Hello I wrote a patch for pgsql RA to manage PostgreSQL 9.0 streaming replication(PGSR) which is new feature in Postgres 9.0.
I add new parameters, "rep_mode", "trigger_file" and "tmpdir".
"rep_mode" means replication mode
Default is "none" which do the same thing as before.
You have to set "async" to use PGSR.
"trigger_file" must be set same parameter of recovery.conf.
This file is created to promote hot standby to primary.
Default is /var/lib/pgsql/PGSQL.5432.trigger.
"tmpdir" is temporary directory used for trigger_file and flag files.
Default is /var/lib/pgsql.
* Details
PGSR's transition differs from Pacemaker's one as follows.
******* transition of Pacemaker ******************************
start promote
----------> ------------>
Stop Slave Master
<--------- <------------
stop demote
**************************************************************
******* transition of PGSR ***********************************
pg_ctl start
------------------------------------------>
pg_ctl start create trigger_file
---------------> ------------------------>
Stop Hot standby Primary
<----------------
pg_ctl stop
<------------------------------------------
pg_ctl stop
Note
PostgreSQL developer says that it's not normal operation
to start primary through hot standby initially.
***************************************************************
Therefore this patch has 4 states as follows.
STATE1 | STATE2 | STATE3 | STATE4
Stop | Slave | Slave | Master <- Pacemaker's state
Stop | Stop | HotStandby | Primary <- PGSR's state
The STATE2 is steppingstone to transit to the STATE4
so primary's initial start transits as follows.
STATE1 ---> STATE2 ---> STATE4
start promote
In the STATE2, PostgreSQL is stopping, so resource agent creates a flag
file($REPRESS_MONITOR) to cheat monitor operation, and waites for promote op.
* Method of starting
You have to start only primary server's pacemaker with primary configuration
and waits for promote op (STATE4).
After that, you backup data from primary server and restore it
to hot standby server and starts hot standby server's pacemaker with
hot standby configuration.
* About repressing start flag file
Hot standby server must be started with new restored data.
But if restored data is old, hot standby server fails to replication,
but resource
agent can't notice it.
At this time primary server is broken, hot standby server is promoted with very
old data, so I bring in repressing start flag file($REPRESS_START)
to repress restarting PostgreSQL automatically with old data.
Therefore you have to backup and restore new data and remove this flag
before starting PostgreSQL if it exists.
How do you think these implementation ?
Any comment ?
Regards,
Takatoshi MATSUO
Linux-HA Japan
pgsql.patch
Description: Binary data
_______________________________________________________ Linux-HA-Dev: Linux-HA-Dev@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev Home Page: http://linux-ha.org/
