Обсуждение: Re: Keycloak and Postgres
On Thu, 30 Mar 2017 13:58:36 +0000 Marc Tempelmeier <marc.tempelmeier@flane.de> wrote: > Hi, > > I have a replication question, we have some big Cisco UCS VM thingy, where VMs are snapshotted, the drives are abstractedetc. If a VM crashes it will be resumed in 1 min from another rack. What brings us master slave replication orsome other kind of replication in this setup? Should we do it because of other failures? Because of how Postgres caches changes, you may find that a failover requires some time in recovery mode. Those VM snapshot systems are great, but they aren't quite perfect if they don't know what is being done with the data on the drives. Whether it's good enough depends heavily on what your expectation is. Before trusting it to meet your needs, I would spend some time simulating failures and seeing what actually happens. -- Bill Moran <wmoran@potentialtech.com>
Hi, Thanks for your answer! Can you elaborate a bit on this part: " Because of how Postgres caches changes, you may find that a failover requires some time in recovery mode." Thanks! -----Ursprüngliche Nachricht----- Von: Bill Moran [mailto:wmoran@potentialtech.com] Gesendet: Saturday, April 1, 2017 12:57 PM An: Marc Tempelmeier <marc.tempelmeier@flane.de> Cc: pgsql-general@postgresql.org Betreff: Re: [GENERAL] Keycloak and Postgres On Thu, 30 Mar 2017 13:58:36 +0000 Marc Tempelmeier <marc.tempelmeier@flane.de> wrote: > Hi, > > I have a replication question, we have some big Cisco UCS VM thingy, where VMs are snapshotted, the drives are abstractedetc. If a VM crashes it will be resumed in 1 min from another rack. What brings us master slave replication orsome other kind of replication in this setup? Should we do it because of other failures? Because of how Postgres caches changes, you may find that a failover requires some time in recovery mode. Those VM snapshotsystems are great, but they aren't quite perfect if they don't know what is being done with the data on the drives. Whether it's good enough depends heavily on what your expectation is. Before trusting it to meet your needs, I would spend some time simulating failures and seeing what actually happens. -- Bill Moran <wmoran@potentialtech.com>
On Wed, 5 Apr 2017 07:24:32 +0000 Marc Tempelmeier <marc.tempelmeier@flane.de> wrote: > > Can you elaborate a bit on this part: > " Because of how Postgres caches changes, you may find that a failover requires some time in recovery mode." https://www.postgresql.org/docs/9.6/static/wal-intro.html The WAL requires that any unexpected shutdown of Postgres (where it doesn't get to explicitly flush data pages to disk) go through a recovery cycle to fix anything in the WAL that is not yet in the data pages. Doing disk level replication and using that as a failover essentially duplicates a crash on the PostgreSQL end when you failover. -- Bill Moran <wmoran@potentialtech.com>