICATION %s FOR ALL TABLES",
1489 : ipubname_esc);
1490 :
1491 10 : pg_log_debug("command is: %s", str->data);
1492 :
1493 10 : if (!dry_run)
1494 : {
1495 4 : res = PQexec(conn, str->data);
1496 4 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1497 : {
1498 0 : pg_log_error("could not create publication \"%s\" on database \"%s\": %s",
1499 : dbinfo->pubname, dbinfo->dbname, PQresultErrorMessage(res));
1500 0 : disconnect_database(conn, true);
1501 : }
1502 4 : PQclear(res);
1503 : }
1504 :
1505 : /* For cleanup purposes */
1506 10 : dbinfo->made_publication = true;
1507 :
1508 10 : pg_free(ipubname_esc);
1509 10 : pg_free(spubname_esc);
1510 10 : destroyPQExpBuffer(str);
1511 10 : }
1512 :
1513 : /*
1514 : * Remove publication if it couldn't finish all steps.
1515 : */
1516 : static void
1517 10 : drop_publication(PGconn *conn, struct LogicalRepInfo *dbinfo)
1518 : {
1519 10 : PQExpBuffer str = createPQExpBuffer();
1520 : PGresult *res;
1521 : char *pubname_esc;
1522 :
1523 : Assert(conn != NULL);
1524 :
1525 10 : pubname_esc = PQescapeIdentifier(conn, dbinfo->pubname, strlen(dbinfo->pubname));
1526 :
1527 10 : pg_log_info("dropping publication \"%s\" on database \"%s\"",
1528 : dbinfo->pubname, dbinfo->dbname);
1529 :
1530 10 : appendPQExpBuffer(str, "DROP PUBLICATION %s", pubname_esc);
1531 :
1532 10 : pg_free(pubname_esc);
1533 :
1534 10 : pg_log_debug("command is: %s", str->data);
1535 :
1536 10 : if (!dry_run)
1537 : {
1538 4 : res = PQexec(conn, str->data);
1539 4 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1540 : {
1541 0 : pg_log_error("could not drop publication \"%s\" on database \"%s\": %s",
1542 : dbinfo->pubname, dbinfo->dbname, PQresultErrorMessage(res));
1543 0 : dbinfo->made_publication = false; /* don't try again. */
1544 :
1545 : /*
1546 : * Don't disconnect and exit here. This routine is used by primary
1547 : * (cleanup publication / replication slot due to an error) and
1548 : * subscriber (remove the replicated publications). In both cases,
1549 : * it can continue and provide instructions for the user to remove
1550 : * it later if cleanup fails.
1551 : */
1552 : }
1553 4 : PQclear(res);
1554 : }
1555 :
1556 10 : destroyPQExpBuffer(str);
1557 10 : }
1558 :
1559 : /*
1560 : * Create a subscription with some predefined options.
1561 : *
1562 : * A replication slot was already created in a previous step. Let's use it. It
1563 : * is not required to copy data. The subscription will be created but it will
1564 : * not be enabled now. That's because the replication progress must be set and
1565 : * the replication origin name (one of the function arguments) contains the
1566 : * subscription OID in its name. Once the subscription is created,
1567 : * set_replication_progress() can obtain the chosen origin name and set up its
1568 : * initial location.
1569 : */
1570 : static void
1571 10 : create_subscription(PGconn *conn, const struct LogicalRepInfo *dbinfo)
1572 : {
1573 10 : PQExpBuffer str = createPQExpBuffer();
1574 : PGresult *res;
1575 : char *pubname_esc;
1576 : char *subname_esc;
1577 : char *pubconninfo_esc;
1578 : char *replslotname_esc;
1579 :
1580 : Assert(conn != NULL);
1581 :
1582 10 : pubname_esc = PQescapeIdentifier(conn, dbinfo->pubname, strlen(dbinfo->pubname));
1583 10 : subname_esc = PQescapeIdentifier(conn, dbinfo->subname, strlen(dbinfo->subname));
1584 10 : pubconninfo_esc = PQescapeLiteral(conn, dbinfo->pubconninfo, strlen(dbinfo->pubconninfo));
1585 10 : replslotname_esc = PQescapeLiteral(conn, dbinfo->replslotname, strlen(dbinfo->replslotname));
1586 :
1587 10 : pg_log_info("creating subscription \"%s\" on database \"%s\"",
1588 : dbinfo->subname, dbinfo->dbname);
1589 :
1590 10 : appendPQExpBuffer(str,
1591 : "CREATE SUBSCRIPTION %s CONNECTION %s PUBLICATION %s "
1592 : "WITH (create_slot = false, enabled = false, "
1593 : "slot_name = %s, copy_data = false)",
1594 : subname_esc, pubconninfo_esc, pubname_esc, replslotname_esc);
1595 :
1596 10 : pg_free(pubname_esc);
1597 10 : pg_free(subname_esc);
1598 10 : pg_free(pubconninfo_esc);
1599 10 : pg_free(replslotname_esc);
1600 :
1601 10 : pg_log_debug("command is: %s", str->data);
1602 :
1603 10 : if (!dry_run)
1604 : {
1605 4 : res = PQexec(conn, str->data);
1606 4 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1607 : {
1608 0 : pg_log_error("could not create subscription \"%s\" on database \"%s\": %s",
1609 : dbinfo->subname, dbinfo->dbname, PQresultErrorMessage(res));
1610 0 : disconnect_database(conn, true);
1611 : }
1612 4 : PQclear(res);
1613 : }
1614 :
1615 10 : destroyPQExpBuffer(str);
1616 10 : }
1617 :
1618 : /*
1619 : * Sets the replication progress to the consistent LSN.
1620 : *
1621 : * The subscriber caught up to the consistent LSN provided by the last
1622 : * replication slot that was created. The goal is to set up the initial
1623 : * location for the logical replication that is the exact LSN that the
1624 : * subscriber was promoted. Once the subscription is enabled it will start
1625 : * streaming from that location onwards. In dry run mode, the subscription OID
1626 : * and LSN are set to invalid values for printing purposes.
1627 : */
1628 : static void
1629 10 : set_replication_progress(PGconn *conn, const struct LogicalRepInfo *dbinfo, const char *lsn)
1630 : {
1631 10 : PQExpBuffer str = createPQExpBuffer();
1632 : PGresult *res;
1633 : Oid suboid;
1634 : char *subname;
1635 : char *dbname;
1636 : char *originname;
1637 : char *lsnstr;
1638 :
1639 : Assert(conn != NULL);
1640 :
1641 10 : subname = PQescapeLiteral(conn, dbinfo->subname, strlen(dbinfo->subname));
1642 10 : dbname = PQescapeLiteral(conn, dbinfo->dbname, strlen(dbinfo->dbname));
1643 :
1644 10 : appendPQExpBuffer(str,
1645 : "SELECT s.oid FROM pg_catalog.pg_subscription s "
1646 : "INNER JOIN pg_catalog.pg_database d ON (s.subdbid = d.oid) "
1647 : "WHERE s.subname = %s AND d.datname = %s",
1648 : subname, dbname);
1649 :
1650 10 : res = PQexec(conn, str->data);
1651 10 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1652 : {
1653 0 : pg_log_error("could not obtain subscription OID: %s",
1654 : PQresultErrorMessage(res));
1655 0 : disconnect_database(conn, true);
1656 : }
1657 :
1658 10 : if (PQntuples(res) != 1 && !dry_run)
1659 : {
1660 0 : pg_log_error("could not obtain subscription OID: got %d rows, expected %d row",
1661 : PQntuples(res), 1);
1662 0 : disconnect_database(conn, true);
1663 : }
1664 :
1665 10 : if (dry_run)
1666 : {
1667 6 : suboid = InvalidOid;
1668 6 : lsnstr = psprintf("%X/%X", LSN_FORMAT_ARGS((XLogRecPtr) InvalidXLogRecPtr));
1669 : }
1670 : else
1671 : {
1672 4 : suboid = strtoul(PQgetvalue(res, 0, 0), NULL, 10);
1673 4 : lsnstr = psprintf("%s", lsn);
1674 : }
1675 :
1676 10 : PQclear(res);
1677 :
1678 : /*
1679 : * The origin name is defined as pg_%u. %u is the subscription OID. See
1680 : * ApplyWorkerMain().
1681 : */
1682 10 : originname = psprintf("pg_%u", suboid);
1683 :
1684 10 : pg_log_info("setting the replication progress (node name \"%s\" ; LSN %s) on database \"%s\"",
1685 : originname, lsnstr, dbinfo->dbname);
1686 :
1687 10 : resetPQExpBuffer(str);
1688 10 : appendPQExpBuffer(str,
1689 : "SELECT pg_catalog.pg_replication_origin_advance('%s', '%s')",
1690 : originname, lsnstr);
1691 :
1692 10 : pg_log_debug("command is: %s", str->data);
1693 :
1694 10 : if (!dry_run)
1695 : {
1696 4 : res = PQexec(conn, str->data);
1697 4 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1698 : {
1699 0 : pg_log_error("could not set replication progress for the subscription \"%s\": %s",
1700 : dbinfo->subname, PQresultErrorMessage(res));
1701 0 : disconnect_database(conn, true);
1702 : }
1703 4 : PQclear(res);
1704 : }
1705 :
1706 10 : pg_free(subname);
1707 10 : pg_free(dbname);
1708 10 : pg_free(originname);
1709 10 : pg_free(lsnstr);
1710 10 : destroyPQExpBuffer(str);
1711 10 : }
1712 :
1713 : /*
1714 : * Enables the subscription.
1715 : *
1716 : * The subscription was created in a previous step but it was disabled. After
1717 : * adjusting the initial logical replication location, enable the subscription.
1718 : */
1719 : static void
1720 10 : enable_subscription(PGconn *conn, const struct LogicalRepInfo *dbinfo)
1721 : {
1722 10 : PQExpBuffer str = createPQExpBuffer();
1723 : PGresult *res;
1724 : char *subname;
1725 :
1726 : Assert(conn != NULL);
1727 :
1728 10 : subname = PQescapeIdentifier(conn, dbinfo->subname, strlen(dbinfo->subname));
1729 :
1730 10 : pg_log_info("enabling subscription \"%s\" on database \"%s\"",
1731 : dbinfo->subname, dbinfo->dbname);
1732 :
1733 10 : appendPQExpBuffer(str, "ALTER SUBSCRIPTION %s ENABLE", subname);
1734 :
1735 10 : pg_log_debug("command is: %s", str->data);
1736 :
1737 10 : if (!dry_run)
1738 : {
1739 4 : res = PQexec(conn, str->data);
1740 4 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1741 : {
1742 0 : pg_log_error("could not enable subscription \"%s\": %s",
1743 : dbinfo->subname, PQresultErrorMessage(res));
1744 0 : disconnect_database(conn, true);
1745 : }
1746 :
1747 4 : PQclear(res);
1748 : }
1749 :
1750 10 : pg_free(subname);
1751 10 : destroyPQExpBuffer(str);
1752 10 : }
1753 :
1754 : int
1755 40 : main(int argc, char **argv)
1756 : {
1757 : static struct option long_options[] =
1758 : {
1759 : {"database", required_argument, NULL, 'd'},
1760 : {"pgdata", required_argument, NULL, 'D'},
1761 : {"dry-run", no_argument, NULL, 'n'},
1762 : {"subscriber-port", required_argument, NULL, 'p'},
1763 : {"publisher-server", required_argument, NULL, 'P'},
1764 : {"socket-directory", required_argument, NULL, 's'},
1765 : {"recovery-timeout", required_argument, NULL, 't'},
1766 : {"subscriber-username", required_argument, NULL, 'U'},
1767 : {"verbose", no_argument, NULL, 'v'},
1768 : {"version", no_argument, NULL, 'V'},
1769 : {"help", no_argument, NULL, '?'},
1770 : {"config-file", required_argument, NULL, 1},
1771 : {"publication", required_argument, NULL, 2},
1772 : {"replication-slot", required_argument, NULL, 3},
1773 : {"subscription", required_argument, NULL, 4},
1774 : {NULL, 0, NULL, 0}
1775 : };
1776 :
1777 40 : struct CreateSubscriberOptions opt = {0};
1778 :
1779 : int c;
1780 : int option_index;
1781 :
1782 : char *pub_base_conninfo;
1783 : char *sub_base_conninfo;
1784 40 : char *dbname_conninfo = NULL;
1785 :
1786 : uint64 pub_sysid;
1787 : uint64 sub_sysid;
1788 : struct stat statbuf;
1789 :
1790 : char *consistent_lsn;
1791 :
1792 : char pidfile[MAXPGPATH];
1793 :
1794 40 : pg_logging_init(argv[0]);
1795 40 : pg_logging_set_level(PG_LOG_WARNING);
1796 40 : progname = get_progname(argv[0]);
1797 40 : set_pglocale_pgservice(argv[0], PG_TEXTDOMAIN("pg_createsubscriber"));
1798 :
1799 40 : if (argc > 1)
1800 : {
1801 38 : if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
1802 : {
1803 2 : usage();
1804 2 : exit(0);
1805 : }
1806 36 : else if (strcmp(argv[1], "-V") == 0
1807 36 : || strcmp(argv[1], "--version") == 0)
1808 : {
1809 2 : puts("pg_createsubscriber (PostgreSQL) " PG_VERSION);
1810 2 : exit(0);
1811 : }
1812 : }
1813 :
1814 : /* Default settings */
1815 36 : subscriber_dir = NULL;
1816 36 : opt.config_file = NULL;
1817 36 : opt.pub_conninfo_str = NULL;
1818 36 : opt.socket_dir = NULL;
1819 36 : opt.sub_port = DEFAULT_SUB_PORT;
1820 36 : opt.sub_username = NULL;
1821 36 : opt.database_names = (SimpleStringList)
1822 : {
1823 : 0
1824 : };
1825 36 : opt.recovery_timeout = 0;
1826 :
1827 : /*
1828 : * Don't allow it to be run as root. It uses pg_ctl which does not allow
1829 : * it either.
1830 : */
1831 : #ifndef WIN32
1832 36 : if (geteuid() == 0)
1833 : {
1834 0 : pg_log_error("cannot be executed by \"root\"");
1835 0 : pg_log_error_hint("You must run %s as the PostgreSQL superuser.",
1836 : progname);
1837 0 : exit(1);
1838 : }
1839 : #endif
1840 :
1841 36 : get_restricted_token();
1842 :
1843 264 : while ((c = getopt_long(argc, argv, "d:D:np:P:s:t:U:v",
1844 : long_options, &option_index)) != -1)
1845 : {
1846 234 : switch (c)
1847 : {
1848 48 : case 'd':
1849 48 : if (!simple_string_list_member(&opt.database_names, optarg))
1850 : {
1851 46 : simple_string_list_append(&opt.database_names, optarg);
1852 46 : num_dbs++;
1853 : }
1854 : else
1855 : {
1856 2 : pg_log_error("duplicate database \"%s\"", optarg);
1857 2 : exit(1);
1858 : }
1859 46 : break;
1860 32 : case 'D':
1861 32 : subscriber_dir = pg_strdup(optarg);
1862 32 : canonicalize_path(subscriber_dir);
1863 32 : break;
1864 14 : case 'n':
1865 14 : dry_run = true;
1866 14 : break;
1867 18 : case 'p':
1868 18 : opt.sub_port = pg_strdup(optarg);
1869 18 : break;
1870 30 : case 'P':
1871 30 : opt.pub_conninfo_str = pg_strdup(optarg);
1872 30 : break;
1873 18 : case 's':
1874 18 : opt.socket_dir = pg_strdup(optarg);
1875 18 : canonicalize_path(opt.socket_dir);
1876 18 : break;
1877 0 : case 't':
1878 0 : opt.recovery_timeout = atoi(optarg);
1879 0 : break;
1880 0 : case 'U':
1881 0 : opt.sub_username = pg_strdup(optarg);
1882 0 : break;
1883 32 : case 'v':
1884 32 : pg_logging_increase_verbosity();
1885 32 : break;
1886 0 : case 1:
1887 0 : opt.config_file = pg_strdup(optarg);
1888 0 : break;
1889 22 : case 2:
1890 22 : if (!simple_string_list_member(&opt.pub_names, optarg))
1891 : {
1892 20 : simple_string_list_append(&opt.pub_names, optarg);
1893 20 : num_pubs++;
1894 : }
1895 : else
1896 : {
1897 2 : pg_log_error("duplicate publication \"%s\"", optarg);
1898 2 : exit(1);
1899 : }
1900 20 : break;
1901 8 : case 3:
1902 8 : if (!simple_string_list_member(&opt.replslot_names, optarg))
1903 : {
1904 8 : simple_string_list_append(&opt.replslot_names, optarg);
1905 8 : num_replslots++;
1906 : }
1907 : else
1908 : {
1909 0 : pg_log_error("duplicate replication slot \"%s\"", optarg);
1910 0 : exit(1);
1911 : }
1912 8 : break;
1913 10 : case 4:
1914 10 : if (!simple_string_list_member(&opt.sub_names, optarg))
1915 : {
1916 10 : simple_string_list_append(&opt.sub_names, optarg);
1917 10 : num_subs++;
1918 : }
1919 : else
1920 : {
1921 0 : pg_log_error("duplicate subscription \"%s\"", optarg);
1922 0 : exit(1);
1923 : }
1924 10 : break;
1925 2 : default:
1926 : /* getopt_long already emitted a complaint */
1927 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
1928 2 : exit(1);
1929 : }
1930 : }
1931 :
1932 : /* Any non-option arguments? */
1933 30 : if (optind < argc)
1934 : {
1935 0 : pg_log_error("too many command-line arguments (first is \"%s\")",
1936 : argv[optind]);
1937 0 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
1938 0 : exit(1);
1939 : }
1940 :
1941 : /* Required arguments */
1942 30 : if (subscriber_dir == NULL)
1943 : {
1944 2 : pg_log_error("no subscriber data directory specified");
1945 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
1946 2 : exit(1);
1947 : }
1948 :
1949 : /* If socket directory is not provided, use the current directory */
1950 28 : if (opt.socket_dir == NULL)
1951 : {
1952 : char cwd[MAXPGPATH];
1953 :
1954 10 : if (!getcwd(cwd, MAXPGPATH))
1955 0 : pg_fatal("could not determine current directory");
1956 10 : opt.socket_dir = pg_strdup(cwd);
1957 10 : canonicalize_path(opt.socket_dir);
1958 : }
1959 :
1960 : /*
1961 : * Parse connection string. Build a base connection string that might be
1962 : * reused by multiple databases.
1963 : */
1964 28 : if (opt.pub_conninfo_str == NULL)
1965 : {
1966 : /*
1967 : * TODO use primary_conninfo (if available) from subscriber and
1968 : * extract publisher connection string. Assume that there are
1969 : * identical entries for physical and logical replication. If there is
1970 : * not, we would fail anyway.
1971 : */
1972 2 : pg_log_error("no publisher connection string specified");
1973 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
1974 2 : exit(1);
1975 : }
1976 26 : pg_log_info("validating connection string on publisher");
1977 26 : pub_base_conninfo = get_base_conninfo(opt.pub_conninfo_str,
1978 : &dbname_conninfo);
1979 26 : if (pub_base_conninfo == NULL)
1980 0 : exit(1);
1981 :
1982 26 : pg_log_info("validating connection string on subscriber");
1983 26 : sub_base_conninfo = get_sub_conninfo(&opt);
1984 :
1985 26 : if (opt.database_names.head == NULL)
1986 : {
1987 4 : pg_log_info("no database was specified");
1988 :
1989 : /*
1990 : * If --database option is not provided, try to obtain the dbname from
1991 : * the publisher conninfo. If dbname parameter is not available, error
1992 : * out.
1993 : */
1994 4 : if (dbname_conninfo)
1995 : {
1996 2 : simple_string_list_append(&opt.database_names, dbname_conninfo);
1997 2 : num_dbs++;
1998 :
1999 2 : pg_log_info("database \"%s\" was extracted from the publisher connection string",
2000 : dbname_conninfo);
2001 : }
2002 : else
2003 : {
2004 2 : pg_log_error("no database name specified");
2005 2 : pg_log_error_hint("Try \"%s --help\" for more information.",
2006 : progname);
2007 2 : exit(1);
2008 : }
2009 : }
2010 :
2011 : /* Number of object names must match number of databases */
2012 24 : if (num_pubs > 0 && num_pubs != num_dbs)
2013 : {
2014 2 : pg_log_error("wrong number of publication names");
2015 2 : pg_log_error_hint("Number of publication names (%d) must match number of database names (%d).",
2016 : num_pubs, num_dbs);
2017 2 : exit(1);
2018 : }
2019 22 : if (num_subs > 0 && num_subs != num_dbs)
2020 : {
2021 2 : pg_log_error("wrong number of subscription names");
2022 2 : pg_log_error_hint("Number of subscription names (%d) must match number of database names (%d).",
2023 : num_subs, num_dbs);
2024 2 : exit(1);
2025 : }
2026 20 : if (num_replslots > 0 && num_replslots != num_dbs)
2027 : {
2028 2 : pg_log_error("wrong number of replication slot names");
2029 2 : pg_log_error_hint("Number of replication slot names (%d) must match number of database names (%d).",
2030 : num_replslots, num_dbs);
2031 2 : exit(1);
2032 : }
2033 :
2034 : /* Get the absolute path of pg_ctl and pg_resetwal on the subscriber */
2035 18 : pg_ctl_path = get_exec_path(argv[0], "pg_ctl");
2036 18 : pg_resetwal_path = get_exec_path(argv[0], "pg_resetwal");
2037 :
2038 : /* Rudimentary check for a data directory */
2039 18 : check_data_directory(subscriber_dir);
2040 :
2041 : /*
2042 : * Store database information for publisher and subscriber. It should be
2043 : * called before atexit() because its return is used in the
2044 : * cleanup_objects_atexit().
2045 : */
2046 18 : dbinfo = store_pub_sub_info(&opt, pub_base_conninfo, sub_base_conninfo);
2047 :
2048 : /* Register a function to clean up objects in case of failure */
2049 18 : atexit(cleanup_objects_atexit);
2050 :
2051 : /*
2052 : * Check if the subscriber data directory has the same system identifier
2053 : * than the publisher data directory.
2054 : */
2055 18 : pub_sysid = get_primary_sysid(dbinfo[0].pubconninfo);
2056 18 : sub_sysid = get_standby_sysid(subscriber_dir);
2057 18 : if (pub_sysid != sub_sysid)
2058 2 : pg_fatal("subscriber data directory is not a copy of the source database cluster");
2059 :
2060 : /* Subscriber PID file */
2061 16 : snprintf(pidfile, MAXPGPATH, "%s/postmaster.pid", subscriber_dir);
2062 :
2063 : /*
2064 : * The standby server must not be running. If the server is started under
2065 : * service manager and pg_createsubscriber stops it, the service manager
2066 : * might react to this action and start the server again. Therefore,
2067 : * refuse to proceed if the server is running to avoid possible failures.
2068 : */
2069 16 : if (stat(pidfile, &statbuf) == 0)
2070 : {
2071 2 : pg_log_error("standby is up and running");
2072 2 : pg_log_error_hint("Stop the standby and try again.");
2073 2 : exit(1);
2074 : }
2075 :
2076 : /*
2077 : * Start a short-lived standby server with temporary parameters (provided
2078 : * by command-line options). The goal is to avoid connections during the
2079 : * transformation steps.
2080 : */
2081 14 : pg_log_info("starting the standby with command-line options");
2082 14 : start_standby_server(&opt, true);
2083 :
2084 : /* Check if the standby server is ready for logical replication */
2085 14 : check_subscriber(dbinfo);
2086 :
2087 : /*
2088 : * Check if the primary server is ready for logical replication. This
2089 : * routine checks if a replication slot is in use on primary so it relies
2090 : * on check_subscriber() to obtain the primary_slot_name. That's why it is
2091 : * called after it.
2092 : */
2093 10 : check_publisher(dbinfo);
2094 :
2095 : /*
2096 : * Stop the target server. The recovery process requires that the server
2097 : * reaches a consistent state before targeting the recovery stop point.
2098 : * Make sure a consistent state is reached (stop the target server
2099 : * guarantees it) *before* creating the replication slots in
2100 : * setup_publisher().
2101 : */
2102 6 : pg_log_info("stopping the subscriber");
2103 6 : stop_standby_server(subscriber_dir);
2104 :
2105 : /*
2106 : * Create the required objects for each database on publisher. This step
2107 : * is here mainly because if we stop the standby we cannot verify if the
2108 : * primary slot is in use. We could use an extra connection for it but it
2109 : * doesn't seem worth.
2110 : */
2111 6 : consistent_lsn = setup_publisher(dbinfo);
2112 :
2113 : /* Write the required recovery parameters */
2114 6 : setup_recovery(dbinfo, subscriber_dir, consistent_lsn);
2115 :
2116 : /*
2117 : * Start subscriber so the recovery parameters will take effect. Wait
2118 : * until accepting connections.
2119 : */
2120 6 : pg_log_info("starting the subscriber");
2121 6 : start_standby_server(&opt, true);
2122 :
2123 : /* Waiting the subscriber to be promoted */
2124 6 : wait_for_end_recovery(dbinfo[0].subconninfo, &opt);
2125 :
2126 : /*
2127 : * Create the subscription for each database on subscriber. It does not
2128 : * enable it immediately because it needs to adjust the replication start
2129 : * point to the LSN reported by setup_publisher(). It also cleans up
2130 : * publications created by this tool and replication to the standby.
2131 : */
2132 6 : setup_subscriber(dbinfo, consistent_lsn);
2133 :
2134 : /* Remove primary_slot_name if it exists on primary */
2135 6 : drop_primary_replication_slot(dbinfo, primary_slot_name);
2136 :
2137 : /* Stop the subscriber */
2138 6 : pg_log_info("stopping the subscriber");
2139 6 : stop_standby_server(subscriber_dir);
2140 :
2141 : /* Change system identifier from subscriber */
2142 6 : modify_subscriber_sysid(&opt);
2143 :
2144 6 : success = true;
2145 :
2146 6 : pg_log_info("Done!");
2147 :
2148 6 : return 0;
2149 : }
|
