Christopher Browne [Mon, 15 Dec 2014 16:57:47 +0000 (11:57 -0500)]
Remove a no longer used variable
Christopher Browne [Fri, 12 Dec 2014 23:54:48 +0000 (18:54 -0500)]
Initial swing at bug #352, resolving:
a) set_config_option() has an 8 parm version in Postgres HEAD (9.5?)
b) resolution of the INT64_FORMAT refactoring in PG HEAD (9.5?)
Christopher Browne [Fri, 12 Dec 2014 23:42:29 +0000 (18:42 -0500)]
Merge branch 'master' of ssh://git.postgresql.org/slony1-engine
Steve Singer [Tue, 9 Dec 2014 02:48:06 +0000 (21:48 -0500)]
Fix comment and log message so it no longer
refers to helper threads(which were removed in 2.2)
Steve Singer [Tue, 9 Dec 2014 02:41:18 +0000 (21:41 -0500)]
Fix comment to reflect the actual default value
Patch by Kazuki Uehara <
[email protected]>
Steve Singer [Fri, 25 Jul 2014 15:35:59 +0000 (11:35 -0400)]
Bug 345
The strtok_r on the node list was adding replacing the comma
with a NULL causing only 1 node to get inserted into sl_event.
This mean any nodes that where cascaded only saw 1 of the nodes
in the drop node list.
Call strtok_r on a copy of the list so it can safely be munged.
Also add in a DropNode cluster test to demonstrate test this issue
Steve Singer [Thu, 21 Aug 2014 12:01:03 +0000 (08:01 -0400)]
Include alloca.h
Fix build problems on Solaris/SPARC in slonik due to an
undefined alloca symbol.
Report and fix from Venkata Balaji N
Jan Wieck [Tue, 12 Aug 2014 17:30:04 +0000 (13:30 -0400)]
Remove warning message for possible unsupported version up to 9.4.
Christopher Browne [Thu, 17 Jul 2014 20:26:20 +0000 (16:26 -0400)]
Reset RELEASE notes to be empty, as new development begins on a new branch
Jan Wieck [Mon, 30 Jun 2014 14:34:29 +0000 (10:34 -0400)]
Continue setting options even if a non-existing one
was encountered. Fixes bug 343.
Steve Singer [Tue, 13 May 2014 13:49:37 +0000 (09:49 -0400)]
fixing generate load origin node parameter mismatch
Steve Singer [Mon, 12 May 2014 01:37:45 +0000 (21:37 -0400)]
Bug 342 - Fix various failover issues
This commit fixes some related failover issues
1. The query slonik used to get a node list during failover
was ignoring non-failover targets. This meant that we
would never call preFailover on a node that wasn't a failover
target. This node might then still be trying to use
a failed provider.
2. A node (say node 3) might process a FAILOVER_NODE
event from node 2 (the new-origin) but might not
have yet processed the last event from node 1, the
failed node. We need to make sure that the
subscription paths and listen networks are updated
so we don't only listen for events and data from
failed nodes.
Steve Singer [Mon, 30 Jun 2014 12:37:54 +0000 (08:37 -0400)]
Remove HAVE_POSIX_SIGNALS from config.h
Compiling slony against PG 9.4 gives compiler warnings because the
definition of HAVE_POSIX_SIGNALS has changed. This is defined in pg_config
so we do not really need to redefine it in the config.h slony provides.
We leave the configure check in place to we just don't include it in
the header
Steve Singer [Fri, 2 May 2014 18:27:33 +0000 (14:27 -0400)]
add another multi-node failover configuration to the test
David Christensen [Fri, 25 Apr 2014 18:36:32 +0000 (13:36 -0500)]
Fix typo in preamble for regression tests
Jan Wieck [Sat, 26 Apr 2014 02:41:25 +0000 (22:41 -0400)]
Changing version number of master to 2.3.0
Steve Singer [Thu, 17 Apr 2014 13:54:14 +0000 (09:54 -0400)]
Document the EXECUTE ONLY ON list of nodes syntax
Also add a regression test to test this
Steve Singer [Tue, 8 Apr 2014 01:52:19 +0000 (21:52 -0400)]
Bug 338 - Have ddlScript return a bigint since the action_seq is a bigint
Rose Nancy [Tue, 11 Mar 2014 14:37:00 +0000 (10:37 -0400)]
fixing Deadlock with application during minor version slony upgrade
Rose Nancy [Tue, 11 Mar 2014 14:10:36 +0000 (10:10 -0400)]
fixing Deadlock with application during minor version slony upgrade
Christopher Browne [Wed, 5 Mar 2014 21:48:05 +0000 (16:48 -0500)]
Merge branch 'master' of ssh://git.postgresql.org/slony1-engine
Jan Wieck [Wed, 29 Jan 2014 17:11:57 +0000 (12:11 -0500)]
logApply trigger free()s data from wrong cache entry on evict.
Steve Singer [Wed, 5 Feb 2014 23:33:49 +0000 (18:33 -0500)]
Fixing typo in docs
Thanks Jeff Frost
Rose Nancy [Wed, 29 Jan 2014 17:21:34 +0000 (12:21 -0500)]
Clarify namespace setup in documentation
Steve Singer [Wed, 25 Dec 2013 21:04:06 +0000 (16:04 -0500)]
Bug 326 Run perl tidy on the altperl tools
Formatting changes (perltidy) on the altperl tools
Patch by David Fetter
Steve Singer [Wed, 25 Dec 2013 20:58:46 +0000 (15:58 -0500)]
Bug 325 Make perltool pass level 5 of perlcrit
Patch from David Fetter
Christopher Browne [Thu, 2 Jan 2014 21:15:29 +0000 (16:15 -0500)]
Fix SGML tagging of man page volnum - bug #328
Steve Singer [Thu, 12 Dec 2013 14:26:04 +0000 (09:26 -0500)]
Use server includes for pgport when building slonik
When building with --with-pgport slonik needs to include port.h
port.h is normally found under the server include directory but
some distributions (ie debian libpq-dev) also include it in
/usr/include/postgresql
The fix for bug 315 removed the server include paths from
all client programs including slonik. This went too far,
the actual issue with bug 315 was src/backend failing because
it pulled in the client headers.
If we are building with pgport we will pull in the server
include directories for slonik
Christopher Browne [Mon, 18 Nov 2013 17:02:34 +0000 (12:02 -0500)]
Merge branch 'master' of ssh://git.postgresql.org/slony1-engine
Jan Wieck [Wed, 30 Oct 2013 16:12:11 +0000 (12:12 -0400)]
Disable the "known failing" disorder tests.
Jan Wieck [Tue, 29 Oct 2013 22:48:25 +0000 (18:48 -0400)]
Folding slon.stop() and coordinator.join(slon) into one loop.
Stopping the slons and joining their processes in separate
loops apparently can cause the regression test to hang under
some rare race conditions.
Steve Singer [Thu, 31 Oct 2013 13:05:20 +0000 (09:05 -0400)]
bug 321 - fix to the previous patch for when we have multiple real providers
If the provider list contains multiple providers that actually provide
a set (set_head != NULL) then we want to use those providers and
not skip them
Steve Singer [Wed, 30 Oct 2013 19:51:23 +0000 (15:51 -0400)]
bug 321 - Stop reconnecting to providers
Event origins that are not set origins will have
multiple li_origin/li_receiver pairs in sl_listen
we listen for events from these origins from multiple
sources. This means a node such as node 2 might
receive an event from ev_origin=1 via event provider 1
then receive the next event with ev_origin=1 from
event provider 2. Slon keeps connecting and
disconnecting from these providers.
Instead keep of dropping the connections to
event providers we keep them around unless
a reconfiguration has taken place. This lets
us use the connection for the event_provider
we received the event from.
Steve Singer [Mon, 28 Oct 2013 18:11:10 +0000 (14:11 -0400)]
add a regression test for EXECUTE SCRIPT using a non-provider event node
Jan Wieck [Fri, 25 Oct 2013 04:39:53 +0000 (00:39 -0400)]
Disable the "not expect lag" check from disorder ExecuteScript.
"Apparent" lag of 10 seconds can happen, the way confirm
messages propagate back to the origin.
Jan Wieck [Fri, 25 Oct 2013 04:34:11 +0000 (00:34 -0400)]
From
fccfb06687588c6155f0f77665be358837dd47e1 Mon Sep 17 00:00:00 2001
From: Steve Singer <
[email protected]>
Date: Thu, 24 Oct 2013 21:29:32 -0400
Subject: [PATCH] fixing the failover unit test.
It is sometimes possible for node 4 to get unsubscribed from
set 1 because it doesn't have a subscription to set 2 but
might be the farthest ahead node (but it isn't a failover candidate)
Resubscribed node 4 before continuing on with the next test
Steve Singer [Tue, 29 Oct 2013 20:22:37 +0000 (16:22 -0400)]
Additional changes for bug322.
cloneNodePrepare() and many other places need to grab
the central config lock. They heavily rely on checking
things via "not found", which inherently has a race
condition.
Conflicts:
src/slon/remote_worker.c
note: removed src/slon/remote_worker.c changes from
the cherry-picked patch they aren't actually part of 322
Steve Singer [Thu, 17 Oct 2013 21:21:36 +0000 (17:21 -0400)]
bug 322 allow CLONE PREPARE processing to deal with an earlier STORE PATH
If a node sees a STORE PATH event before it sees the CLONE PREPARE
that creates a node involved in that path then the CLONE PREPARE
event was failing because the node was already in sl_node.
Check for the existance for rows and update them instead
Conflicts:
src/slon/remote_worker.c
Jan Wieck [Mon, 14 Oct 2013 18:23:11 +0000 (14:23 -0400)]
Fix for bug 319, dereferencing of NULL pointers.
The "provider" pointer at that point in the code is
always NULL due to the for() loop above. The query can
only fail if the connection to the backend is lost in
that very moment.
Steve Singer [Wed, 16 Oct 2013 13:10:01 +0000 (09:10 -0400)]
adding in tests for bug 318
Jan Wieck [Mon, 14 Oct 2013 18:10:45 +0000 (14:10 -0400)]
Fix for endless restart loop on FAILOVER (bug 318).
If the backup node is not the most advanced, function
failedNode() will trigger a slon restart even if it
did not make any changes to the configuration because
of a missing WHERE clause.
Jan Wieck [Wed, 16 Oct 2013 03:05:56 +0000 (23:05 -0400)]
The UPDATE FUNCTIONS upgrade from a 2.1.x (or 2.0.x) Slony to 2.2.0 does not
create the no_failed column in sl_node.
Christopher Browne [Thu, 10 Oct 2013 18:00:33 +0000 (14:00 -0400)]
Note that upgrade process implies an application outage
Steve Singer [Thu, 3 Oct 2013 17:32:08 +0000 (13:32 -0400)]
Only include the libpq header directory OR the server header directory but not both
This addresses bug 315 where a debian install had libpq-dev 9.3 installed
and postgresql-server-dev 8.4 installed. The libpq headers are in /usr/include/postgresql
but the server headers are in /usr/lib/postgresql/8.4/server.
configure checks were failing because we were pulling in a mixutre of headers
from both versions. The only slony component that needs the server
headers are the backend functions and they shouldn't need to use libpq headers.
This patch sets CPPFLAGS as approriate for one or the other but not both
Christopher Browne [Thu, 26 Sep 2013 16:49:13 +0000 (12:49 -0400)]
Bug #281 - improve documentation on performance implications of PATHs
Add in material describing the effects of having fully versus less-fully connected path networks
Christopher Browne [Wed, 25 Sep 2013 16:56:48 +0000 (12:56 -0400)]
Merge branch 'master' of github.com:cbbrowne/slony1-engine
Christopher Browne [Mon, 16 Sep 2013 17:01:03 +0000 (13:01 -0400)]
Merge branch 'master' of git://git.postgresql.org/git/slony1-engine
Conflicts:
RELEASE
doc/adminguide/ddlchanges.sgml
Christopher Browne [Mon, 9 Sep 2013 21:49:25 +0000 (17:49 -0400)]
Merge branch 'master' of ssh://git.postgresql.org/slony1-engine
Christopher Browne [Mon, 9 Sep 2013 21:47:34 +0000 (17:47 -0400)]
Merge branch 'master' of ssh://git.postgresql.org/slony1-engine
Conflicts:
RELEASE
Steve Singer [Mon, 9 Sep 2013 01:08:39 +0000 (21:08 -0400)]
documentation updates - fixing out of data information
Steve Singer [Mon, 9 Sep 2013 00:49:36 +0000 (20:49 -0400)]
adding PG 9.3 to version list
Steve Singer [Mon, 9 Sep 2013 00:48:45 +0000 (20:48 -0400)]
updating copyright year
Steve Singer [Wed, 4 Sep 2013 19:53:31 +0000 (15:53 -0400)]
bug 313 - don't error out in applying EXECUTE SCRIPT sequence updates
Since bug 304 was fixed we have been including the list of replicated
sequences and current values in the arguments on sl_log_script rows.
If some sequences are replicated to some nodes but not others then the
execute script will fail to apply on the nodes that don't have those
sequences.
This patch ignores missing sequences during an execute script
instead of erroring out.
Christopher Browne [Tue, 3 Sep 2013 20:04:08 +0000 (16:04 -0400)]
Minor tweaks to docs
Steve Singer [Tue, 3 Sep 2013 14:16:07 +0000 (10:16 -0400)]
removing stray '_' that was introduced in bug 309
Steve Singer [Tue, 3 Sep 2013 13:06:36 +0000 (09:06 -0400)]
preparing for 2.2.0 release - updating version numbers
Steve Singer [Tue, 3 Sep 2013 12:59:55 +0000 (08:59 -0400)]
removing some bug references from the release notes
These bugs were introduced in 2.2.0 and fixed later in the 2.2.0 cycle
Christopher Browne [Fri, 30 Aug 2013 19:17:44 +0000 (15:17 -0400)]
Reorganize release notes for 2.2
Christopher Browne [Fri, 30 Aug 2013 19:07:00 +0000 (15:07 -0400)]
Merge branch 'master' of ssh://git.postgresql.org/slony1-engine
Conflicts:
RELEASE
doc/adminguide/ddlchanges.sgml
Rose Nancy [Tue, 27 Aug 2013 21:38:49 +0000 (17:38 -0400)]
BUG308 : added TCP KEEP ALIVE configurations in slon.conf-sample
Steve Singer [Tue, 20 Aug 2013 19:31:02 +0000 (15:31 -0400)]
release notes update
Steve Singer [Tue, 20 Aug 2013 14:06:22 +0000 (10:06 -0400)]
variable declarations must be at the start of the block (VC does still enforces this)
Steve Singer [Mon, 19 Aug 2013 20:55:46 +0000 (16:55 -0400)]
Updating version numbers to 2.2.0 rc1
Steve Singer [Mon, 19 Aug 2013 20:37:29 +0000 (16:37 -0400)]
release notes update
Steve Singer [Mon, 19 Aug 2013 20:31:12 +0000 (16:31 -0400)]
Merge branch 'bug310'
Steve Singer [Mon, 19 Aug 2013 20:29:45 +0000 (16:29 -0400)]
Bug 311 - Don't include performance.sgml ( 'man') file twice in the tar
Steve Singer [Thu, 15 Aug 2013 00:37:22 +0000 (20:37 -0400)]
bug 309 - make sure sl_failover_targets gets created when upgrading to 2.2
Steve Singer [Wed, 14 Aug 2013 21:47:30 +0000 (17:47 -0400)]
Merge branch 'bug304'
Steve Singer [Wed, 14 Aug 2013 21:30:13 +0000 (17:30 -0400)]
bug 310 - only issue the notify Restart if we actually change configuration
The FAILOVER_NODE event is processed in two parts by slon.
First failedNode(..) is called which reconfigures sl_subscribe,
sl_path and then the listen network. Once this is done notify
is used to signal the slon to restart.
Then the slon commits this transactions and waits for any
pending events from the failed origin to arrive from a third
provider. Slon then continues with the failover (failoverSet_int).
The problem was that when slon restarts it repeats the processing
the FAILOVER_NODE. It was possible for slon to keep restarting
before it gets to be caught up and recording the FAILOVER_NODE
event as processed.
In this patch we only restart slon if the configuration actually
is changed. After the first restart the configuration shouldn't
need changing so there should not be anymore restarts.
Steve Singer [Wed, 14 Aug 2013 14:46:45 +0000 (10:46 -0400)]
deconstruct_array wants the OID of the element type, not the array type
Christopher Browne [Mon, 12 Aug 2013 22:09:02 +0000 (18:09 -0400)]
Merge branch 'master' of git://git.postgresql.org/git/slony1-engine
Steve Singer [Fri, 9 Aug 2013 16:52:08 +0000 (12:52 -0400)]
create string_agg in slony1_base.sql not slony1_funcs.sql
If the CREATE AGGREGATE is in slony1_funcs.sql it will fail on an update functions
since the aggregate already exists. As part of the upgrade we will check
to see if we need to create the aggregate.
Christopher Browne [Fri, 2 Aug 2013 21:43:36 +0000 (17:43 -0400)]
Some erroneous queries in date test
Christopher Browne [Fri, 2 Aug 2013 21:31:30 +0000 (17:31 -0400)]
Fix typo in sample data
Christopher Browne [Fri, 2 Aug 2013 21:20:41 +0000 (17:20 -0400)]
Merge branch 'bug304' of github.com:ssinger/slony1-engine into bug304
Christopher Browne [Thu, 1 Aug 2013 21:53:16 +0000 (17:53 -0400)]
Add release note on #304
Christopher Browne [Thu, 1 Aug 2013 21:53:16 +0000 (17:53 -0400)]
Add release note on #304
Christopher Browne [Thu, 1 Aug 2013 21:32:12 +0000 (17:32 -0400)]
Revisions to DDL docs
Steve Singer [Thu, 1 Aug 2013 20:03:48 +0000 (16:03 -0400)]
Allow the apply trigger to work (ie not fail) on sl_log_script rows from earlier betas of 2.2.0
Steve Singer [Thu, 1 Aug 2013 19:09:45 +0000 (15:09 -0400)]
bug 304 - record sequence values as part of a EXECUTE_SCRIPT
Prior to 2.2 the EXECUTE_SCRIPT command result in a DDL_SCRIPT event
and a SYNC event generated as part of the same transaction. That
SYNC event also replicates the value of any sequences to the replica
before the DDL SCRIPT command.
In 2.2 the DDL script is replicated as part of a normal SYNC
as rows in sl_log_script. This commit stores any sequence changes
as extra elements in the cmdargs array. These sequence values are
then set before the DDL_SCRIPT is executed on the replica
This commit includes a unit test change to reproduce this issue
the test now passes
Steve Singer [Mon, 29 Jul 2013 21:10:51 +0000 (17:10 -0400)]
updating release notes
Steve Singer [Mon, 29 Jul 2013 21:00:44 +0000 (17:00 -0400)]
Merge branch 'bug305'
Christopher Browne [Mon, 29 Jul 2013 15:19:11 +0000 (11:19 -0400)]
Fixes to DDL test (old regression)
Steve Singer [Fri, 26 Jul 2013 20:10:47 +0000 (16:10 -0400)]
C requires variable declarations to be at the start of the block
Steve Singer [Fri, 26 Jul 2013 19:57:32 +0000 (15:57 -0400)]
upating win32 symbol definitions for 2.2.0 b5
Steve Singer [Fri, 26 Jul 2013 19:27:54 +0000 (15:27 -0400)]
updating version to for 2.2.0 b5
Steve Singer [Fri, 26 Jul 2013 19:26:41 +0000 (15:26 -0400)]
fixing sytnax errors in documentation SGML
Steve Singer [Fri, 26 Jul 2013 19:15:38 +0000 (15:15 -0400)]
bug 306 release notes update
Steve Singer [Fri, 26 Jul 2013 19:14:35 +0000 (15:14 -0400)]
updating versions for 2.2.0 beta 5
Steve Singer [Fri, 26 Jul 2013 17:51:54 +0000 (13:51 -0400)]
Merge branch 'bug306'
Conflicts:
doc/adminguide/slonyupgrade.sgml
Steve Singer [Fri, 26 Jul 2013 14:41:13 +0000 (10:41 -0400)]
release notes update
Steve Singer [Fri, 26 Jul 2013 14:34:37 +0000 (10:34 -0400)]
bug 301 - adding a missing PQclear to fix a memory leak
This PQclear was added by Jan to the patch he posted in bugzilla.
Free the result set when dealing with the error.
Steve Singer [Fri, 26 Jul 2013 14:23:55 +0000 (10:23 -0400)]
Merge branch 'bug301'
Steve Singer [Fri, 26 Jul 2013 14:23:48 +0000 (10:23 -0400)]
Merge branch 'bug299'
Conflicts:
doc/adminguide/slonyupgrade.sgml
Steve Singer [Thu, 25 Jul 2013 19:37:17 +0000 (15:37 -0400)]
add in a check/test to ensure that the backup node in a failover is a forwarding subscriber
If the backup node is not a forwarding subscriber to all of the sets from the origin
then the failover can't complete. It is best to check this early on and warn the user
before the failover actually starts.
Steve Singer [Fri, 19 Jul 2013 22:22:38 +0000 (18:22 -0400)]
add documentation that mentions that the filenames of installable files have changed
Steve Singer [Fri, 26 Jul 2013 14:17:48 +0000 (10:17 -0400)]
set session_replication_role to local in the apply trigger when performing DDL
This fixes bug 305 where user defined (non-replication) triggers don't fire during
the remote/replica exection of EXECUTE_SCRIPT commands
Steve Singer [Fri, 26 Jul 2013 12:50:51 +0000 (08:50 -0400)]
The table definitions we create on a 2.1 -> 2.2 upgrade should match the updated DDL in slony1_base.sql
This is to fix bug 306
Steve Singer [Thu, 25 Jul 2013 19:37:17 +0000 (15:37 -0400)]
add in a check/test to ensure that the backup node in a failover is a forwarding subscriber
If the backup node is not a forwarding subscriber to all of the sets from the origin
then the failover can't complete. It is best to check this early on and warn the user
before the failover actually starts.
Steve Singer [Thu, 25 Jul 2013 19:30:06 +0000 (15:30 -0400)]
additional dstring_free() to fix an additional memory leak in sync_helper