Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pg_createsubscriber.c
4 : * Create a new logical replica from a standby server
5 : *
6 : * Copyright (c) 2024-2026, 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/stat.h>
17 : #include <sys/time.h>
18 : #include <sys/wait.h>
19 : #include <time.h>
20 :
21 : #include "common/connect.h"
22 : #include "common/controldata_utils.h"
23 : #include "common/file_utils.h"
24 : #include "common/logging.h"
25 : #include "common/pg_prng.h"
26 : #include "common/restricted_token.h"
27 : #include "datatype/timestamp.h"
28 : #include "fe_utils/recovery_gen.h"
29 : #include "fe_utils/simple_list.h"
30 : #include "fe_utils/string_utils.h"
31 : #include "fe_utils/version.h"
32 : #include "getopt_long.h"
33 :
34 : #define DEFAULT_SUB_PORT "50432"
35 : #define OBJECTTYPE_PUBLICATIONS 0x0001
36 :
37 : /*
38 : * Configuration files for recovery parameters.
39 : *
40 : * The recovery parameters are set in INCLUDED_CONF_FILE, itself loaded by
41 : * the server through an include_if_exists in postgresql.auto.conf.
42 : *
43 : * INCLUDED_CONF_FILE is renamed to INCLUDED_CONF_FILE_DISABLED when exiting,
44 : * so as the recovery parameters set by this tool never take effect on node
45 : * restart. The contents of INCLUDED_CONF_FILE_DISABLED can be useful for
46 : * debugging.
47 : */
48 : #define PG_AUTOCONF_FILENAME "postgresql.auto.conf"
49 : #define INCLUDED_CONF_FILE "pg_createsubscriber.conf"
50 : #define INCLUDED_CONF_FILE_DISABLED INCLUDED_CONF_FILE ".disabled"
51 :
52 : /* Command-line options */
53 : struct CreateSubscriberOptions
54 : {
55 : char *config_file; /* configuration file */
56 : char *pub_conninfo_str; /* publisher connection string */
57 : char *socket_dir; /* directory for Unix-domain socket, if any */
58 : char *sub_port; /* subscriber port number */
59 : const char *sub_username; /* subscriber username */
60 : bool two_phase; /* enable-two-phase option */
61 : SimpleStringList database_names; /* list of database names */
62 : SimpleStringList pub_names; /* list of publication names */
63 : SimpleStringList sub_names; /* list of subscription names */
64 : SimpleStringList replslot_names; /* list of replication slot names */
65 : int recovery_timeout; /* stop recovery after this time */
66 : bool all_dbs; /* all option */
67 : SimpleStringList objecttypes_to_clean; /* list of object types to cleanup */
68 : };
69 :
70 : /* per-database publication/subscription info */
71 : struct LogicalRepInfo
72 : {
73 : char *dbname; /* database name */
74 : char *pubconninfo; /* publisher connection string */
75 : char *subconninfo; /* subscriber connection string */
76 : char *pubname; /* publication name */
77 : char *subname; /* subscription name */
78 : char *replslotname; /* replication slot name */
79 :
80 : bool made_replslot; /* replication slot was created */
81 : bool made_publication; /* publication was created */
82 : };
83 :
84 : /*
85 : * Information shared across all the databases (or publications and
86 : * subscriptions).
87 : */
88 : struct LogicalRepInfos
89 : {
90 : struct LogicalRepInfo *dbinfo;
91 : bool two_phase; /* enable-two-phase option */
92 : bits32 objecttypes_to_clean; /* flags indicating which object types
93 : * to clean up on subscriber */
94 : };
95 :
96 : static void cleanup_objects_atexit(void);
97 : static void usage(void);
98 : static char *get_base_conninfo(const char *conninfo, char **dbname);
99 : static char *get_sub_conninfo(const struct CreateSubscriberOptions *opt);
100 : static char *get_exec_path(const char *argv0, const char *progname);
101 : static void check_data_directory(const char *datadir);
102 : static char *concat_conninfo_dbname(const char *conninfo, const char *dbname);
103 : static struct LogicalRepInfo *store_pub_sub_info(const struct CreateSubscriberOptions *opt,
104 : const char *pub_base_conninfo,
105 : const char *sub_base_conninfo);
106 : static PGconn *connect_database(const char *conninfo, bool exit_on_error);
107 : static void disconnect_database(PGconn *conn, bool exit_on_error);
108 : static uint64 get_primary_sysid(const char *conninfo);
109 : static uint64 get_standby_sysid(const char *datadir);
110 : static void modify_subscriber_sysid(const struct CreateSubscriberOptions *opt);
111 : static bool server_is_in_recovery(PGconn *conn);
112 : static char *generate_object_name(PGconn *conn);
113 : static void check_publisher(const struct LogicalRepInfo *dbinfo);
114 : static char *setup_publisher(struct LogicalRepInfo *dbinfo);
115 : static void check_subscriber(const struct LogicalRepInfo *dbinfo);
116 : static void setup_subscriber(struct LogicalRepInfo *dbinfo,
117 : const char *consistent_lsn);
118 : static void setup_recovery(const struct LogicalRepInfo *dbinfo, const char *datadir,
119 : const char *lsn);
120 : static void drop_primary_replication_slot(struct LogicalRepInfo *dbinfo,
121 : const char *slotname);
122 : static void drop_failover_replication_slots(struct LogicalRepInfo *dbinfo);
123 : static char *create_logical_replication_slot(PGconn *conn,
124 : struct LogicalRepInfo *dbinfo);
125 : static void drop_replication_slot(PGconn *conn, struct LogicalRepInfo *dbinfo,
126 : const char *slot_name);
127 : static void pg_ctl_status(const char *pg_ctl_cmd, int rc);
128 : static void start_standby_server(const struct CreateSubscriberOptions *opt,
129 : bool restricted_access,
130 : bool restrict_logical_worker);
131 : static void stop_standby_server(const char *datadir);
132 : static void wait_for_end_recovery(const char *conninfo,
133 : const struct CreateSubscriberOptions *opt);
134 : static void create_publication(PGconn *conn, struct LogicalRepInfo *dbinfo);
135 : static bool find_publication(PGconn *conn, const char *pubname, const char *dbname);
136 : static void drop_publication(PGconn *conn, const char *pubname,
137 : const char *dbname, bool *made_publication);
138 : static void check_and_drop_publications(PGconn *conn, struct LogicalRepInfo *dbinfo);
139 : static void create_subscription(PGconn *conn, const struct LogicalRepInfo *dbinfo);
140 : static void set_replication_progress(PGconn *conn, const struct LogicalRepInfo *dbinfo,
141 : const char *lsn);
142 : static void enable_subscription(PGconn *conn, const struct LogicalRepInfo *dbinfo);
143 : static void check_and_drop_existing_subscriptions(PGconn *conn,
144 : const struct LogicalRepInfo *dbinfo);
145 : static void drop_existing_subscription(PGconn *conn, const char *subname,
146 : const char *dbname);
147 : static void get_publisher_databases(struct CreateSubscriberOptions *opt,
148 : bool dbnamespecified);
149 :
150 : #define WAIT_INTERVAL 1 /* 1 second */
151 :
152 : static const char *progname;
153 :
154 : static char *primary_slot_name = NULL;
155 : static bool dry_run = false;
156 :
157 : static bool success = false;
158 :
159 : static struct LogicalRepInfos dbinfos;
160 : static int num_dbs = 0; /* number of specified databases */
161 : static int num_pubs = 0; /* number of specified publications */
162 : static int num_subs = 0; /* number of specified subscriptions */
163 : static int num_replslots = 0; /* number of specified replication slots */
164 :
165 : static pg_prng_state prng_state;
166 :
167 : static char *pg_ctl_path = NULL;
168 : static char *pg_resetwal_path = NULL;
169 :
170 : /* standby / subscriber data directory */
171 : static char *subscriber_dir = NULL;
172 :
173 : static bool recovery_ended = false;
174 : static bool standby_running = false;
175 : static bool recovery_params_set = false;
176 :
177 :
178 : /*
179 : * Clean up objects created by pg_createsubscriber.
180 : *
181 : * Publications and replication slots are created on the primary. Depending
182 : * on the step where it failed, already-created objects should be removed if
183 : * possible (sometimes this won't work due to a connection issue).
184 : * There is no cleanup on the target server *after* its promotion, because any
185 : * failure at this point means recreating the physical replica and starting
186 : * again.
187 : *
188 : * The recovery configuration is always removed, by renaming the included
189 : * configuration file out of the way.
190 : */
191 : static void
192 10 : cleanup_objects_atexit(void)
193 : {
194 : /* Rename the included configuration file, if necessary. */
195 10 : if (recovery_params_set)
196 : {
197 : char conf_filename[MAXPGPATH];
198 : char conf_filename_disabled[MAXPGPATH];
199 :
200 1 : snprintf(conf_filename, MAXPGPATH, "%s/%s", subscriber_dir,
201 : INCLUDED_CONF_FILE);
202 1 : snprintf(conf_filename_disabled, MAXPGPATH, "%s/%s", subscriber_dir,
203 : INCLUDED_CONF_FILE_DISABLED);
204 :
205 1 : if (durable_rename(conf_filename, conf_filename_disabled) != 0)
206 : {
207 : /* durable_rename() has already logged something. */
208 0 : pg_log_warning_hint("A manual removal of the recovery parameters may be required.");
209 : }
210 : }
211 :
212 10 : if (success)
213 4 : return;
214 :
215 : /*
216 : * If the server is promoted, there is no way to use the current setup
217 : * again. Warn the user that a new replication setup should be done before
218 : * trying again.
219 : */
220 6 : if (recovery_ended)
221 : {
222 0 : pg_log_warning("failed after the end of recovery");
223 0 : pg_log_warning_hint("The target server cannot be used as a physical replica anymore. "
224 : "You must recreate the physical replica before continuing.");
225 : }
226 :
227 18 : for (int i = 0; i < num_dbs; i++)
228 : {
229 12 : struct LogicalRepInfo *dbinfo = &dbinfos.dbinfo[i];
230 :
231 12 : if (dbinfo->made_publication || dbinfo->made_replslot)
232 : {
233 : PGconn *conn;
234 :
235 0 : conn = connect_database(dbinfo->pubconninfo, false);
236 0 : if (conn != NULL)
237 : {
238 0 : if (dbinfo->made_publication)
239 0 : drop_publication(conn, dbinfo->pubname, dbinfo->dbname,
240 : &dbinfo->made_publication);
241 0 : if (dbinfo->made_replslot)
242 0 : drop_replication_slot(conn, dbinfo, dbinfo->replslotname);
243 0 : disconnect_database(conn, false);
244 : }
245 : else
246 : {
247 : /*
248 : * If a connection could not be established, inform the user
249 : * that some objects were left on primary and should be
250 : * removed before trying again.
251 : */
252 0 : if (dbinfo->made_publication)
253 : {
254 0 : pg_log_warning("publication \"%s\" created in database \"%s\" on primary was left behind",
255 : dbinfo->pubname,
256 : dbinfo->dbname);
257 0 : pg_log_warning_hint("Drop this publication before trying again.");
258 : }
259 0 : if (dbinfo->made_replslot)
260 : {
261 0 : pg_log_warning("replication slot \"%s\" created in database \"%s\" on primary was left behind",
262 : dbinfo->replslotname,
263 : dbinfo->dbname);
264 0 : pg_log_warning_hint("Drop this replication slot soon to avoid retention of WAL files.");
265 : }
266 : }
267 : }
268 : }
269 :
270 6 : if (standby_running)
271 4 : stop_standby_server(subscriber_dir);
272 : }
273 :
274 : static void
275 1 : usage(void)
276 : {
277 1 : printf(_("%s creates a new logical replica from a standby server.\n\n"),
278 : progname);
279 1 : printf(_("Usage:\n"));
280 1 : printf(_(" %s [OPTION]...\n"), progname);
281 1 : printf(_("\nOptions:\n"));
282 1 : printf(_(" -a, --all create subscriptions for all databases except template\n"
283 : " databases and databases that don't allow connections\n"));
284 1 : printf(_(" -d, --database=DBNAME database in which to create a subscription\n"));
285 1 : printf(_(" -D, --pgdata=DATADIR location for the subscriber data directory\n"));
286 1 : printf(_(" -n, --dry-run dry run, just show what would be done\n"));
287 1 : printf(_(" -p, --subscriber-port=PORT subscriber port number (default %s)\n"), DEFAULT_SUB_PORT);
288 1 : printf(_(" -P, --publisher-server=CONNSTR publisher connection string\n"));
289 1 : printf(_(" -s, --socketdir=DIR socket directory to use (default current dir.)\n"));
290 1 : printf(_(" -t, --recovery-timeout=SECS seconds to wait for recovery to end\n"));
291 1 : printf(_(" -T, --enable-two-phase enable two-phase commit for all subscriptions\n"));
292 1 : printf(_(" -U, --subscriber-username=NAME user name for subscriber connection\n"));
293 1 : printf(_(" -v, --verbose output verbose messages\n"));
294 1 : printf(_(" --clean=OBJECTTYPE drop all objects of the specified type from specified\n"
295 : " databases on the subscriber; accepts: \"%s\"\n"), "publications");
296 1 : printf(_(" --config-file=FILENAME use specified main server configuration\n"
297 : " file when running target cluster\n"));
298 1 : printf(_(" --publication=NAME publication name\n"));
299 1 : printf(_(" --replication-slot=NAME replication slot name\n"));
300 1 : printf(_(" --subscription=NAME subscription name\n"));
301 1 : printf(_(" -V, --version output version information, then exit\n"));
302 1 : printf(_(" -?, --help show this help, then exit\n"));
303 1 : printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT);
304 1 : printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL);
305 1 : }
306 :
307 : /*
308 : * Subroutine to append "keyword=value" to a connection string,
309 : * with proper quoting of the value. (We assume keywords don't need that.)
310 : */
311 : static void
312 107 : appendConnStrItem(PQExpBuffer buf, const char *keyword, const char *val)
313 : {
314 107 : if (buf->len > 0)
315 79 : appendPQExpBufferChar(buf, ' ');
316 107 : appendPQExpBufferStr(buf, keyword);
317 107 : appendPQExpBufferChar(buf, '=');
318 107 : appendConnStrVal(buf, val);
319 107 : }
320 :
321 : /*
322 : * Validate a connection string. Returns a base connection string that is a
323 : * connection string without a database name.
324 : *
325 : * Since we might process multiple databases, each database name will be
326 : * appended to this base connection string to provide a final connection
327 : * string. If the second argument (dbname) is not null, returns dbname if the
328 : * provided connection string contains it.
329 : *
330 : * It is the caller's responsibility to free the returned connection string and
331 : * dbname.
332 : */
333 : static char *
334 14 : get_base_conninfo(const char *conninfo, char **dbname)
335 : {
336 : PQExpBuffer buf;
337 : PQconninfoOption *conn_opts;
338 : PQconninfoOption *conn_opt;
339 14 : char *errmsg = NULL;
340 : char *ret;
341 :
342 14 : conn_opts = PQconninfoParse(conninfo, &errmsg);
343 14 : if (conn_opts == NULL)
344 : {
345 0 : pg_log_error("could not parse connection string: %s", errmsg);
346 0 : PQfreemem(errmsg);
347 0 : return NULL;
348 : }
349 :
350 14 : buf = createPQExpBuffer();
351 728 : for (conn_opt = conn_opts; conn_opt->keyword != NULL; conn_opt++)
352 : {
353 714 : if (conn_opt->val != NULL && conn_opt->val[0] != '\0')
354 : {
355 33 : if (strcmp(conn_opt->keyword, "dbname") == 0)
356 : {
357 9 : if (dbname)
358 9 : *dbname = pg_strdup(conn_opt->val);
359 9 : continue;
360 : }
361 24 : appendConnStrItem(buf, conn_opt->keyword, conn_opt->val);
362 : }
363 : }
364 :
365 14 : ret = pg_strdup(buf->data);
366 :
367 14 : destroyPQExpBuffer(buf);
368 14 : PQconninfoFree(conn_opts);
369 :
370 14 : return ret;
371 : }
372 :
373 : /*
374 : * Build a subscriber connection string. Only a few parameters are supported
375 : * since it starts a server with restricted access.
376 : */
377 : static char *
378 14 : get_sub_conninfo(const struct CreateSubscriberOptions *opt)
379 : {
380 14 : PQExpBuffer buf = createPQExpBuffer();
381 : char *ret;
382 :
383 14 : appendConnStrItem(buf, "port", opt->sub_port);
384 : #if !defined(WIN32)
385 14 : appendConnStrItem(buf, "host", opt->socket_dir);
386 : #endif
387 14 : if (opt->sub_username != NULL)
388 0 : appendConnStrItem(buf, "user", opt->sub_username);
389 14 : appendConnStrItem(buf, "fallback_application_name", progname);
390 :
391 14 : ret = pg_strdup(buf->data);
392 :
393 14 : destroyPQExpBuffer(buf);
394 :
395 14 : return ret;
396 : }
397 :
398 : /*
399 : * Verify if a PostgreSQL binary (progname) is available in the same directory as
400 : * pg_createsubscriber and it has the same version. It returns the absolute
401 : * path of the progname.
402 : */
403 : static char *
404 20 : get_exec_path(const char *argv0, const char *progname)
405 : {
406 : char *versionstr;
407 : char *exec_path;
408 : int ret;
409 :
410 20 : versionstr = psprintf("%s (PostgreSQL) %s\n", progname, PG_VERSION);
411 20 : exec_path = pg_malloc(MAXPGPATH);
412 20 : ret = find_other_exec(argv0, progname, versionstr, exec_path);
413 :
414 20 : if (ret < 0)
415 : {
416 : char full_path[MAXPGPATH];
417 :
418 0 : if (find_my_exec(argv0, full_path) < 0)
419 0 : strlcpy(full_path, progname, sizeof(full_path));
420 :
421 0 : if (ret == -1)
422 0 : pg_fatal("program \"%s\" is needed by %s but was not found in the same directory as \"%s\"",
423 : progname, "pg_createsubscriber", full_path);
424 : else
425 0 : pg_fatal("program \"%s\" was found by \"%s\" but was not the same version as %s",
426 : progname, full_path, "pg_createsubscriber");
427 : }
428 :
429 20 : pg_log_debug("%s path is: %s", progname, exec_path);
430 :
431 20 : return exec_path;
432 : }
433 :
434 : /*
435 : * Is it a cluster directory? These are preliminary checks. It is far from
436 : * making an accurate check. If it is not a clone from the publisher, it will
437 : * eventually fail in a future step.
438 : */
439 : static void
440 10 : check_data_directory(const char *datadir)
441 : {
442 : struct stat statbuf;
443 : uint32 major_version;
444 : char *version_str;
445 :
446 10 : pg_log_info("checking if directory \"%s\" is a cluster data directory",
447 : datadir);
448 :
449 10 : if (stat(datadir, &statbuf) != 0)
450 : {
451 0 : if (errno == ENOENT)
452 0 : pg_fatal("data directory \"%s\" does not exist", datadir);
453 : else
454 0 : pg_fatal("could not access directory \"%s\": %m", datadir);
455 : }
456 :
457 : /*
458 : * Retrieve the contents of this cluster's PG_VERSION. We require
459 : * compatibility with the same major version as the one this tool is
460 : * compiled with.
461 : */
462 10 : major_version = GET_PG_MAJORVERSION_NUM(get_pg_version(datadir, &version_str));
463 10 : if (major_version != PG_MAJORVERSION_NUM)
464 : {
465 0 : pg_log_error("data directory is of wrong version");
466 0 : pg_log_error_detail("File \"%s\" contains \"%s\", which is not compatible with this program's version \"%s\".",
467 : "PG_VERSION", version_str, PG_MAJORVERSION);
468 0 : exit(1);
469 : }
470 10 : }
471 :
472 : /*
473 : * Append database name into a base connection string.
474 : *
475 : * dbname is the only parameter that changes so it is not included in the base
476 : * connection string. This function concatenates dbname to build a "real"
477 : * connection string.
478 : */
479 : static char *
480 41 : concat_conninfo_dbname(const char *conninfo, const char *dbname)
481 : {
482 41 : PQExpBuffer buf = createPQExpBuffer();
483 : char *ret;
484 :
485 : Assert(conninfo != NULL);
486 :
487 41 : appendPQExpBufferStr(buf, conninfo);
488 41 : appendConnStrItem(buf, "dbname", dbname);
489 :
490 41 : ret = pg_strdup(buf->data);
491 41 : destroyPQExpBuffer(buf);
492 :
493 41 : return ret;
494 : }
495 :
496 : /*
497 : * Store publication and subscription information.
498 : *
499 : * If publication, replication slot and subscription names were specified,
500 : * store it here. Otherwise, a generated name will be assigned to the object in
501 : * setup_publisher().
502 : */
503 : static struct LogicalRepInfo *
504 10 : store_pub_sub_info(const struct CreateSubscriberOptions *opt,
505 : const char *pub_base_conninfo,
506 : const char *sub_base_conninfo)
507 : {
508 : struct LogicalRepInfo *dbinfo;
509 10 : SimpleStringListCell *pubcell = NULL;
510 10 : SimpleStringListCell *subcell = NULL;
511 10 : SimpleStringListCell *replslotcell = NULL;
512 10 : int i = 0;
513 :
514 10 : dbinfo = pg_malloc_array(struct LogicalRepInfo, num_dbs);
515 :
516 10 : if (num_pubs > 0)
517 2 : pubcell = opt->pub_names.head;
518 10 : if (num_subs > 0)
519 1 : subcell = opt->sub_names.head;
520 10 : if (num_replslots > 0)
521 2 : replslotcell = opt->replslot_names.head;
522 :
523 30 : for (SimpleStringListCell *cell = opt->database_names.head; cell; cell = cell->next)
524 : {
525 : char *conninfo;
526 :
527 : /* Fill publisher attributes */
528 20 : conninfo = concat_conninfo_dbname(pub_base_conninfo, cell->val);
529 20 : dbinfo[i].pubconninfo = conninfo;
530 20 : dbinfo[i].dbname = cell->val;
531 20 : if (num_pubs > 0)
532 4 : dbinfo[i].pubname = pubcell->val;
533 : else
534 16 : dbinfo[i].pubname = NULL;
535 20 : if (num_replslots > 0)
536 3 : dbinfo[i].replslotname = replslotcell->val;
537 : else
538 17 : dbinfo[i].replslotname = NULL;
539 20 : dbinfo[i].made_replslot = false;
540 20 : dbinfo[i].made_publication = false;
541 : /* Fill subscriber attributes */
542 20 : conninfo = concat_conninfo_dbname(sub_base_conninfo, cell->val);
543 20 : dbinfo[i].subconninfo = conninfo;
544 20 : if (num_subs > 0)
545 2 : dbinfo[i].subname = subcell->val;
546 : else
547 18 : dbinfo[i].subname = NULL;
548 : /* Other fields will be filled later */
549 :
550 20 : pg_log_debug("publisher(%d): publication: %s ; replication slot: %s ; connection string: %s", i,
551 : dbinfo[i].pubname ? dbinfo[i].pubname : "(auto)",
552 : dbinfo[i].replslotname ? dbinfo[i].replslotname : "(auto)",
553 : dbinfo[i].pubconninfo);
554 20 : pg_log_debug("subscriber(%d): subscription: %s ; connection string: %s, two_phase: %s", i,
555 : dbinfo[i].subname ? dbinfo[i].subname : "(auto)",
556 : dbinfo[i].subconninfo,
557 : dbinfos.two_phase ? "true" : "false");
558 :
559 20 : if (num_pubs > 0)
560 4 : pubcell = pubcell->next;
561 20 : if (num_subs > 0)
562 2 : subcell = subcell->next;
563 20 : if (num_replslots > 0)
564 3 : replslotcell = replslotcell->next;
565 :
566 20 : i++;
567 : }
568 :
569 10 : return dbinfo;
570 : }
571 :
572 : /*
573 : * Open a new connection. If exit_on_error is true, it has an undesired
574 : * condition and it should exit immediately.
575 : */
576 : static PGconn *
577 57 : connect_database(const char *conninfo, bool exit_on_error)
578 : {
579 : PGconn *conn;
580 : PGresult *res;
581 :
582 57 : conn = PQconnectdb(conninfo);
583 57 : if (PQstatus(conn) != CONNECTION_OK)
584 : {
585 0 : pg_log_error("connection to database failed: %s",
586 : PQerrorMessage(conn));
587 0 : PQfinish(conn);
588 :
589 0 : if (exit_on_error)
590 0 : exit(1);
591 0 : return NULL;
592 : }
593 :
594 : /* Secure search_path */
595 57 : res = PQexec(conn, ALWAYS_SECURE_SEARCH_PATH_SQL);
596 57 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
597 : {
598 0 : pg_log_error("could not clear \"search_path\": %s",
599 : PQresultErrorMessage(res));
600 0 : PQclear(res);
601 0 : PQfinish(conn);
602 :
603 0 : if (exit_on_error)
604 0 : exit(1);
605 0 : return NULL;
606 : }
607 57 : PQclear(res);
608 :
609 57 : return conn;
610 : }
611 :
612 : /*
613 : * Close the connection. If exit_on_error is true, it has an undesired
614 : * condition and it should exit immediately.
615 : */
616 : static void
617 57 : disconnect_database(PGconn *conn, bool exit_on_error)
618 : {
619 : Assert(conn != NULL);
620 :
621 57 : PQfinish(conn);
622 :
623 57 : if (exit_on_error)
624 2 : exit(1);
625 55 : }
626 :
627 : /*
628 : * Obtain the system identifier using the provided connection. It will be used
629 : * to compare if a data directory is a clone of another one.
630 : */
631 : static uint64
632 10 : get_primary_sysid(const char *conninfo)
633 : {
634 : PGconn *conn;
635 : PGresult *res;
636 : uint64 sysid;
637 :
638 10 : pg_log_info("getting system identifier from publisher");
639 :
640 10 : conn = connect_database(conninfo, true);
641 :
642 10 : res = PQexec(conn, "SELECT system_identifier FROM pg_catalog.pg_control_system()");
643 10 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
644 : {
645 0 : pg_log_error("could not get system identifier: %s",
646 : PQresultErrorMessage(res));
647 0 : disconnect_database(conn, true);
648 : }
649 10 : if (PQntuples(res) != 1)
650 : {
651 0 : pg_log_error("could not get system identifier: got %d rows, expected %d row",
652 : PQntuples(res), 1);
653 0 : disconnect_database(conn, true);
654 : }
655 :
656 10 : sysid = strtou64(PQgetvalue(res, 0, 0), NULL, 10);
657 :
658 10 : pg_log_info("system identifier is %" PRIu64 " on publisher", sysid);
659 :
660 10 : PQclear(res);
661 10 : disconnect_database(conn, false);
662 :
663 10 : return sysid;
664 : }
665 :
666 : /*
667 : * Obtain the system identifier from control file. It will be used to compare
668 : * if a data directory is a clone of another one. This routine is used locally
669 : * and avoids a connection.
670 : */
671 : static uint64
672 10 : get_standby_sysid(const char *datadir)
673 : {
674 : ControlFileData *cf;
675 : bool crc_ok;
676 : uint64 sysid;
677 :
678 10 : pg_log_info("getting system identifier from subscriber");
679 :
680 10 : cf = get_controlfile(datadir, &crc_ok);
681 10 : if (!crc_ok)
682 0 : pg_fatal("control file appears to be corrupt");
683 :
684 10 : sysid = cf->system_identifier;
685 :
686 10 : pg_log_info("system identifier is %" PRIu64 " on subscriber", sysid);
687 :
688 10 : pg_free(cf);
689 :
690 10 : return sysid;
691 : }
692 :
693 : /*
694 : * Modify the system identifier. Since a standby server preserves the system
695 : * identifier, it makes sense to change it to avoid situations in which WAL
696 : * files from one of the systems might be used in the other one.
697 : */
698 : static void
699 4 : modify_subscriber_sysid(const struct CreateSubscriberOptions *opt)
700 : {
701 : ControlFileData *cf;
702 : bool crc_ok;
703 : struct timeval tv;
704 :
705 : char *cmd_str;
706 :
707 4 : pg_log_info("modifying system identifier of subscriber");
708 :
709 4 : cf = get_controlfile(subscriber_dir, &crc_ok);
710 4 : if (!crc_ok)
711 0 : pg_fatal("control file appears to be corrupt");
712 :
713 : /*
714 : * Select a new system identifier.
715 : *
716 : * XXX this code was extracted from BootStrapXLOG().
717 : */
718 4 : gettimeofday(&tv, NULL);
719 4 : cf->system_identifier = ((uint64) tv.tv_sec) << 32;
720 4 : cf->system_identifier |= ((uint64) tv.tv_usec) << 12;
721 4 : cf->system_identifier |= getpid() & 0xFFF;
722 :
723 4 : if (dry_run)
724 3 : pg_log_info("dry-run: would set system identifier to %" PRIu64 " on subscriber",
725 : cf->system_identifier);
726 : else
727 : {
728 1 : update_controlfile(subscriber_dir, cf, true);
729 1 : pg_log_info("system identifier is %" PRIu64 " on subscriber",
730 : cf->system_identifier);
731 : }
732 :
733 4 : if (dry_run)
734 3 : pg_log_info("dry-run: would run pg_resetwal on the subscriber");
735 : else
736 1 : pg_log_info("running pg_resetwal on the subscriber");
737 :
738 4 : cmd_str = psprintf("\"%s\" -D \"%s\" > \"%s\"", pg_resetwal_path,
739 : subscriber_dir, DEVNULL);
740 :
741 4 : pg_log_debug("pg_resetwal command is: %s", cmd_str);
742 :
743 4 : if (!dry_run)
744 : {
745 1 : int rc = system(cmd_str);
746 :
747 1 : if (rc == 0)
748 1 : pg_log_info("successfully reset WAL on the subscriber");
749 : else
750 0 : pg_fatal("could not reset WAL on subscriber: %s", wait_result_to_str(rc));
751 : }
752 :
753 4 : pg_free(cf);
754 4 : }
755 :
756 : /*
757 : * Generate an object name using a prefix, database oid and a random integer.
758 : * It is used in case the user does not specify an object name (publication,
759 : * subscription, replication slot).
760 : */
761 : static char *
762 8 : generate_object_name(PGconn *conn)
763 : {
764 : PGresult *res;
765 : Oid oid;
766 : uint32 rand;
767 : char *objname;
768 :
769 8 : res = PQexec(conn,
770 : "SELECT oid FROM pg_catalog.pg_database "
771 : "WHERE datname = pg_catalog.current_database()");
772 8 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
773 : {
774 0 : pg_log_error("could not obtain database OID: %s",
775 : PQresultErrorMessage(res));
776 0 : disconnect_database(conn, true);
777 : }
778 :
779 8 : if (PQntuples(res) != 1)
780 : {
781 0 : pg_log_error("could not obtain database OID: got %d rows, expected %d row",
782 : PQntuples(res), 1);
783 0 : disconnect_database(conn, true);
784 : }
785 :
786 : /* Database OID */
787 8 : oid = strtoul(PQgetvalue(res, 0, 0), NULL, 10);
788 :
789 8 : PQclear(res);
790 :
791 : /* Random unsigned integer */
792 8 : rand = pg_prng_uint32(&prng_state);
793 :
794 : /*
795 : * Build the object name. The name must not exceed NAMEDATALEN - 1. This
796 : * current schema uses a maximum of 40 characters (20 + 10 + 1 + 8 +
797 : * '\0').
798 : */
799 8 : objname = psprintf("pg_createsubscriber_%u_%x", oid, rand);
800 :
801 8 : return objname;
802 : }
803 :
804 : /*
805 : * Does the publication exist in the specified database?
806 : */
807 : static bool
808 8 : find_publication(PGconn *conn, const char *pubname, const char *dbname)
809 : {
810 8 : PQExpBuffer str = createPQExpBuffer();
811 : PGresult *res;
812 8 : bool found = false;
813 8 : char *pubname_esc = PQescapeLiteral(conn, pubname, strlen(pubname));
814 :
815 8 : appendPQExpBuffer(str,
816 : "SELECT 1 FROM pg_catalog.pg_publication "
817 : "WHERE pubname = %s",
818 : pubname_esc);
819 8 : res = PQexec(conn, str->data);
820 8 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
821 : {
822 0 : pg_log_error("could not find publication \"%s\" in database \"%s\": %s",
823 : pubname, dbname, PQerrorMessage(conn));
824 0 : disconnect_database(conn, true);
825 : }
826 :
827 8 : if (PQntuples(res) == 1)
828 1 : found = true;
829 :
830 8 : PQclear(res);
831 8 : PQfreemem(pubname_esc);
832 8 : destroyPQExpBuffer(str);
833 :
834 8 : return found;
835 : }
836 :
837 : /*
838 : * Create the publications and replication slots in preparation for logical
839 : * replication. Returns the LSN from latest replication slot. It will be the
840 : * replication start point that is used to adjust the subscriptions (see
841 : * set_replication_progress).
842 : */
843 : static char *
844 4 : setup_publisher(struct LogicalRepInfo *dbinfo)
845 : {
846 4 : char *lsn = NULL;
847 :
848 4 : pg_prng_seed(&prng_state, (uint64) (getpid() ^ time(NULL)));
849 :
850 12 : for (int i = 0; i < num_dbs; i++)
851 : {
852 : PGconn *conn;
853 8 : char *genname = NULL;
854 :
855 8 : conn = connect_database(dbinfo[i].pubconninfo, true);
856 :
857 : /*
858 : * If an object name was not specified as command-line options, assign
859 : * a generated object name. The replication slot has a different rule.
860 : * The subscription name is assigned to the replication slot name if
861 : * no replication slot is specified. It follows the same rule as
862 : * CREATE SUBSCRIPTION.
863 : */
864 8 : if (num_pubs == 0 || num_subs == 0 || num_replslots == 0)
865 8 : genname = generate_object_name(conn);
866 8 : if (num_pubs == 0)
867 4 : dbinfo[i].pubname = pg_strdup(genname);
868 8 : if (num_subs == 0)
869 6 : dbinfo[i].subname = pg_strdup(genname);
870 8 : if (num_replslots == 0)
871 5 : dbinfo[i].replslotname = pg_strdup(dbinfo[i].subname);
872 :
873 8 : if (find_publication(conn, dbinfo[i].pubname, dbinfo[i].dbname))
874 : {
875 : /* Reuse existing publication on publisher. */
876 1 : pg_log_info("use existing publication \"%s\" in database \"%s\"",
877 : dbinfo[i].pubname, dbinfo[i].dbname);
878 : /* Don't remove pre-existing publication if an error occurs. */
879 1 : dbinfo[i].made_publication = false;
880 : }
881 : else
882 : {
883 : /*
884 : * Create publication on publisher. This step should be executed
885 : * *before* promoting the subscriber to avoid any transactions
886 : * between consistent LSN and the new publication rows (such
887 : * transactions wouldn't see the new publication rows resulting in
888 : * an error).
889 : */
890 7 : create_publication(conn, &dbinfo[i]);
891 : }
892 :
893 : /* Create replication slot on publisher */
894 8 : if (lsn)
895 1 : pg_free(lsn);
896 8 : lsn = create_logical_replication_slot(conn, &dbinfo[i]);
897 8 : if (lsn == NULL && !dry_run)
898 0 : exit(1);
899 :
900 : /*
901 : * Since we are using the LSN returned by the last replication slot as
902 : * recovery_target_lsn, this LSN is ahead of the current WAL position
903 : * and the recovery waits until the publisher writes a WAL record to
904 : * reach the target and ends the recovery. On idle systems, this wait
905 : * time is unpredictable and could lead to failure in promoting the
906 : * subscriber. To avoid that, insert a harmless WAL record.
907 : */
908 8 : if (i == num_dbs - 1 && !dry_run)
909 : {
910 : PGresult *res;
911 :
912 1 : res = PQexec(conn, "SELECT pg_log_standby_snapshot()");
913 1 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
914 : {
915 0 : pg_log_error("could not write an additional WAL record: %s",
916 : PQresultErrorMessage(res));
917 0 : disconnect_database(conn, true);
918 : }
919 1 : PQclear(res);
920 : }
921 :
922 8 : disconnect_database(conn, false);
923 : }
924 :
925 4 : return lsn;
926 : }
927 :
928 : /*
929 : * Is recovery still in progress?
930 : */
931 : static bool
932 15 : server_is_in_recovery(PGconn *conn)
933 : {
934 : PGresult *res;
935 : int ret;
936 :
937 15 : res = PQexec(conn, "SELECT pg_catalog.pg_is_in_recovery()");
938 :
939 15 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
940 : {
941 0 : pg_log_error("could not obtain recovery progress: %s",
942 : PQresultErrorMessage(res));
943 0 : disconnect_database(conn, true);
944 : }
945 :
946 :
947 15 : ret = strcmp("t", PQgetvalue(res, 0, 0));
948 :
949 15 : PQclear(res);
950 :
951 15 : return ret == 0;
952 : }
953 :
954 : /*
955 : * Is the primary server ready for logical replication?
956 : *
957 : * XXX Does it not allow a synchronous replica?
958 : */
959 : static void
960 6 : check_publisher(const struct LogicalRepInfo *dbinfo)
961 : {
962 : PGconn *conn;
963 : PGresult *res;
964 6 : bool failed = false;
965 :
966 : char *wal_level;
967 : int max_repslots;
968 : int cur_repslots;
969 : int max_walsenders;
970 : int cur_walsenders;
971 : int max_prepared_transactions;
972 : char *max_slot_wal_keep_size;
973 :
974 6 : pg_log_info("checking settings on publisher");
975 :
976 6 : conn = connect_database(dbinfo[0].pubconninfo, true);
977 :
978 : /*
979 : * If the primary server is in recovery (i.e. cascading replication),
980 : * objects (publication) cannot be created because it is read only.
981 : */
982 6 : if (server_is_in_recovery(conn))
983 : {
984 1 : pg_log_error("primary server cannot be in recovery");
985 1 : disconnect_database(conn, true);
986 : }
987 :
988 : /*------------------------------------------------------------------------
989 : * Logical replication requires a few parameters to be set on publisher.
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 : * - wal_level >= replica
994 : * - max_replication_slots >= current + number of dbs to be converted
995 : * - max_wal_senders >= current + number of dbs to be converted
996 : * - max_slot_wal_keep_size = -1 (to prevent deletion of required WAL files)
997 : * -----------------------------------------------------------------------
998 : */
999 5 : res = PQexec(conn,
1000 : "SELECT pg_catalog.current_setting('wal_level'),"
1001 : " pg_catalog.current_setting('max_replication_slots'),"
1002 : " (SELECT count(*) FROM pg_catalog.pg_replication_slots),"
1003 : " pg_catalog.current_setting('max_wal_senders'),"
1004 : " (SELECT count(*) FROM pg_catalog.pg_stat_activity WHERE backend_type = 'walsender'),"
1005 : " pg_catalog.current_setting('max_prepared_transactions'),"
1006 : " pg_catalog.current_setting('max_slot_wal_keep_size')");
1007 :
1008 5 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1009 : {
1010 0 : pg_log_error("could not obtain publisher settings: %s",
1011 : PQresultErrorMessage(res));
1012 0 : disconnect_database(conn, true);
1013 : }
1014 :
1015 5 : wal_level = pg_strdup(PQgetvalue(res, 0, 0));
1016 5 : max_repslots = atoi(PQgetvalue(res, 0, 1));
1017 5 : cur_repslots = atoi(PQgetvalue(res, 0, 2));
1018 5 : max_walsenders = atoi(PQgetvalue(res, 0, 3));
1019 5 : cur_walsenders = atoi(PQgetvalue(res, 0, 4));
1020 5 : max_prepared_transactions = atoi(PQgetvalue(res, 0, 5));
1021 5 : max_slot_wal_keep_size = pg_strdup(PQgetvalue(res, 0, 6));
1022 :
1023 5 : PQclear(res);
1024 :
1025 5 : pg_log_debug("publisher: wal_level: %s", wal_level);
1026 5 : pg_log_debug("publisher: max_replication_slots: %d", max_repslots);
1027 5 : pg_log_debug("publisher: current replication slots: %d", cur_repslots);
1028 5 : pg_log_debug("publisher: max_wal_senders: %d", max_walsenders);
1029 5 : pg_log_debug("publisher: current wal senders: %d", cur_walsenders);
1030 5 : pg_log_debug("publisher: max_prepared_transactions: %d",
1031 : max_prepared_transactions);
1032 5 : pg_log_debug("publisher: max_slot_wal_keep_size: %s",
1033 : max_slot_wal_keep_size);
1034 :
1035 5 : disconnect_database(conn, false);
1036 :
1037 5 : if (strcmp(wal_level, "minimal") == 0)
1038 : {
1039 0 : pg_log_error("publisher requires \"wal_level\" >= \"replica\"");
1040 0 : failed = true;
1041 : }
1042 :
1043 5 : if (max_repslots - cur_repslots < num_dbs)
1044 : {
1045 1 : pg_log_error("publisher requires %d replication slots, but only %d remain",
1046 : num_dbs, max_repslots - cur_repslots);
1047 1 : pg_log_error_hint("Increase the configuration parameter \"%s\" to at least %d.",
1048 : "max_replication_slots", cur_repslots + num_dbs);
1049 1 : failed = true;
1050 : }
1051 :
1052 5 : if (max_walsenders - cur_walsenders < num_dbs)
1053 : {
1054 1 : pg_log_error("publisher requires %d WAL sender processes, but only %d remain",
1055 : num_dbs, max_walsenders - cur_walsenders);
1056 1 : pg_log_error_hint("Increase the configuration parameter \"%s\" to at least %d.",
1057 : "max_wal_senders", cur_walsenders + num_dbs);
1058 1 : failed = true;
1059 : }
1060 :
1061 5 : if (max_prepared_transactions != 0 && !dbinfos.two_phase)
1062 : {
1063 0 : pg_log_warning("two_phase option will not be enabled for replication slots");
1064 0 : pg_log_warning_detail("Subscriptions will be created with the two_phase option disabled. "
1065 : "Prepared transactions will be replicated at COMMIT PREPARED.");
1066 0 : pg_log_warning_hint("You can use the command-line option --enable-two-phase to enable two_phase.");
1067 : }
1068 :
1069 : /*
1070 : * In dry-run mode, validate 'max_slot_wal_keep_size'. If this parameter
1071 : * is set to a non-default value, it may cause replication failures due to
1072 : * required WAL files being prematurely removed.
1073 : */
1074 5 : if (dry_run && (strcmp(max_slot_wal_keep_size, "-1") != 0))
1075 : {
1076 0 : pg_log_warning("required WAL could be removed from the publisher");
1077 0 : pg_log_warning_hint("Set the configuration parameter \"%s\" to -1 to ensure that required WAL files are not prematurely removed.",
1078 : "max_slot_wal_keep_size");
1079 : }
1080 :
1081 5 : pg_free(wal_level);
1082 :
1083 5 : if (failed)
1084 1 : exit(1);
1085 4 : }
1086 :
1087 : /*
1088 : * Is the standby server ready for logical replication?
1089 : *
1090 : * XXX Does it not allow a time-delayed replica?
1091 : *
1092 : * XXX In a cascaded replication scenario (P -> S -> C), if the target server
1093 : * is S, it cannot detect there is a replica (server C) because server S starts
1094 : * accepting only local connections and server C cannot connect to it. Hence,
1095 : * there is not a reliable way to provide a suitable error saying the server C
1096 : * will be broken at the end of this process (due to pg_resetwal).
1097 : */
1098 : static void
1099 8 : check_subscriber(const struct LogicalRepInfo *dbinfo)
1100 : {
1101 : PGconn *conn;
1102 : PGresult *res;
1103 8 : bool failed = false;
1104 :
1105 : int max_lrworkers;
1106 : int max_replorigins;
1107 : int max_wprocs;
1108 :
1109 8 : pg_log_info("checking settings on subscriber");
1110 :
1111 8 : conn = connect_database(dbinfo[0].subconninfo, true);
1112 :
1113 : /* The target server must be a standby */
1114 8 : if (!server_is_in_recovery(conn))
1115 : {
1116 1 : pg_log_error("target server must be a standby");
1117 1 : disconnect_database(conn, true);
1118 : }
1119 :
1120 : /*------------------------------------------------------------------------
1121 : * Logical replication requires a few parameters to be set on subscriber.
1122 : * Since these parameters are not a requirement for physical replication,
1123 : * we should check it to make sure it won't fail.
1124 : *
1125 : * - max_active_replication_origins >= number of dbs to be converted
1126 : * - max_logical_replication_workers >= number of dbs to be converted
1127 : * - max_worker_processes >= 1 + number of dbs to be converted
1128 : *------------------------------------------------------------------------
1129 : */
1130 7 : res = PQexec(conn,
1131 : "SELECT setting FROM pg_catalog.pg_settings WHERE name IN ("
1132 : "'max_logical_replication_workers', "
1133 : "'max_active_replication_origins', "
1134 : "'max_worker_processes', "
1135 : "'primary_slot_name') "
1136 : "ORDER BY name");
1137 :
1138 7 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1139 : {
1140 0 : pg_log_error("could not obtain subscriber settings: %s",
1141 : PQresultErrorMessage(res));
1142 0 : disconnect_database(conn, true);
1143 : }
1144 :
1145 7 : max_replorigins = atoi(PQgetvalue(res, 0, 0));
1146 7 : max_lrworkers = atoi(PQgetvalue(res, 1, 0));
1147 7 : max_wprocs = atoi(PQgetvalue(res, 2, 0));
1148 7 : if (strcmp(PQgetvalue(res, 3, 0), "") != 0)
1149 6 : primary_slot_name = pg_strdup(PQgetvalue(res, 3, 0));
1150 :
1151 7 : pg_log_debug("subscriber: max_logical_replication_workers: %d",
1152 : max_lrworkers);
1153 7 : pg_log_debug("subscriber: max_active_replication_origins: %d", max_replorigins);
1154 7 : pg_log_debug("subscriber: max_worker_processes: %d", max_wprocs);
1155 7 : if (primary_slot_name)
1156 6 : pg_log_debug("subscriber: primary_slot_name: %s", primary_slot_name);
1157 :
1158 7 : PQclear(res);
1159 :
1160 7 : disconnect_database(conn, false);
1161 :
1162 7 : if (max_replorigins < num_dbs)
1163 : {
1164 1 : pg_log_error("subscriber requires %d active replication origins, but only %d remain",
1165 : num_dbs, max_replorigins);
1166 1 : pg_log_error_hint("Increase the configuration parameter \"%s\" to at least %d.",
1167 : "max_active_replication_origins", num_dbs);
1168 1 : failed = true;
1169 : }
1170 :
1171 7 : if (max_lrworkers < num_dbs)
1172 : {
1173 1 : pg_log_error("subscriber requires %d logical replication workers, but only %d remain",
1174 : num_dbs, max_lrworkers);
1175 1 : pg_log_error_hint("Increase the configuration parameter \"%s\" to at least %d.",
1176 : "max_logical_replication_workers", num_dbs);
1177 1 : failed = true;
1178 : }
1179 :
1180 7 : if (max_wprocs < num_dbs + 1)
1181 : {
1182 1 : pg_log_error("subscriber requires %d worker processes, but only %d remain",
1183 : num_dbs + 1, max_wprocs);
1184 1 : pg_log_error_hint("Increase the configuration parameter \"%s\" to at least %d.",
1185 : "max_worker_processes", num_dbs + 1);
1186 1 : failed = true;
1187 : }
1188 :
1189 7 : if (failed)
1190 1 : exit(1);
1191 6 : }
1192 :
1193 : /*
1194 : * Drop a specified subscription. This is to avoid duplicate subscriptions on
1195 : * the primary (publisher node) and the newly created subscriber. We
1196 : * shouldn't drop the associated slot as that would be used by the publisher
1197 : * node.
1198 : */
1199 : static void
1200 4 : drop_existing_subscription(PGconn *conn, const char *subname, const char *dbname)
1201 : {
1202 4 : PQExpBuffer query = createPQExpBuffer();
1203 : PGresult *res;
1204 :
1205 : Assert(conn != NULL);
1206 :
1207 : /*
1208 : * Construct a query string. These commands are allowed to be executed
1209 : * within a transaction.
1210 : */
1211 4 : appendPQExpBuffer(query, "ALTER SUBSCRIPTION %s DISABLE;",
1212 : subname);
1213 4 : appendPQExpBuffer(query, " ALTER SUBSCRIPTION %s SET (slot_name = NONE);",
1214 : subname);
1215 4 : appendPQExpBuffer(query, " DROP SUBSCRIPTION %s;", subname);
1216 :
1217 4 : if (dry_run)
1218 3 : pg_log_info("dry-run: would drop subscription \"%s\" in database \"%s\"",
1219 : subname, dbname);
1220 : else
1221 : {
1222 1 : pg_log_info("dropping subscription \"%s\" in database \"%s\"",
1223 : subname, dbname);
1224 :
1225 1 : res = PQexec(conn, query->data);
1226 :
1227 1 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1228 : {
1229 0 : pg_log_error("could not drop subscription \"%s\": %s",
1230 : subname, PQresultErrorMessage(res));
1231 0 : disconnect_database(conn, true);
1232 : }
1233 :
1234 1 : PQclear(res);
1235 : }
1236 :
1237 4 : destroyPQExpBuffer(query);
1238 4 : }
1239 :
1240 : /*
1241 : * Retrieve and drop the pre-existing subscriptions.
1242 : */
1243 : static void
1244 8 : check_and_drop_existing_subscriptions(PGconn *conn,
1245 : const struct LogicalRepInfo *dbinfo)
1246 : {
1247 8 : PQExpBuffer query = createPQExpBuffer();
1248 : char *dbname;
1249 : PGresult *res;
1250 :
1251 : Assert(conn != NULL);
1252 :
1253 8 : dbname = PQescapeLiteral(conn, dbinfo->dbname, strlen(dbinfo->dbname));
1254 :
1255 8 : appendPQExpBuffer(query,
1256 : "SELECT s.subname FROM pg_catalog.pg_subscription s "
1257 : "INNER JOIN pg_catalog.pg_database d ON (s.subdbid = d.oid) "
1258 : "WHERE d.datname = %s",
1259 : dbname);
1260 8 : res = PQexec(conn, query->data);
1261 :
1262 8 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1263 : {
1264 0 : pg_log_error("could not obtain pre-existing subscriptions: %s",
1265 : PQresultErrorMessage(res));
1266 0 : disconnect_database(conn, true);
1267 : }
1268 :
1269 12 : for (int i = 0; i < PQntuples(res); i++)
1270 4 : drop_existing_subscription(conn, PQgetvalue(res, i, 0),
1271 4 : dbinfo->dbname);
1272 :
1273 8 : PQclear(res);
1274 8 : destroyPQExpBuffer(query);
1275 8 : PQfreemem(dbname);
1276 8 : }
1277 :
1278 : /*
1279 : * Create the subscriptions, adjust the initial location for logical
1280 : * replication and enable the subscriptions. That's the last step for logical
1281 : * replication setup.
1282 : */
1283 : static void
1284 4 : setup_subscriber(struct LogicalRepInfo *dbinfo, const char *consistent_lsn)
1285 : {
1286 12 : for (int i = 0; i < num_dbs; i++)
1287 : {
1288 : PGconn *conn;
1289 :
1290 : /* Connect to subscriber. */
1291 8 : conn = connect_database(dbinfo[i].subconninfo, true);
1292 :
1293 : /*
1294 : * We don't need the pre-existing subscriptions on the newly formed
1295 : * subscriber. They can connect to other publisher nodes and either
1296 : * get some unwarranted data or can lead to ERRORs in connecting to
1297 : * such nodes.
1298 : */
1299 8 : check_and_drop_existing_subscriptions(conn, &dbinfo[i]);
1300 :
1301 : /* Check and drop the required publications in the given database. */
1302 8 : check_and_drop_publications(conn, &dbinfo[i]);
1303 :
1304 8 : create_subscription(conn, &dbinfo[i]);
1305 :
1306 : /* Set the replication progress to the correct LSN */
1307 8 : set_replication_progress(conn, &dbinfo[i], consistent_lsn);
1308 :
1309 : /* Enable subscription */
1310 8 : enable_subscription(conn, &dbinfo[i]);
1311 :
1312 8 : disconnect_database(conn, false);
1313 : }
1314 4 : }
1315 :
1316 : /*
1317 : * Write the required recovery parameters.
1318 : */
1319 : static void
1320 4 : setup_recovery(const struct LogicalRepInfo *dbinfo, const char *datadir, const char *lsn)
1321 : {
1322 : PGconn *conn;
1323 : PQExpBuffer recoveryconfcontents;
1324 :
1325 : /*
1326 : * Despite of the recovery parameters will be written to the subscriber,
1327 : * use a publisher connection. The primary_conninfo is generated using the
1328 : * connection settings.
1329 : */
1330 4 : conn = connect_database(dbinfo[0].pubconninfo, true);
1331 :
1332 : /*
1333 : * Write recovery parameters.
1334 : *
1335 : * The subscriber is not running yet. In dry run mode, the recovery
1336 : * parameters *won't* be written. An invalid LSN is used for printing
1337 : * purposes. Additional recovery parameters are added here. It avoids
1338 : * unexpected behavior such as end of recovery as soon as a consistent
1339 : * state is reached (recovery_target) and failure due to multiple recovery
1340 : * targets (name, time, xid, LSN).
1341 : */
1342 4 : recoveryconfcontents = GenerateRecoveryConfig(conn, NULL, NULL);
1343 4 : appendPQExpBufferStr(recoveryconfcontents, "recovery_target = ''\n");
1344 4 : appendPQExpBufferStr(recoveryconfcontents,
1345 : "recovery_target_timeline = 'latest'\n");
1346 :
1347 : /*
1348 : * Set recovery_target_inclusive = false to avoid reapplying the
1349 : * transaction committed at 'lsn' after subscription is enabled. This is
1350 : * because the provided 'lsn' is also used as the replication start point
1351 : * for the subscription. So, the server can send the transaction committed
1352 : * at that 'lsn' after replication is started which can lead to applying
1353 : * the same transaction twice if we keep recovery_target_inclusive = true.
1354 : */
1355 4 : appendPQExpBufferStr(recoveryconfcontents,
1356 : "recovery_target_inclusive = false\n");
1357 4 : appendPQExpBufferStr(recoveryconfcontents,
1358 : "recovery_target_action = promote\n");
1359 4 : appendPQExpBufferStr(recoveryconfcontents, "recovery_target_name = ''\n");
1360 4 : appendPQExpBufferStr(recoveryconfcontents, "recovery_target_time = ''\n");
1361 4 : appendPQExpBufferStr(recoveryconfcontents, "recovery_target_xid = ''\n");
1362 :
1363 4 : if (dry_run)
1364 : {
1365 3 : appendPQExpBufferStr(recoveryconfcontents, "# dry run mode\n");
1366 3 : appendPQExpBuffer(recoveryconfcontents,
1367 : "recovery_target_lsn = '%X/%08X'\n",
1368 : LSN_FORMAT_ARGS((XLogRecPtr) InvalidXLogRecPtr));
1369 : }
1370 : else
1371 : {
1372 1 : appendPQExpBuffer(recoveryconfcontents, "recovery_target_lsn = '%s'\n",
1373 : lsn);
1374 : }
1375 :
1376 4 : pg_log_debug("recovery parameters:\n%s", recoveryconfcontents->data);
1377 :
1378 4 : if (!dry_run)
1379 : {
1380 : char conf_filename[MAXPGPATH];
1381 : FILE *fd;
1382 :
1383 : /* Write the recovery parameters to INCLUDED_CONF_FILE */
1384 1 : snprintf(conf_filename, MAXPGPATH, "%s/%s", datadir,
1385 : INCLUDED_CONF_FILE);
1386 1 : fd = fopen(conf_filename, "w");
1387 1 : if (fd == NULL)
1388 0 : pg_fatal("could not open file \"%s\": %m", conf_filename);
1389 :
1390 1 : if (fwrite(recoveryconfcontents->data, recoveryconfcontents->len, 1, fd) != 1)
1391 0 : pg_fatal("could not write to file \"%s\": %m", conf_filename);
1392 :
1393 1 : fclose(fd);
1394 1 : recovery_params_set = true;
1395 :
1396 : /* Include conditionally the recovery parameters. */
1397 1 : resetPQExpBuffer(recoveryconfcontents);
1398 1 : appendPQExpBufferStr(recoveryconfcontents,
1399 : "include_if_exists '" INCLUDED_CONF_FILE "'\n");
1400 1 : WriteRecoveryConfig(conn, datadir, recoveryconfcontents);
1401 : }
1402 :
1403 4 : disconnect_database(conn, false);
1404 4 : }
1405 :
1406 : /*
1407 : * Drop physical replication slot on primary if the standby was using it. After
1408 : * the transformation, it has no use.
1409 : *
1410 : * XXX we might not fail here. Instead, we provide a warning so the user
1411 : * eventually drops this replication slot later.
1412 : */
1413 : static void
1414 4 : drop_primary_replication_slot(struct LogicalRepInfo *dbinfo, const char *slotname)
1415 : {
1416 : PGconn *conn;
1417 :
1418 : /* Replication slot does not exist, do nothing */
1419 4 : if (!primary_slot_name)
1420 0 : return;
1421 :
1422 4 : conn = connect_database(dbinfo[0].pubconninfo, false);
1423 4 : if (conn != NULL)
1424 : {
1425 4 : drop_replication_slot(conn, &dbinfo[0], slotname);
1426 4 : disconnect_database(conn, false);
1427 : }
1428 : else
1429 : {
1430 0 : pg_log_warning("could not drop replication slot \"%s\" on primary",
1431 : slotname);
1432 0 : pg_log_warning_hint("Drop this replication slot soon to avoid retention of WAL files.");
1433 : }
1434 : }
1435 :
1436 : /*
1437 : * Drop failover replication slots on subscriber. After the transformation,
1438 : * they have no use.
1439 : *
1440 : * XXX We do not fail here. Instead, we provide a warning so the user can drop
1441 : * them later.
1442 : */
1443 : static void
1444 4 : drop_failover_replication_slots(struct LogicalRepInfo *dbinfo)
1445 : {
1446 : PGconn *conn;
1447 : PGresult *res;
1448 :
1449 4 : conn = connect_database(dbinfo[0].subconninfo, false);
1450 4 : if (conn != NULL)
1451 : {
1452 : /* Get failover replication slot names */
1453 4 : res = PQexec(conn,
1454 : "SELECT slot_name FROM pg_catalog.pg_replication_slots WHERE failover");
1455 :
1456 4 : if (PQresultStatus(res) == PGRES_TUPLES_OK)
1457 : {
1458 : /* Remove failover replication slots from subscriber */
1459 8 : for (int i = 0; i < PQntuples(res); i++)
1460 4 : drop_replication_slot(conn, &dbinfo[0], PQgetvalue(res, i, 0));
1461 : }
1462 : else
1463 : {
1464 0 : pg_log_warning("could not obtain failover replication slot information: %s",
1465 : PQresultErrorMessage(res));
1466 0 : pg_log_warning_hint("Drop the failover replication slots on subscriber soon to avoid retention of WAL files.");
1467 : }
1468 :
1469 4 : PQclear(res);
1470 4 : disconnect_database(conn, false);
1471 : }
1472 : else
1473 : {
1474 0 : pg_log_warning("could not drop failover replication slot");
1475 0 : pg_log_warning_hint("Drop the failover replication slots on subscriber soon to avoid retention of WAL files.");
1476 : }
1477 4 : }
1478 :
1479 : /*
1480 : * Create a logical replication slot and returns a LSN.
1481 : *
1482 : * CreateReplicationSlot() is not used because it does not provide the one-row
1483 : * result set that contains the LSN.
1484 : */
1485 : static char *
1486 8 : create_logical_replication_slot(PGconn *conn, struct LogicalRepInfo *dbinfo)
1487 : {
1488 8 : PQExpBuffer str = createPQExpBuffer();
1489 8 : PGresult *res = NULL;
1490 8 : const char *slot_name = dbinfo->replslotname;
1491 : char *slot_name_esc;
1492 8 : char *lsn = NULL;
1493 :
1494 : Assert(conn != NULL);
1495 :
1496 8 : if (dry_run)
1497 6 : pg_log_info("dry-run: would create the replication slot \"%s\" in database \"%s\" on publisher",
1498 : slot_name, dbinfo->dbname);
1499 : else
1500 2 : pg_log_info("creating the replication slot \"%s\" in database \"%s\" on publisher",
1501 : slot_name, dbinfo->dbname);
1502 :
1503 8 : slot_name_esc = PQescapeLiteral(conn, slot_name, strlen(slot_name));
1504 :
1505 8 : appendPQExpBuffer(str,
1506 : "SELECT lsn FROM pg_catalog.pg_create_logical_replication_slot(%s, 'pgoutput', false, %s, false)",
1507 : slot_name_esc,
1508 8 : dbinfos.two_phase ? "true" : "false");
1509 :
1510 8 : PQfreemem(slot_name_esc);
1511 :
1512 8 : pg_log_debug("command is: %s", str->data);
1513 :
1514 8 : if (!dry_run)
1515 : {
1516 2 : res = PQexec(conn, str->data);
1517 2 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1518 : {
1519 0 : pg_log_error("could not create replication slot \"%s\" in database \"%s\": %s",
1520 : slot_name, dbinfo->dbname,
1521 : PQresultErrorMessage(res));
1522 0 : PQclear(res);
1523 0 : destroyPQExpBuffer(str);
1524 0 : return NULL;
1525 : }
1526 :
1527 2 : lsn = pg_strdup(PQgetvalue(res, 0, 0));
1528 2 : PQclear(res);
1529 : }
1530 :
1531 : /* For cleanup purposes */
1532 8 : dbinfo->made_replslot = true;
1533 :
1534 8 : destroyPQExpBuffer(str);
1535 :
1536 8 : return lsn;
1537 : }
1538 :
1539 : static void
1540 8 : drop_replication_slot(PGconn *conn, struct LogicalRepInfo *dbinfo,
1541 : const char *slot_name)
1542 : {
1543 8 : PQExpBuffer str = createPQExpBuffer();
1544 : char *slot_name_esc;
1545 : PGresult *res;
1546 :
1547 : Assert(conn != NULL);
1548 :
1549 8 : if (dry_run)
1550 6 : pg_log_info("dry-run: would drop the replication slot \"%s\" in database \"%s\"",
1551 : slot_name, dbinfo->dbname);
1552 : else
1553 2 : pg_log_info("dropping the replication slot \"%s\" in database \"%s\"",
1554 : slot_name, dbinfo->dbname);
1555 :
1556 8 : slot_name_esc = PQescapeLiteral(conn, slot_name, strlen(slot_name));
1557 :
1558 8 : appendPQExpBuffer(str, "SELECT pg_catalog.pg_drop_replication_slot(%s)", slot_name_esc);
1559 :
1560 8 : PQfreemem(slot_name_esc);
1561 :
1562 8 : pg_log_debug("command is: %s", str->data);
1563 :
1564 8 : if (!dry_run)
1565 : {
1566 2 : res = PQexec(conn, str->data);
1567 2 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1568 : {
1569 0 : pg_log_error("could not drop replication slot \"%s\" in database \"%s\": %s",
1570 : slot_name, dbinfo->dbname, PQresultErrorMessage(res));
1571 0 : dbinfo->made_replslot = false; /* don't try again. */
1572 : }
1573 :
1574 2 : PQclear(res);
1575 : }
1576 :
1577 8 : destroyPQExpBuffer(str);
1578 8 : }
1579 :
1580 : /*
1581 : * Reports a suitable message if pg_ctl fails.
1582 : */
1583 : static void
1584 24 : pg_ctl_status(const char *pg_ctl_cmd, int rc)
1585 : {
1586 24 : if (rc != 0)
1587 : {
1588 0 : if (WIFEXITED(rc))
1589 : {
1590 0 : pg_log_error("pg_ctl failed with exit code %d", WEXITSTATUS(rc));
1591 : }
1592 0 : else if (WIFSIGNALED(rc))
1593 : {
1594 : #if defined(WIN32)
1595 : pg_log_error("pg_ctl was terminated by exception 0x%X",
1596 : WTERMSIG(rc));
1597 : pg_log_error_detail("See C include file \"ntstatus.h\" for a description of the hexadecimal value.");
1598 : #else
1599 0 : pg_log_error("pg_ctl was terminated by signal %d: %s",
1600 : WTERMSIG(rc), pg_strsignal(WTERMSIG(rc)));
1601 : #endif
1602 : }
1603 : else
1604 : {
1605 0 : pg_log_error("pg_ctl exited with unrecognized status %d", rc);
1606 : }
1607 :
1608 0 : pg_log_error_detail("The failed command was: %s", pg_ctl_cmd);
1609 0 : exit(1);
1610 : }
1611 24 : }
1612 :
1613 : static void
1614 12 : start_standby_server(const struct CreateSubscriberOptions *opt, bool restricted_access,
1615 : bool restrict_logical_worker)
1616 : {
1617 12 : PQExpBuffer pg_ctl_cmd = createPQExpBuffer();
1618 : int rc;
1619 :
1620 12 : appendPQExpBuffer(pg_ctl_cmd, "\"%s\" start -D ", pg_ctl_path);
1621 12 : appendShellString(pg_ctl_cmd, subscriber_dir);
1622 12 : appendPQExpBufferStr(pg_ctl_cmd, " -s -o \"-c sync_replication_slots=off\"");
1623 :
1624 : /* Prevent unintended slot invalidation */
1625 12 : appendPQExpBufferStr(pg_ctl_cmd, " -o \"-c idle_replication_slot_timeout=0\"");
1626 :
1627 12 : if (restricted_access)
1628 : {
1629 12 : appendPQExpBuffer(pg_ctl_cmd, " -o \"-p %s\"", opt->sub_port);
1630 : #if !defined(WIN32)
1631 :
1632 : /*
1633 : * An empty listen_addresses list means the server does not listen on
1634 : * any IP interfaces; only Unix-domain sockets can be used to connect
1635 : * to the server. Prevent external connections to minimize the chance
1636 : * of failure.
1637 : */
1638 12 : appendPQExpBufferStr(pg_ctl_cmd, " -o \"-c listen_addresses='' -c unix_socket_permissions=0700");
1639 12 : if (opt->socket_dir)
1640 12 : appendPQExpBuffer(pg_ctl_cmd, " -c unix_socket_directories='%s'",
1641 12 : opt->socket_dir);
1642 12 : appendPQExpBufferChar(pg_ctl_cmd, '"');
1643 : #endif
1644 : }
1645 12 : if (opt->config_file != NULL)
1646 0 : appendPQExpBuffer(pg_ctl_cmd, " -o \"-c config_file=%s\"",
1647 0 : opt->config_file);
1648 :
1649 : /* Suppress to start logical replication if requested */
1650 12 : if (restrict_logical_worker)
1651 4 : appendPQExpBufferStr(pg_ctl_cmd, " -o \"-c max_logical_replication_workers=0\"");
1652 :
1653 12 : pg_log_debug("pg_ctl command is: %s", pg_ctl_cmd->data);
1654 12 : rc = system(pg_ctl_cmd->data);
1655 12 : pg_ctl_status(pg_ctl_cmd->data, rc);
1656 12 : standby_running = true;
1657 12 : destroyPQExpBuffer(pg_ctl_cmd);
1658 12 : pg_log_info("server was started");
1659 12 : }
1660 :
1661 : static void
1662 12 : stop_standby_server(const char *datadir)
1663 : {
1664 : char *pg_ctl_cmd;
1665 : int rc;
1666 :
1667 12 : pg_ctl_cmd = psprintf("\"%s\" stop -D \"%s\" -s", pg_ctl_path,
1668 : datadir);
1669 12 : pg_log_debug("pg_ctl command is: %s", pg_ctl_cmd);
1670 12 : rc = system(pg_ctl_cmd);
1671 12 : pg_ctl_status(pg_ctl_cmd, rc);
1672 12 : standby_running = false;
1673 12 : pg_log_info("server was stopped");
1674 12 : }
1675 :
1676 : /*
1677 : * Returns after the server finishes the recovery process.
1678 : *
1679 : * If recovery_timeout option is set, terminate abnormally without finishing
1680 : * the recovery process. By default, it waits forever.
1681 : *
1682 : * XXX Is the recovery process still in progress? When recovery process has a
1683 : * better progress reporting mechanism, it should be added here.
1684 : */
1685 : static void
1686 4 : wait_for_end_recovery(const char *conninfo, const struct CreateSubscriberOptions *opt)
1687 : {
1688 : PGconn *conn;
1689 4 : bool ready = false;
1690 4 : int timer = 0;
1691 :
1692 4 : pg_log_info("waiting for the target server to reach the consistent state");
1693 :
1694 4 : conn = connect_database(conninfo, true);
1695 :
1696 : for (;;)
1697 : {
1698 : /* Did the recovery process finish? We're done if so. */
1699 4 : if (dry_run || !server_is_in_recovery(conn))
1700 : {
1701 4 : ready = true;
1702 4 : recovery_ended = true;
1703 4 : break;
1704 : }
1705 :
1706 : /* Bail out after recovery_timeout seconds if this option is set */
1707 0 : if (opt->recovery_timeout > 0 && timer >= opt->recovery_timeout)
1708 : {
1709 0 : stop_standby_server(subscriber_dir);
1710 0 : pg_log_error("recovery timed out");
1711 0 : disconnect_database(conn, true);
1712 : }
1713 :
1714 : /* Keep waiting */
1715 0 : pg_usleep(WAIT_INTERVAL * USECS_PER_SEC);
1716 0 : timer += WAIT_INTERVAL;
1717 : }
1718 :
1719 4 : disconnect_database(conn, false);
1720 :
1721 4 : if (!ready)
1722 0 : pg_fatal("server did not end recovery");
1723 :
1724 4 : pg_log_info("target server reached the consistent state");
1725 4 : pg_log_info_hint("If pg_createsubscriber fails after this point, you must recreate the physical replica before continuing.");
1726 4 : }
1727 :
1728 : /*
1729 : * Create a publication that includes all tables in the database.
1730 : */
1731 : static void
1732 7 : create_publication(PGconn *conn, struct LogicalRepInfo *dbinfo)
1733 : {
1734 7 : PQExpBuffer str = createPQExpBuffer();
1735 : PGresult *res;
1736 : char *ipubname_esc;
1737 : char *spubname_esc;
1738 :
1739 : Assert(conn != NULL);
1740 :
1741 7 : ipubname_esc = PQescapeIdentifier(conn, dbinfo->pubname, strlen(dbinfo->pubname));
1742 7 : spubname_esc = PQescapeLiteral(conn, dbinfo->pubname, strlen(dbinfo->pubname));
1743 :
1744 : /* Check if the publication already exists */
1745 7 : appendPQExpBuffer(str,
1746 : "SELECT 1 FROM pg_catalog.pg_publication "
1747 : "WHERE pubname = %s",
1748 : spubname_esc);
1749 7 : res = PQexec(conn, str->data);
1750 7 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1751 : {
1752 0 : pg_log_error("could not obtain publication information: %s",
1753 : PQresultErrorMessage(res));
1754 0 : disconnect_database(conn, true);
1755 : }
1756 :
1757 7 : if (PQntuples(res) == 1)
1758 : {
1759 : /*
1760 : * Unfortunately, if it reaches this code path, it will always fail
1761 : * (unless you decide to change the existing publication name). That's
1762 : * bad but it is very unlikely that the user will choose a name with
1763 : * pg_createsubscriber_ prefix followed by the exact database oid and
1764 : * a random number.
1765 : */
1766 0 : pg_log_error("publication \"%s\" already exists", dbinfo->pubname);
1767 0 : pg_log_error_hint("Consider renaming this publication before continuing.");
1768 0 : disconnect_database(conn, true);
1769 : }
1770 :
1771 7 : PQclear(res);
1772 7 : resetPQExpBuffer(str);
1773 :
1774 7 : if (dry_run)
1775 6 : pg_log_info("dry-run: would create publication \"%s\" in database \"%s\"",
1776 : dbinfo->pubname, dbinfo->dbname);
1777 : else
1778 1 : pg_log_info("creating publication \"%s\" in database \"%s\"",
1779 : dbinfo->pubname, dbinfo->dbname);
1780 :
1781 7 : appendPQExpBuffer(str, "CREATE PUBLICATION %s FOR ALL TABLES",
1782 : ipubname_esc);
1783 :
1784 7 : pg_log_debug("command is: %s", str->data);
1785 :
1786 7 : if (!dry_run)
1787 : {
1788 1 : res = PQexec(conn, str->data);
1789 1 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1790 : {
1791 0 : pg_log_error("could not create publication \"%s\" in database \"%s\": %s",
1792 : dbinfo->pubname, dbinfo->dbname, PQresultErrorMessage(res));
1793 0 : disconnect_database(conn, true);
1794 : }
1795 1 : PQclear(res);
1796 : }
1797 :
1798 : /* For cleanup purposes */
1799 7 : dbinfo->made_publication = true;
1800 :
1801 7 : PQfreemem(ipubname_esc);
1802 7 : PQfreemem(spubname_esc);
1803 7 : destroyPQExpBuffer(str);
1804 7 : }
1805 :
1806 : /*
1807 : * Drop the specified publication in the given database.
1808 : */
1809 : static void
1810 10 : drop_publication(PGconn *conn, const char *pubname, const char *dbname,
1811 : bool *made_publication)
1812 : {
1813 10 : PQExpBuffer str = createPQExpBuffer();
1814 : PGresult *res;
1815 : char *pubname_esc;
1816 :
1817 : Assert(conn != NULL);
1818 :
1819 10 : pubname_esc = PQescapeIdentifier(conn, pubname, strlen(pubname));
1820 :
1821 10 : if (dry_run)
1822 6 : pg_log_info("dry-run: would drop publication \"%s\" in database \"%s\"",
1823 : pubname, dbname);
1824 : else
1825 4 : pg_log_info("dropping publication \"%s\" in database \"%s\"",
1826 : pubname, dbname);
1827 :
1828 10 : appendPQExpBuffer(str, "DROP PUBLICATION %s", pubname_esc);
1829 :
1830 10 : PQfreemem(pubname_esc);
1831 :
1832 10 : pg_log_debug("command is: %s", str->data);
1833 :
1834 10 : if (!dry_run)
1835 : {
1836 4 : res = PQexec(conn, str->data);
1837 4 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1838 : {
1839 0 : pg_log_error("could not drop publication \"%s\" in database \"%s\": %s",
1840 : pubname, dbname, PQresultErrorMessage(res));
1841 0 : *made_publication = false; /* don't try again. */
1842 :
1843 : /*
1844 : * Don't disconnect and exit here. This routine is used by primary
1845 : * (cleanup publication / replication slot due to an error) and
1846 : * subscriber (remove the replicated publications). In both cases,
1847 : * it can continue and provide instructions for the user to remove
1848 : * it later if cleanup fails.
1849 : */
1850 : }
1851 4 : PQclear(res);
1852 : }
1853 :
1854 10 : destroyPQExpBuffer(str);
1855 10 : }
1856 :
1857 : /*
1858 : * Retrieve and drop the publications.
1859 : *
1860 : * Publications copied during physical replication remain on the subscriber
1861 : * after promotion. If --clean=publications is specified, drop all existing
1862 : * publications in the subscriber database. Otherwise, only drop publications
1863 : * that were created by pg_createsubscriber during this operation.
1864 : */
1865 : static void
1866 8 : check_and_drop_publications(PGconn *conn, struct LogicalRepInfo *dbinfo)
1867 : {
1868 : PGresult *res;
1869 8 : bool drop_all_pubs = dbinfos.objecttypes_to_clean & OBJECTTYPE_PUBLICATIONS;
1870 :
1871 : Assert(conn != NULL);
1872 :
1873 8 : if (drop_all_pubs)
1874 : {
1875 2 : pg_log_info("dropping all existing publications in database \"%s\"",
1876 : dbinfo->dbname);
1877 :
1878 : /* Fetch all publication names */
1879 2 : res = PQexec(conn, "SELECT pubname FROM pg_catalog.pg_publication;");
1880 2 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1881 : {
1882 0 : pg_log_error("could not obtain publication information: %s",
1883 : PQresultErrorMessage(res));
1884 0 : PQclear(res);
1885 0 : disconnect_database(conn, true);
1886 : }
1887 :
1888 : /* Drop each publication */
1889 6 : for (int i = 0; i < PQntuples(res); i++)
1890 4 : drop_publication(conn, PQgetvalue(res, i, 0), dbinfo->dbname,
1891 : &dbinfo->made_publication);
1892 :
1893 2 : PQclear(res);
1894 : }
1895 : else
1896 : {
1897 : /* Drop publication only if it was created by this tool */
1898 6 : if (dbinfo->made_publication)
1899 : {
1900 6 : drop_publication(conn, dbinfo->pubname, dbinfo->dbname,
1901 : &dbinfo->made_publication);
1902 : }
1903 : else
1904 : {
1905 0 : if (dry_run)
1906 0 : pg_log_info("dry-run: would preserve existing publication \"%s\" in database \"%s\"",
1907 : dbinfo->pubname, dbinfo->dbname);
1908 : else
1909 0 : pg_log_info("preserve existing publication \"%s\" in database \"%s\"",
1910 : dbinfo->pubname, dbinfo->dbname);
1911 : }
1912 : }
1913 8 : }
1914 :
1915 : /*
1916 : * Create a subscription with some predefined options.
1917 : *
1918 : * A replication slot was already created in a previous step. Let's use it. It
1919 : * is not required to copy data. The subscription will be created but it will
1920 : * not be enabled now. That's because the replication progress must be set and
1921 : * the replication origin name (one of the function arguments) contains the
1922 : * subscription OID in its name. Once the subscription is created,
1923 : * set_replication_progress() can obtain the chosen origin name and set up its
1924 : * initial location.
1925 : */
1926 : static void
1927 8 : create_subscription(PGconn *conn, const struct LogicalRepInfo *dbinfo)
1928 : {
1929 8 : PQExpBuffer str = createPQExpBuffer();
1930 : PGresult *res;
1931 : char *pubname_esc;
1932 : char *subname_esc;
1933 : char *pubconninfo_esc;
1934 : char *replslotname_esc;
1935 :
1936 : Assert(conn != NULL);
1937 :
1938 8 : pubname_esc = PQescapeIdentifier(conn, dbinfo->pubname, strlen(dbinfo->pubname));
1939 8 : subname_esc = PQescapeIdentifier(conn, dbinfo->subname, strlen(dbinfo->subname));
1940 8 : pubconninfo_esc = PQescapeLiteral(conn, dbinfo->pubconninfo, strlen(dbinfo->pubconninfo));
1941 8 : replslotname_esc = PQescapeLiteral(conn, dbinfo->replslotname, strlen(dbinfo->replslotname));
1942 :
1943 8 : if (dry_run)
1944 6 : pg_log_info("dry-run: would create subscription \"%s\" in database \"%s\"",
1945 : dbinfo->subname, dbinfo->dbname);
1946 : else
1947 2 : pg_log_info("creating subscription \"%s\" in database \"%s\"",
1948 : dbinfo->subname, dbinfo->dbname);
1949 :
1950 8 : appendPQExpBuffer(str,
1951 : "CREATE SUBSCRIPTION %s CONNECTION %s PUBLICATION %s "
1952 : "WITH (create_slot = false, enabled = false, "
1953 : "slot_name = %s, copy_data = false, two_phase = %s)",
1954 : subname_esc, pubconninfo_esc, pubname_esc, replslotname_esc,
1955 8 : dbinfos.two_phase ? "true" : "false");
1956 :
1957 8 : PQfreemem(pubname_esc);
1958 8 : PQfreemem(subname_esc);
1959 8 : PQfreemem(pubconninfo_esc);
1960 8 : PQfreemem(replslotname_esc);
1961 :
1962 8 : pg_log_debug("command is: %s", str->data);
1963 :
1964 8 : if (!dry_run)
1965 : {
1966 2 : res = PQexec(conn, str->data);
1967 2 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1968 : {
1969 0 : pg_log_error("could not create subscription \"%s\" in database \"%s\": %s",
1970 : dbinfo->subname, dbinfo->dbname, PQresultErrorMessage(res));
1971 0 : disconnect_database(conn, true);
1972 : }
1973 2 : PQclear(res);
1974 : }
1975 :
1976 8 : destroyPQExpBuffer(str);
1977 8 : }
1978 :
1979 : /*
1980 : * Sets the replication progress to the consistent LSN.
1981 : *
1982 : * The subscriber caught up to the consistent LSN provided by the last
1983 : * replication slot that was created. The goal is to set up the initial
1984 : * location for the logical replication that is the exact LSN that the
1985 : * subscriber was promoted. Once the subscription is enabled it will start
1986 : * streaming from that location onwards. In dry run mode, the subscription OID
1987 : * and LSN are set to invalid values for printing purposes.
1988 : */
1989 : static void
1990 8 : set_replication_progress(PGconn *conn, const struct LogicalRepInfo *dbinfo, const char *lsn)
1991 : {
1992 8 : PQExpBuffer str = createPQExpBuffer();
1993 : PGresult *res;
1994 : Oid suboid;
1995 : char *subname;
1996 : char *dbname;
1997 : char *originname;
1998 : char *lsnstr;
1999 :
2000 : Assert(conn != NULL);
2001 :
2002 8 : subname = PQescapeLiteral(conn, dbinfo->subname, strlen(dbinfo->subname));
2003 8 : dbname = PQescapeLiteral(conn, dbinfo->dbname, strlen(dbinfo->dbname));
2004 :
2005 8 : appendPQExpBuffer(str,
2006 : "SELECT s.oid FROM pg_catalog.pg_subscription s "
2007 : "INNER JOIN pg_catalog.pg_database d ON (s.subdbid = d.oid) "
2008 : "WHERE s.subname = %s AND d.datname = %s",
2009 : subname, dbname);
2010 :
2011 8 : res = PQexec(conn, str->data);
2012 8 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
2013 : {
2014 0 : pg_log_error("could not obtain subscription OID: %s",
2015 : PQresultErrorMessage(res));
2016 0 : disconnect_database(conn, true);
2017 : }
2018 :
2019 8 : if (PQntuples(res) != 1 && !dry_run)
2020 : {
2021 0 : pg_log_error("could not obtain subscription OID: got %d rows, expected %d row",
2022 : PQntuples(res), 1);
2023 0 : disconnect_database(conn, true);
2024 : }
2025 :
2026 8 : if (dry_run)
2027 : {
2028 6 : suboid = InvalidOid;
2029 6 : lsnstr = psprintf("%X/%08X", LSN_FORMAT_ARGS((XLogRecPtr) InvalidXLogRecPtr));
2030 : }
2031 : else
2032 : {
2033 2 : suboid = strtoul(PQgetvalue(res, 0, 0), NULL, 10);
2034 2 : lsnstr = psprintf("%s", lsn);
2035 : }
2036 :
2037 8 : PQclear(res);
2038 :
2039 : /*
2040 : * The origin name is defined as pg_%u. %u is the subscription OID. See
2041 : * ApplyWorkerMain().
2042 : */
2043 8 : originname = psprintf("pg_%u", suboid);
2044 :
2045 8 : if (dry_run)
2046 6 : pg_log_info("dry-run: would set the replication progress (node name \"%s\", LSN %s) in database \"%s\"",
2047 : originname, lsnstr, dbinfo->dbname);
2048 : else
2049 2 : pg_log_info("setting the replication progress (node name \"%s\", LSN %s) in database \"%s\"",
2050 : originname, lsnstr, dbinfo->dbname);
2051 :
2052 8 : resetPQExpBuffer(str);
2053 8 : appendPQExpBuffer(str,
2054 : "SELECT pg_catalog.pg_replication_origin_advance('%s', '%s')",
2055 : originname, lsnstr);
2056 :
2057 8 : pg_log_debug("command is: %s", str->data);
2058 :
2059 8 : if (!dry_run)
2060 : {
2061 2 : res = PQexec(conn, str->data);
2062 2 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
2063 : {
2064 0 : pg_log_error("could not set replication progress for subscription \"%s\": %s",
2065 : dbinfo->subname, PQresultErrorMessage(res));
2066 0 : disconnect_database(conn, true);
2067 : }
2068 2 : PQclear(res);
2069 : }
2070 :
2071 8 : PQfreemem(subname);
2072 8 : PQfreemem(dbname);
2073 8 : pg_free(originname);
2074 8 : pg_free(lsnstr);
2075 8 : destroyPQExpBuffer(str);
2076 8 : }
2077 :
2078 : /*
2079 : * Enables the subscription.
2080 : *
2081 : * The subscription was created in a previous step but it was disabled. After
2082 : * adjusting the initial logical replication location, enable the subscription.
2083 : */
2084 : static void
2085 8 : enable_subscription(PGconn *conn, const struct LogicalRepInfo *dbinfo)
2086 : {
2087 8 : PQExpBuffer str = createPQExpBuffer();
2088 : PGresult *res;
2089 : char *subname;
2090 :
2091 : Assert(conn != NULL);
2092 :
2093 8 : subname = PQescapeIdentifier(conn, dbinfo->subname, strlen(dbinfo->subname));
2094 :
2095 8 : if (dry_run)
2096 6 : pg_log_info("dry-run: would enable subscription \"%s\" in database \"%s\"",
2097 : dbinfo->subname, dbinfo->dbname);
2098 : else
2099 2 : pg_log_info("enabling subscription \"%s\" in database \"%s\"",
2100 : dbinfo->subname, dbinfo->dbname);
2101 :
2102 8 : appendPQExpBuffer(str, "ALTER SUBSCRIPTION %s ENABLE", subname);
2103 :
2104 8 : pg_log_debug("command is: %s", str->data);
2105 :
2106 8 : if (!dry_run)
2107 : {
2108 2 : res = PQexec(conn, str->data);
2109 2 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
2110 : {
2111 0 : pg_log_error("could not enable subscription \"%s\": %s",
2112 : dbinfo->subname, PQresultErrorMessage(res));
2113 0 : disconnect_database(conn, true);
2114 : }
2115 :
2116 2 : PQclear(res);
2117 : }
2118 :
2119 8 : PQfreemem(subname);
2120 8 : destroyPQExpBuffer(str);
2121 8 : }
2122 :
2123 : /*
2124 : * Fetch a list of all connectable non-template databases from the source server
2125 : * and form a list such that they appear as if the user has specified multiple
2126 : * --database options, one for each source database.
2127 : */
2128 : static void
2129 1 : get_publisher_databases(struct CreateSubscriberOptions *opt,
2130 : bool dbnamespecified)
2131 : {
2132 : PGconn *conn;
2133 : PGresult *res;
2134 :
2135 : /* If a database name was specified, just connect to it. */
2136 1 : if (dbnamespecified)
2137 0 : conn = connect_database(opt->pub_conninfo_str, true);
2138 : else
2139 : {
2140 : /* Otherwise, try postgres first and then template1. */
2141 : char *conninfo;
2142 :
2143 1 : conninfo = concat_conninfo_dbname(opt->pub_conninfo_str, "postgres");
2144 1 : conn = connect_database(conninfo, false);
2145 1 : pg_free(conninfo);
2146 1 : if (!conn)
2147 : {
2148 0 : conninfo = concat_conninfo_dbname(opt->pub_conninfo_str, "template1");
2149 0 : conn = connect_database(conninfo, true);
2150 0 : pg_free(conninfo);
2151 : }
2152 : }
2153 :
2154 1 : res = PQexec(conn, "SELECT datname FROM pg_database WHERE datistemplate = false AND datallowconn AND datconnlimit <> -2 ORDER BY 1");
2155 1 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
2156 : {
2157 0 : pg_log_error("could not obtain a list of databases: %s", PQresultErrorMessage(res));
2158 0 : PQclear(res);
2159 0 : disconnect_database(conn, true);
2160 : }
2161 :
2162 4 : for (int i = 0; i < PQntuples(res); i++)
2163 : {
2164 3 : const char *dbname = PQgetvalue(res, i, 0);
2165 :
2166 3 : simple_string_list_append(&opt->database_names, dbname);
2167 :
2168 : /* Increment num_dbs to reflect multiple --database options */
2169 3 : num_dbs++;
2170 : }
2171 :
2172 1 : PQclear(res);
2173 1 : disconnect_database(conn, false);
2174 1 : }
2175 :
2176 : int
2177 23 : main(int argc, char **argv)
2178 : {
2179 : static struct option long_options[] =
2180 : {
2181 : {"all", no_argument, NULL, 'a'},
2182 : {"database", required_argument, NULL, 'd'},
2183 : {"pgdata", required_argument, NULL, 'D'},
2184 : {"dry-run", no_argument, NULL, 'n'},
2185 : {"subscriber-port", required_argument, NULL, 'p'},
2186 : {"publisher-server", required_argument, NULL, 'P'},
2187 : {"socketdir", required_argument, NULL, 's'},
2188 : {"recovery-timeout", required_argument, NULL, 't'},
2189 : {"enable-two-phase", no_argument, NULL, 'T'},
2190 : {"subscriber-username", required_argument, NULL, 'U'},
2191 : {"verbose", no_argument, NULL, 'v'},
2192 : {"version", no_argument, NULL, 'V'},
2193 : {"help", no_argument, NULL, '?'},
2194 : {"config-file", required_argument, NULL, 1},
2195 : {"publication", required_argument, NULL, 2},
2196 : {"replication-slot", required_argument, NULL, 3},
2197 : {"subscription", required_argument, NULL, 4},
2198 : {"clean", required_argument, NULL, 5},
2199 : {NULL, 0, NULL, 0}
2200 : };
2201 :
2202 23 : struct CreateSubscriberOptions opt = {0};
2203 :
2204 : int c;
2205 : int option_index;
2206 :
2207 : char *pub_base_conninfo;
2208 : char *sub_base_conninfo;
2209 23 : char *dbname_conninfo = NULL;
2210 :
2211 : uint64 pub_sysid;
2212 : uint64 sub_sysid;
2213 : struct stat statbuf;
2214 :
2215 : char *consistent_lsn;
2216 :
2217 : char pidfile[MAXPGPATH];
2218 :
2219 23 : pg_logging_init(argv[0]);
2220 23 : pg_logging_set_level(PG_LOG_WARNING);
2221 23 : progname = get_progname(argv[0]);
2222 23 : set_pglocale_pgservice(argv[0], PG_TEXTDOMAIN("pg_basebackup"));
2223 :
2224 23 : if (argc > 1)
2225 : {
2226 22 : if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
2227 : {
2228 1 : usage();
2229 1 : exit(0);
2230 : }
2231 21 : else if (strcmp(argv[1], "-V") == 0
2232 21 : || strcmp(argv[1], "--version") == 0)
2233 : {
2234 1 : puts("pg_createsubscriber (PostgreSQL) " PG_VERSION);
2235 1 : exit(0);
2236 : }
2237 : }
2238 :
2239 : /* Default settings */
2240 21 : subscriber_dir = NULL;
2241 21 : opt.config_file = NULL;
2242 21 : opt.pub_conninfo_str = NULL;
2243 21 : opt.socket_dir = NULL;
2244 21 : opt.sub_port = DEFAULT_SUB_PORT;
2245 21 : opt.sub_username = NULL;
2246 21 : opt.two_phase = false;
2247 21 : opt.database_names = (SimpleStringList)
2248 : {
2249 : 0
2250 : };
2251 21 : opt.recovery_timeout = 0;
2252 21 : opt.all_dbs = false;
2253 :
2254 : /*
2255 : * Don't allow it to be run as root. It uses pg_ctl which does not allow
2256 : * it either.
2257 : */
2258 : #ifndef WIN32
2259 21 : if (geteuid() == 0)
2260 : {
2261 0 : pg_log_error("cannot be executed by \"root\"");
2262 0 : pg_log_error_hint("You must run %s as the PostgreSQL superuser.",
2263 : progname);
2264 0 : exit(1);
2265 : }
2266 : #endif
2267 :
2268 21 : get_restricted_token();
2269 :
2270 162 : while ((c = getopt_long(argc, argv, "ad:D:np:P:s:t:TU:v",
2271 162 : long_options, &option_index)) != -1)
2272 : {
2273 144 : switch (c)
2274 : {
2275 3 : case 'a':
2276 3 : opt.all_dbs = true;
2277 3 : break;
2278 25 : case 'd':
2279 25 : if (!simple_string_list_member(&opt.database_names, optarg))
2280 : {
2281 24 : simple_string_list_append(&opt.database_names, optarg);
2282 24 : num_dbs++;
2283 : }
2284 : else
2285 1 : pg_fatal("database \"%s\" specified more than once for -d/--database", optarg);
2286 24 : break;
2287 19 : case 'D':
2288 19 : subscriber_dir = pg_strdup(optarg);
2289 19 : canonicalize_path(subscriber_dir);
2290 19 : break;
2291 9 : case 'n':
2292 9 : dry_run = true;
2293 9 : break;
2294 12 : case 'p':
2295 12 : opt.sub_port = pg_strdup(optarg);
2296 12 : break;
2297 18 : case 'P':
2298 18 : opt.pub_conninfo_str = pg_strdup(optarg);
2299 18 : break;
2300 12 : case 's':
2301 12 : opt.socket_dir = pg_strdup(optarg);
2302 12 : canonicalize_path(opt.socket_dir);
2303 12 : break;
2304 3 : case 't':
2305 3 : opt.recovery_timeout = atoi(optarg);
2306 3 : break;
2307 1 : case 'T':
2308 1 : opt.two_phase = true;
2309 1 : break;
2310 0 : case 'U':
2311 0 : opt.sub_username = pg_strdup(optarg);
2312 0 : break;
2313 19 : case 'v':
2314 19 : pg_logging_increase_verbosity();
2315 19 : break;
2316 0 : case 1:
2317 0 : opt.config_file = pg_strdup(optarg);
2318 0 : break;
2319 12 : case 2:
2320 12 : if (!simple_string_list_member(&opt.pub_names, optarg))
2321 : {
2322 11 : simple_string_list_append(&opt.pub_names, optarg);
2323 11 : num_pubs++;
2324 : }
2325 : else
2326 1 : pg_fatal("publication \"%s\" specified more than once for --publication", optarg);
2327 11 : break;
2328 4 : case 3:
2329 4 : if (!simple_string_list_member(&opt.replslot_names, optarg))
2330 : {
2331 4 : simple_string_list_append(&opt.replslot_names, optarg);
2332 4 : num_replslots++;
2333 : }
2334 : else
2335 0 : pg_fatal("replication slot \"%s\" specified more than once for --replication-slot", optarg);
2336 4 : break;
2337 5 : case 4:
2338 5 : if (!simple_string_list_member(&opt.sub_names, optarg))
2339 : {
2340 5 : simple_string_list_append(&opt.sub_names, optarg);
2341 5 : num_subs++;
2342 : }
2343 : else
2344 0 : pg_fatal("subscription \"%s\" specified more than once for --subscription", optarg);
2345 5 : break;
2346 1 : case 5:
2347 1 : if (!simple_string_list_member(&opt.objecttypes_to_clean, optarg))
2348 1 : simple_string_list_append(&opt.objecttypes_to_clean, optarg);
2349 : else
2350 0 : pg_fatal("object type \"%s\" specified more than once for --clean", optarg);
2351 1 : break;
2352 1 : default:
2353 : /* getopt_long already emitted a complaint */
2354 1 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
2355 1 : exit(1);
2356 : }
2357 : }
2358 :
2359 : /* Validate that --all is not used with incompatible options */
2360 18 : if (opt.all_dbs)
2361 : {
2362 3 : char *bad_switch = NULL;
2363 :
2364 3 : if (num_dbs > 0)
2365 1 : bad_switch = "--database";
2366 2 : else if (num_pubs > 0)
2367 1 : bad_switch = "--publication";
2368 1 : else if (num_replslots > 0)
2369 0 : bad_switch = "--replication-slot";
2370 1 : else if (num_subs > 0)
2371 0 : bad_switch = "--subscription";
2372 :
2373 3 : if (bad_switch)
2374 : {
2375 2 : pg_log_error("options %s and %s cannot be used together",
2376 : bad_switch, "-a/--all");
2377 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
2378 2 : exit(1);
2379 : }
2380 : }
2381 :
2382 : /* Any non-option arguments? */
2383 16 : if (optind < argc)
2384 : {
2385 0 : pg_log_error("too many command-line arguments (first is \"%s\")",
2386 : argv[optind]);
2387 0 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
2388 0 : exit(1);
2389 : }
2390 :
2391 : /* Required arguments */
2392 16 : if (subscriber_dir == NULL)
2393 : {
2394 1 : pg_log_error("no subscriber data directory specified");
2395 1 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
2396 1 : exit(1);
2397 : }
2398 :
2399 : /* If socket directory is not provided, use the current directory */
2400 15 : if (opt.socket_dir == NULL)
2401 : {
2402 : char cwd[MAXPGPATH];
2403 :
2404 5 : if (!getcwd(cwd, MAXPGPATH))
2405 0 : pg_fatal("could not determine current directory");
2406 5 : opt.socket_dir = pg_strdup(cwd);
2407 5 : canonicalize_path(opt.socket_dir);
2408 : }
2409 :
2410 : /*
2411 : * Parse connection string. Build a base connection string that might be
2412 : * reused by multiple databases.
2413 : */
2414 15 : if (opt.pub_conninfo_str == NULL)
2415 : {
2416 : /*
2417 : * TODO use primary_conninfo (if available) from subscriber and
2418 : * extract publisher connection string. Assume that there are
2419 : * identical entries for physical and logical replication. If there is
2420 : * not, we would fail anyway.
2421 : */
2422 1 : pg_log_error("no publisher connection string specified");
2423 1 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
2424 1 : exit(1);
2425 : }
2426 :
2427 14 : if (dry_run)
2428 8 : pg_log_info("Executing in dry-run mode.\n"
2429 : "The target directory will not be modified.");
2430 :
2431 14 : pg_log_info("validating publisher connection string");
2432 14 : pub_base_conninfo = get_base_conninfo(opt.pub_conninfo_str,
2433 : &dbname_conninfo);
2434 14 : if (pub_base_conninfo == NULL)
2435 0 : exit(1);
2436 :
2437 14 : pg_log_info("validating subscriber connection string");
2438 14 : sub_base_conninfo = get_sub_conninfo(&opt);
2439 :
2440 : /*
2441 : * Fetch all databases from the source (publisher) and treat them as if
2442 : * the user specified has multiple --database options, one for each source
2443 : * database.
2444 : */
2445 14 : if (opt.all_dbs)
2446 : {
2447 1 : bool dbnamespecified = (dbname_conninfo != NULL);
2448 :
2449 1 : get_publisher_databases(&opt, dbnamespecified);
2450 : }
2451 :
2452 14 : if (opt.database_names.head == NULL)
2453 : {
2454 2 : pg_log_info("no database was specified");
2455 :
2456 : /*
2457 : * Try to obtain the dbname from the publisher conninfo. If dbname
2458 : * parameter is not available, error out.
2459 : */
2460 2 : if (dbname_conninfo)
2461 : {
2462 1 : simple_string_list_append(&opt.database_names, dbname_conninfo);
2463 1 : num_dbs++;
2464 :
2465 1 : pg_log_info("database name \"%s\" was extracted from the publisher connection string",
2466 : dbname_conninfo);
2467 : }
2468 : else
2469 : {
2470 1 : pg_log_error("no database name specified");
2471 1 : pg_log_error_hint("Try \"%s --help\" for more information.",
2472 : progname);
2473 1 : exit(1);
2474 : }
2475 : }
2476 :
2477 : /* Number of object names must match number of databases */
2478 13 : if (num_pubs > 0 && num_pubs != num_dbs)
2479 : {
2480 1 : pg_log_error("wrong number of publication names specified");
2481 1 : pg_log_error_detail("The number of specified publication names (%d) must match the number of specified database names (%d).",
2482 : num_pubs, num_dbs);
2483 1 : exit(1);
2484 : }
2485 12 : if (num_subs > 0 && num_subs != num_dbs)
2486 : {
2487 1 : pg_log_error("wrong number of subscription names specified");
2488 1 : pg_log_error_detail("The number of specified subscription names (%d) must match the number of specified database names (%d).",
2489 : num_subs, num_dbs);
2490 1 : exit(1);
2491 : }
2492 11 : if (num_replslots > 0 && num_replslots != num_dbs)
2493 : {
2494 1 : pg_log_error("wrong number of replication slot names specified");
2495 1 : pg_log_error_detail("The number of specified replication slot names (%d) must match the number of specified database names (%d).",
2496 : num_replslots, num_dbs);
2497 1 : exit(1);
2498 : }
2499 :
2500 : /* Verify the object types specified for removal from the subscriber */
2501 11 : for (SimpleStringListCell *cell = opt.objecttypes_to_clean.head; cell; cell = cell->next)
2502 : {
2503 1 : if (pg_strcasecmp(cell->val, "publications") == 0)
2504 1 : dbinfos.objecttypes_to_clean |= OBJECTTYPE_PUBLICATIONS;
2505 : else
2506 : {
2507 0 : pg_log_error("invalid object type \"%s\" specified for %s",
2508 : cell->val, "--clean");
2509 0 : pg_log_error_hint("The valid value is: \"%s\"", "publications");
2510 0 : exit(1);
2511 : }
2512 : }
2513 :
2514 : /* Get the absolute path of pg_ctl and pg_resetwal on the subscriber */
2515 10 : pg_ctl_path = get_exec_path(argv[0], "pg_ctl");
2516 10 : pg_resetwal_path = get_exec_path(argv[0], "pg_resetwal");
2517 :
2518 : /* Rudimentary check for a data directory */
2519 10 : check_data_directory(subscriber_dir);
2520 :
2521 10 : dbinfos.two_phase = opt.two_phase;
2522 :
2523 : /*
2524 : * Store database information for publisher and subscriber. It should be
2525 : * called before atexit() because its return is used in the
2526 : * cleanup_objects_atexit().
2527 : */
2528 10 : dbinfos.dbinfo = store_pub_sub_info(&opt, pub_base_conninfo, sub_base_conninfo);
2529 :
2530 : /* Register a function to clean up objects in case of failure */
2531 10 : atexit(cleanup_objects_atexit);
2532 :
2533 : /*
2534 : * Check if the subscriber data directory has the same system identifier
2535 : * than the publisher data directory.
2536 : */
2537 10 : pub_sysid = get_primary_sysid(dbinfos.dbinfo[0].pubconninfo);
2538 10 : sub_sysid = get_standby_sysid(subscriber_dir);
2539 10 : if (pub_sysid != sub_sysid)
2540 1 : pg_fatal("subscriber data directory is not a copy of the source database cluster");
2541 :
2542 : /* Subscriber PID file */
2543 9 : snprintf(pidfile, MAXPGPATH, "%s/postmaster.pid", subscriber_dir);
2544 :
2545 : /*
2546 : * The standby server must not be running. If the server is started under
2547 : * service manager and pg_createsubscriber stops it, the service manager
2548 : * might react to this action and start the server again. Therefore,
2549 : * refuse to proceed if the server is running to avoid possible failures.
2550 : */
2551 9 : if (stat(pidfile, &statbuf) == 0)
2552 : {
2553 1 : pg_log_error("standby server is running");
2554 1 : pg_log_error_hint("Stop the standby server and try again.");
2555 1 : exit(1);
2556 : }
2557 :
2558 : /*
2559 : * Start a short-lived standby server with temporary parameters (provided
2560 : * by command-line options). The goal is to avoid connections during the
2561 : * transformation steps.
2562 : */
2563 8 : pg_log_info("starting the standby server with command-line options");
2564 8 : start_standby_server(&opt, true, false);
2565 :
2566 : /* Check if the standby server is ready for logical replication */
2567 8 : check_subscriber(dbinfos.dbinfo);
2568 :
2569 : /* Check if the primary server is ready for logical replication */
2570 6 : check_publisher(dbinfos.dbinfo);
2571 :
2572 : /*
2573 : * Stop the target server. The recovery process requires that the server
2574 : * reaches a consistent state before targeting the recovery stop point.
2575 : * Make sure a consistent state is reached (stop the target server
2576 : * guarantees it) *before* creating the replication slots in
2577 : * setup_publisher().
2578 : */
2579 4 : pg_log_info("stopping the subscriber");
2580 4 : stop_standby_server(subscriber_dir);
2581 :
2582 : /* Create the required objects for each database on publisher */
2583 4 : consistent_lsn = setup_publisher(dbinfos.dbinfo);
2584 :
2585 : /* Write the required recovery parameters */
2586 4 : setup_recovery(dbinfos.dbinfo, subscriber_dir, consistent_lsn);
2587 :
2588 : /*
2589 : * Start subscriber so the recovery parameters will take effect. Wait
2590 : * until accepting connections. We don't want to start logical replication
2591 : * during setup.
2592 : */
2593 4 : pg_log_info("starting the subscriber");
2594 4 : start_standby_server(&opt, true, true);
2595 :
2596 : /* Waiting the subscriber to be promoted */
2597 4 : wait_for_end_recovery(dbinfos.dbinfo[0].subconninfo, &opt);
2598 :
2599 : /*
2600 : * Create the subscription for each database on subscriber. It does not
2601 : * enable it immediately because it needs to adjust the replication start
2602 : * point to the LSN reported by setup_publisher(). It also cleans up
2603 : * publications created by this tool and replication to the standby.
2604 : */
2605 4 : setup_subscriber(dbinfos.dbinfo, consistent_lsn);
2606 :
2607 : /* Remove primary_slot_name if it exists on primary */
2608 4 : drop_primary_replication_slot(dbinfos.dbinfo, primary_slot_name);
2609 :
2610 : /* Remove failover replication slots if they exist on subscriber */
2611 4 : drop_failover_replication_slots(dbinfos.dbinfo);
2612 :
2613 : /* Stop the subscriber */
2614 4 : pg_log_info("stopping the subscriber");
2615 4 : stop_standby_server(subscriber_dir);
2616 :
2617 : /* Change system identifier from subscriber */
2618 4 : modify_subscriber_sysid(&opt);
2619 :
2620 4 : success = true;
2621 :
2622 4 : pg_log_info("Done!");
2623 :
2624 4 : return 0;
2625 : }
|
