keep_wal_segments is a minimum setting when replication slots are used, and in the event communications are lost, the number may grow as the slave still needs them. The growth may be such that it could fill up a filesystem, so precautions should be made by monitoring or having a way to redirect those segments to another growth area.
Correct. Always good to have some minimum number of wal files at any time, hence why wal_keep_segments is mention in that section.
On Wed, Sep 7, 2016 at 10:37 AM, John Scalia <jayknowsunix@xxxxxxxxx> wrote:
Just to reiterate what you've explained, my understanding is now that keep_wal_segments is a minimum setting when replication slots are used, and in the event communications are lost, the number may grow as the slave still needs them. The growth may be such that it could fill up a filesystem, so precautions should be made by monitoring or having a way to redirect those segments to another growth area. My original thinking this morning was more likely the keep_wal_segments settings was a maximum not a minimum. Thanks for clearing that up for me.On Wed, Sep 7, 2016 at 9:14 AM, Payal Singh <payal@xxxxxxxxxx> wrote:Then, the second paragraph begins "In lieu of using replication slots... use keep_wal_segments"This is talking about using keep_wal_segments to always have certain minimum number of WALs even if replication is up-to-date and doesn't need those WALs anymore.Also, if the slot is managing the storage of WAL segments, then isn't there a possibility of filling up a disk in the event of an outage?Yes, which is why you should have disk alerts set in place and a quick way to redirect WAL files to another machine, possibly one that stores backups.And would keep_wal_segments prevent that?No. If slave is out, slots will keep all WALs irrespective of what the wal_keep_segments is set to. So in a way, wal_keep_segments value is the lowest minimum number of WALs you will have on master at any moment, and with slots it is possible that number will increase if slave goes down.On Wed, Sep 7, 2016 at 9:12 AM, John Scalia <jayknowsunix@xxxxxxxxx> wrote:Also, if the slot is managing the storage of WAL segments, then isn't there a possibility of filling up a disk in the event of an outage? And would keep_wal_segments prevent that? This whole section just seems to be a little vague to me and like it wasn't updated cleanly since slots were envisioned.Hi all,I've been reading section 25.2.6 Replication Slots in the docs, and I'm just a little confused. This section clearly states that replication slots automates the management of WAL slots so the master will not remove them prior to receipt by all standbys. Then, the second paragraph begins "In lieu of using replication slots... use keep_wal_segments". So, if I have a primary-standby cluster using a replication slot, then what, if anything, should keep_wal_segments be set to? Should I just comment it out?Maybe it's just too early in the morning for me :-)Jay
