Squashed commit of the following:

commit e7ef17117efe6679e154a4905d587c808b48df50 Merge: cd3a280... 43268f2... Author: Greg Smith <greg@2ndQuadrant.com> Date: Tue Jun 7 01:40:08 2011 -0400 Merge commit 'origin/master' into autofailover Conflicts: repmgr.c commit cd3a280804a01c5270c5c743e5822c7beb9ac77a Merge: 72ad378... 8200b68... Author: Greg Smith <greg@2ndQuadrant.com> Date: Tue Jun 7 00:52:42 2011 -0400 Merge commit 'origin/master' into autofailover Conflicts: config.c commit 72ad378bed21d74dab743fec411fe10b19007481 Merge: 17bafa1... 367d0b1... Author: Greg Smith <greg@2ndQuadrant.com> Date: Tue Jun 7 00:38:01 2011 -0400 Merge commit 'origin/master' into autofailover Conflicts: config.c dbutils.c repmgr.c repmgrd.c commit 17bafa1ca509c1f6614810bab2538e570ebc599e Author: Greg Smith <greg@2ndQuadrant.com> Date: Tue Jun 7 00:31:28 2011 -0400 Run astyle to fix recent changes commit a5fbbaecce8fe86bc17c0ebeb1324f9262967316 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Tue May 10 00:46:58 2011 +0200 Fix a crititcal bug in the decision process If the postgresql on the first node returned by the query to find candidates in do_failover is down then the initialization of the bestCandidate is done with non assigned variables. Fix the situation by moving the initialization in the loop above. And loop until we have a find_best. Added a log message if no candidate is found commit 42b21475ac248db8f0e50f5956ef96808e92c68c Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Mon May 9 22:39:21 2011 +0200 Add test_ssh_connection The feature was written by Jaime and reworked to fix https://github.com/greg2ndQuadrant/repmgr/issues/5 commit 86f01afae631e9541600af6578e649d88c3ece98 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Mon May 9 21:39:42 2011 +0200 Improve log output commit db2f29fc1c8ea03a8ff85717873f8a876846b844 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Mon May 9 01:41:34 2011 -0500 Only compare getenv("USER") when it's actually set, otherwise it will segfault commit ea4f3f20747e2e0294551d5e61869bdde6d3cd7b Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Mon May 9 01:03:39 2011 -0500 Fix a message to only show when log_info is requested and the verbose flag is set. This is because it needs a calculation that is only done when the verbose flag is set, so if i have requested log INFO level but haven't set the flag it shows a null commit 35a53bac7e341cfdbb64d2c15fa77c9c4e18bd40 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Mon May 9 01:00:54 2011 -0500 Use log_* functions in do_witness_create() commit 8c526f758a46ad53b4d391fc76360561d4ff8bdd Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Sun May 8 19:30:34 2011 -0500 Add a fallback_application_name parameter to the conninfo identify the connection if application_name is not set commit 01057fc12cbc1fb656d619f483044f28a5f08d37 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri May 6 23:57:27 2011 +0200 Fix the best_candidate loop there was an overflow in the loop, already fixed but loosed during merge. commit e80effa3daf56f08005704fc1a5bbe69c1324212 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri May 6 23:55:15 2011 +0200 Fix check in do_failover (merge faillure) And also remove an unused variable as I was here. commit 79ba37e2933f4e87523a77375dfda1d96150e7d3 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Thu May 5 21:15:46 2011 +0200 Fix compile error commit 67c7b5d68c95a60bb4cd0cfb750b4c8d047fa2a0 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Sun Apr 24 23:27:57 2011 +0200 And apply astyle .... commit 9a321722537d96983b8162227ff629a267b6ed67 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Sun Apr 24 23:27:09 2011 +0200 Cosmetic change to reduce diff with master commit 09037efea3fa2c31896b5dc78b0340516a743ba6 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Sun Apr 24 22:26:03 2011 +0200 Apply astyle commit 7c4786f662943558be967be4a8dad976f52155dd Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Sun Apr 24 02:22:12 2011 +0200 Improve the standby clone action By default, all config files and directories are cloned from the master in the same place in the slave. If a destination directory is provided (-D), everything is copied in the provided dir, and if the master have tablespaces repmgr exit without cloning. commit a6d7f765b9403a2cff7e2e1df8ae45a5a7ee1665 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri Apr 22 23:31:09 2011 +0200 Add success message for repmgr standby register commit 26bf3b08e661137dd3f3c0d3c00fd6b3b90b08b3 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri Apr 22 22:51:28 2011 +0200 Change the exit to a return in config.c commit 1bd8f4c119e1dbf9a94b2eaec884abce96eeb174 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri Apr 22 22:32:57 2011 +0200 Reduce duplicate code commit db553fab45ca075f95f09bdb2147de68948b60c8 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri Apr 22 22:24:04 2011 +0200 Some cosmetic commit f19d0ad714ebcf7df7726772e887c873d005d350 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri Apr 22 22:23:06 2011 +0200 Move a function declaration into header file commit 1f328bc438c896a9f2067069d756f901b58d41f2 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Mon Apr 11 00:38:30 2011 -0500 We don't use conninfo as a separate variable anymore commit f6ade0d63b8a5dd43377f546f5311b4a151b2bfb Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Sun Apr 10 20:53:22 2011 -0500 Fix a few typos commit ceca9fa983c8dbde61a7a78da29a1e1871756d8c Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Sun Apr 10 19:32:57 2011 -0500 Fix code to allow the code to compile: - some log_* had problems with parenthesis - some uses of variables without the runtime_options prefix commit 73431f955afd77560bca5370924e09329566c4b7 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Sun Apr 10 23:21:37 2011 +0200 Fix the debian package name commit 688eab371110083ae8715b35f414e29c6d87e1ac Merge: 5c23375... 7995c42... Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Sun Apr 10 23:17:58 2011 +0200 Merge branch 'autofailover' of git.2ndquadrant.it:repmgr into autofailover commit 5c23375f88a53ed469e9d13934d618f7a74669be Merge: cc3315c... c4ae574... Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Sun Apr 10 23:08:36 2011 +0200 Merge branch 'master' into autofailover Conflicts: repmgr.c commit 7995c428161566cfc54a67eb16f9134c859e7381 Merge: 788ff98... 1303e49... Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Sun Apr 10 16:14:30 2011 -0500 Merge branch 'autofailover' of git+ssh://git.2ndquadrant.it/git/repmgr into autofailover commit cc3315ce235b898711c34fd1f2fa1116dbee4e16 Merge: 1303e49... d77186c... Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Sun Apr 10 23:03:11 2011 +0200 Merge commit 'd77186c90444b9c5ca2de201651841f56a7ded02' into autofailover commit 1303e49852705046e15ef64f5f7ab739a1689431 Merge: 7ff621b... 4c792c8... Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Sun Apr 10 22:28:08 2011 +0200 Merge commit '4c792c8013f5713589f53dbdb47721febf139a85' into autofailover commit 788ff98e94311a33e3e6f7d85a303cbc61288e5f Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Tue Mar 15 19:00:56 2011 -0500 A few fixes after merge to unbroke what the merge broke, and to make the new logging system more consistent through the system commit 7ff621b96784dfaf40baab4f0f8e7857b4aed6ce Author: Dan Farina <drfarina@acm.org> Date: Tue Dec 7 21:30:44 2010 -0800 Install install/uninstall SQL also. Signed-off-by: Dan Farina <drfarina@acm.org> Signed-off-by: Peter van Hardenberg <pvh@heroku.com> commit c9147dad8223eff20bf5d52ced8a35eed6d82110 Author: Dan Farina <drfarina@acm.org> Date: Tue Dec 7 21:30:20 2010 -0800 Split up install/uninstall actions more like a standard contrib Signed-off-by: Dan Farina <drfarina@acm.org> Signed-off-by: Peter van Hardenberg <pvh@heroku.com> commit c8028780b50f2c7fb4384cb9891796647f356e19 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Sat Feb 12 13:29:32 2011 +0100 Fixing SLEEP times and RETRY commit 39a1bf3d29f3e33fbf0e1b066a311e8a72f2dc38 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Sat Feb 12 01:17:37 2011 +0100 Add a pause after update_shared_memory() in do_failover we pause for SLEEP_MONITOR+1 to let other nodes update themselves. commit 527af2baa945e3b640352c01c6dd181d93c9529a Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri Feb 11 21:14:22 2011 +0100 change the debian package filename too commit c8cb27c7039b2b3a838554187a8add850a42027a Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri Feb 11 15:14:40 2011 +0100 Change package name for the automatic fail-over branch of repmgr commit 7427988628f754e57069453d65a71f79117c3a3d Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri Feb 11 14:28:03 2011 +0100 Exit 1 when SIGINT commit af366fe731b70e24ead056e50b69269392bd15a1 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri Feb 11 14:27:46 2011 +0100 Improve log output when reloading configuration commit 6cc18ce081d7bf55ba9993e9d87567879da35c4d Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri Feb 11 14:20:36 2011 +0100 Add reload conf on (re)start commit 4259e2c410fd0ef1273c7d1b4ab8fcf1e778e968 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri Feb 11 14:01:37 2011 +0100 avoid double free on repmgrd exit as master Per commit from Charles Duffy <charles@tippr.com> and faillure to cherry-pick it correctly. Conflicts: repmgrd.c commit 431e27b1c005e000f9a346d982419979b4363d77 Author: Greg Smith <greg@2ndQuadrant.com> Date: Thu Feb 10 15:09:18 2011 -0500 Tweak .gitignore to ignore more doc build artifacts commit b725fa9ae65c7bd5fea7a4e944db5685dee2e8bd Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Sun Mar 13 15:16:27 2011 -0500 Delete a paragraph that appears twice, because a merge problem commit d990c77b327a282c1903b7a339f35a22b6a89958 Author: trbs <trbs@trbs.net> Date: Tue Jan 11 18:24:17 2011 +0100 added note about postgresql-server-dev-9.0 and use libxslt-dev instead of version specific package name commit 69bc1cd3772103b529598978160327e1f9025157 Author: trbs <trbs@trbs.net> Date: Fri Jan 7 01:32:31 2011 +0100 fix line commit f7b1d1e5e3764c85cec7afa81c164fac3679e1ea Author: trbs <trbs@trbs.net> Date: Thu Dec 23 15:02:23 2010 +0100 Updated README with Debian/Ubuntu install information commit 77d28960ff78c3936be0e1029305b0b578e260a9 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri Feb 11 13:34:49 2011 +0100 Create the function used for shared memory access in create_schema, note that this is incompatible with current master commit 4a73043f232f0a143ede898841530f4d7442c95b Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Fri Feb 11 10:00:34 2011 +0100 improve log output commit 62c90a4e86b2cd56ec14255adcfef564945d0769 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Fri Feb 11 00:40:05 2011 -0500 Close local connection on witness before exit on error of primary commit e5156865e05670fa9944d74d472127082556d0a0 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Fri Feb 11 00:34:25 2011 -0500 Remove a semicolon which is just a typo commit 7586a09bc321241932adacf6a1431029964dc46f Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Fri Feb 11 00:07:02 2011 -0500 Fix the computation of quorum, we need to count master and the division should not be an integer division commit a19c0ad2059a00e9e7415fc6ea280c109c809c9c Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Thu Feb 10 23:54:35 2011 +0100 move the functions back into public schema commit 19fc8ffb1dc0fd9acddad5d22bf5c01704687474 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Thu Feb 10 00:48:00 2011 -0500 A few more fixes. Make repmgr functions exists in repmgr schema and fix a typo that caused a seg fault. commit c6d2b8c6421f93074d7d616980feb0175ee4ef36 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Wed Feb 9 17:56:44 2011 -0500 A few places where i forgot to update the priority field commit 0ff0bb8d981b868693c6a751e7e80473b25f2399 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Wed Feb 9 14:24:43 2011 -0500 Fix a few bugs from last commit and make reload configuration also update registration on repl_nodes commit 508c34e9dfb2bfb7e47d5c6836ead7992e6112fe Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Wed Feb 9 13:45:20 2011 -0500 Add a way for the user to indicate it's preference about which node should be promoted in case of a conflict (ie: two nodes with the same wal location). This will be provided as a parameter in repmgr.conf called priority, andd will be registered in the repl_nodes table. commit 6005f1bbf90de61b4c5ebc34302307fa05b019a7 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Wed Feb 9 11:15:30 2011 -0500 Add a heartbeat for the witness, this should write to repl_monitor table so we can see the witness in repl_status and monitor if it is working. Also close connection at the end of do_witness_create in repmgr.c commit ac1c6367ab689aeae2eff3dda22db42337f300c1 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Wed Feb 9 01:26:41 2011 -0500 Add a sighup handler to reload the configuration commit 7df2fb7b74a3c5287319e56112840d9c2a3e7d5b Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Thu Feb 3 18:42:36 2011 +0100 Change the is_pgup () check test remove spurious 'return' commit 7e58e6aa91ab3f681854a44fe282b44da81768fa Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Thu Feb 3 16:53:17 2011 +0100 Add constant for the sleep times and retry, rework monitor functions Rename MonitorExecute() to StandbyMonitor() Add WitnessMonitor() # very simple version to start service mode isolation commit 1b270dab2e2c3c60527b86a33cd0fc9c0d11c08c Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Thu Feb 3 16:23:01 2011 +0100 Improve PrimaryCheck add a function "bool is_pgup()" Now, repmgrd-master can work. commit c6f07229713c8f2b77596459c06184edddd8d77e Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Wed Feb 2 19:31:06 2011 +0100 Fix strcmp in config parser, now failover parameter should be set correctly commit 0b690698a0d9aa87d3e8f1e462ee0771aa2ae9e8 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Wed Feb 2 16:23:50 2011 +0100 fix sprintf extra param commit 6050da315824048661be9c425ae6005576e5870f Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Wed Feb 2 13:53:29 2011 +0100 Add some other files to ignore commit a146dd581b46ea0e26b7b56b087d6b0d4ae15d44 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Wed Feb 2 13:53:04 2011 +0100 Fix SQL query commit 8f5db0f9c0f68ce2519afda72b6a778536427eab Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Wed Feb 2 00:51:54 2011 +0100 Some more minor fix and remove TODO commit c9299ad74e8f929bdc24804a6a834f24b66b7074 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Wed Feb 2 00:39:18 2011 +0100 fix some memory leak and fix testlogic for is_standby is_witness * is_standby() must be tested *after* is_witness else we think we are in a master * remove SELECT * in favor of SELECT witness commit cc5d06ea8bf1dcde4c264e95eb90f7fb1e821af3 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Tue Feb 1 23:40:15 2011 +0100 Forgot to remove a param from fprintf commit 426e22fa8dfd78f0c256bda1b166a31807de9ec6 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Tue Feb 1 22:47:58 2011 +0100 Restore previous usage of --force and rsync tablespace before data_dir The --force option is used to reduce the time needed to restore a failed node: it will overwrite existing files thanks to rsync --delete option The tablespaces need to be coyed first, because there are symlinks to them from the data_directory commit 1937973fced703d14159e6aae1cbdabb5619accb Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Tue Feb 1 21:09:12 2011 +0100 Improve message of repmgrd commit 035a9bcc1eea55cd95790bc72276727cc492694a Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Tue Feb 1 21:08:38 2011 +0100 Fix (bool *)PQgetval commit bf9181654213f898949e9c8f094b974915f82258 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Tue Feb 1 01:54:49 2011 +0100 Fix pg_hba on witness and connection * Copy the pg_hba.conf file from master to witness server * createdb and createuser in witness if they are different from getenv(USER) commit a2d8dcb2fd105d8f02bd76856969aca6605c66fa Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Tue Feb 1 01:01:43 2011 +0100 Improve initialization of repmgr (+ critical bug and minor fixes) * standby clone now *clone* the master files and dir to the *same* place on the standby if destination_directory is not provided * add preload library to the witness configuration * sleep 2 seconds after starting the witness postgresql to let it start enough to be able to connect to it. * Fix rsync files * Fix insert configuration into witness commit bc1a265d272e4805ac7859c208b51b57edd10fc7 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Mon Jan 31 12:25:20 2011 +0100 Fix some error message new line commit e087bd5de5ab43ffac90c6a20df6ef3fb19eed6d Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Mon Jan 31 11:37:08 2011 +0100 Guess data_directory from master in 'standby clone' and remove --force for dir --force does not overwrite directories anymore (it was not working very well anyway) dest_dir is the same as the master's one by default. Move down the tablespace check directories process commit 0a961e7ef05f26c87af1946b8141a639076fc488 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Mon Jan 31 11:21:40 2011 +0100 Add new function: create_pgdir (and fix 2 bugs in the process). It also fix function create_schema. Reduce repmgr code commit 7e5958dcc1daa9b54cb6f295af96fbef750c7952 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Mon Jan 31 10:34:58 2011 +0100 Improve an ERROR message commit f3a66a65a361f919727fc2d0ff9bf9544a10a822 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Mon Jan 31 10:25:45 2011 +0100 Improve error message about 'wal_keep_segments' commit 150dbcc0fe53ce4eff08797210fd2e9e4dd0e17a Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Sat Jan 29 23:35:00 2011 -0500 Add witness server support commit 6281e22a9c467da883ad960567f8ab6bdbc155ba Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Thu Jan 27 21:32:11 2011 +0100 Build all at once and update debian makefile to include the sql/ commit 50d752bf1ead7c9343900d4b494844284b7aac6c Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Thu Jan 27 02:10:31 2011 +0100 Adding information for debian and --version test commit 16d56dbfa05314eea69869ee2a7a705636432ad9 Author: Cédric Villemain <cedric@2ndQuadrant.fr> Date: Thu Jan 27 02:03:20 2011 +0100 Add a hint at the end of the standby clone and minor typo and message shuffle commit 6404ba247de1e2e3b995f30b6e7626e459849136 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Wed Jan 26 06:13:30 2011 -0500 Fix compiler warning about variables beign used unintialized commit a4f48993d5fe3b22bdd2aaefcff315115f8764b7 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Fri Jan 21 21:09:03 2011 -0500 Fix a new typo commit 904e61c9edcbbce6b1027c80ff77317d7cbd4919 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Fri Jan 21 19:30:56 2011 -0500 Use a function to make the call to repmgr_update_standby_location() so i avoid typos like the one i fixed in a previous commit. It also makes the code cleaner. commit 4ed388726f4bc0a52cc88d044d1f81697f348a7c Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Wed Jan 19 09:17:16 2011 -0500 Fix a typo when calling the sql function that writes shared memory commit d9232266561306eabef90e13c084c051a0e7f458 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Tue Jan 18 01:25:23 2011 -0500 Define the variable that we are using to test the result status of the system() call. commit 4d131c212b91e40ca027f76637c182456ab12514 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Tue Jan 18 01:04:12 2011 -0500 Makes repmgrd warn if promote_command or follow_command fails, add a "still alive" check for primary. Add a few messages and fix a bug in do_failover() in which we were using a closed PGresult. commit a5189e68cf4c8cf84259ea667a35e96de56fa4c9 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Thu Jan 13 15:45:50 2011 -0500 Initial attempt to get autofailover commit d0e09010a9d4610997c900b62ea1df2a71b01015 Author: Jaime Casanova <jaime@2ndQuadrant.com> Date: Wed Jan 12 14:40:29 2011 -0500 Add options failover, promote_command and follow_command to repmgr.conf, in pass also rename sample repmgr.conf to repmgr.conf.sample promote_command and follow_command allows to use a custom script for those actions.
2026-03-23 07:06:30 +00:00 · 2011-06-07 01:42:15 -04:00
parent 43268f2345
commit 3a950c9f8b
18 changed files with 1612 additions and 306 deletions
--- a/.gitignore
+++ b/.gitignore
@@ -1,6 +1,9 @@
 *~
 *.o
+*.so
 repmgr
 repmgrd
 README.htm*
 README.pdf
+sql/repmgr_funcs.so
+sql/repmgr_funcs.sql
--- a/10
+++ b/10
@@ -11,9 +11,11 @@ PG_CPPFLAGS = -I$(libpq_srcdir)
 PG_LIBS = $(libpq_pgport)

 all:  repmgrd repmgr
+	$(MAKE) -C sql

 repmgrd: $(repmgrd_OBJS)
 	$(CC) $(CFLAGS) $(repmgrd_OBJS) $(PG_LIBS) $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o repmgrd
+	$(MAKE) -C sql

 repmgr: $(repmgr_OBJS)
 	$(CC) $(CFLAGS) $(repmgr_OBJS) $(PG_LIBS) $(LDFLAGS) $(LDFLAGS_EX) $(LIBS) -o repmgr
@@ -33,6 +35,7 @@ endif
 install:
 	$(INSTALL_PROGRAM) repmgrd$(X) '$(DESTDIR)$(bindir)'
 	$(INSTALL_PROGRAM) repmgr$(X) '$(DESTDIR)$(bindir)'
+	$(MAKE) -C sql install

 ifneq (,$(DATA)$(DATA_built))
 	@for file in $(addprefix $(srcdir)/, $(DATA)) $(DATA_built); do \
@@ -45,10 +48,17 @@ clean:
 	rm -f *.o
 	rm -f repmgrd
 	rm -f repmgr
+	$(MAKE) -C sql clean

 deb: repmgrd repmgr
 	mkdir -p ./debian/usr/bin
 	cp repmgrd repmgr ./debian/usr/bin/
+	mkdir -p ./debian/usr/share/postgresql/9.0/contrib/
+	cp sql/repmgr_funcs.sql ./debian/usr/share/postgresql/9.0/contrib/
+	cp sql/uninstall_repmgr_funcs.sql ./debian/usr/share/postgresql/9.0/contrib/
+	mkdir -p ./debian/usr/lib/postgresql/9.0/lib/
+	cp sql/repmgr_funcs.so ./debian/usr/lib/postgresql/9.0/lib/
 	dpkg-deb --build debian
 	mv debian.deb ../postgresql-repmgr-9.0_1.0.0.deb
+	rm -rf ./debian/usr

--- a/check_dir.c
+++ b/check_dir.c
@@ -31,8 +31,6 @@
 #include "strutil.h"
 #include "log.h"

-static int mkdir_p(char *path, mode_t omode);
-
 /*
 * make sure the directory either doesn't exist or is empty
 * we use this function to check the new data directory and
@@ -124,7 +122,7 @@ set_directory_permissions(char *dir)
 * note that on failure, the path arg has been modified to show the particular
 * directory level we had problems with.
 */
-static int
+int
 mkdir_p(char *path, mode_t omode)
 {
 	struct stat sb;
@@ -225,8 +223,85 @@ is_pg_dir(char *dir)
 	const size_t buf_sz = 8192;
 	char		 path[buf_sz];
 	struct stat	 sb;
+	int		r;

+	// test pgdata
 	xsnprintf(path, buf_sz, "%s/PG_VERSION", dir);
+	if (stat(path, &sb) == 0)
+		return true;

-	return (stat(path, &sb) == 0) ? true : false;
+	// test tablespace dir
+	sprintf(path, "ls %s/PG_9.0_*/ -I*", dir);
+	r = system(path);
+	if (r == 0)
+		return true;
+
+	return false;
+}
+
+
+bool
+create_pgdir(char *dir, bool force)
+{
+	bool	pg_dir = false;
+
+	/* Check this directory could be used as a PGDATA dir */
+	switch (check_dir(dir))
+	{
+	case 0:
+		/* dir not there, must create it */
+		log_info(_("creating directory \"%s\"...\n"), dir);
+
+		if (!create_directory(dir))
+		{
+			log_err(_("couldn't create directory \"%s\"...\n"),
+			        dir);
+			exit(ERR_BAD_CONFIG);
+		}
+		break;
+	case 1:
+		/* Present but empty, fix permissions and use it */
+		log_info(_("checking and correcting permissions on existing directory %s ...\n"),
+		         dir);
+
+		if (!set_directory_permissions(dir))
+		{
+			log_err(_("could not change permissions of directory \"%s\": %s\n"),
+			        dir, strerror(errno));
+			exit(ERR_BAD_CONFIG);
+		}
+		break;
+	case 2:
+		/* Present and not empty */
+		log_warning(_("directory \"%s\" exists but is not empty\n"),
+		            dir);
+
+		pg_dir = is_pg_dir(dir);
+
+		/*
+		 * we use force to reduce the time needed to restore a node which
+		 * turn async after a failover or anything else
+		 */
+		if (pg_dir && force)
+		{
+			/* Let it continue */
+			break;
+		}
+		else if (pg_dir && !force)
+		{
+			log_warning(_("\nThis looks like a PostgreSQL directory.\n"
+			              "If you are sure you want to clone here, "
+			              "please check there is no PostgreSQL server "
+			              "running and use the --force option\n"));
+			exit(ERR_BAD_CONFIG);
+		}
+
+		return false;
+	default:
+		/* Trouble accessing directory */
+		log_err(_("could not access directory \"%s\": %s\n"),
+		        dir, strerror(errno));
+		exit(ERR_BAD_CONFIG);
+	}
+	return true;
 }
--- a/check_dir.h
+++ b/check_dir.h
@@ -20,9 +20,11 @@
 #ifndef _REPMGR_CHECK_DIR_H_
 #define _REPMGR_CHECK_DIR_H_

+int mkdir_p(char *path, mode_t omode);
 int  check_dir(char *dir);
 bool create_directory(char *dir);
 bool set_directory_permissions(char *dir);
 bool is_pg_dir(char *dir);
+bool create_pgdir(char *dir, bool force);

 #endif
--- a/config.c
+++ b/config.c
@@ -35,6 +35,10 @@ parse_config(const char *config_file, t_configuration_options *options)
 	memset(options->cluster_name, 0, sizeof(options->cluster_name));
 	options->node = -1;
 	memset(options->conninfo, 0, sizeof(options->conninfo));
+	options->failover = MANUAL_FAILOVER;
+	options->priority = 0;
+	memset(options->promote_command, 0, sizeof(options->promote_command));
+	memset(options->follow_command, 0, sizeof(options->follow_command));
 	memset(options->rsync_options, 0, sizeof(options->rsync_options));

 	/*
@@ -70,6 +74,27 @@ parse_config(const char *config_file, t_configuration_options *options)
 			strncpy (options->loglevel, value, MAXLEN);
 		else if (strcmp(name, "logfacility") == 0)
 			strncpy (options->logfacility, value, MAXLEN);
+		else if (strcmp(name, "failover") == 0)
+		{
+			char failoverstr[MAXLEN];
+			strncpy(failoverstr, value, MAXLEN);
+
+			if (strcmp(failoverstr, "manual") == 0)
+				options->failover = MANUAL_FAILOVER;
+			else if (strcmp(failoverstr, "automatic") == 0)
+				options->failover = AUTOMATIC_FAILOVER;
+			else
+			{
+				log_warning(_("value for failover option is incorrect, it should be automatic or manual. Defaulting to manual.\n"));
+				options->failover = MANUAL_FAILOVER;
+			}
+		}
+		else if (strcmp(name, "priority") == 0)
+			options->priority = atoi(value);
+		else if (strcmp(name, "promote_command") == 0)
+			strncpy(options->promote_command, value, MAXLEN);
+		else if (strcmp(name, "follow_command") == 0)
+			strncpy(options->follow_command, value, MAXLEN);
 		else
 			log_warning(_("%s/%s: Unknown name/value pair!\n"), name, value);
 	}
@@ -145,3 +170,72 @@ parse_line(char *buff, char *name, char *value)
 	trim(value);
 }

+<<<<<<< HEAD:config.c
+
+bool
+reload_configuration(char *config_file, t_configuration_options *orig_options)
+{
+	PGconn	*conn;
+
+	t_configuration_options new_options;
+
+	/*
+	 * Re-read the configuration file: repmgr.conf
+	 */
+	log_info(_("Reloading configuration file and updating repmgr tables\n"));
+	parse_config(config_file, &new_options);
+	if (new_options.node == -1)
+	{
+		log_warning(_("\nCannot load new configuration, will keep current one.\n"));
+		return false;
+	}
+
+	if (strcmp(new_options.cluster_name, orig_options->cluster_name) != 0)
+	{
+		log_warning(_("\nCannot change cluster name, will keep current configuration.\n"));
+		return false;
+	}
+
+	if (new_options.node != orig_options->node)
+	{
+		log_warning(_("\nCannot change node number, will keep current configuration.\n"));
+		return false;
+	}
+
+	if (new_options.failover != MANUAL_FAILOVER && new_options.failover != AUTOMATIC_FAILOVER)
+	{
+		log_warning(_("\nNew value for failover is not valid. Should be MANUAL or AUTOMATIC.\n"));
+		return false;
+	}
+
+	/* Test conninfo string */
+	conn = establishDBConnection(new_options.conninfo, false);
+	if (!conn || (PQstatus(conn) != CONNECTION_OK))
+	{
+		log_warning(_("\nconninfo string is not valid, will keep current configuration.\n"));
+		return false;
+	}
+	PQfinish(conn);
+
+	/* Configuration seems ok, will load new values */
+	strcpy(orig_options->cluster_name, new_options.cluster_name);
+	orig_options->node = new_options.node;
+	strcpy(orig_options->conninfo, new_options.conninfo);
+	orig_options->failover = new_options.failover;
+	orig_options->priority = new_options.priority;
+	strcpy(orig_options->promote_command, new_options.promote_command);
+	strcpy(orig_options->follow_command, new_options.follow_command);
+	strcpy(orig_options->rsync_options, new_options.rsync_options);
+	/*
+	 * XXX These ones can change with a simple SIGHUP?
+
+		strcpy (orig_options->loglevel, new_options.loglevel);
+		strcpy (orig_options->logfacility, new_options.logfacility);
+
+		logger_shutdown();
+		XXX do we have progname here ?
+		logger_init(progname, orig_options.loglevel, orig_options.logfacility);
+	*/
+
+	return true;
+}
--- a/config.h
+++ b/config.h
@@ -28,6 +28,10 @@ typedef struct
 	char cluster_name[MAXLEN];
 	int node;
 	char conninfo[MAXLEN];
+	int failover;
+	int priority;
+	char promote_command[MAXLEN];
+	char follow_command[MAXLEN];
 	char loglevel[MAXLEN];
 	char logfacility[MAXLEN];
 	char rsync_options[QUERY_STR_LEN];
@@ -36,5 +40,6 @@ typedef struct
 void parse_config(const char *config_file, t_configuration_options *options);
 void parse_line(char *buff, char *name, char *value);
 char *trim(char *s);
+bool reload_configuration(char *config_file, t_configuration_options *orig_options);

 #endif
--- a/dbutils.c
+++ b/dbutils.c
@@ -26,7 +26,7 @@ establishDBConnection(const char *conninfo, const bool exit_on_error)
 {
 	/* Make a connection to the database */
 	PGconn *conn = NULL;
-	char	connection_string[MAXLEN];
+	char    connection_string[MAXLEN];

 	strcpy(connection_string, conninfo);
 	strcat(connection_string, " fallback_application_name='repmgr'");
@@ -96,6 +96,84 @@ is_standby(PGconn *conn)
 }


+
+bool
+is_witness(PGconn *conn, char *schema, char *cluster, int node_id)
+{
+	PGresult   *res;
+	bool		result;
+	char		sqlquery[QUERY_STR_LEN];
+
+	sqlquery_snprintf(sqlquery, "SELECT witness from %s.repl_nodes where cluster = '%s' and id = %d",
+	                  schema, cluster, node_id);
+	res = PQexec(conn, sqlquery);
+	if (PQresultStatus(res) != PGRES_TUPLES_OK)
+	{
+		log_err(_("Can't query server mode: %s"), PQerrorMessage(conn));
+		PQclear(res);
+		PQfinish(conn);
+		exit(ERR_DB_QUERY);
+	}
+
+	if (strcmp(PQgetvalue(res, 0, 0), "f") == 0)
+		result = false;
+	else
+		result = true;
+
+	PQclear(res);
+	return result;
+}
+
+
+/* check the PQStatus and try to 'select 1' to confirm good connection */
+bool
+is_pgup(PGconn *conn)
+{
+	PGresult   *res;
+	char		sqlquery[QUERY_STR_LEN];
+	/* Check the connection status twice in case it changes after reset */
+	bool		twice = false;
+
+	/* Check the connection status twice in case it changes after reset */
+	for (;;)
+	{
+		if (PQstatus(conn) != CONNECTION_OK)
+		{
+			if (twice)
+				return false;
+			PQreset(conn);  // reconnect
+			twice = true;
+		}
+		else
+		{
+			/*
+			* Send a SELECT 1 just to check if connection is OK
+			* the PQstatus() won't catch disconnected connection
+			* XXXX
+			* the error message can be used by repmgrd
+			*/
+			sqlquery_snprintf(sqlquery, "SELECT 1");
+			res = PQexec(conn, sqlquery);
+			// we need to retry, because we might just have loose the connection once
+			if (PQresultStatus(res) != PGRES_TUPLES_OK)
+			{
+				log_err(_("PQexec failed: %s\n"), PQerrorMessage(conn));
+				PQclear(res);
+				if (twice)
+					return false;
+				PQreset(conn);  // reconnect
+				twice = true;
+			}
+			else
+			{
+				PQclear(res);
+				return true;
+			}
+		}
+	}
+}
+
+
 /*
 * If postgreSQL version is 9 or superior returns the major version
 * if 8 or inferior returns an empty string
@@ -207,7 +285,7 @@ get_cluster_size(PGconn *conn)
 * connection string is placed there.
 */
 PGconn *
-getMasterConnection(PGconn *standby_conn, int id, char *cluster,
+getMasterConnection(PGconn *standby_conn, char *schema, int id, char *cluster,
                    int *master_id, char *master_conninfo_out)
 {
 	PGconn		*master_conn	 = NULL;
@@ -216,7 +294,6 @@ getMasterConnection(PGconn *standby_conn, int id, char *cluster,
 	char		 sqlquery[QUERY_STR_LEN];
 	char		 master_conninfo_stack[MAXCONNINFO];
 	char		*master_conninfo = &*master_conninfo_stack;
-	char		 schema_str[MAXLEN];
 	char		 schema_quoted[MAXLEN];

 	int		 i;
@@ -233,10 +310,9 @@ getMasterConnection(PGconn *standby_conn, int id, char *cluster,
 	 *
 	 * Assemble the unquoted schema name
 	 */
-	maxlen_snprintf(schema_str, "repmgr_%s", cluster);
 	{
-		char *identifier = PQescapeIdentifier(standby_conn, schema_str,
-		                                      strlen(schema_str));
+		char *identifier = PQescapeIdentifier(standby_conn, schema,
+		                                      strlen(schema));

 		maxlen_snprintf(schema_quoted, "%s", identifier);
 		PQfreemem(identifier);
@@ -247,7 +323,7 @@ getMasterConnection(PGconn *standby_conn, int id, char *cluster,
 	         cluster);

 	sqlquery_snprintf(sqlquery, "SELECT * FROM %s.repl_nodes "
-	                  " WHERE cluster = '%s' and id <> %d",
+	                  " WHERE cluster = '%s' and id <> %d and not witness",
 	                  schema_quoted, cluster, id);

 	res1 = PQexec(standby_conn, sqlquery);
--- a/dbutils.h
+++ b/dbutils.h
@@ -24,12 +24,14 @@ PGconn *establishDBConnection(const char *conninfo, const bool exit_on_error);
 PGconn *establishDBConnectionByParams(const char *keywords[],
                                      const char *values[],
                                      const bool exit_on_error);
-bool	is_standby(PGconn *conn);
+bool    is_standby(PGconn *conn);
+bool    is_witness(PGconn *conn, char *schema, char *cluster, int node_id);
+bool	is_pgup(PGconn *conn);
 char   *pg_version(PGconn *conn, char* major_version);
 bool	guc_setted(PGconn *conn, const char *parameter, const char *op,
                const char *value);
 const char	 *get_cluster_size(PGconn *conn);
-PGconn *getMasterConnection(PGconn *standby_conn, int id, char *cluster,
+PGconn *getMasterConnection(PGconn *standby_conn, char *schema, int id, char *cluster,
                            int *master_id, char *master_conninfo_out);

 #endif
--- a/debian/DEBIAN/control
+++ b/debian/DEBIAN/control
@@ -1,4 +1,4 @@
-Package: repmgr
+Package: repmgr-auto
 Version: 1.0-1
 Section: database
 Priority: optional
--- a/errcode.h
+++ b/errcode.h
@@ -33,5 +33,6 @@
 #define ERR_PROMOTED 8
 #define ERR_BAD_PASSWORD 9
 #define ERR_STR_OVERFLOW 10
+#define ERR_FAILOVER_FAIL 11

 #endif	/* _ERRCODE_H_ */
--- a/repmgr.c
+++ b/repmgr.c
--- a/repmgr.conf.sample
+++ b/repmgr.conf.sample
@@ -12,6 +12,12 @@ node=2
 conninfo='host=192.168.204.104'
 rsync_options=--archive --checksum --compress --progress --rsh=ssh

+# Autofailover options
+failover=automatic
+priority=-1
+promote_command='repmgr promote'
+follow_command='repmgr follow'
+
 # Log level: possible values are DEBUG, INFO, NOTICE, WARNING, ERR, ALERT, CRIT or EMERG
 # Default: NOTICE
 loglevel=NOTICE
--- a/repmgr.h
+++ b/repmgr.h
@@ -30,6 +30,7 @@

 #define PRIMARY_MODE		0
 #define STANDBY_MODE		1
+#define WITNESS_MODE		2

 #include "config.h"
 #define MAXFILENAME		1024
@@ -42,6 +43,9 @@
 #define DEFAULT_DBNAME			"postgres"
 #define DEFAULT_REPMGR_SCHEMA_PREFIX	"repmgr_"

+#define MANUAL_FAILOVER		0
+#define AUTOMATIC_FAILOVER	1
+
 /* Run time options type */
 typedef struct
 {
@@ -58,7 +62,14 @@ typedef struct
 	bool ignore_rsync_warn;

 	char masterport[MAXLEN];
+	char localport[MAXLEN];

 } t_runtime_options;

+#define SLEEP_MONITOR		2
+#define SLEEP_RETRY			3
+#define NUM_RETRY			40
+
+
+
 #endif
--- a/repmgrd.c
+++ b/repmgrd.c
@@ -31,8 +31,22 @@
 #include "log.h"
 #include "strutil.h"

+#include "access/xlogdefs.h"
 #include "libpq/pqsignal.h"

+/*
+ * Struct to keep info about the nodes, used in the voting process in
+ * do_failover()
+ */
+typedef struct nodeInfo
+{
+	int nodeId;
+	XLogRecPtr xlog_location;
+	bool is_ready;
+} nodeInfo;
+
+
+char    myClusterName[MAXLEN];

 /* Local info */
 t_configuration_options local_options;
@@ -60,16 +74,28 @@ t_configuration_options config = {};

 static void help(const char* progname);
 static void usage(void);
-static void checkClusterConfiguration(PGconn *conn,PGconn *primary);
+static void checkClusterConfiguration(PGconn *conn, PGconn *primary);
 static void checkNodeConfiguration(char *conninfo);
 static void CancelQuery(void);

-static void MonitorExecute(void);
+static void StandbyMonitor(void);
+static void WitnessMonitor(void);
+static bool CheckPrimaryConnection(void);
+static void update_shared_memory(char *last_wal_standby_applied);
+static void update_registration(void);
+static void do_failover(void);

 static unsigned long long int walLocationToBytes(char *wal_location);

+/*
+ * Flag to mark SIGHUP. Whenever the main loop comes around it
+ * will reread the configuration file.
+ */
+static volatile sig_atomic_t got_SIGHUP = false;
+
+static void handle_sighup(SIGNAL_ARGS);
 static void handle_sigint(SIGNAL_ARGS);
-static void setup_cancel_handler(void);
+static void setup_event_handlers(void);

 #define CloseConnections()	\
 	if (PQisBusy(primaryConn) == 1) \
@@ -79,16 +105,6 @@ static void setup_cancel_handler(void);
 	if (primaryConn != NULL && primaryConn != myLocalConn) \
 		PQfinish(primaryConn);

-/*
- * Every 3 seconds, insert monitor info
- */
-#define MonitorCheck()						  \
-	for (;;)								  \
-	{										  \
-		MonitorExecute();					  \
-		sleep(3);							  \
-	}
-

 int
 main(int argc, char **argv)
@@ -137,7 +153,7 @@ main(int argc, char **argv)
 		}
 	}

-	setup_cancel_handler();
+	setup_event_handlers();

 	/*
 	 * Read the configuration file: repmgr.conf
@@ -173,38 +189,127 @@ main(int argc, char **argv)
 	 * Set my server mode, establish a connection to primary
 	 * and start monitor
 	 */
-	myLocalMode = is_standby(myLocalConn) ? STANDBY_MODE : PRIMARY_MODE;
-	if (myLocalMode == PRIMARY_MODE)
+	if (is_witness(myLocalConn, repmgr_schema, local_options.cluster_name, local_options.node))
+		myLocalMode = WITNESS_MODE;
+	else if (is_standby(myLocalConn))
+		myLocalMode = STANDBY_MODE;
+	else /* is the master */
+		myLocalMode = PRIMARY_MODE;
+
+	switch (myLocalMode)
 	{
+	case PRIMARY_MODE:
 		primary_options.node = local_options.node;
 		strncpy(primary_options.conninfo, local_options.conninfo, MAXLEN);
 		primaryConn = myLocalConn;
-	}
-	else
-	{
+
+		checkClusterConfiguration(myLocalConn, primaryConn);
+		checkNodeConfiguration(local_options.conninfo);
+
+		if (reload_configuration(config_file, &local_options))
+		{
+			PQfinish(myLocalConn);
+			myLocalConn = establishDBConnection(local_options.conninfo, true);
+			update_registration();
+		}
+
+		log_info(_("%s Starting continuous primary connection check\n"), progname);
+		/* Check that primary is still alive, and standbies are sending info */
+		/*
+		 * Every SLEEP_MONITOR seconds, do master checks
+		 * XXX
+		 * Check that standbies are sending info
+		*/
+		for (;;)
+		{
+			if (CheckPrimaryConnection())
+			{
+				/*
+									CheckActiveStandbiesConnections();
+									CheckInactiveStandbies();
+				*/
+				sleep(SLEEP_MONITOR);
+			}
+			else
+			{
+				/* XXX
+				 * May we do something more verbose ?
+				 */
+				exit (1);
+			}
+
+			if (got_SIGHUP)
+			{
+				/* if we can reload, then could need to change myLocalConn */
+				if (reload_configuration(config_file, &local_options))
+				{
+					PQfinish(myLocalConn);
+					myLocalConn = establishDBConnection(local_options.conninfo, true);
+					update_registration();
+				}
+				got_SIGHUP = false;
+			}
+		}
+		break;
+	case WITNESS_MODE:
+	case STANDBY_MODE:
 		/* I need the id of the primary as well as a connection to it */
 		log_info(_("%s Connecting to primary for cluster '%s'\n"),
 		         progname, local_options.cluster_name);
-		primaryConn = getMasterConnection(myLocalConn, local_options.node,
+		primaryConn = getMasterConnection(myLocalConn, repmgr_schema, local_options.node,
 		                                  local_options.cluster_name,
-		                                  &primary_options.node,NULL);
+		                                  &primary_options.node, NULL);
 		if (primaryConn == NULL)
 		{
 			CloseConnections();
 			exit(ERR_BAD_CONFIG);
 		}
-	}

-	checkClusterConfiguration(myLocalConn,primaryConn);
-	checkNodeConfiguration(local_options.conninfo);
-	if (myLocalMode == STANDBY_MODE)
-	{
-		log_info(_("%s Starting continuous standby node monitoring\n"), progname);
-		MonitorCheck();
-	}
-	else
-	{
-		log_info(_("%s This is a primary node, program not needed here; exiting'\n"), progname);
+		checkClusterConfiguration(myLocalConn, primaryConn);
+		checkNodeConfiguration(local_options.conninfo);
+
+		if (reload_configuration(config_file, &local_options))
+		{
+			PQfinish(myLocalConn);
+			myLocalConn = establishDBConnection(local_options.conninfo, true);
+			update_registration();
+		}
+
+		/*
+		 * Every SLEEP_MONITOR seconds, do checks
+		 */
+		if (myLocalMode == WITNESS_MODE)
+		{
+			log_info(_("%s Starting continuous witness node monitoring\n"), progname);
+		}
+		else if (myLocalMode == STANDBY_MODE)
+		{
+			log_info(_("%s Starting continuous standby node monitoring\n"), progname);
+		}
+
+		for (;;)
+		{
+			if (myLocalMode == WITNESS_MODE)
+				WitnessMonitor();
+			else if (myLocalMode == STANDBY_MODE)
+				StandbyMonitor();
+			sleep(SLEEP_MONITOR);
+
+			if (got_SIGHUP)
+			{
+				/* if we can reload, then could need to change myLocalConn */
+				if (reload_configuration(config_file, &local_options))
+				{
+					PQfinish(myLocalConn);
+					myLocalConn = establishDBConnection(local_options.conninfo, true);
+					update_registration();
+				}
+				got_SIGHUP = false;
+			}
+		}
+		break;
+	default:
+		log_err(_("%s: Unrecognized mode for node %d\n"), progname, local_options.node);
 	}

 	/* Prevent a double-free */
@@ -220,6 +325,73 @@ main(int argc, char **argv)
 	return 0;
 }

+/*
+ *
+ */
+static void
+WitnessMonitor(void)
+{
+	char monitor_witness_timestamp[MAXLEN];
+	PGresult	*res;
+
+	/*
+	 * Check if the master is still available, if after 5 minutes of retries
+	 * we cannot reconnect, return false.
+	 */
+	CheckPrimaryConnection(); // this take up to NUM_RETRY * SLEEP_RETRY seconds
+
+	if (PQstatus(primaryConn) != CONNECTION_OK)
+	{
+		/*
+		 * If we can't reconnect, just exit...
+		 * XXX we need to make witness connect to the new master
+		 */
+		PQfinish(myLocalConn);
+		exit(0);
+	}
+
+	/*
+	 * first check if there is a command being executed,
+	 * and if that is the case, cancel the query so i can
+	 * insert the current record
+	 */
+	if (PQisBusy(primaryConn) == 1)
+		CancelQuery();
+
+	/* Get local xlog info */
+	sqlquery_snprintf(sqlquery, "SELECT CURRENT_TIMESTAMP ");
+
+	res = PQexec(myLocalConn, sqlquery);
+	if (PQresultStatus(res) != PGRES_TUPLES_OK)
+	{
+		log_err(_("PQexec failed: %s\n"), PQerrorMessage(myLocalConn));
+		PQclear(res);
+		/* if there is any error just let it be and retry in next loop */
+		return;
+	}
+
+	strcpy(monitor_witness_timestamp, PQgetvalue(res, 0, 0));
+	PQclear(res);
+
+	/*
+	 * Build the SQL to execute on primary
+	 */
+	sqlquery_snprintf(sqlquery,
+	                  "INSERT INTO %s.repl_monitor "
+	                  "VALUES(%d, %d, '%s'::timestamp with time zone, "
+	                  " pg_current_xlog_location(), null,  "
+	                  " 0, 0)",
+	                  repmgr_schema, primary_options.node, local_options.node, monitor_witness_timestamp);
+
+	/*
+	 * Execute the query asynchronously, but don't check for a result. We
+	 * will check the result next time we pause for a monitor step.
+	 */
+	if (PQsendQuery(primaryConn, sqlquery) == 0)
+		log_warning(_("Query could not be sent to primary. %s\n"),
+		            PQerrorMessage(primaryConn));
+}
+

 /*
 * Insert monitor info, this is basically the time and xlog replayed,
@@ -227,7 +399,7 @@ main(int argc, char **argv)
 * Also do the math to see how far are we in bytes for being uptodate
 */
 static void
-MonitorExecute(void)
+StandbyMonitor(void)
 {
 	PGresult *res;
 	char monitor_standby_timestamp[MAXLEN];
@@ -245,50 +417,45 @@ MonitorExecute(void)
 	 * Check if the master is still available, if after 5 minutes of retries
 	 * we cannot reconnect, try to get a new master.
 	 */
-	for (connection_retries = 0; connection_retries < 15; connection_retries++)
-	{
-		if (PQstatus(primaryConn) != CONNECTION_OK)
-		{
-			log_warning(_("Connection to master has been lost, trying to recover...\n"));
-			/* wait 20 seconds between retries */
-			sleep(20);
+	CheckPrimaryConnection(); // this take up to NUM_RETRY * SLEEP_RETRY seconds

-			PQreset(primaryConn);
-		}
-		else
-		{
-			if (connection_retries > 0)
-			{
-				log_notice(_("Connection to master has been restored, continue monitoring.\n"));
-			}
-			break;
-		}
-	}
 	if (PQstatus(primaryConn) != CONNECTION_OK)
 	{
-		log_err(_("We couldn't reconnect to master. Now checking if another node has been promoted.\n"));
-		for (connection_retries = 0; connection_retries < 6; connection_retries++)
+		if (local_options.failover == MANUAL_FAILOVER)
 		{
-			primaryConn = getMasterConnection(myLocalConn, local_options.node,
-			                                  local_options.cluster_name, &primary_options.node,NULL);
-			if (PQstatus(primaryConn) == CONNECTION_OK)
+			log_err(_("We couldn't reconnect to master. Now checking if another node has been promoted.\n"));
+			for (connection_retries = 0; connection_retries < 6; connection_retries++)
 			{
-				/* Connected, we can continue the process so break the loop */
-				log_err(_("Connected to node %d, continue monitoring.\n"), primary_options.node);
-				break;
+				primaryConn = getMasterConnection(myLocalConn, repmgr_schema, local_options.node,
+				                                  local_options.cluster_name, &primary_options.node, NULL);
+				if (PQstatus(primaryConn) == CONNECTION_OK)
+				{
+					/* Connected, we can continue the process so break the loop */
+					log_err(_("Connected to node %d, continue monitoring.\n"), primary_options.node);
+					break;
+				}
+				else
+				{
+					log_err(_("We haven't found a new master, waiting before retry...\n"));
+					/* wait 5 minutes before retries, after 6 failures (30 minutes) we stop trying */
+					sleep(300);
+				}
 			}
-			else
+
+			if (PQstatus(primaryConn) != CONNECTION_OK)
 			{
-				log_err(_("We haven't found a new master, waiting before retry...\n"));
-				/* wait 5 minutes before retries, after 6 failures (30 minutes) we stop trying */
-				sleep(300);
+				log_err(_("We couldn't reconnect for long enough, exiting...\n"));
+				exit(ERR_DB_CON);
 			}
 		}
-	}
-	if (PQstatus(primaryConn) != CONNECTION_OK)
-	{
-		log_err(_("We couldn't reconnect for long enough, exiting...\n"));
-		exit(ERR_DB_CON);
+		else if (local_options.failover == AUTOMATIC_FAILOVER)
+		{
+			/*
+			 * When we returns from this function we will have a new primary and
+			 * a new primaryConn
+			 */
+			do_failover();
+		}
 	}

 	/* Check if we still are a standby, we could have been promoted */
@@ -370,6 +537,241 @@ MonitorExecute(void)
 }


+static void
+do_failover(void)
+{
+	PGresult *res1;
+	PGresult *res2;
+	char 	sqlquery[8192];
+
+	int		total_nodes = 0;
+	int		visible_nodes = 0;
+	bool	find_best = false;
+
+	int		i;
+	int		r;
+
+	int 	node;
+	char	nodeConninfo[MAXLEN];
+
+	unsigned int uxlogid;
+	unsigned int uxrecoff;
+	char last_wal_standby_applied[MAXLEN];
+
+	PGconn	*nodeConn = NULL;
+
+	/*
+	 * will get info about until 50 nodes,
+	 * which seems to be large enough for most scenarios
+	 */
+	nodeInfo nodes[50];
+	nodeInfo best_candidate;
+
+	/* first we get info about this node, and update shared memory */
+	sprintf(sqlquery, "SELECT pg_last_xlog_replay_location()");
+	res1 = PQexec(myLocalConn, sqlquery);
+	if (PQresultStatus(res1) != PGRES_TUPLES_OK)
+	{
+		log_err(_("PQexec failed: %s.\nReport an invalid value to not be considered as new primary and exit.\n"), PQerrorMessage(myLocalConn));
+		PQclear(res1);
+		sprintf(last_wal_standby_applied, "'%X/%X'", 0, 0);
+		update_shared_memory(last_wal_standby_applied);
+		exit(ERR_DB_QUERY);
+	}
+
+	/* write last location in shared memory */
+	update_shared_memory(PQgetvalue(res1, 0, 0));
+
+	/*
+	 * we sleep the monitor time + one second
+	 * we bet it should be enough for other repmgrd to update their own data
+	 */
+	sleep(SLEEP_MONITOR + 1);
+
+	/* get a list of standby nodes, including myself */
+	sprintf(sqlquery, "SELECT * "
+	        "  FROM %s.repl_nodes "
+	        " WHERE id IN (SELECT standby_node FROM %s.repl_status) "
+	        "   AND cluster = '%s' "
+	        " ORDER BY priority ",
+	        repmgr_schema, repmgr_schema, local_options.cluster_name);
+
+	res1 = PQexec(myLocalConn, sqlquery);
+	if (PQresultStatus(res1) != PGRES_TUPLES_OK)
+	{
+		log_err(_("Can't get nodes info: %s\n"), PQerrorMessage(myLocalConn));
+		PQclear(res1);
+		PQfinish(myLocalConn);
+		exit(ERR_DB_QUERY);
+	}
+
+	/* ask for the locations */
+	for (i = 0; i < PQntuples(res1); i++)
+	{
+		node = atoi(PQgetvalue(res1, i, 0));
+		/* Initialize on false so if we can't reach this node we know that later */
+		nodes[i].is_ready = false;
+		strncpy(nodeConninfo, PQgetvalue(res1, i, 2), MAXLEN);
+		nodeConn = establishDBConnection(nodeConninfo, false);
+		/* if we can't see the node just skip it */
+		if (PQstatus(nodeConn) != CONNECTION_OK)
+			continue;
+
+		sqlquery_snprintf(sqlquery, "SELECT repmgr_get_last_standby_location()");
+		res2 = PQexec(nodeConn, sqlquery);
+		if (PQresultStatus(res2) != PGRES_TUPLES_OK)
+		{
+			log_info(_("Can't get node's last standby location: %s\n"), PQerrorMessage(nodeConn));
+			log_info(_("Connection details: %s\n"), nodeConninfo);
+			PQclear(res2);
+			PQfinish(nodeConn);
+			continue;
+		}
+
+		visible_nodes++;
+
+		if (sscanf(PQgetvalue(res2, 0, 0), "%X/%X", &uxlogid, &uxrecoff) != 2)
+			log_info(_("could not parse transaction log location \"%s\"\n"), PQgetvalue(res2, 0, 0));
+
+		nodes[i].nodeId = node;
+		nodes[i].xlog_location.xlogid = uxlogid;
+		nodes[i].xlog_location.xrecoff = uxrecoff;
+		nodes[i].is_ready = true;
+
+		PQclear(res2);
+		PQfinish(nodeConn);
+	}
+	PQclear(res1);
+	/* Close the connection to this server */
+	PQfinish(myLocalConn);
+
+	/*
+	 * total nodes that are registered, include master which is a node but was
+	 * not counted because it's not a standby
+	 */
+	total_nodes = i + 1;
+
+	/*
+	 * am i on the group that should keep alive?
+	 * if i see less than half of total_nodes then i should do nothing
+	 */
+	if (visible_nodes < (total_nodes / 2.0))
+	{
+		log_err(_("Can't reach most of the nodes.\n"
+		          "Let the other standby servers decide which one will be the primary.\n"
+		          "Manual action will be needed to readd this node to the cluster.\n"));
+		exit(ERR_FAILOVER_FAIL);
+	}
+
+	/*
+	 * determine which one is the best candidate to promote to primary
+	 */
+	for (i = 0; i < total_nodes - 1; i++)
+	{
+		if (!nodes[i].is_ready)
+			continue;
+		else if (!find_best)
+		{
+			/* start with the first ready node, and then move on to the next one */
+			best_candidate.nodeId                = nodes[i].nodeId;
+			best_candidate.xlog_location.xlogid  = nodes[i].xlog_location.xlogid;
+			best_candidate.xlog_location.xrecoff = nodes[i].xlog_location.xrecoff;
+			best_candidate.is_ready              = nodes[i].is_ready;
+			find_best = true;
+		}
+
+		/* we use the macros provided by xlogdefs.h to compare XLogPtr */
+		/*
+		 * Nodes are retrieved ordered by priority, so if the current
+		 * best candidate is lower or equal to the next node's wal location
+		 * then assign next node as the new best candidate.
+		 */
+		if (XLByteLE(best_candidate.xlog_location, nodes[i].xlog_location))
+		{
+			best_candidate.nodeId                = nodes[i].nodeId;
+			best_candidate.xlog_location.xlogid  = nodes[i].xlog_location.xlogid;
+			best_candidate.xlog_location.xrecoff = nodes[i].xlog_location.xrecoff;
+			best_candidate.is_ready              = nodes[i].is_ready;
+		}
+	}
+
+	/* once we know who is the best candidate, promote it */
+	if (find_best && (best_candidate.nodeId == local_options.node))
+	{
+		if (verbose)
+			log_info(_("%s: This node is the best candidate to be the new primary, promoting...\n"),
+			         progname);
+		log_debug(_("promote command is: \"%s\"\n"), local_options.promote_command);
+		r = system(local_options.promote_command);
+		if (r != 0)
+		{
+			log_err(_("%s: promote command failed. You could check and try it manually.\n"), progname);
+			exit(ERR_BAD_CONFIG);
+		}
+	}
+	else if (find_best)
+	{
+		if (verbose)
+			log_info(_("%s: Node %d is the best candidate to be the new primary, we should follow it...\n"),
+			         progname, best_candidate.nodeId);
+		log_debug(_("follow command is: \"%s\"\n"), local_options.follow_command);
+		/*
+		 * New Primary need some time to be promoted.
+		 * The follow command should take care of that.
+		 */
+		r = system(local_options.follow_command);
+		if (r != 0)
+		{
+			log_err(_("%s: follow command failed. You could check and try it manually.\n"), progname);
+			exit(ERR_BAD_CONFIG);
+		}
+	}
+	else
+	{
+		log_err(_("%s: Did not find candidates. You should check and try manually.\n"), progname);
+		exit(ERR_FAILOVER_FAIL);
+	}
+
+	/* and reconnect to the local database */
+	myLocalConn = establishDBConnection(local_options.conninfo, true);
+}
+
+
+static bool
+CheckPrimaryConnection(void)
+{
+	int	connection_retries;
+
+	/*
+	 * Check if the master is still available
+	 * if after NUM_RETRY * SLEEP_RETRY seconds of retries
+	 * we cannot reconnect
+	 * return false
+	 */
+	for (connection_retries = 0; connection_retries < NUM_RETRY; connection_retries++)
+	{
+		if (!is_pgup(primaryConn))
+		{
+			log_warning(_("%s: Connection to master has been lost, trying to recover... %i seconds before failover decision\n"), progname, (SLEEP_RETRY*(NUM_RETRY-connection_retries)));
+			/* wait SLEEP_RETRY seconds between retries */
+			sleep(SLEEP_RETRY);
+		}
+		else
+		{
+			log_info(_("%s: Connection to master has been restored.\n"), progname);
+			break;
+		}
+	}
+	if (!is_pgup(primaryConn))
+	{
+		log_err(_("%s: We couldn't reconnect for long enough, exiting...\n"), progname);
+		/* XXX Anything else to do here? */
+		return false;
+	}
+	return true;
+}
+
+
 static void
 checkClusterConfiguration(PGconn *conn, PGconn *primary)
 {
@@ -433,17 +835,25 @@ checkNodeConfiguration(char *conninfo)

 	/*
 	 * If there isn't any results then we have not configured this node yet
-	 * in repmgr, if that is the case we will insert the node to the cluster
+	 * in repmgr, if that is the case we will insert the node to the cluster,
+	 * except if it is a witness
 	 */
 	if (PQntuples(res) == 0)
 	{
 		PQclear(res);

+		if (myLocalMode == WITNESS_MODE)
+		{
+			log_err(_("The witness is not configured\n"));
+			CloseConnections();
+			exit(ERR_BAD_CONFIG);
+		}
+
 		/* Adding the node */
 		log_info(_("%s Adding node %d to cluster '%s'\n"),
 		         progname, local_options.node, local_options.cluster_name);
 		sqlquery_snprintf(sqlquery, "INSERT INTO %s.repl_nodes "
-		                  "VALUES (%d, '%s', '%s')",
+		                  "VALUES (%d, '%s', '%s', 'f')",
 		                  repmgr_schema, local_options.node,
 		                  local_options.cluster_name,
 		                  local_options.conninfo);
@@ -500,12 +910,20 @@ static void
 handle_sigint(SIGNAL_ARGS)
 {
 	CloseConnections();
+	exit(1);
 }

+/* SIGHUP: set flag to re-read config file at next convenient time */
+static void
+handle_sighup(SIGNAL_ARGS)
+{
+	got_SIGHUP = true;
+}

 static void
-setup_cancel_handler(void)
+setup_event_handlers(void)
 {
+	pqsignal(SIGHUP, handle_sighup);
 	pqsignal(SIGINT, handle_sigint);
 }
 #endif
@@ -524,3 +942,43 @@ CancelQuery(void)

 	PQfreeCancel(pgcancel);
 }
+
+
+static void
+update_shared_memory(char *last_wal_standby_applied)
+{
+	PGresult *res;
+
+	sprintf(sqlquery, "SELECT repmgr_update_standby_location('%s')",
+	        last_wal_standby_applied);
+
+	/* If an error happens, just inform about that and continue */
+	res = PQexec(myLocalConn, sqlquery);
+	if (PQresultStatus(res) != PGRES_TUPLES_OK)
+	{
+		log_warning(_("Cannot update this standby's shared memory: %s\n"), PQerrorMessage(myLocalConn));
+		/* XXX is this enough reason to terminate this repmgrd? */
+	}
+	PQclear(res);
+}
+
+static void
+update_registration(void)
+{
+	PGresult *res;
+
+	sqlquery_snprintf(sqlquery, "UPDATE %s.repl_nodes "
+	                  "   SET conninfo = '%s', "
+	                  "       priority = %d "
+	                  " WHERE id = %d",
+	                  repmgr_schema, local_options.conninfo, local_options.priority, local_options.node);
+
+	res = PQexec(primaryConn, sqlquery);
+	if (PQresultStatus(res) != PGRES_COMMAND_OK)
+	{
+		log_err(_("Cannot update registration: %s\n"), PQerrorMessage(primaryConn));
+		CloseConnections();
+		exit(ERR_DB_CON);
+	}
+	PQclear(res);
+}
--- a/sql/Makefile
+++ b/sql/Makefile
@@ -0,0 +1,19 @@
+#
+# Makefile
+# Copyright (c) 2ndQuadrant, 2010
+#
+
+MODULE_big = repmgr_funcs
+DATA_built=repmgr_funcs.sql
+DATA=uninstall_repmgr_funcs.sql
+OBJS=repmgr_funcs.o 
+
+ifdef USE_PGXS
+PGXS := $(shell pg_config --pgxs)
+include $(PGXS)
+else
+subdir = contrib/repmgr/sql
+top_builddir = ../../..
+include $(top_builddir)/src/Makefile.global
+include $(top_srcdir)/contrib/contrib-global.mk
+endif
--- a/sql/repmgr_funcs.c
+++ b/sql/repmgr_funcs.c
@@ -0,0 +1,189 @@
+/*
+ * repmgr_funcs.c
+ * Copyright (c) 2ndQuadrant, 2010
+ *
+ * Shared memory state management and some backend functions in SQL
+ */
+
+#include "postgres.h"
+#include "fmgr.h"
+#include "access/xlog.h"
+#include "miscadmin.h"
+#include "storage/ipc.h"
+#include "storage/lwlock.h"
+#include "storage/procarray.h"
+#include "storage/shmem.h"
+#include "storage/spin.h"
+#include "utils/builtins.h"
+
+/* same definition as the one in xlog_internal.h */
+#define MAXFNAMELEN 	64
+
+PG_MODULE_MAGIC;
+
+/*
+ * Global shared state
+ */
+typedef struct repmgrSharedState
+{
+    LWLockId    lock;           		/* protects search/modification */
+    char		location[MAXFNAMELEN];	/* last known xlog location */
+} repmgrSharedState;
+
+/* Links to shared memory state */
+static repmgrSharedState *shared_state = NULL;
+
+static shmem_startup_hook_type prev_shmem_startup_hook = NULL;
+
+void        _PG_init(void);
+void        _PG_fini(void);
+
+static void repmgr_shmem_startup(void);
+static Size repmgr_memsize(void);
+
+static bool repmgr_set_standby_location(char *locationstr);
+
+Datum repmgr_update_standby_location(PG_FUNCTION_ARGS);
+Datum repmgr_get_last_standby_location(PG_FUNCTION_ARGS);
+
+PG_FUNCTION_INFO_V1(repmgr_update_standby_location);
+PG_FUNCTION_INFO_V1(repmgr_get_last_standby_location);
+
+
+/*
+ * Module load callback
+ */
+void
+_PG_init(void)
+{
+	/*
+	 * In order to create our shared memory area, we have to be loaded via
+	 * shared_preload_libraries.  If not, fall out without hooking into any of
+	 * the main system.  (We don't throw error here because it seems useful to
+	 * allow the repmgr functions to be created even when the
+	 * module isn't active.  The functions must protect themselves against
+	 * being called then, however.)
+	 */
+	if (!process_shared_preload_libraries_in_progress)
+		return;
+
+	/*
+	 * Request additional shared resources.  (These are no-ops if we're not in
+	 * the postmaster process.)  We'll allocate or attach to the shared
+	 * resources in repmgr_shmem_startup().
+	 */
+	RequestAddinShmemSpace(repmgr_memsize());
+	RequestAddinLWLocks(1);
+
+	/*
+	 * Install hooks.
+	 */
+	prev_shmem_startup_hook = shmem_startup_hook;
+	shmem_startup_hook = repmgr_shmem_startup;
+}
+
+/*
+ * Module unload callback
+ */
+void
+_PG_fini(void)
+{
+	/* Uninstall hooks. */
+	shmem_startup_hook = prev_shmem_startup_hook;
+}
+
+/*
+ * shmem_startup hook: allocate or attach to shared memory,
+ */
+static void
+repmgr_shmem_startup(void)
+{
+	bool		found;
+
+	if (prev_shmem_startup_hook)
+		prev_shmem_startup_hook();
+
+	/* reset in case this is a restart within the postmaster */
+	shared_state = NULL;
+
+	/*
+	 * Create or attach to the shared memory state, including hash table
+	 */
+	LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);
+
+	shared_state = ShmemInitStruct("repmgr shared state",
+						   sizeof(repmgrSharedState),
+						   &found);
+
+	if (!found)
+	{
+		/* First time through ... */
+		shared_state->lock = LWLockAssign();
+		snprintf(shared_state->location, 	
+				sizeof(shared_state->location), "%X/%X", 0, 0);
+	}
+
+	LWLockRelease(AddinShmemInitLock);
+}
+
+
+/*
+ * Estimate shared memory space needed.
+ */
+static Size
+repmgr_memsize(void)
+{
+    return MAXALIGN(sizeof(repmgrSharedState));
+}
+
+
+static bool
+repmgr_set_standby_location(char *locationstr)
+{
+    /* Safety check... */
+    if (!shared_state)
+        return false;
+
+    LWLockAcquire(shared_state->lock, LW_EXCLUSIVE);
+	strncpy(shared_state->location, locationstr, MAXFNAMELEN);
+    LWLockRelease(shared_state->lock);
+
+	return true;
+}
+
+
+/* SQL Functions */
+
+/* Read last xlog location reported by this standby from shared memory */
+Datum 
+repmgr_get_last_standby_location(PG_FUNCTION_ARGS)
+{
+	char location[MAXFNAMELEN];
+
+	/* Safety check... */
+	if (!shared_state)
+		PG_RETURN_NULL();
+
+	LWLockAcquire(shared_state->lock, LW_SHARED);
+	strncpy(location, shared_state->location, MAXFNAMELEN);
+	LWLockRelease(shared_state->lock);
+
+	PG_RETURN_TEXT_P(cstring_to_text(location));
+}
+
+
+/* Set update last xlog location reported by this standby to shared memory */ 
+Datum       
+repmgr_update_standby_location(PG_FUNCTION_ARGS)
+{
+	text       *location = PG_GETARG_TEXT_P(0);
+	char       *locationstr;
+
+    /* Safety check... */
+    if (!shared_state)
+        PG_RETURN_BOOL(false);
+
+    locationstr = text_to_cstring(location);
+
+	PG_RETURN_BOOL(repmgr_set_standby_location(locationstr));
+}
--- a/sql/repmgr_funcs.sql.in
+++ b/sql/repmgr_funcs.sql.in
@@ -0,0 +1,15 @@
+/*
+ * repmgr_function.sql
+ * Copyright (c) 2ndQuadrant, 2010
+ *
+ */
+
+-- SET SEARCH_PATH TO 'repmgr';
+
+CREATE FUNCTION repmgr_update_standby_location(text) RETURNS boolean
+AS 'MODULE_PATHNAME', 'repmgr_update_standby_location'
+LANGUAGE C STRICT;
+
+CREATE FUNCTION repmgr_get_last_standby_location() RETURNS text
+AS 'MODULE_PATHNAME', 'repmgr_get_last_standby_location'
+LANGUAGE C STRICT;
--- a/sql/uninstall_repmgr_funcs.sql
+++ b/sql/uninstall_repmgr_funcs.sql
@@ -0,0 +1,2 @@
+DROP FUNCTION repmgr_update_standby_location(text);
+DROP FUNCTION repmgr_get_last_standby_location();