Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pg_createsubscriber.c
4 : * Create a new logical replica from a standby server
5 : *
6 : * Copyright (c) 2024-2025, 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/logging.h"
24 : #include "common/pg_prng.h"
25 : #include "common/restricted_token.h"
26 : #include "fe_utils/recovery_gen.h"
27 : #include "fe_utils/simple_list.h"
28 : #include "fe_utils/string_utils.h"
29 : #include "fe_utils/version.h"
30 : #include "getopt_long.h"
31 :
32 : #define DEFAULT_SUB_PORT "50432"
33 : #define OBJECTTYPE_PUBLICATIONS 0x0001
34 :
35 : /* Command-line options */
36 : struct CreateSubscriberOptions
37 : {
38 : char *config_file; /* configuration file */
39 : char *pub_conninfo_str; /* publisher connection string */
40 : char *socket_dir; /* directory for Unix-domain socket, if any */
41 : char *sub_port; /* subscriber port number */
42 : const char *sub_username; /* subscriber username */
43 : bool two_phase; /* enable-two-phase option */
44 : SimpleStringList database_names; /* list of database names */
45 : SimpleStringList pub_names; /* list of publication names */
46 : SimpleStringList sub_names; /* list of subscription names */
47 : SimpleStringList replslot_names; /* list of replication slot names */
48 : int recovery_timeout; /* stop recovery after this time */
49 : bool all_dbs; /* all option */
50 : SimpleStringList objecttypes_to_clean; /* list of object types to cleanup */
51 : };
52 :
53 : /* per-database publication/subscription info */
54 : struct LogicalRepInfo
55 : {
56 : char *dbname; /* database name */
57 : char *pubconninfo; /* publisher connection string */
58 : char *subconninfo; /* subscriber connection string */
59 : char *pubname; /* publication name */
60 : char *subname; /* subscription name */
61 : char *replslotname; /* replication slot name */
62 :
63 : bool made_replslot; /* replication slot was created */
64 : bool made_publication; /* publication was created */
65 : };
66 :
67 : /*
68 : * Information shared across all the databases (or publications and
69 : * subscriptions).
70 : */
71 : struct LogicalRepInfos
72 : {
73 : struct LogicalRepInfo *dbinfo;
74 : bool two_phase; /* enable-two-phase option */
75 : bits32 objecttypes_to_clean; /* flags indicating which object types
76 : * to clean up on subscriber */
77 : };
78 :
79 : static void cleanup_objects_atexit(void);
80 : static void usage();
81 : static char *get_base_conninfo(const char *conninfo, char **dbname);
82 : static char *get_sub_conninfo(const struct CreateSubscriberOptions *opt);
83 : static char *get_exec_path(const char *argv0, const char *progname);
84 : static void check_data_directory(const char *datadir);
85 : static char *concat_conninfo_dbname(const char *conninfo, const char *dbname);
86 : static struct LogicalRepInfo *store_pub_sub_info(const struct CreateSubscriberOptions *opt,
87 : const char *pub_base_conninfo,
88 : const char *sub_base_conninfo);
89 : static PGconn *connect_database(const char *conninfo, bool exit_on_error);
90 : static void disconnect_database(PGconn *conn, bool exit_on_error);
91 : static uint64 get_primary_sysid(const char *conninfo);
92 : static uint64 get_standby_sysid(const char *datadir);
93 : static void modify_subscriber_sysid(const struct CreateSubscriberOptions *opt);
94 : static bool server_is_in_recovery(PGconn *conn);
95 : static char *generate_object_name(PGconn *conn);
96 : static void check_publisher(const struct LogicalRepInfo *dbinfo);
97 : static char *setup_publisher(struct LogicalRepInfo *dbinfo);
98 : static void check_subscriber(const struct LogicalRepInfo *dbinfo);
99 : static void setup_subscriber(struct LogicalRepInfo *dbinfo,
100 : const char *consistent_lsn);
101 : static void setup_recovery(const struct LogicalRepInfo *dbinfo, const char *datadir,
102 : const char *lsn);
103 : static void drop_primary_replication_slot(struct LogicalRepInfo *dbinfo,
104 : const char *slotname);
105 : static void drop_failover_replication_slots(struct LogicalRepInfo *dbinfo);
106 : static char *create_logical_replication_slot(PGconn *conn,
107 : struct LogicalRepInfo *dbinfo);
108 : static void drop_replication_slot(PGconn *conn, struct LogicalRepInfo *dbinfo,
109 : const char *slot_name);
110 : static void pg_ctl_status(const char *pg_ctl_cmd, int rc);
111 : static void start_standby_server(const struct CreateSubscriberOptions *opt,
112 : bool restricted_access,
113 : bool restrict_logical_worker);
114 : static void stop_standby_server(const char *datadir);
115 : static void wait_for_end_recovery(const char *conninfo,
116 : const struct CreateSubscriberOptions *opt);
117 : static void create_publication(PGconn *conn, struct LogicalRepInfo *dbinfo);
118 : static void drop_publication(PGconn *conn, const char *pubname,
119 : const char *dbname, bool *made_publication);
120 : static void check_and_drop_publications(PGconn *conn, struct LogicalRepInfo *dbinfo);
121 : static void create_subscription(PGconn *conn, const struct LogicalRepInfo *dbinfo);
122 : static void set_replication_progress(PGconn *conn, const struct LogicalRepInfo *dbinfo,
123 : const char *lsn);
124 : static void enable_subscription(PGconn *conn, const struct LogicalRepInfo *dbinfo);
125 : static void check_and_drop_existing_subscriptions(PGconn *conn,
126 : const struct LogicalRepInfo *dbinfo);
127 : static void drop_existing_subscriptions(PGconn *conn, const char *subname,
128 : const char *dbname);
129 : static void get_publisher_databases(struct CreateSubscriberOptions *opt,
130 : bool dbnamespecified);
131 :
132 : #define USEC_PER_SEC 1000000
133 : #define WAIT_INTERVAL 1 /* 1 second */
134 :
135 : static const char *progname;
136 :
137 : static char *primary_slot_name = NULL;
138 : static bool dry_run = false;
139 :
140 : static bool success = false;
141 :
142 : static struct LogicalRepInfos dbinfos;
143 : static int num_dbs = 0; /* number of specified databases */
144 : static int num_pubs = 0; /* number of specified publications */
145 : static int num_subs = 0; /* number of specified subscriptions */
146 : static int num_replslots = 0; /* number of specified replication slots */
147 :
148 : static pg_prng_state prng_state;
149 :
150 : static char *pg_ctl_path = NULL;
151 : static char *pg_resetwal_path = NULL;
152 :
153 : /* standby / subscriber data directory */
154 : static char *subscriber_dir = NULL;
155 :
156 : static bool recovery_ended = false;
157 : static bool standby_running = false;
158 :
159 : enum WaitPMResult
160 : {
161 : POSTMASTER_READY,
162 : POSTMASTER_STILL_STARTING
163 : };
164 :
165 :
166 : /*
167 : * Cleanup objects that were created by pg_createsubscriber if there is an
168 : * error.
169 : *
170 : * Publications and replication slots are created on primary. Depending on the
171 : * step it failed, it should remove the already created objects if it is
172 : * possible (sometimes it won't work due to a connection issue).
173 : * There is no cleanup on the target server. The steps on the target server are
174 : * executed *after* promotion, hence, at this point, a failure means recreate
175 : * the physical replica and start again.
176 : */
177 : static void
178 20 : cleanup_objects_atexit(void)
179 : {
180 20 : if (success)
181 8 : return;
182 :
183 : /*
184 : * If the server is promoted, there is no way to use the current setup
185 : * again. Warn the user that a new replication setup should be done before
186 : * trying again.
187 : */
188 12 : if (recovery_ended)
189 : {
190 0 : pg_log_warning("failed after the end of recovery");
191 0 : pg_log_warning_hint("The target server cannot be used as a physical replica anymore. "
192 : "You must recreate the physical replica before continuing.");
193 : }
194 :
195 36 : for (int i = 0; i < num_dbs; i++)
196 : {
197 24 : struct LogicalRepInfo *dbinfo = &dbinfos.dbinfo[i];
198 :
199 24 : if (dbinfo->made_publication || dbinfo->made_replslot)
200 : {
201 : PGconn *conn;
202 :
203 0 : conn = connect_database(dbinfo->pubconninfo, false);
204 0 : if (conn != NULL)
205 : {
206 0 : if (dbinfo->made_publication)
207 0 : drop_publication(conn, dbinfo->pubname, dbinfo->dbname,
208 : &dbinfo->made_publication);
209 0 : if (dbinfo->made_replslot)
210 0 : drop_replication_slot(conn, dbinfo, dbinfo->replslotname);
211 0 : disconnect_database(conn, false);
212 : }
213 : else
214 : {
215 : /*
216 : * If a connection could not be established, inform the user
217 : * that some objects were left on primary and should be
218 : * removed before trying again.
219 : */
220 0 : if (dbinfo->made_publication)
221 : {
222 0 : pg_log_warning("publication \"%s\" created in database \"%s\" on primary was left behind",
223 : dbinfo->pubname,
224 : dbinfo->dbname);
225 0 : pg_log_warning_hint("Drop this publication before trying again.");
226 : }
227 0 : if (dbinfo->made_replslot)
228 : {
229 0 : pg_log_warning("replication slot \"%s\" created in database \"%s\" on primary was left behind",
230 : dbinfo->replslotname,
231 : dbinfo->dbname);
232 0 : pg_log_warning_hint("Drop this replication slot soon to avoid retention of WAL files.");
233 : }
234 : }
235 : }
236 : }
237 :
238 12 : if (standby_running)
239 8 : stop_standby_server(subscriber_dir);
240 : }
241 :
242 : static void
243 2 : usage(void)
244 : {
245 2 : printf(_("%s creates a new logical replica from a standby server.\n\n"),
246 : progname);
247 2 : printf(_("Usage:\n"));
248 2 : printf(_(" %s [OPTION]...\n"), progname);
249 2 : printf(_("\nOptions:\n"));
250 2 : printf(_(" -a, --all create subscriptions for all databases except template\n"
251 : " databases and databases that don't allow connections\n"));
252 2 : printf(_(" -d, --database=DBNAME database in which to create a subscription\n"));
253 2 : printf(_(" -D, --pgdata=DATADIR location for the subscriber data directory\n"));
254 2 : printf(_(" -n, --dry-run dry run, just show what would be done\n"));
255 2 : printf(_(" -p, --subscriber-port=PORT subscriber port number (default %s)\n"), DEFAULT_SUB_PORT);
256 2 : printf(_(" -P, --publisher-server=CONNSTR publisher connection string\n"));
257 2 : printf(_(" -s, --socketdir=DIR socket directory to use (default current dir.)\n"));
258 2 : printf(_(" -t, --recovery-timeout=SECS seconds to wait for recovery to end\n"));
259 2 : printf(_(" -T, --enable-two-phase enable two-phase commit for all subscriptions\n"));
260 2 : printf(_(" -U, --subscriber-username=NAME user name for subscriber connection\n"));
261 2 : printf(_(" -v, --verbose output verbose messages\n"));
262 2 : printf(_(" --clean=OBJECTTYPE drop all objects of the specified type from specified\n"
263 : " databases on the subscriber; accepts: \"%s\"\n"), "publications");
264 2 : printf(_(" --config-file=FILENAME use specified main server configuration\n"
265 : " file when running target cluster\n"));
266 2 : printf(_(" --publication=NAME publication name\n"));
267 2 : printf(_(" --replication-slot=NAME replication slot name\n"));
268 2 : printf(_(" --subscription=NAME subscription name\n"));
269 2 : printf(_(" -V, --version output version information, then exit\n"));
270 2 : printf(_(" -?, --help show this help, then exit\n"));
271 2 : printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT);
272 2 : printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL);
273 2 : }
274 :
275 : /*
276 : * Subroutine to append "keyword=value" to a connection string,
277 : * with proper quoting of the value. (We assume keywords don't need that.)
278 : */
279 : static void
280 214 : appendConnStrItem(PQExpBuffer buf, const char *keyword, const char *val)
281 : {
282 214 : if (buf->len > 0)
283 158 : appendPQExpBufferChar(buf, ' ');
284 214 : appendPQExpBufferStr(buf, keyword);
285 214 : appendPQExpBufferChar(buf, '=');
286 214 : appendConnStrVal(buf, val);
287 214 : }
288 :
289 : /*
290 : * Validate a connection string. Returns a base connection string that is a
291 : * connection string without a database name.
292 : *
293 : * Since we might process multiple databases, each database name will be
294 : * appended to this base connection string to provide a final connection
295 : * string. If the second argument (dbname) is not null, returns dbname if the
296 : * provided connection string contains it.
297 : *
298 : * It is the caller's responsibility to free the returned connection string and
299 : * dbname.
300 : */
301 : static char *
302 28 : get_base_conninfo(const char *conninfo, char **dbname)
303 : {
304 : PQExpBuffer buf;
305 : PQconninfoOption *conn_opts;
306 : PQconninfoOption *conn_opt;
307 28 : char *errmsg = NULL;
308 : char *ret;
309 :
310 28 : conn_opts = PQconninfoParse(conninfo, &errmsg);
311 28 : if (conn_opts == NULL)
312 : {
313 0 : pg_log_error("could not parse connection string: %s", errmsg);
314 0 : PQfreemem(errmsg);
315 0 : return NULL;
316 : }
317 :
318 28 : buf = createPQExpBuffer();
319 1456 : for (conn_opt = conn_opts; conn_opt->keyword != NULL; conn_opt++)
320 : {
321 1428 : if (conn_opt->val != NULL && conn_opt->val[0] != '\0')
322 : {
323 66 : if (strcmp(conn_opt->keyword, "dbname") == 0)
324 : {
325 18 : if (dbname)
326 18 : *dbname = pg_strdup(conn_opt->val);
327 18 : continue;
328 : }
329 48 : appendConnStrItem(buf, conn_opt->keyword, conn_opt->val);
330 : }
331 : }
332 :
333 28 : ret = pg_strdup(buf->data);
334 :
335 28 : destroyPQExpBuffer(buf);
336 28 : PQconninfoFree(conn_opts);
337 :
338 28 : return ret;
339 : }
340 :
341 : /*
342 : * Build a subscriber connection string. Only a few parameters are supported
343 : * since it starts a server with restricted access.
344 : */
345 : static char *
346 28 : get_sub_conninfo(const struct CreateSubscriberOptions *opt)
347 : {
348 28 : PQExpBuffer buf = createPQExpBuffer();
349 : char *ret;
350 :
351 28 : appendConnStrItem(buf, "port", opt->sub_port);
352 : #if !defined(WIN32)
353 28 : appendConnStrItem(buf, "host", opt->socket_dir);
354 : #endif
355 28 : if (opt->sub_username != NULL)
356 0 : appendConnStrItem(buf, "user", opt->sub_username);
357 28 : appendConnStrItem(buf, "fallback_application_name", progname);
358 :
359 28 : ret = pg_strdup(buf->data);
360 :
361 28 : destroyPQExpBuffer(buf);
362 :
363 28 : return ret;
364 : }
365 :
366 : /*
367 : * Verify if a PostgreSQL binary (progname) is available in the same directory as
368 : * pg_createsubscriber and it has the same version. It returns the absolute
369 : * path of the progname.
370 : */
371 : static char *
372 40 : get_exec_path(const char *argv0, const char *progname)
373 : {
374 : char *versionstr;
375 : char *exec_path;
376 : int ret;
377 :
378 40 : versionstr = psprintf("%s (PostgreSQL) %s\n", progname, PG_VERSION);
379 40 : exec_path = pg_malloc(MAXPGPATH);
380 40 : ret = find_other_exec(argv0, progname, versionstr, exec_path);
381 :
382 40 : if (ret < 0)
383 : {
384 : char full_path[MAXPGPATH];
385 :
386 0 : if (find_my_exec(argv0, full_path) < 0)
387 0 : strlcpy(full_path, progname, sizeof(full_path));
388 :
389 0 : if (ret == -1)
390 0 : pg_fatal("program \"%s\" is needed by %s but was not found in the same directory as \"%s\"",
391 : progname, "pg_createsubscriber", full_path);
392 : else
393 0 : pg_fatal("program \"%s\" was found by \"%s\" but was not the same version as %s",
394 : progname, full_path, "pg_createsubscriber");
395 : }
396 :
397 40 : pg_log_debug("%s path is: %s", progname, exec_path);
398 :
399 40 : return exec_path;
400 : }
401 :
402 : /*
403 : * Is it a cluster directory? These are preliminary checks. It is far from
404 : * making an accurate check. If it is not a clone from the publisher, it will
405 : * eventually fail in a future step.
406 : */
407 : static void
408 20 : check_data_directory(const char *datadir)
409 : {
410 : struct stat statbuf;
411 : uint32 major_version;
412 : char *version_str;
413 :
414 20 : pg_log_info("checking if directory \"%s\" is a cluster data directory",
415 : datadir);
416 :
417 20 : if (stat(datadir, &statbuf) != 0)
418 : {
419 0 : if (errno == ENOENT)
420 0 : pg_fatal("data directory \"%s\" does not exist", datadir);
421 : else
422 0 : pg_fatal("could not access directory \"%s\": %m", datadir);
423 : }
424 :
425 : /*
426 : * Retrieve the contents of this cluster's PG_VERSION. We require
427 : * compatibility with the same major version as the one this tool is
428 : * compiled with.
429 : */
430 20 : major_version = GET_PG_MAJORVERSION_NUM(get_pg_version(datadir, &version_str));
431 20 : if (major_version != PG_MAJORVERSION_NUM)
432 : {
433 0 : pg_log_error("data directory is of wrong version");
434 0 : pg_log_error_detail("File \"%s\" contains \"%s\", which is not compatible with this program's version \"%s\".",
435 : "PG_VERSION", version_str, PG_MAJORVERSION);
436 0 : exit(1);
437 : }
438 20 : }
439 :
440 : /*
441 : * Append database name into a base connection string.
442 : *
443 : * dbname is the only parameter that changes so it is not included in the base
444 : * connection string. This function concatenates dbname to build a "real"
445 : * connection string.
446 : */
447 : static char *
448 82 : concat_conninfo_dbname(const char *conninfo, const char *dbname)
449 : {
450 82 : PQExpBuffer buf = createPQExpBuffer();
451 : char *ret;
452 :
453 : Assert(conninfo != NULL);
454 :
455 82 : appendPQExpBufferStr(buf, conninfo);
456 82 : appendConnStrItem(buf, "dbname", dbname);
457 :
458 82 : ret = pg_strdup(buf->data);
459 82 : destroyPQExpBuffer(buf);
460 :
461 82 : return ret;
462 : }
463 :
464 : /*
465 : * Store publication and subscription information.
466 : *
467 : * If publication, replication slot and subscription names were specified,
468 : * store it here. Otherwise, a generated name will be assigned to the object in
469 : * setup_publisher().
470 : */
471 : static struct LogicalRepInfo *
472 20 : store_pub_sub_info(const struct CreateSubscriberOptions *opt,
473 : const char *pub_base_conninfo,
474 : const char *sub_base_conninfo)
475 : {
476 : struct LogicalRepInfo *dbinfo;
477 20 : SimpleStringListCell *pubcell = NULL;
478 20 : SimpleStringListCell *subcell = NULL;
479 20 : SimpleStringListCell *replslotcell = NULL;
480 20 : int i = 0;
481 :
482 20 : dbinfo = pg_malloc_array(struct LogicalRepInfo, num_dbs);
483 :
484 20 : if (num_pubs > 0)
485 4 : pubcell = opt->pub_names.head;
486 20 : if (num_subs > 0)
487 2 : subcell = opt->sub_names.head;
488 20 : if (num_replslots > 0)
489 4 : replslotcell = opt->replslot_names.head;
490 :
491 60 : for (SimpleStringListCell *cell = opt->database_names.head; cell; cell = cell->next)
492 : {
493 : char *conninfo;
494 :
495 : /* Fill publisher attributes */
496 40 : conninfo = concat_conninfo_dbname(pub_base_conninfo, cell->val);
497 40 : dbinfo[i].pubconninfo = conninfo;
498 40 : dbinfo[i].dbname = cell->val;
499 40 : if (num_pubs > 0)
500 8 : dbinfo[i].pubname = pubcell->val;
501 : else
502 32 : dbinfo[i].pubname = NULL;
503 40 : if (num_replslots > 0)
504 6 : dbinfo[i].replslotname = replslotcell->val;
505 : else
506 34 : dbinfo[i].replslotname = NULL;
507 40 : dbinfo[i].made_replslot = false;
508 40 : dbinfo[i].made_publication = false;
509 : /* Fill subscriber attributes */
510 40 : conninfo = concat_conninfo_dbname(sub_base_conninfo, cell->val);
511 40 : dbinfo[i].subconninfo = conninfo;
512 40 : if (num_subs > 0)
513 4 : dbinfo[i].subname = subcell->val;
514 : else
515 36 : dbinfo[i].subname = NULL;
516 : /* Other fields will be filled later */
517 :
518 40 : pg_log_debug("publisher(%d): publication: %s ; replication slot: %s ; connection string: %s", i,
519 : dbinfo[i].pubname ? dbinfo[i].pubname : "(auto)",
520 : dbinfo[i].replslotname ? dbinfo[i].replslotname : "(auto)",
521 : dbinfo[i].pubconninfo);
522 40 : pg_log_debug("subscriber(%d): subscription: %s ; connection string: %s, two_phase: %s", i,
523 : dbinfo[i].subname ? dbinfo[i].subname : "(auto)",
524 : dbinfo[i].subconninfo,
525 : dbinfos.two_phase ? "true" : "false");
526 :
527 40 : if (num_pubs > 0)
528 8 : pubcell = pubcell->next;
529 40 : if (num_subs > 0)
530 4 : subcell = subcell->next;
531 40 : if (num_replslots > 0)
532 6 : replslotcell = replslotcell->next;
533 :
534 40 : i++;
535 : }
536 :
537 20 : return dbinfo;
538 : }
539 :
540 : /*
541 : * Open a new connection. If exit_on_error is true, it has an undesired
542 : * condition and it should exit immediately.
543 : */
544 : static PGconn *
545 114 : connect_database(const char *conninfo, bool exit_on_error)
546 : {
547 : PGconn *conn;
548 : PGresult *res;
549 :
550 114 : conn = PQconnectdb(conninfo);
551 114 : if (PQstatus(conn) != CONNECTION_OK)
552 : {
553 0 : pg_log_error("connection to database failed: %s",
554 : PQerrorMessage(conn));
555 0 : PQfinish(conn);
556 :
557 0 : if (exit_on_error)
558 0 : exit(1);
559 0 : return NULL;
560 : }
561 :
562 : /* Secure search_path */
563 114 : res = PQexec(conn, ALWAYS_SECURE_SEARCH_PATH_SQL);
564 114 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
565 : {
566 0 : pg_log_error("could not clear \"search_path\": %s",
567 : PQresultErrorMessage(res));
568 0 : PQclear(res);
569 0 : PQfinish(conn);
570 :
571 0 : if (exit_on_error)
572 0 : exit(1);
573 0 : return NULL;
574 : }
575 114 : PQclear(res);
576 :
577 114 : return conn;
578 : }
579 :
580 : /*
581 : * Close the connection. If exit_on_error is true, it has an undesired
582 : * condition and it should exit immediately.
583 : */
584 : static void
585 114 : disconnect_database(PGconn *conn, bool exit_on_error)
586 : {
587 : Assert(conn != NULL);
588 :
589 114 : PQfinish(conn);
590 :
591 114 : if (exit_on_error)
592 4 : exit(1);
593 110 : }
594 :
595 : /*
596 : * Obtain the system identifier using the provided connection. It will be used
597 : * to compare if a data directory is a clone of another one.
598 : */
599 : static uint64
600 20 : get_primary_sysid(const char *conninfo)
601 : {
602 : PGconn *conn;
603 : PGresult *res;
604 : uint64 sysid;
605 :
606 20 : pg_log_info("getting system identifier from publisher");
607 :
608 20 : conn = connect_database(conninfo, true);
609 :
610 20 : res = PQexec(conn, "SELECT system_identifier FROM pg_catalog.pg_control_system()");
611 20 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
612 : {
613 0 : pg_log_error("could not get system identifier: %s",
614 : PQresultErrorMessage(res));
615 0 : disconnect_database(conn, true);
616 : }
617 20 : if (PQntuples(res) != 1)
618 : {
619 0 : pg_log_error("could not get system identifier: got %d rows, expected %d row",
620 : PQntuples(res), 1);
621 0 : disconnect_database(conn, true);
622 : }
623 :
624 20 : sysid = strtou64(PQgetvalue(res, 0, 0), NULL, 10);
625 :
626 20 : pg_log_info("system identifier is %" PRIu64 " on publisher", sysid);
627 :
628 20 : PQclear(res);
629 20 : disconnect_database(conn, false);
630 :
631 20 : return sysid;
632 : }
633 :
634 : /*
635 : * Obtain the system identifier from control file. It will be used to compare
636 : * if a data directory is a clone of another one. This routine is used locally
637 : * and avoids a connection.
638 : */
639 : static uint64
640 20 : get_standby_sysid(const char *datadir)
641 : {
642 : ControlFileData *cf;
643 : bool crc_ok;
644 : uint64 sysid;
645 :
646 20 : pg_log_info("getting system identifier from subscriber");
647 :
648 20 : cf = get_controlfile(datadir, &crc_ok);
649 20 : if (!crc_ok)
650 0 : pg_fatal("control file appears to be corrupt");
651 :
652 20 : sysid = cf->system_identifier;
653 :
654 20 : pg_log_info("system identifier is %" PRIu64 " on subscriber", sysid);
655 :
656 20 : pg_free(cf);
657 :
658 20 : return sysid;
659 : }
660 :
661 : /*
662 : * Modify the system identifier. Since a standby server preserves the system
663 : * identifier, it makes sense to change it to avoid situations in which WAL
664 : * files from one of the systems might be used in the other one.
665 : */
666 : static void
667 8 : modify_subscriber_sysid(const struct CreateSubscriberOptions *opt)
668 : {
669 : ControlFileData *cf;
670 : bool crc_ok;
671 : struct timeval tv;
672 :
673 : char *cmd_str;
674 :
675 8 : pg_log_info("modifying system identifier of subscriber");
676 :
677 8 : cf = get_controlfile(subscriber_dir, &crc_ok);
678 8 : if (!crc_ok)
679 0 : pg_fatal("control file appears to be corrupt");
680 :
681 : /*
682 : * Select a new system identifier.
683 : *
684 : * XXX this code was extracted from BootStrapXLOG().
685 : */
686 8 : gettimeofday(&tv, NULL);
687 8 : cf->system_identifier = ((uint64) tv.tv_sec) << 32;
688 8 : cf->system_identifier |= ((uint64) tv.tv_usec) << 12;
689 8 : cf->system_identifier |= getpid() & 0xFFF;
690 :
691 8 : if (!dry_run)
692 2 : update_controlfile(subscriber_dir, cf, true);
693 :
694 8 : pg_log_info("system identifier is %" PRIu64 " on subscriber",
695 : cf->system_identifier);
696 :
697 8 : pg_log_info("running pg_resetwal on the subscriber");
698 :
699 8 : cmd_str = psprintf("\"%s\" -D \"%s\" > \"%s\"", pg_resetwal_path,
700 : subscriber_dir, DEVNULL);
701 :
702 8 : pg_log_debug("pg_resetwal command is: %s", cmd_str);
703 :
704 8 : if (!dry_run)
705 : {
706 2 : int rc = system(cmd_str);
707 :
708 2 : if (rc == 0)
709 2 : pg_log_info("subscriber successfully changed the system identifier");
710 : else
711 0 : pg_fatal("could not change system identifier of subscriber: %s", wait_result_to_str(rc));
712 : }
713 :
714 8 : pg_free(cf);
715 8 : }
716 :
717 : /*
718 : * Generate an object name using a prefix, database oid and a random integer.
719 : * It is used in case the user does not specify an object name (publication,
720 : * subscription, replication slot).
721 : */
722 : static char *
723 16 : generate_object_name(PGconn *conn)
724 : {
725 : PGresult *res;
726 : Oid oid;
727 : uint32 rand;
728 : char *objname;
729 :
730 16 : res = PQexec(conn,
731 : "SELECT oid FROM pg_catalog.pg_database "
732 : "WHERE datname = pg_catalog.current_database()");
733 16 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
734 : {
735 0 : pg_log_error("could not obtain database OID: %s",
736 : PQresultErrorMessage(res));
737 0 : disconnect_database(conn, true);
738 : }
739 :
740 16 : if (PQntuples(res) != 1)
741 : {
742 0 : pg_log_error("could not obtain database OID: got %d rows, expected %d row",
743 : PQntuples(res), 1);
744 0 : disconnect_database(conn, true);
745 : }
746 :
747 : /* Database OID */
748 16 : oid = strtoul(PQgetvalue(res, 0, 0), NULL, 10);
749 :
750 16 : PQclear(res);
751 :
752 : /* Random unsigned integer */
753 16 : rand = pg_prng_uint32(&prng_state);
754 :
755 : /*
756 : * Build the object name. The name must not exceed NAMEDATALEN - 1. This
757 : * current schema uses a maximum of 40 characters (20 + 10 + 1 + 8 +
758 : * '\0').
759 : */
760 16 : objname = psprintf("pg_createsubscriber_%u_%x", oid, rand);
761 :
762 16 : return objname;
763 : }
764 :
765 : /*
766 : * Create the publications and replication slots in preparation for logical
767 : * replication. Returns the LSN from latest replication slot. It will be the
768 : * replication start point that is used to adjust the subscriptions (see
769 : * set_replication_progress).
770 : */
771 : static char *
772 8 : setup_publisher(struct LogicalRepInfo *dbinfo)
773 : {
774 8 : char *lsn = NULL;
775 :
776 8 : pg_prng_seed(&prng_state, (uint64) (getpid() ^ time(NULL)));
777 :
778 24 : for (int i = 0; i < num_dbs; i++)
779 : {
780 : PGconn *conn;
781 16 : char *genname = NULL;
782 :
783 16 : conn = connect_database(dbinfo[i].pubconninfo, true);
784 :
785 : /*
786 : * If an object name was not specified as command-line options, assign
787 : * a generated object name. The replication slot has a different rule.
788 : * The subscription name is assigned to the replication slot name if
789 : * no replication slot is specified. It follows the same rule as
790 : * CREATE SUBSCRIPTION.
791 : */
792 16 : if (num_pubs == 0 || num_subs == 0 || num_replslots == 0)
793 16 : genname = generate_object_name(conn);
794 16 : if (num_pubs == 0)
795 8 : dbinfo[i].pubname = pg_strdup(genname);
796 16 : if (num_subs == 0)
797 12 : dbinfo[i].subname = pg_strdup(genname);
798 16 : if (num_replslots == 0)
799 10 : dbinfo[i].replslotname = pg_strdup(dbinfo[i].subname);
800 :
801 : /*
802 : * Create publication on publisher. This step should be executed
803 : * *before* promoting the subscriber to avoid any transactions between
804 : * consistent LSN and the new publication rows (such transactions
805 : * wouldn't see the new publication rows resulting in an error).
806 : */
807 16 : create_publication(conn, &dbinfo[i]);
808 :
809 : /* Create replication slot on publisher */
810 16 : if (lsn)
811 2 : pg_free(lsn);
812 16 : lsn = create_logical_replication_slot(conn, &dbinfo[i]);
813 16 : if (lsn == NULL && !dry_run)
814 0 : exit(1);
815 :
816 : /*
817 : * Since we are using the LSN returned by the last replication slot as
818 : * recovery_target_lsn, this LSN is ahead of the current WAL position
819 : * and the recovery waits until the publisher writes a WAL record to
820 : * reach the target and ends the recovery. On idle systems, this wait
821 : * time is unpredictable and could lead to failure in promoting the
822 : * subscriber. To avoid that, insert a harmless WAL record.
823 : */
824 16 : if (i == num_dbs - 1 && !dry_run)
825 : {
826 : PGresult *res;
827 :
828 2 : res = PQexec(conn, "SELECT pg_log_standby_snapshot()");
829 2 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
830 : {
831 0 : pg_log_error("could not write an additional WAL record: %s",
832 : PQresultErrorMessage(res));
833 0 : disconnect_database(conn, true);
834 : }
835 2 : PQclear(res);
836 : }
837 :
838 16 : disconnect_database(conn, false);
839 : }
840 :
841 8 : return lsn;
842 : }
843 :
844 : /*
845 : * Is recovery still in progress?
846 : */
847 : static bool
848 36 : server_is_in_recovery(PGconn *conn)
849 : {
850 : PGresult *res;
851 : int ret;
852 :
853 36 : res = PQexec(conn, "SELECT pg_catalog.pg_is_in_recovery()");
854 :
855 36 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
856 : {
857 0 : pg_log_error("could not obtain recovery progress: %s",
858 : PQresultErrorMessage(res));
859 0 : disconnect_database(conn, true);
860 : }
861 :
862 :
863 36 : ret = strcmp("t", PQgetvalue(res, 0, 0));
864 :
865 36 : PQclear(res);
866 :
867 36 : return ret == 0;
868 : }
869 :
870 : /*
871 : * Is the primary server ready for logical replication?
872 : *
873 : * XXX Does it not allow a synchronous replica?
874 : */
875 : static void
876 12 : check_publisher(const struct LogicalRepInfo *dbinfo)
877 : {
878 : PGconn *conn;
879 : PGresult *res;
880 12 : bool failed = false;
881 :
882 : char *wal_level;
883 : int max_repslots;
884 : int cur_repslots;
885 : int max_walsenders;
886 : int cur_walsenders;
887 : int max_prepared_transactions;
888 : char *max_slot_wal_keep_size;
889 :
890 12 : pg_log_info("checking settings on publisher");
891 :
892 12 : conn = connect_database(dbinfo[0].pubconninfo, true);
893 :
894 : /*
895 : * If the primary server is in recovery (i.e. cascading replication),
896 : * objects (publication) cannot be created because it is read only.
897 : */
898 12 : if (server_is_in_recovery(conn))
899 : {
900 2 : pg_log_error("primary server cannot be in recovery");
901 2 : disconnect_database(conn, true);
902 : }
903 :
904 : /*------------------------------------------------------------------------
905 : * Logical replication requires a few parameters to be set on publisher.
906 : * Since these parameters are not a requirement for physical replication,
907 : * we should check it to make sure it won't fail.
908 : *
909 : * - wal_level = logical
910 : * - max_replication_slots >= current + number of dbs to be converted
911 : * - max_wal_senders >= current + number of dbs to be converted
912 : * - max_slot_wal_keep_size = -1 (to prevent deletion of required WAL files)
913 : * -----------------------------------------------------------------------
914 : */
915 10 : res = PQexec(conn,
916 : "SELECT pg_catalog.current_setting('wal_level'),"
917 : " pg_catalog.current_setting('max_replication_slots'),"
918 : " (SELECT count(*) FROM pg_catalog.pg_replication_slots),"
919 : " pg_catalog.current_setting('max_wal_senders'),"
920 : " (SELECT count(*) FROM pg_catalog.pg_stat_activity WHERE backend_type = 'walsender'),"
921 : " pg_catalog.current_setting('max_prepared_transactions'),"
922 : " pg_catalog.current_setting('max_slot_wal_keep_size')");
923 :
924 10 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
925 : {
926 0 : pg_log_error("could not obtain publisher settings: %s",
927 : PQresultErrorMessage(res));
928 0 : disconnect_database(conn, true);
929 : }
930 :
931 10 : wal_level = pg_strdup(PQgetvalue(res, 0, 0));
932 10 : max_repslots = atoi(PQgetvalue(res, 0, 1));
933 10 : cur_repslots = atoi(PQgetvalue(res, 0, 2));
934 10 : max_walsenders = atoi(PQgetvalue(res, 0, 3));
935 10 : cur_walsenders = atoi(PQgetvalue(res, 0, 4));
936 10 : max_prepared_transactions = atoi(PQgetvalue(res, 0, 5));
937 10 : max_slot_wal_keep_size = pg_strdup(PQgetvalue(res, 0, 6));
938 :
939 10 : PQclear(res);
940 :
941 10 : pg_log_debug("publisher: wal_level: %s", wal_level);
942 10 : pg_log_debug("publisher: max_replication_slots: %d", max_repslots);
943 10 : pg_log_debug("publisher: current replication slots: %d", cur_repslots);
944 10 : pg_log_debug("publisher: max_wal_senders: %d", max_walsenders);
945 10 : pg_log_debug("publisher: current wal senders: %d", cur_walsenders);
946 10 : pg_log_debug("publisher: max_prepared_transactions: %d",
947 : max_prepared_transactions);
948 10 : pg_log_debug("publisher: max_slot_wal_keep_size: %s",
949 : max_slot_wal_keep_size);
950 :
951 10 : disconnect_database(conn, false);
952 :
953 10 : if (strcmp(wal_level, "logical") != 0)
954 : {
955 2 : pg_log_error("publisher requires \"wal_level\" >= \"logical\"");
956 2 : failed = true;
957 : }
958 :
959 10 : if (max_repslots - cur_repslots < num_dbs)
960 : {
961 2 : pg_log_error("publisher requires %d replication slots, but only %d remain",
962 : num_dbs, max_repslots - cur_repslots);
963 2 : pg_log_error_hint("Increase the configuration parameter \"%s\" to at least %d.",
964 : "max_replication_slots", cur_repslots + num_dbs);
965 2 : failed = true;
966 : }
967 :
968 10 : if (max_walsenders - cur_walsenders < num_dbs)
969 : {
970 2 : pg_log_error("publisher requires %d WAL sender processes, but only %d remain",
971 : num_dbs, max_walsenders - cur_walsenders);
972 2 : pg_log_error_hint("Increase the configuration parameter \"%s\" to at least %d.",
973 : "max_wal_senders", cur_walsenders + num_dbs);
974 2 : failed = true;
975 : }
976 :
977 10 : if (max_prepared_transactions != 0 && !dbinfos.two_phase)
978 : {
979 0 : pg_log_warning("two_phase option will not be enabled for replication slots");
980 0 : pg_log_warning_detail("Subscriptions will be created with the two_phase option disabled. "
981 : "Prepared transactions will be replicated at COMMIT PREPARED.");
982 0 : pg_log_warning_hint("You can use the command-line option --enable-two-phase to enable two_phase.");
983 : }
984 :
985 : /*
986 : * Validate 'max_slot_wal_keep_size'. If this parameter is set to a
987 : * non-default value, it may cause replication failures due to required
988 : * WAL files being prematurely removed.
989 : */
990 10 : if (dry_run && (strcmp(max_slot_wal_keep_size, "-1") != 0))
991 : {
992 0 : pg_log_warning("required WAL could be removed from the publisher");
993 0 : pg_log_warning_hint("Set the configuration parameter \"%s\" to -1 to ensure that required WAL files are not prematurely removed.",
994 : "max_slot_wal_keep_size");
995 : }
996 :
997 10 : pg_free(wal_level);
998 :
999 10 : if (failed)
1000 2 : exit(1);
1001 8 : }
1002 :
1003 : /*
1004 : * Is the standby server ready for logical replication?
1005 : *
1006 : * XXX Does it not allow a time-delayed replica?
1007 : *
1008 : * XXX In a cascaded replication scenario (P -> S -> C), if the target server
1009 : * is S, it cannot detect there is a replica (server C) because server S starts
1010 : * accepting only local connections and server C cannot connect to it. Hence,
1011 : * there is not a reliable way to provide a suitable error saying the server C
1012 : * will be broken at the end of this process (due to pg_resetwal).
1013 : */
1014 : static void
1015 16 : check_subscriber(const struct LogicalRepInfo *dbinfo)
1016 : {
1017 : PGconn *conn;
1018 : PGresult *res;
1019 16 : bool failed = false;
1020 :
1021 : int max_lrworkers;
1022 : int max_reporigins;
1023 : int max_wprocs;
1024 :
1025 16 : pg_log_info("checking settings on subscriber");
1026 :
1027 16 : conn = connect_database(dbinfo[0].subconninfo, true);
1028 :
1029 : /* The target server must be a standby */
1030 16 : if (!server_is_in_recovery(conn))
1031 : {
1032 2 : pg_log_error("target server must be a standby");
1033 2 : disconnect_database(conn, true);
1034 : }
1035 :
1036 : /*------------------------------------------------------------------------
1037 : * Logical replication requires a few parameters to be set on subscriber.
1038 : * Since these parameters are not a requirement for physical replication,
1039 : * we should check it to make sure it won't fail.
1040 : *
1041 : * - max_active_replication_origins >= number of dbs to be converted
1042 : * - max_logical_replication_workers >= number of dbs to be converted
1043 : * - max_worker_processes >= 1 + number of dbs to be converted
1044 : *------------------------------------------------------------------------
1045 : */
1046 14 : res = PQexec(conn,
1047 : "SELECT setting FROM pg_catalog.pg_settings WHERE name IN ("
1048 : "'max_logical_replication_workers', "
1049 : "'max_active_replication_origins', "
1050 : "'max_worker_processes', "
1051 : "'primary_slot_name') "
1052 : "ORDER BY name");
1053 :
1054 14 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1055 : {
1056 0 : pg_log_error("could not obtain subscriber settings: %s",
1057 : PQresultErrorMessage(res));
1058 0 : disconnect_database(conn, true);
1059 : }
1060 :
1061 14 : max_reporigins = atoi(PQgetvalue(res, 0, 0));
1062 14 : max_lrworkers = atoi(PQgetvalue(res, 1, 0));
1063 14 : max_wprocs = atoi(PQgetvalue(res, 2, 0));
1064 14 : if (strcmp(PQgetvalue(res, 3, 0), "") != 0)
1065 12 : primary_slot_name = pg_strdup(PQgetvalue(res, 3, 0));
1066 :
1067 14 : pg_log_debug("subscriber: max_logical_replication_workers: %d",
1068 : max_lrworkers);
1069 14 : pg_log_debug("subscriber: max_active_replication_origins: %d", max_reporigins);
1070 14 : pg_log_debug("subscriber: max_worker_processes: %d", max_wprocs);
1071 14 : if (primary_slot_name)
1072 12 : pg_log_debug("subscriber: primary_slot_name: %s", primary_slot_name);
1073 :
1074 14 : PQclear(res);
1075 :
1076 14 : disconnect_database(conn, false);
1077 :
1078 14 : if (max_reporigins < num_dbs)
1079 : {
1080 2 : pg_log_error("subscriber requires %d active replication origins, but only %d remain",
1081 : num_dbs, max_reporigins);
1082 2 : pg_log_error_hint("Increase the configuration parameter \"%s\" to at least %d.",
1083 : "max_active_replication_origins", num_dbs);
1084 2 : failed = true;
1085 : }
1086 :
1087 14 : if (max_lrworkers < num_dbs)
1088 : {
1089 2 : pg_log_error("subscriber requires %d logical replication workers, but only %d remain",
1090 : num_dbs, max_lrworkers);
1091 2 : pg_log_error_hint("Increase the configuration parameter \"%s\" to at least %d.",
1092 : "max_logical_replication_workers", num_dbs);
1093 2 : failed = true;
1094 : }
1095 :
1096 14 : if (max_wprocs < num_dbs + 1)
1097 : {
1098 2 : pg_log_error("subscriber requires %d worker processes, but only %d remain",
1099 : num_dbs + 1, max_wprocs);
1100 2 : pg_log_error_hint("Increase the configuration parameter \"%s\" to at least %d.",
1101 : "max_worker_processes", num_dbs + 1);
1102 2 : failed = true;
1103 : }
1104 :
1105 14 : if (failed)
1106 2 : exit(1);
1107 12 : }
1108 :
1109 : /*
1110 : * Drop a specified subscription. This is to avoid duplicate subscriptions on
1111 : * the primary (publisher node) and the newly created subscriber. We
1112 : * shouldn't drop the associated slot as that would be used by the publisher
1113 : * node.
1114 : */
1115 : static void
1116 8 : drop_existing_subscriptions(PGconn *conn, const char *subname, const char *dbname)
1117 : {
1118 8 : PQExpBuffer query = createPQExpBuffer();
1119 : PGresult *res;
1120 :
1121 : Assert(conn != NULL);
1122 :
1123 : /*
1124 : * Construct a query string. These commands are allowed to be executed
1125 : * within a transaction.
1126 : */
1127 8 : appendPQExpBuffer(query, "ALTER SUBSCRIPTION %s DISABLE;",
1128 : subname);
1129 8 : appendPQExpBuffer(query, " ALTER SUBSCRIPTION %s SET (slot_name = NONE);",
1130 : subname);
1131 8 : appendPQExpBuffer(query, " DROP SUBSCRIPTION %s;", subname);
1132 :
1133 8 : pg_log_info("dropping subscription \"%s\" in database \"%s\"",
1134 : subname, dbname);
1135 :
1136 8 : if (!dry_run)
1137 : {
1138 2 : res = PQexec(conn, query->data);
1139 :
1140 2 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1141 : {
1142 0 : pg_log_error("could not drop subscription \"%s\": %s",
1143 : subname, PQresultErrorMessage(res));
1144 0 : disconnect_database(conn, true);
1145 : }
1146 :
1147 2 : PQclear(res);
1148 : }
1149 :
1150 8 : destroyPQExpBuffer(query);
1151 8 : }
1152 :
1153 : /*
1154 : * Retrieve and drop the pre-existing subscriptions.
1155 : */
1156 : static void
1157 16 : check_and_drop_existing_subscriptions(PGconn *conn,
1158 : const struct LogicalRepInfo *dbinfo)
1159 : {
1160 16 : PQExpBuffer query = createPQExpBuffer();
1161 : char *dbname;
1162 : PGresult *res;
1163 :
1164 : Assert(conn != NULL);
1165 :
1166 16 : dbname = PQescapeLiteral(conn, dbinfo->dbname, strlen(dbinfo->dbname));
1167 :
1168 16 : appendPQExpBuffer(query,
1169 : "SELECT s.subname FROM pg_catalog.pg_subscription s "
1170 : "INNER JOIN pg_catalog.pg_database d ON (s.subdbid = d.oid) "
1171 : "WHERE d.datname = %s",
1172 : dbname);
1173 16 : res = PQexec(conn, query->data);
1174 :
1175 16 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1176 : {
1177 0 : pg_log_error("could not obtain pre-existing subscriptions: %s",
1178 : PQresultErrorMessage(res));
1179 0 : disconnect_database(conn, true);
1180 : }
1181 :
1182 24 : for (int i = 0; i < PQntuples(res); i++)
1183 8 : drop_existing_subscriptions(conn, PQgetvalue(res, i, 0),
1184 8 : dbinfo->dbname);
1185 :
1186 16 : PQclear(res);
1187 16 : destroyPQExpBuffer(query);
1188 16 : PQfreemem(dbname);
1189 16 : }
1190 :
1191 : /*
1192 : * Create the subscriptions, adjust the initial location for logical
1193 : * replication and enable the subscriptions. That's the last step for logical
1194 : * replication setup.
1195 : */
1196 : static void
1197 8 : setup_subscriber(struct LogicalRepInfo *dbinfo, const char *consistent_lsn)
1198 : {
1199 24 : for (int i = 0; i < num_dbs; i++)
1200 : {
1201 : PGconn *conn;
1202 :
1203 : /* Connect to subscriber. */
1204 16 : conn = connect_database(dbinfo[i].subconninfo, true);
1205 :
1206 : /*
1207 : * We don't need the pre-existing subscriptions on the newly formed
1208 : * subscriber. They can connect to other publisher nodes and either
1209 : * get some unwarranted data or can lead to ERRORs in connecting to
1210 : * such nodes.
1211 : */
1212 16 : check_and_drop_existing_subscriptions(conn, &dbinfo[i]);
1213 :
1214 : /* Check and drop the required publications in the given database. */
1215 16 : check_and_drop_publications(conn, &dbinfo[i]);
1216 :
1217 16 : create_subscription(conn, &dbinfo[i]);
1218 :
1219 : /* Set the replication progress to the correct LSN */
1220 16 : set_replication_progress(conn, &dbinfo[i], consistent_lsn);
1221 :
1222 : /* Enable subscription */
1223 16 : enable_subscription(conn, &dbinfo[i]);
1224 :
1225 16 : disconnect_database(conn, false);
1226 : }
1227 8 : }
1228 :
1229 : /*
1230 : * Write the required recovery parameters.
1231 : */
1232 : static void
1233 8 : setup_recovery(const struct LogicalRepInfo *dbinfo, const char *datadir, const char *lsn)
1234 : {
1235 : PGconn *conn;
1236 : PQExpBuffer recoveryconfcontents;
1237 :
1238 : /*
1239 : * Despite of the recovery parameters will be written to the subscriber,
1240 : * use a publisher connection. The primary_conninfo is generated using the
1241 : * connection settings.
1242 : */
1243 8 : conn = connect_database(dbinfo[0].pubconninfo, true);
1244 :
1245 : /*
1246 : * Write recovery parameters.
1247 : *
1248 : * The subscriber is not running yet. In dry run mode, the recovery
1249 : * parameters *won't* be written. An invalid LSN is used for printing
1250 : * purposes. Additional recovery parameters are added here. It avoids
1251 : * unexpected behavior such as end of recovery as soon as a consistent
1252 : * state is reached (recovery_target) and failure due to multiple recovery
1253 : * targets (name, time, xid, LSN).
1254 : */
1255 8 : recoveryconfcontents = GenerateRecoveryConfig(conn, NULL, NULL);
1256 8 : appendPQExpBufferStr(recoveryconfcontents, "recovery_target = ''\n");
1257 8 : appendPQExpBufferStr(recoveryconfcontents,
1258 : "recovery_target_timeline = 'latest'\n");
1259 :
1260 : /*
1261 : * Set recovery_target_inclusive = false to avoid reapplying the
1262 : * transaction committed at 'lsn' after subscription is enabled. This is
1263 : * because the provided 'lsn' is also used as the replication start point
1264 : * for the subscription. So, the server can send the transaction committed
1265 : * at that 'lsn' after replication is started which can lead to applying
1266 : * the same transaction twice if we keep recovery_target_inclusive = true.
1267 : */
1268 8 : appendPQExpBufferStr(recoveryconfcontents,
1269 : "recovery_target_inclusive = false\n");
1270 8 : appendPQExpBufferStr(recoveryconfcontents,
1271 : "recovery_target_action = promote\n");
1272 8 : appendPQExpBufferStr(recoveryconfcontents, "recovery_target_name = ''\n");
1273 8 : appendPQExpBufferStr(recoveryconfcontents, "recovery_target_time = ''\n");
1274 8 : appendPQExpBufferStr(recoveryconfcontents, "recovery_target_xid = ''\n");
1275 :
1276 8 : if (dry_run)
1277 : {
1278 6 : appendPQExpBufferStr(recoveryconfcontents, "# dry run mode");
1279 6 : appendPQExpBuffer(recoveryconfcontents,
1280 : "recovery_target_lsn = '%X/%08X'\n",
1281 6 : LSN_FORMAT_ARGS((XLogRecPtr) InvalidXLogRecPtr));
1282 : }
1283 : else
1284 : {
1285 2 : appendPQExpBuffer(recoveryconfcontents, "recovery_target_lsn = '%s'\n",
1286 : lsn);
1287 2 : WriteRecoveryConfig(conn, datadir, recoveryconfcontents);
1288 : }
1289 8 : disconnect_database(conn, false);
1290 :
1291 8 : pg_log_debug("recovery parameters:\n%s", recoveryconfcontents->data);
1292 8 : }
1293 :
1294 : /*
1295 : * Drop physical replication slot on primary if the standby was using it. After
1296 : * the transformation, it has no use.
1297 : *
1298 : * XXX we might not fail here. Instead, we provide a warning so the user
1299 : * eventually drops this replication slot later.
1300 : */
1301 : static void
1302 8 : drop_primary_replication_slot(struct LogicalRepInfo *dbinfo, const char *slotname)
1303 : {
1304 : PGconn *conn;
1305 :
1306 : /* Replication slot does not exist, do nothing */
1307 8 : if (!primary_slot_name)
1308 0 : return;
1309 :
1310 8 : conn = connect_database(dbinfo[0].pubconninfo, false);
1311 8 : if (conn != NULL)
1312 : {
1313 8 : drop_replication_slot(conn, &dbinfo[0], slotname);
1314 8 : disconnect_database(conn, false);
1315 : }
1316 : else
1317 : {
1318 0 : pg_log_warning("could not drop replication slot \"%s\" on primary",
1319 : slotname);
1320 0 : pg_log_warning_hint("Drop this replication slot soon to avoid retention of WAL files.");
1321 : }
1322 : }
1323 :
1324 : /*
1325 : * Drop failover replication slots on subscriber. After the transformation,
1326 : * they have no use.
1327 : *
1328 : * XXX We do not fail here. Instead, we provide a warning so the user can drop
1329 : * them later.
1330 : */
1331 : static void
1332 8 : drop_failover_replication_slots(struct LogicalRepInfo *dbinfo)
1333 : {
1334 : PGconn *conn;
1335 : PGresult *res;
1336 :
1337 8 : conn = connect_database(dbinfo[0].subconninfo, false);
1338 8 : if (conn != NULL)
1339 : {
1340 : /* Get failover replication slot names */
1341 8 : res = PQexec(conn,
1342 : "SELECT slot_name FROM pg_catalog.pg_replication_slots WHERE failover");
1343 :
1344 8 : if (PQresultStatus(res) == PGRES_TUPLES_OK)
1345 : {
1346 : /* Remove failover replication slots from subscriber */
1347 16 : for (int i = 0; i < PQntuples(res); i++)
1348 8 : drop_replication_slot(conn, &dbinfo[0], PQgetvalue(res, i, 0));
1349 : }
1350 : else
1351 : {
1352 0 : pg_log_warning("could not obtain failover replication slot information: %s",
1353 : PQresultErrorMessage(res));
1354 0 : pg_log_warning_hint("Drop the failover replication slots on subscriber soon to avoid retention of WAL files.");
1355 : }
1356 :
1357 8 : PQclear(res);
1358 8 : disconnect_database(conn, false);
1359 : }
1360 : else
1361 : {
1362 0 : pg_log_warning("could not drop failover replication slot");
1363 0 : pg_log_warning_hint("Drop the failover replication slots on subscriber soon to avoid retention of WAL files.");
1364 : }
1365 8 : }
1366 :
1367 : /*
1368 : * Create a logical replication slot and returns a LSN.
1369 : *
1370 : * CreateReplicationSlot() is not used because it does not provide the one-row
1371 : * result set that contains the LSN.
1372 : */
1373 : static char *
1374 16 : create_logical_replication_slot(PGconn *conn, struct LogicalRepInfo *dbinfo)
1375 : {
1376 16 : PQExpBuffer str = createPQExpBuffer();
1377 16 : PGresult *res = NULL;
1378 16 : const char *slot_name = dbinfo->replslotname;
1379 : char *slot_name_esc;
1380 16 : char *lsn = NULL;
1381 :
1382 : Assert(conn != NULL);
1383 :
1384 16 : pg_log_info("creating the replication slot \"%s\" in database \"%s\" on publisher",
1385 : slot_name, dbinfo->dbname);
1386 :
1387 16 : slot_name_esc = PQescapeLiteral(conn, slot_name, strlen(slot_name));
1388 :
1389 16 : appendPQExpBuffer(str,
1390 : "SELECT lsn FROM pg_catalog.pg_create_logical_replication_slot(%s, 'pgoutput', false, %s, false)",
1391 : slot_name_esc,
1392 16 : dbinfos.two_phase ? "true" : "false");
1393 :
1394 16 : PQfreemem(slot_name_esc);
1395 :
1396 16 : pg_log_debug("command is: %s", str->data);
1397 :
1398 16 : if (!dry_run)
1399 : {
1400 4 : res = PQexec(conn, str->data);
1401 4 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1402 : {
1403 0 : pg_log_error("could not create replication slot \"%s\" in database \"%s\": %s",
1404 : slot_name, dbinfo->dbname,
1405 : PQresultErrorMessage(res));
1406 0 : PQclear(res);
1407 0 : destroyPQExpBuffer(str);
1408 0 : return NULL;
1409 : }
1410 :
1411 4 : lsn = pg_strdup(PQgetvalue(res, 0, 0));
1412 4 : PQclear(res);
1413 : }
1414 :
1415 : /* For cleanup purposes */
1416 16 : dbinfo->made_replslot = true;
1417 :
1418 16 : destroyPQExpBuffer(str);
1419 :
1420 16 : return lsn;
1421 : }
1422 :
1423 : static void
1424 16 : drop_replication_slot(PGconn *conn, struct LogicalRepInfo *dbinfo,
1425 : const char *slot_name)
1426 : {
1427 16 : PQExpBuffer str = createPQExpBuffer();
1428 : char *slot_name_esc;
1429 : PGresult *res;
1430 :
1431 : Assert(conn != NULL);
1432 :
1433 16 : pg_log_info("dropping the replication slot \"%s\" in database \"%s\"",
1434 : slot_name, dbinfo->dbname);
1435 :
1436 16 : slot_name_esc = PQescapeLiteral(conn, slot_name, strlen(slot_name));
1437 :
1438 16 : appendPQExpBuffer(str, "SELECT pg_catalog.pg_drop_replication_slot(%s)", slot_name_esc);
1439 :
1440 16 : PQfreemem(slot_name_esc);
1441 :
1442 16 : pg_log_debug("command is: %s", str->data);
1443 :
1444 16 : if (!dry_run)
1445 : {
1446 4 : res = PQexec(conn, str->data);
1447 4 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1448 : {
1449 0 : pg_log_error("could not drop replication slot \"%s\" in database \"%s\": %s",
1450 : slot_name, dbinfo->dbname, PQresultErrorMessage(res));
1451 0 : dbinfo->made_replslot = false; /* don't try again. */
1452 : }
1453 :
1454 4 : PQclear(res);
1455 : }
1456 :
1457 16 : destroyPQExpBuffer(str);
1458 16 : }
1459 :
1460 : /*
1461 : * Reports a suitable message if pg_ctl fails.
1462 : */
1463 : static void
1464 48 : pg_ctl_status(const char *pg_ctl_cmd, int rc)
1465 : {
1466 48 : if (rc != 0)
1467 : {
1468 0 : if (WIFEXITED(rc))
1469 : {
1470 0 : pg_log_error("pg_ctl failed with exit code %d", WEXITSTATUS(rc));
1471 : }
1472 0 : else if (WIFSIGNALED(rc))
1473 : {
1474 : #if defined(WIN32)
1475 : pg_log_error("pg_ctl was terminated by exception 0x%X",
1476 : WTERMSIG(rc));
1477 : pg_log_error_detail("See C include file \"ntstatus.h\" for a description of the hexadecimal value.");
1478 : #else
1479 0 : pg_log_error("pg_ctl was terminated by signal %d: %s",
1480 : WTERMSIG(rc), pg_strsignal(WTERMSIG(rc)));
1481 : #endif
1482 : }
1483 : else
1484 : {
1485 0 : pg_log_error("pg_ctl exited with unrecognized status %d", rc);
1486 : }
1487 :
1488 0 : pg_log_error_detail("The failed command was: %s", pg_ctl_cmd);
1489 0 : exit(1);
1490 : }
1491 48 : }
1492 :
1493 : static void
1494 24 : start_standby_server(const struct CreateSubscriberOptions *opt, bool restricted_access,
1495 : bool restrict_logical_worker)
1496 : {
1497 24 : PQExpBuffer pg_ctl_cmd = createPQExpBuffer();
1498 : int rc;
1499 :
1500 24 : appendPQExpBuffer(pg_ctl_cmd, "\"%s\" start -D ", pg_ctl_path);
1501 24 : appendShellString(pg_ctl_cmd, subscriber_dir);
1502 24 : appendPQExpBufferStr(pg_ctl_cmd, " -s -o \"-c sync_replication_slots=off\"");
1503 :
1504 : /* Prevent unintended slot invalidation */
1505 24 : appendPQExpBufferStr(pg_ctl_cmd, " -o \"-c idle_replication_slot_timeout=0\"");
1506 :
1507 24 : if (restricted_access)
1508 : {
1509 24 : appendPQExpBuffer(pg_ctl_cmd, " -o \"-p %s\"", opt->sub_port);
1510 : #if !defined(WIN32)
1511 :
1512 : /*
1513 : * An empty listen_addresses list means the server does not listen on
1514 : * any IP interfaces; only Unix-domain sockets can be used to connect
1515 : * to the server. Prevent external connections to minimize the chance
1516 : * of failure.
1517 : */
1518 24 : appendPQExpBufferStr(pg_ctl_cmd, " -o \"-c listen_addresses='' -c unix_socket_permissions=0700");
1519 24 : if (opt->socket_dir)
1520 24 : appendPQExpBuffer(pg_ctl_cmd, " -c unix_socket_directories='%s'",
1521 24 : opt->socket_dir);
1522 24 : appendPQExpBufferChar(pg_ctl_cmd, '"');
1523 : #endif
1524 : }
1525 24 : if (opt->config_file != NULL)
1526 0 : appendPQExpBuffer(pg_ctl_cmd, " -o \"-c config_file=%s\"",
1527 0 : opt->config_file);
1528 :
1529 : /* Suppress to start logical replication if requested */
1530 24 : if (restrict_logical_worker)
1531 8 : appendPQExpBufferStr(pg_ctl_cmd, " -o \"-c max_logical_replication_workers=0\"");
1532 :
1533 24 : pg_log_debug("pg_ctl command is: %s", pg_ctl_cmd->data);
1534 24 : rc = system(pg_ctl_cmd->data);
1535 24 : pg_ctl_status(pg_ctl_cmd->data, rc);
1536 24 : standby_running = true;
1537 24 : destroyPQExpBuffer(pg_ctl_cmd);
1538 24 : pg_log_info("server was started");
1539 24 : }
1540 :
1541 : static void
1542 24 : stop_standby_server(const char *datadir)
1543 : {
1544 : char *pg_ctl_cmd;
1545 : int rc;
1546 :
1547 24 : pg_ctl_cmd = psprintf("\"%s\" stop -D \"%s\" -s", pg_ctl_path,
1548 : datadir);
1549 24 : pg_log_debug("pg_ctl command is: %s", pg_ctl_cmd);
1550 24 : rc = system(pg_ctl_cmd);
1551 24 : pg_ctl_status(pg_ctl_cmd, rc);
1552 24 : standby_running = false;
1553 24 : pg_log_info("server was stopped");
1554 24 : }
1555 :
1556 : /*
1557 : * Returns after the server finishes the recovery process.
1558 : *
1559 : * If recovery_timeout option is set, terminate abnormally without finishing
1560 : * the recovery process. By default, it waits forever.
1561 : *
1562 : * XXX Is the recovery process still in progress? When recovery process has a
1563 : * better progress reporting mechanism, it should be added here.
1564 : */
1565 : static void
1566 8 : wait_for_end_recovery(const char *conninfo, const struct CreateSubscriberOptions *opt)
1567 : {
1568 : PGconn *conn;
1569 8 : int status = POSTMASTER_STILL_STARTING;
1570 8 : int timer = 0;
1571 :
1572 8 : pg_log_info("waiting for the target server to reach the consistent state");
1573 :
1574 8 : conn = connect_database(conninfo, true);
1575 :
1576 : for (;;)
1577 0 : {
1578 8 : bool in_recovery = server_is_in_recovery(conn);
1579 :
1580 : /*
1581 : * Does the recovery process finish? In dry run mode, there is no
1582 : * recovery mode. Bail out as the recovery process has ended.
1583 : */
1584 8 : if (!in_recovery || dry_run)
1585 : {
1586 8 : status = POSTMASTER_READY;
1587 8 : recovery_ended = true;
1588 8 : break;
1589 : }
1590 :
1591 : /* Bail out after recovery_timeout seconds if this option is set */
1592 0 : if (opt->recovery_timeout > 0 && timer >= opt->recovery_timeout)
1593 : {
1594 0 : stop_standby_server(subscriber_dir);
1595 0 : pg_log_error("recovery timed out");
1596 0 : disconnect_database(conn, true);
1597 : }
1598 :
1599 : /* Keep waiting */
1600 0 : pg_usleep(WAIT_INTERVAL * USEC_PER_SEC);
1601 :
1602 0 : timer += WAIT_INTERVAL;
1603 : }
1604 :
1605 8 : disconnect_database(conn, false);
1606 :
1607 8 : if (status == POSTMASTER_STILL_STARTING)
1608 0 : pg_fatal("server did not end recovery");
1609 :
1610 8 : pg_log_info("target server reached the consistent state");
1611 8 : pg_log_info_hint("If pg_createsubscriber fails after this point, you must recreate the physical replica before continuing.");
1612 8 : }
1613 :
1614 : /*
1615 : * Create a publication that includes all tables in the database.
1616 : */
1617 : static void
1618 16 : create_publication(PGconn *conn, struct LogicalRepInfo *dbinfo)
1619 : {
1620 16 : PQExpBuffer str = createPQExpBuffer();
1621 : PGresult *res;
1622 : char *ipubname_esc;
1623 : char *spubname_esc;
1624 :
1625 : Assert(conn != NULL);
1626 :
1627 16 : ipubname_esc = PQescapeIdentifier(conn, dbinfo->pubname, strlen(dbinfo->pubname));
1628 16 : spubname_esc = PQescapeLiteral(conn, dbinfo->pubname, strlen(dbinfo->pubname));
1629 :
1630 : /* Check if the publication already exists */
1631 16 : appendPQExpBuffer(str,
1632 : "SELECT 1 FROM pg_catalog.pg_publication "
1633 : "WHERE pubname = %s",
1634 : spubname_esc);
1635 16 : res = PQexec(conn, str->data);
1636 16 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1637 : {
1638 0 : pg_log_error("could not obtain publication information: %s",
1639 : PQresultErrorMessage(res));
1640 0 : disconnect_database(conn, true);
1641 : }
1642 :
1643 16 : if (PQntuples(res) == 1)
1644 : {
1645 : /*
1646 : * Unfortunately, if it reaches this code path, it will always fail
1647 : * (unless you decide to change the existing publication name). That's
1648 : * bad but it is very unlikely that the user will choose a name with
1649 : * pg_createsubscriber_ prefix followed by the exact database oid and
1650 : * a random number.
1651 : */
1652 0 : pg_log_error("publication \"%s\" already exists", dbinfo->pubname);
1653 0 : pg_log_error_hint("Consider renaming this publication before continuing.");
1654 0 : disconnect_database(conn, true);
1655 : }
1656 :
1657 16 : PQclear(res);
1658 16 : resetPQExpBuffer(str);
1659 :
1660 16 : pg_log_info("creating publication \"%s\" in database \"%s\"",
1661 : dbinfo->pubname, dbinfo->dbname);
1662 :
1663 16 : appendPQExpBuffer(str, "CREATE PUBLICATION %s FOR ALL TABLES",
1664 : ipubname_esc);
1665 :
1666 16 : pg_log_debug("command is: %s", str->data);
1667 :
1668 16 : if (!dry_run)
1669 : {
1670 4 : res = PQexec(conn, str->data);
1671 4 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1672 : {
1673 0 : pg_log_error("could not create publication \"%s\" in database \"%s\": %s",
1674 : dbinfo->pubname, dbinfo->dbname, PQresultErrorMessage(res));
1675 0 : disconnect_database(conn, true);
1676 : }
1677 4 : PQclear(res);
1678 : }
1679 :
1680 : /* For cleanup purposes */
1681 16 : dbinfo->made_publication = true;
1682 :
1683 16 : PQfreemem(ipubname_esc);
1684 16 : PQfreemem(spubname_esc);
1685 16 : destroyPQExpBuffer(str);
1686 16 : }
1687 :
1688 : /*
1689 : * Drop the specified publication in the given database.
1690 : */
1691 : static void
1692 20 : drop_publication(PGconn *conn, const char *pubname, const char *dbname,
1693 : bool *made_publication)
1694 : {
1695 20 : PQExpBuffer str = createPQExpBuffer();
1696 : PGresult *res;
1697 : char *pubname_esc;
1698 :
1699 : Assert(conn != NULL);
1700 :
1701 20 : pubname_esc = PQescapeIdentifier(conn, pubname, strlen(pubname));
1702 :
1703 20 : pg_log_info("dropping publication \"%s\" in database \"%s\"",
1704 : pubname, dbname);
1705 :
1706 20 : appendPQExpBuffer(str, "DROP PUBLICATION %s", pubname_esc);
1707 :
1708 20 : PQfreemem(pubname_esc);
1709 :
1710 20 : pg_log_debug("command is: %s", str->data);
1711 :
1712 20 : if (!dry_run)
1713 : {
1714 8 : res = PQexec(conn, str->data);
1715 8 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1716 : {
1717 0 : pg_log_error("could not drop publication \"%s\" in database \"%s\": %s",
1718 : pubname, dbname, PQresultErrorMessage(res));
1719 0 : *made_publication = false; /* don't try again. */
1720 :
1721 : /*
1722 : * Don't disconnect and exit here. This routine is used by primary
1723 : * (cleanup publication / replication slot due to an error) and
1724 : * subscriber (remove the replicated publications). In both cases,
1725 : * it can continue and provide instructions for the user to remove
1726 : * it later if cleanup fails.
1727 : */
1728 : }
1729 8 : PQclear(res);
1730 : }
1731 :
1732 20 : destroyPQExpBuffer(str);
1733 20 : }
1734 :
1735 : /*
1736 : * Retrieve and drop the publications.
1737 : *
1738 : * Since the publications were created before the consistent LSN, they
1739 : * remain on the subscriber even after the physical replica is
1740 : * promoted. Remove these publications from the subscriber because
1741 : * they have no use. Additionally, if requested, drop all pre-existing
1742 : * publications.
1743 : */
1744 : static void
1745 16 : check_and_drop_publications(PGconn *conn, struct LogicalRepInfo *dbinfo)
1746 : {
1747 : PGresult *res;
1748 16 : bool drop_all_pubs = dbinfos.objecttypes_to_clean & OBJECTTYPE_PUBLICATIONS;
1749 :
1750 : Assert(conn != NULL);
1751 :
1752 16 : if (drop_all_pubs)
1753 : {
1754 4 : pg_log_info("dropping all existing publications in database \"%s\"",
1755 : dbinfo->dbname);
1756 :
1757 : /* Fetch all publication names */
1758 4 : res = PQexec(conn, "SELECT pubname FROM pg_catalog.pg_publication;");
1759 4 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1760 : {
1761 0 : pg_log_error("could not obtain publication information: %s",
1762 : PQresultErrorMessage(res));
1763 0 : PQclear(res);
1764 0 : disconnect_database(conn, true);
1765 : }
1766 :
1767 : /* Drop each publication */
1768 12 : for (int i = 0; i < PQntuples(res); i++)
1769 8 : drop_publication(conn, PQgetvalue(res, i, 0), dbinfo->dbname,
1770 : &dbinfo->made_publication);
1771 :
1772 4 : PQclear(res);
1773 : }
1774 :
1775 : /*
1776 : * In dry-run mode, we don't create publications, but we still try to drop
1777 : * those to provide necessary information to the user.
1778 : */
1779 16 : if (!drop_all_pubs || dry_run)
1780 12 : drop_publication(conn, dbinfo->pubname, dbinfo->dbname,
1781 : &dbinfo->made_publication);
1782 16 : }
1783 :
1784 : /*
1785 : * Create a subscription with some predefined options.
1786 : *
1787 : * A replication slot was already created in a previous step. Let's use it. It
1788 : * is not required to copy data. The subscription will be created but it will
1789 : * not be enabled now. That's because the replication progress must be set and
1790 : * the replication origin name (one of the function arguments) contains the
1791 : * subscription OID in its name. Once the subscription is created,
1792 : * set_replication_progress() can obtain the chosen origin name and set up its
1793 : * initial location.
1794 : */
1795 : static void
1796 16 : create_subscription(PGconn *conn, const struct LogicalRepInfo *dbinfo)
1797 : {
1798 16 : PQExpBuffer str = createPQExpBuffer();
1799 : PGresult *res;
1800 : char *pubname_esc;
1801 : char *subname_esc;
1802 : char *pubconninfo_esc;
1803 : char *replslotname_esc;
1804 :
1805 : Assert(conn != NULL);
1806 :
1807 16 : pubname_esc = PQescapeIdentifier(conn, dbinfo->pubname, strlen(dbinfo->pubname));
1808 16 : subname_esc = PQescapeIdentifier(conn, dbinfo->subname, strlen(dbinfo->subname));
1809 16 : pubconninfo_esc = PQescapeLiteral(conn, dbinfo->pubconninfo, strlen(dbinfo->pubconninfo));
1810 16 : replslotname_esc = PQescapeLiteral(conn, dbinfo->replslotname, strlen(dbinfo->replslotname));
1811 :
1812 16 : pg_log_info("creating subscription \"%s\" in database \"%s\"",
1813 : dbinfo->subname, dbinfo->dbname);
1814 :
1815 16 : appendPQExpBuffer(str,
1816 : "CREATE SUBSCRIPTION %s CONNECTION %s PUBLICATION %s "
1817 : "WITH (create_slot = false, enabled = false, "
1818 : "slot_name = %s, copy_data = false, two_phase = %s)",
1819 : subname_esc, pubconninfo_esc, pubname_esc, replslotname_esc,
1820 16 : dbinfos.two_phase ? "true" : "false");
1821 :
1822 16 : PQfreemem(pubname_esc);
1823 16 : PQfreemem(subname_esc);
1824 16 : PQfreemem(pubconninfo_esc);
1825 16 : PQfreemem(replslotname_esc);
1826 :
1827 16 : pg_log_debug("command is: %s", str->data);
1828 :
1829 16 : if (!dry_run)
1830 : {
1831 4 : res = PQexec(conn, str->data);
1832 4 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1833 : {
1834 0 : pg_log_error("could not create subscription \"%s\" in database \"%s\": %s",
1835 : dbinfo->subname, dbinfo->dbname, PQresultErrorMessage(res));
1836 0 : disconnect_database(conn, true);
1837 : }
1838 4 : PQclear(res);
1839 : }
1840 :
1841 16 : destroyPQExpBuffer(str);
1842 16 : }
1843 :
1844 : /*
1845 : * Sets the replication progress to the consistent LSN.
1846 : *
1847 : * The subscriber caught up to the consistent LSN provided by the last
1848 : * replication slot that was created. The goal is to set up the initial
1849 : * location for the logical replication that is the exact LSN that the
1850 : * subscriber was promoted. Once the subscription is enabled it will start
1851 : * streaming from that location onwards. In dry run mode, the subscription OID
1852 : * and LSN are set to invalid values for printing purposes.
1853 : */
1854 : static void
1855 16 : set_replication_progress(PGconn *conn, const struct LogicalRepInfo *dbinfo, const char *lsn)
1856 : {
1857 16 : PQExpBuffer str = createPQExpBuffer();
1858 : PGresult *res;
1859 : Oid suboid;
1860 : char *subname;
1861 : char *dbname;
1862 : char *originname;
1863 : char *lsnstr;
1864 :
1865 : Assert(conn != NULL);
1866 :
1867 16 : subname = PQescapeLiteral(conn, dbinfo->subname, strlen(dbinfo->subname));
1868 16 : dbname = PQescapeLiteral(conn, dbinfo->dbname, strlen(dbinfo->dbname));
1869 :
1870 16 : appendPQExpBuffer(str,
1871 : "SELECT s.oid FROM pg_catalog.pg_subscription s "
1872 : "INNER JOIN pg_catalog.pg_database d ON (s.subdbid = d.oid) "
1873 : "WHERE s.subname = %s AND d.datname = %s",
1874 : subname, dbname);
1875 :
1876 16 : res = PQexec(conn, str->data);
1877 16 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1878 : {
1879 0 : pg_log_error("could not obtain subscription OID: %s",
1880 : PQresultErrorMessage(res));
1881 0 : disconnect_database(conn, true);
1882 : }
1883 :
1884 16 : if (PQntuples(res) != 1 && !dry_run)
1885 : {
1886 0 : pg_log_error("could not obtain subscription OID: got %d rows, expected %d row",
1887 : PQntuples(res), 1);
1888 0 : disconnect_database(conn, true);
1889 : }
1890 :
1891 16 : if (dry_run)
1892 : {
1893 12 : suboid = InvalidOid;
1894 12 : lsnstr = psprintf("%X/%08X", LSN_FORMAT_ARGS((XLogRecPtr) InvalidXLogRecPtr));
1895 : }
1896 : else
1897 : {
1898 4 : suboid = strtoul(PQgetvalue(res, 0, 0), NULL, 10);
1899 4 : lsnstr = psprintf("%s", lsn);
1900 : }
1901 :
1902 16 : PQclear(res);
1903 :
1904 : /*
1905 : * The origin name is defined as pg_%u. %u is the subscription OID. See
1906 : * ApplyWorkerMain().
1907 : */
1908 16 : originname = psprintf("pg_%u", suboid);
1909 :
1910 16 : pg_log_info("setting the replication progress (node name \"%s\", LSN %s) in database \"%s\"",
1911 : originname, lsnstr, dbinfo->dbname);
1912 :
1913 16 : resetPQExpBuffer(str);
1914 16 : appendPQExpBuffer(str,
1915 : "SELECT pg_catalog.pg_replication_origin_advance('%s', '%s')",
1916 : originname, lsnstr);
1917 :
1918 16 : pg_log_debug("command is: %s", str->data);
1919 :
1920 16 : if (!dry_run)
1921 : {
1922 4 : res = PQexec(conn, str->data);
1923 4 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
1924 : {
1925 0 : pg_log_error("could not set replication progress for subscription \"%s\": %s",
1926 : dbinfo->subname, PQresultErrorMessage(res));
1927 0 : disconnect_database(conn, true);
1928 : }
1929 4 : PQclear(res);
1930 : }
1931 :
1932 16 : PQfreemem(subname);
1933 16 : PQfreemem(dbname);
1934 16 : pg_free(originname);
1935 16 : pg_free(lsnstr);
1936 16 : destroyPQExpBuffer(str);
1937 16 : }
1938 :
1939 : /*
1940 : * Enables the subscription.
1941 : *
1942 : * The subscription was created in a previous step but it was disabled. After
1943 : * adjusting the initial logical replication location, enable the subscription.
1944 : */
1945 : static void
1946 16 : enable_subscription(PGconn *conn, const struct LogicalRepInfo *dbinfo)
1947 : {
1948 16 : PQExpBuffer str = createPQExpBuffer();
1949 : PGresult *res;
1950 : char *subname;
1951 :
1952 : Assert(conn != NULL);
1953 :
1954 16 : subname = PQescapeIdentifier(conn, dbinfo->subname, strlen(dbinfo->subname));
1955 :
1956 16 : pg_log_info("enabling subscription \"%s\" in database \"%s\"",
1957 : dbinfo->subname, dbinfo->dbname);
1958 :
1959 16 : appendPQExpBuffer(str, "ALTER SUBSCRIPTION %s ENABLE", subname);
1960 :
1961 16 : pg_log_debug("command is: %s", str->data);
1962 :
1963 16 : if (!dry_run)
1964 : {
1965 4 : res = PQexec(conn, str->data);
1966 4 : if (PQresultStatus(res) != PGRES_COMMAND_OK)
1967 : {
1968 0 : pg_log_error("could not enable subscription \"%s\": %s",
1969 : dbinfo->subname, PQresultErrorMessage(res));
1970 0 : disconnect_database(conn, true);
1971 : }
1972 :
1973 4 : PQclear(res);
1974 : }
1975 :
1976 16 : PQfreemem(subname);
1977 16 : destroyPQExpBuffer(str);
1978 16 : }
1979 :
1980 : /*
1981 : * Fetch a list of all connectable non-template databases from the source server
1982 : * and form a list such that they appear as if the user has specified multiple
1983 : * --database options, one for each source database.
1984 : */
1985 : static void
1986 2 : get_publisher_databases(struct CreateSubscriberOptions *opt,
1987 : bool dbnamespecified)
1988 : {
1989 : PGconn *conn;
1990 : PGresult *res;
1991 :
1992 : /* If a database name was specified, just connect to it. */
1993 2 : if (dbnamespecified)
1994 0 : conn = connect_database(opt->pub_conninfo_str, true);
1995 : else
1996 : {
1997 : /* Otherwise, try postgres first and then template1. */
1998 : char *conninfo;
1999 :
2000 2 : conninfo = concat_conninfo_dbname(opt->pub_conninfo_str, "postgres");
2001 2 : conn = connect_database(conninfo, false);
2002 2 : pg_free(conninfo);
2003 2 : if (!conn)
2004 : {
2005 0 : conninfo = concat_conninfo_dbname(opt->pub_conninfo_str, "template1");
2006 0 : conn = connect_database(conninfo, true);
2007 0 : pg_free(conninfo);
2008 : }
2009 : }
2010 :
2011 2 : res = PQexec(conn, "SELECT datname FROM pg_database WHERE datistemplate = false AND datallowconn AND datconnlimit <> -2 ORDER BY 1");
2012 2 : if (PQresultStatus(res) != PGRES_TUPLES_OK)
2013 : {
2014 0 : pg_log_error("could not obtain a list of databases: %s", PQresultErrorMessage(res));
2015 0 : PQclear(res);
2016 0 : disconnect_database(conn, true);
2017 : }
2018 :
2019 8 : for (int i = 0; i < PQntuples(res); i++)
2020 : {
2021 6 : const char *dbname = PQgetvalue(res, i, 0);
2022 :
2023 6 : simple_string_list_append(&opt->database_names, dbname);
2024 :
2025 : /* Increment num_dbs to reflect multiple --database options */
2026 6 : num_dbs++;
2027 : }
2028 :
2029 2 : PQclear(res);
2030 2 : disconnect_database(conn, false);
2031 2 : }
2032 :
2033 : int
2034 46 : main(int argc, char **argv)
2035 : {
2036 : static struct option long_options[] =
2037 : {
2038 : {"all", no_argument, NULL, 'a'},
2039 : {"database", required_argument, NULL, 'd'},
2040 : {"pgdata", required_argument, NULL, 'D'},
2041 : {"dry-run", no_argument, NULL, 'n'},
2042 : {"subscriber-port", required_argument, NULL, 'p'},
2043 : {"publisher-server", required_argument, NULL, 'P'},
2044 : {"socketdir", required_argument, NULL, 's'},
2045 : {"recovery-timeout", required_argument, NULL, 't'},
2046 : {"enable-two-phase", no_argument, NULL, 'T'},
2047 : {"subscriber-username", required_argument, NULL, 'U'},
2048 : {"verbose", no_argument, NULL, 'v'},
2049 : {"version", no_argument, NULL, 'V'},
2050 : {"help", no_argument, NULL, '?'},
2051 : {"config-file", required_argument, NULL, 1},
2052 : {"publication", required_argument, NULL, 2},
2053 : {"replication-slot", required_argument, NULL, 3},
2054 : {"subscription", required_argument, NULL, 4},
2055 : {"clean", required_argument, NULL, 5},
2056 : {NULL, 0, NULL, 0}
2057 : };
2058 :
2059 46 : struct CreateSubscriberOptions opt = {0};
2060 :
2061 : int c;
2062 : int option_index;
2063 :
2064 : char *pub_base_conninfo;
2065 : char *sub_base_conninfo;
2066 46 : char *dbname_conninfo = NULL;
2067 :
2068 : uint64 pub_sysid;
2069 : uint64 sub_sysid;
2070 : struct stat statbuf;
2071 :
2072 : char *consistent_lsn;
2073 :
2074 : char pidfile[MAXPGPATH];
2075 :
2076 46 : pg_logging_init(argv[0]);
2077 46 : pg_logging_set_level(PG_LOG_WARNING);
2078 46 : progname = get_progname(argv[0]);
2079 46 : set_pglocale_pgservice(argv[0], PG_TEXTDOMAIN("pg_basebackup"));
2080 :
2081 46 : if (argc > 1)
2082 : {
2083 44 : if (strcmp(argv[1], "--help") == 0 || strcmp(argv[1], "-?") == 0)
2084 : {
2085 2 : usage();
2086 2 : exit(0);
2087 : }
2088 42 : else if (strcmp(argv[1], "-V") == 0
2089 42 : || strcmp(argv[1], "--version") == 0)
2090 : {
2091 2 : puts("pg_createsubscriber (PostgreSQL) " PG_VERSION);
2092 2 : exit(0);
2093 : }
2094 : }
2095 :
2096 : /* Default settings */
2097 42 : subscriber_dir = NULL;
2098 42 : opt.config_file = NULL;
2099 42 : opt.pub_conninfo_str = NULL;
2100 42 : opt.socket_dir = NULL;
2101 42 : opt.sub_port = DEFAULT_SUB_PORT;
2102 42 : opt.sub_username = NULL;
2103 42 : opt.two_phase = false;
2104 42 : opt.database_names = (SimpleStringList)
2105 : {
2106 : 0
2107 : };
2108 42 : opt.recovery_timeout = 0;
2109 42 : opt.all_dbs = false;
2110 :
2111 : /*
2112 : * Don't allow it to be run as root. It uses pg_ctl which does not allow
2113 : * it either.
2114 : */
2115 : #ifndef WIN32
2116 42 : if (geteuid() == 0)
2117 : {
2118 0 : pg_log_error("cannot be executed by \"root\"");
2119 0 : pg_log_error_hint("You must run %s as the PostgreSQL superuser.",
2120 : progname);
2121 0 : exit(1);
2122 : }
2123 : #endif
2124 :
2125 42 : get_restricted_token();
2126 :
2127 324 : while ((c = getopt_long(argc, argv, "ad:D:np:P:s:t:TU:v",
2128 324 : long_options, &option_index)) != -1)
2129 : {
2130 288 : switch (c)
2131 : {
2132 6 : case 'a':
2133 6 : opt.all_dbs = true;
2134 6 : break;
2135 50 : case 'd':
2136 50 : if (!simple_string_list_member(&opt.database_names, optarg))
2137 : {
2138 48 : simple_string_list_append(&opt.database_names, optarg);
2139 48 : num_dbs++;
2140 : }
2141 : else
2142 2 : pg_fatal("database \"%s\" specified more than once for -d/--database", optarg);
2143 48 : break;
2144 38 : case 'D':
2145 38 : subscriber_dir = pg_strdup(optarg);
2146 38 : canonicalize_path(subscriber_dir);
2147 38 : break;
2148 18 : case 'n':
2149 18 : dry_run = true;
2150 18 : break;
2151 24 : case 'p':
2152 24 : opt.sub_port = pg_strdup(optarg);
2153 24 : break;
2154 36 : case 'P':
2155 36 : opt.pub_conninfo_str = pg_strdup(optarg);
2156 36 : break;
2157 24 : case 's':
2158 24 : opt.socket_dir = pg_strdup(optarg);
2159 24 : canonicalize_path(opt.socket_dir);
2160 24 : break;
2161 6 : case 't':
2162 6 : opt.recovery_timeout = atoi(optarg);
2163 6 : break;
2164 2 : case 'T':
2165 2 : opt.two_phase = true;
2166 2 : break;
2167 0 : case 'U':
2168 0 : opt.sub_username = pg_strdup(optarg);
2169 0 : break;
2170 38 : case 'v':
2171 38 : pg_logging_increase_verbosity();
2172 38 : break;
2173 0 : case 1:
2174 0 : opt.config_file = pg_strdup(optarg);
2175 0 : break;
2176 24 : case 2:
2177 24 : if (!simple_string_list_member(&opt.pub_names, optarg))
2178 : {
2179 22 : simple_string_list_append(&opt.pub_names, optarg);
2180 22 : num_pubs++;
2181 : }
2182 : else
2183 2 : pg_fatal("publication \"%s\" specified more than once for --publication", optarg);
2184 22 : break;
2185 8 : case 3:
2186 8 : if (!simple_string_list_member(&opt.replslot_names, optarg))
2187 : {
2188 8 : simple_string_list_append(&opt.replslot_names, optarg);
2189 8 : num_replslots++;
2190 : }
2191 : else
2192 0 : pg_fatal("replication slot \"%s\" specified more than once for --replication-slot", optarg);
2193 8 : break;
2194 10 : case 4:
2195 10 : if (!simple_string_list_member(&opt.sub_names, optarg))
2196 : {
2197 10 : simple_string_list_append(&opt.sub_names, optarg);
2198 10 : num_subs++;
2199 : }
2200 : else
2201 0 : pg_fatal("subscription \"%s\" specified more than once for --subscription", optarg);
2202 10 : break;
2203 2 : case 5:
2204 2 : if (!simple_string_list_member(&opt.objecttypes_to_clean, optarg))
2205 2 : simple_string_list_append(&opt.objecttypes_to_clean, optarg);
2206 : else
2207 0 : pg_fatal("object type \"%s\" specified more than once for --clean", optarg);
2208 2 : break;
2209 2 : default:
2210 : /* getopt_long already emitted a complaint */
2211 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
2212 2 : exit(1);
2213 : }
2214 : }
2215 :
2216 : /* Validate that --all is not used with incompatible options */
2217 36 : if (opt.all_dbs)
2218 : {
2219 6 : char *bad_switch = NULL;
2220 :
2221 6 : if (num_dbs > 0)
2222 2 : bad_switch = "--database";
2223 4 : else if (num_pubs > 0)
2224 2 : bad_switch = "--publication";
2225 2 : else if (num_replslots > 0)
2226 0 : bad_switch = "--replication-slot";
2227 2 : else if (num_subs > 0)
2228 0 : bad_switch = "--subscription";
2229 :
2230 6 : if (bad_switch)
2231 : {
2232 4 : pg_log_error("options %s and -a/--all cannot be used together", bad_switch);
2233 4 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
2234 4 : exit(1);
2235 : }
2236 : }
2237 :
2238 : /* Any non-option arguments? */
2239 32 : if (optind < argc)
2240 : {
2241 0 : pg_log_error("too many command-line arguments (first is \"%s\")",
2242 : argv[optind]);
2243 0 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
2244 0 : exit(1);
2245 : }
2246 :
2247 : /* Required arguments */
2248 32 : if (subscriber_dir == NULL)
2249 : {
2250 2 : pg_log_error("no subscriber data directory specified");
2251 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
2252 2 : exit(1);
2253 : }
2254 :
2255 : /* If socket directory is not provided, use the current directory */
2256 30 : if (opt.socket_dir == NULL)
2257 : {
2258 : char cwd[MAXPGPATH];
2259 :
2260 10 : if (!getcwd(cwd, MAXPGPATH))
2261 0 : pg_fatal("could not determine current directory");
2262 10 : opt.socket_dir = pg_strdup(cwd);
2263 10 : canonicalize_path(opt.socket_dir);
2264 : }
2265 :
2266 : /*
2267 : * Parse connection string. Build a base connection string that might be
2268 : * reused by multiple databases.
2269 : */
2270 30 : if (opt.pub_conninfo_str == NULL)
2271 : {
2272 : /*
2273 : * TODO use primary_conninfo (if available) from subscriber and
2274 : * extract publisher connection string. Assume that there are
2275 : * identical entries for physical and logical replication. If there is
2276 : * not, we would fail anyway.
2277 : */
2278 2 : pg_log_error("no publisher connection string specified");
2279 2 : pg_log_error_hint("Try \"%s --help\" for more information.", progname);
2280 2 : exit(1);
2281 : }
2282 28 : pg_log_info("validating publisher connection string");
2283 28 : pub_base_conninfo = get_base_conninfo(opt.pub_conninfo_str,
2284 : &dbname_conninfo);
2285 28 : if (pub_base_conninfo == NULL)
2286 0 : exit(1);
2287 :
2288 28 : pg_log_info("validating subscriber connection string");
2289 28 : sub_base_conninfo = get_sub_conninfo(&opt);
2290 :
2291 : /*
2292 : * Fetch all databases from the source (publisher) and treat them as if
2293 : * the user specified has multiple --database options, one for each source
2294 : * database.
2295 : */
2296 28 : if (opt.all_dbs)
2297 : {
2298 2 : bool dbnamespecified = (dbname_conninfo != NULL);
2299 :
2300 2 : get_publisher_databases(&opt, dbnamespecified);
2301 : }
2302 :
2303 28 : if (opt.database_names.head == NULL)
2304 : {
2305 4 : pg_log_info("no database was specified");
2306 :
2307 : /*
2308 : * Try to obtain the dbname from the publisher conninfo. If dbname
2309 : * parameter is not available, error out.
2310 : */
2311 4 : if (dbname_conninfo)
2312 : {
2313 2 : simple_string_list_append(&opt.database_names, dbname_conninfo);
2314 2 : num_dbs++;
2315 :
2316 2 : pg_log_info("database name \"%s\" was extracted from the publisher connection string",
2317 : dbname_conninfo);
2318 : }
2319 : else
2320 : {
2321 2 : pg_log_error("no database name specified");
2322 2 : pg_log_error_hint("Try \"%s --help\" for more information.",
2323 : progname);
2324 2 : exit(1);
2325 : }
2326 : }
2327 :
2328 : /* Number of object names must match number of databases */
2329 26 : if (num_pubs > 0 && num_pubs != num_dbs)
2330 : {
2331 2 : pg_log_error("wrong number of publication names specified");
2332 2 : pg_log_error_detail("The number of specified publication names (%d) must match the number of specified database names (%d).",
2333 : num_pubs, num_dbs);
2334 2 : exit(1);
2335 : }
2336 24 : if (num_subs > 0 && num_subs != num_dbs)
2337 : {
2338 2 : pg_log_error("wrong number of subscription names specified");
2339 2 : pg_log_error_detail("The number of specified subscription names (%d) must match the number of specified database names (%d).",
2340 : num_subs, num_dbs);
2341 2 : exit(1);
2342 : }
2343 22 : if (num_replslots > 0 && num_replslots != num_dbs)
2344 : {
2345 2 : pg_log_error("wrong number of replication slot names specified");
2346 2 : pg_log_error_detail("The number of specified replication slot names (%d) must match the number of specified database names (%d).",
2347 : num_replslots, num_dbs);
2348 2 : exit(1);
2349 : }
2350 :
2351 : /* Verify the object types specified for removal from the subscriber */
2352 22 : for (SimpleStringListCell *cell = opt.objecttypes_to_clean.head; cell; cell = cell->next)
2353 : {
2354 2 : if (pg_strcasecmp(cell->val, "publications") == 0)
2355 2 : dbinfos.objecttypes_to_clean |= OBJECTTYPE_PUBLICATIONS;
2356 : else
2357 : {
2358 0 : pg_log_error("invalid object type \"%s\" specified for --clean", cell->val);
2359 0 : pg_log_error_hint("The valid value is: \"%s\"", "publications");
2360 0 : exit(1);
2361 : }
2362 : }
2363 :
2364 : /* Get the absolute path of pg_ctl and pg_resetwal on the subscriber */
2365 20 : pg_ctl_path = get_exec_path(argv[0], "pg_ctl");
2366 20 : pg_resetwal_path = get_exec_path(argv[0], "pg_resetwal");
2367 :
2368 : /* Rudimentary check for a data directory */
2369 20 : check_data_directory(subscriber_dir);
2370 :
2371 20 : dbinfos.two_phase = opt.two_phase;
2372 :
2373 : /*
2374 : * Store database information for publisher and subscriber. It should be
2375 : * called before atexit() because its return is used in the
2376 : * cleanup_objects_atexit().
2377 : */
2378 20 : dbinfos.dbinfo = store_pub_sub_info(&opt, pub_base_conninfo, sub_base_conninfo);
2379 :
2380 : /* Register a function to clean up objects in case of failure */
2381 20 : atexit(cleanup_objects_atexit);
2382 :
2383 : /*
2384 : * Check if the subscriber data directory has the same system identifier
2385 : * than the publisher data directory.
2386 : */
2387 20 : pub_sysid = get_primary_sysid(dbinfos.dbinfo[0].pubconninfo);
2388 20 : sub_sysid = get_standby_sysid(subscriber_dir);
2389 20 : if (pub_sysid != sub_sysid)
2390 2 : pg_fatal("subscriber data directory is not a copy of the source database cluster");
2391 :
2392 : /* Subscriber PID file */
2393 18 : snprintf(pidfile, MAXPGPATH, "%s/postmaster.pid", subscriber_dir);
2394 :
2395 : /*
2396 : * The standby server must not be running. If the server is started under
2397 : * service manager and pg_createsubscriber stops it, the service manager
2398 : * might react to this action and start the server again. Therefore,
2399 : * refuse to proceed if the server is running to avoid possible failures.
2400 : */
2401 18 : if (stat(pidfile, &statbuf) == 0)
2402 : {
2403 2 : pg_log_error("standby server is running");
2404 2 : pg_log_error_hint("Stop the standby server and try again.");
2405 2 : exit(1);
2406 : }
2407 :
2408 : /*
2409 : * Start a short-lived standby server with temporary parameters (provided
2410 : * by command-line options). The goal is to avoid connections during the
2411 : * transformation steps.
2412 : */
2413 16 : pg_log_info("starting the standby server with command-line options");
2414 16 : start_standby_server(&opt, true, false);
2415 :
2416 : /* Check if the standby server is ready for logical replication */
2417 16 : check_subscriber(dbinfos.dbinfo);
2418 :
2419 : /* Check if the primary server is ready for logical replication */
2420 12 : check_publisher(dbinfos.dbinfo);
2421 :
2422 : /*
2423 : * Stop the target server. The recovery process requires that the server
2424 : * reaches a consistent state before targeting the recovery stop point.
2425 : * Make sure a consistent state is reached (stop the target server
2426 : * guarantees it) *before* creating the replication slots in
2427 : * setup_publisher().
2428 : */
2429 8 : pg_log_info("stopping the subscriber");
2430 8 : stop_standby_server(subscriber_dir);
2431 :
2432 : /* Create the required objects for each database on publisher */
2433 8 : consistent_lsn = setup_publisher(dbinfos.dbinfo);
2434 :
2435 : /* Write the required recovery parameters */
2436 8 : setup_recovery(dbinfos.dbinfo, subscriber_dir, consistent_lsn);
2437 :
2438 : /*
2439 : * Start subscriber so the recovery parameters will take effect. Wait
2440 : * until accepting connections. We don't want to start logical replication
2441 : * during setup.
2442 : */
2443 8 : pg_log_info("starting the subscriber");
2444 8 : start_standby_server(&opt, true, true);
2445 :
2446 : /* Waiting the subscriber to be promoted */
2447 8 : wait_for_end_recovery(dbinfos.dbinfo[0].subconninfo, &opt);
2448 :
2449 : /*
2450 : * Create the subscription for each database on subscriber. It does not
2451 : * enable it immediately because it needs to adjust the replication start
2452 : * point to the LSN reported by setup_publisher(). It also cleans up
2453 : * publications created by this tool and replication to the standby.
2454 : */
2455 8 : setup_subscriber(dbinfos.dbinfo, consistent_lsn);
2456 :
2457 : /* Remove primary_slot_name if it exists on primary */
2458 8 : drop_primary_replication_slot(dbinfos.dbinfo, primary_slot_name);
2459 :
2460 : /* Remove failover replication slots if they exist on subscriber */
2461 8 : drop_failover_replication_slots(dbinfos.dbinfo);
2462 :
2463 : /* Stop the subscriber */
2464 8 : pg_log_info("stopping the subscriber");
2465 8 : stop_standby_server(subscriber_dir);
2466 :
2467 : /* Change system identifier from subscriber */
2468 8 : modify_subscriber_sysid(&opt);
2469 :
2470 8 : success = true;
2471 :
2472 8 : pg_log_info("Done!");
2473 :
2474 8 : return 0;
2475 : }
|
