Re: [GENERAL] 9.0 replication -- multiple hot_standby servers
On 10/28/10 11:25 PM, Dean Gibson AE7Q wrote: Two days ago I upgraded five DB boxes (for load balancing) from 8.3.0 to 9.0.1 in order to use replication. The replication configuration went reasonably well, and now all the four hot_standby servers are (streaming) replicating just fine from the primary DB server. If the primary fails and I touch the trigger file on one of the standby boxes, that goes into primary mode just as it should. Of course, I have to externally redirect updates to the new server. My question is, how do I configure the other three (still) hot_standby boxes to now use the new primary? Clearly I can change the recovery.conf file on each standby box, but that seems like an unnecessary nuisance. I've not worked with the 9.0.x replication yet, so my comments are of a more general nature... A) keep it super simple. complex schemes have a marvelous way of finding corner cases and biting you in the @%%. B) don't forget corner cases like a 'stoned' server that somehow isn't communicating with the others and decides ITS the master when in fact another node is running just fine. robust cluster management systems like Veritas Cluster insist on redundant inter-node heartbeat communications paths, and hardware fencing support so only one node can possibly be 'master' at any given time. -- Sent via pgsql-general mailing list (pgsql-general@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-general
Re: [GENERAL] 9.0 replication -- multiple hot_standby servers
On October 29, 2010, Dean Gibson (DB Administrator) postgre...@ultimeth.com wrote: My question is, how do I configure the other three (still) hot_standby boxes to now use the new primary? Clearly I can change the recovery.conf file on each standby box, but that seems like an unnecessary nuisance. I'm curious about this too. It seems that currently I'd have to rebuild any additional slaves basically from scratch to use the new master. -- A hybrid Escalade is missing the point much in the same way that having a diet soda with your extra large pepperoni pizza is missing the point. -- Sent via pgsql-general mailing list (pgsql-general@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-general
Re: [GENERAL] 9.0 replication -- multiple hot_standby servers
On 2010-10-29 11:17, Alan Hodgson wrote: I'm curious about this too. It seems that currently I'd have to rebuild any additional slaves basically from scratch to use the new master. I think so long as you pointed (via primary_conninfo) the additional slaves to the new (pending) master, before you touched the pending master's trigger file, you should be OK, as all the DBs should be in sync at that point. When the primary DB server goes down, updates are no longer accepted. In such a situation, the human goal is to get the updates accepted again, and human beings in a hurry are apt to forget the exact sequence of steps to follow. If one forgets, and enables the new master without repointing the other slaves, then you have a situation where you have to backup/restore the new primary to each of the slaves, in order to recover the slaves. The failover shouldn't be that brittle. A similar situation exists when having to reboot all the DB servers (say, for maintenance), or just reboot one in a period where you can afford to suspend updates: As near as I can tell (and I will find out over the weekend), the correct procedure is to stop the primary FIRST, and then reboot whatever slaves you need, and then reboot (if needed) or restart the primary. I wonder if this thread should be in the ADMIN group (and if so, should I repost the original message there) ??? -- Mail to my list address MUST be sent via the mailing list. All other mail to my list address will bounce.
Re: [GENERAL] 9.0 replication -- multiple hot_standby servers
On October 29, 2010, Dean Gibson (DB Administrator) postgre...@ultimeth.com wrote: On 2010-10-29 11:17, Alan Hodgson wrote: I'm curious about this too. It seems that currently I'd have to rebuild any additional slaves basically from scratch to use the new master. I think so long as you pointed (via primary_conninfo) the additional slaves to the new (pending) master, before you touched the pending master's trigger file, you should be OK, as all the DBs should be in sync at that point. Yeah they're in sync data-wise, but do they think they're the same WAL stream for continuity? Would be nice. -- A hybrid Escalade is missing the point much in the same way that having a diet soda with your extra large pepperoni pizza is missing the point. -- Sent via pgsql-general mailing list (pgsql-general@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-general