Ian Barwick
6e270b2faf
repmgrd: catch cases where more than one node has initiated voting
...
The node(s) with higher ID will "yield", leaving the decision making
up to the node with the lower ID.
This happens very rarely, usually when the random delay is close
enough on two or mode nodes that vote initiation is simultaneous.
2017-07-18 17:04:24 +09:00
Ian Barwick
248525ccba
Remove unused PQexpBuffer
...
It was being referenced without being initialised, but the output
which would have been placed there is not used anyway, so discard
completely.
2017-07-18 12:06:26 +09:00
Ian Barwick
2c8dd49831
repmgrd: additional check to ensure only one node handles failover
...
It's possible the "failover" is completed by one repmgrd before the
other has a chance to react, in which case the am_bdr_failover_handler()
check will not apply. Instead check if the node record has already been
set to "inactive".
2017-07-17 16:47:42 +09:00
Ian Barwick
48a0aa3bf7
repmgrd: improve failover handling
...
Make retry frequency/interval configurable as per streaming replication.
2017-07-17 14:56:52 +09:00
Ian Barwick
a56bb41891
Remove redundant fields from node record struct
2017-07-17 14:11:14 +09:00
Ian Barwick
0dcd479322
Store node status in node record struct
2017-07-17 13:50:17 +09:00
Ian Barwick
46acf75286
Fix usage of get_primary_node_record()
2017-07-17 12:12:59 +09:00
Ian Barwick
437cb26b7e
Fixes to function request_vote()
2017-07-17 12:04:56 +09:00
Ian Barwick
ec554e5694
Improve connection handling
...
Set "connect_timeout" and "fallback_application_name" if not present.
2017-07-17 11:10:37 +09:00
Ian Barwick
084e0429fc
Disable non-BDR functions for BDR-only buils
2017-07-17 08:44:49 +09:00
Ian Barwick
ec00202a31
Add configure option --with-bdr-only
...
Builds repmgr with only BDR functionality; other code is disabled
at critical points.
2017-07-16 17:18:34 +09:00
Ian Barwick
a6f8d59f77
Initial support for BDR-only build
2017-07-16 17:18:23 +09:00
Ian Barwick
a29bc3e0fa
Rename config.[ch] to configfile.[ch]
2017-07-16 09:41:26 +09:00
Ian Barwick
b1051bfca2
Minor fixes to BDR code
2017-07-16 08:27:07 +09:00
Ian Barwick
c93eb10ad3
Handle node recovery
2017-07-14 16:15:10 +09:00
Ian Barwick
951c7dbd07
repmgrd: in BDR mode, have each repmgrd monitor each node
...
This will cover both the case when an entire node including
repmgrd goes down, and when one PostgreSQL instance goes down
but repmgrd is still up (in which case only one of the repmgrds
will handle the failover).
2017-07-14 15:01:18 +09:00
Ian Barwick
e3b3fb65f0
repmgrd: restrict BDR monitoring to two node setup
...
It's not safe to have more than two nodes with this kind of
"failover", so we don't need to select alternative nodes by
priority.
2017-07-14 12:56:11 +09:00
Ian Barwick
91c7013024
Remove erroneously added file
2017-07-14 11:25:37 +09:00
Ian Barwick
d653888c65
Support pre-10 WAL functions
2017-07-14 10:40:11 +09:00
Ian Barwick
dfcf85a62f
repmgrd: further BDR sanity checks
2017-07-14 10:27:28 +09:00
Ian Barwick
875cdf5102
repmgrd: don't run in BDR mode on a node marked as inactive
2017-07-14 09:59:56 +09:00
Ian Barwick
aeba2e31f1
repmgrd: fix BDR sanity check
2017-07-14 09:54:34 +09:00
Ian Barwick
d77e8d4d22
repmgrd: split physical and BDR functionality into separate files
2017-07-13 17:21:29 +09:00
Ian Barwick
0320f409aa
Detect BDR capability via presence of extension
2017-07-13 14:13:46 +09:00
Ian Barwick
2594a46a4f
Limit repmgr installation on a BDR cluster to two nodes
2017-07-13 14:10:08 +09:00
Ian Barwick
7eadbf6b17
Various improvements to "repmgr bdr register/unregister"
2017-07-12 22:38:03 +09:00
Ian Barwick
0a1addfdc0
When registering a BDR node, sync repmgr.nodes from another node
...
If a BDR node is added via bdr_group_join(), repmgr.nodes will
start off empty, so we'll need to sync it ourselves before adding
it to the repmgr replication set.
2017-07-12 10:11:25 +09:00
Ian Barwick
1cccb1dd5a
Add "repmgr bdr unregister"
2017-07-12 10:11:21 +09:00
Ian Barwick
71a0871232
Add "repmgr bdr register"
2017-07-11 15:38:58 +09:00
Ian Barwick
2962ffe605
repmgrd: initial BDR monitoring support
2017-07-10 23:58:59 +09:00
Ian Barwick
dddea9814b
Add BDR-related database functions
2017-07-10 21:52:39 +09:00
Ian Barwick
941ed343f8
repmgrd: additional location check
...
If only one promotion candidate is available, check it is in the
same location as the primary.
2017-07-10 15:20:43 +09:00
Ian Barwick
5fbcf3e476
Remove witness server references
2017-07-10 09:31:31 +09:00
Ian Barwick
9e3d942917
Handle various (unlikely) failure states
2017-07-10 09:00:18 +09:00
Ian Barwick
5bf7098139
repmgrd: consolidate clear_node_info_list() calls
2017-07-09 11:10:49 +09:00
Ian Barwick
2787994a6e
Make repmgrd failover settings configurable
2017-07-07 21:11:22 +09:00
Ian Barwick
b08511ec79
When cloning from Barman, use basebackups_directory
...
See: https://github.com/2ndQuadrant/repmgr/issues/312
2017-07-07 13:39:10 +09:00
Ian Barwick
35df85e67d
repmgrd: improve handling of "degraded monitoring"
...
In some cases, the monitored upstream may not be available for a while
(e.g. network split), in which case it makes sense to have repmgrd
keep running and trying to reconnect. Previously it would just keel
over and quit.
2017-07-06 17:19:55 +09:00
Ian Barwick
6b82ecbafc
repmgrd: check for visibility of primary location
...
Also support degraded monitoring status
2017-07-06 11:45:04 +09:00
Ian Barwick
0d226867b4
Add "location" column
2017-07-06 01:17:00 +09:00
Ian Barwick
9351e532b4
Ensure configuration parameter "replication_user" is smaller than NAMEDATALEN
2017-07-06 00:22:23 +09:00
Ian Barwick
ff4cc64a67
repmgrd: refresh upstream node record from upstream itself
2017-07-05 13:23:42 +09:00
Ian Barwick
3c45853b6f
Fixes for cascaded standby failover
2017-07-05 12:43:35 +09:00
Ian Barwick
a31d66f826
repmgr standby follow: add event details
...
Useful to have a confirmation of which node was followed.
2017-07-05 11:36:30 +09:00
Ian Barwick
2f2a983272
Fixes for cascaded standby monitoring
2017-07-05 11:26:09 +09:00
Ian Barwick
614287548d
Fix function get_primary_node_record()
2017-07-05 11:20:32 +09:00
Ian Barwick
617dee6bd6
Add function create_event_record()
...
For logging an event to the event table without generating an external
event notification.
Rename existing create_event_record*() functions to create_event_notification*()
as this describes their function better.
2017-07-05 09:52:22 +09:00
Ian Barwick
d3ec15cd25
Fixes from merge
2017-07-04 23:57:43 +09:00
Ian Barwick
582d0ef363
Rename "logxxx" configuration file parameters to "log_xxx"
...
This is more consistent with other parameters and conforms to
the pattern used by PostgreSQL itself, which uses the prefix "log_"
for logging parameters.
A warning will be emitted if the old version of the parameter name
is detected.
2017-07-04 23:36:47 +09:00
Ian Barwick
9bf264177f
Add checks for use of --upstream-node-id
2017-07-04 23:35:19 +09:00
