Postgres-operator: Is it possible upgrade posgtresql from 9.x to 10.x with patroni k8s operator with near-zero downtime?

[Slach]

i search solution for upgrade databases from 9.6 to 10.0 without or near-zero downtime

i found suggestions like a https://blog.2ndquadrant.com/tag/pglupgrade/, but don't found any source code for this solution, could patroni kubernetes operator help me with this solution?

[sdudoladov]

Hello @Slach

i search solution for upgrade databases from 9.6 to 10.0 without or near-zero downtime

It is possible if you use a recent Spilo image (relevant Spilo PR). Our OSS image already has it.
How much downtime you will have depends on your particular setup.

the support/docs from the operator will follow in the upcoming months

Please note that option is brand new and currently undergoes testing. @CyberDem0n may provide your with more information

[CyberDem0n]

Neither Patroni nor postgres-operator will do in place pg_upgrade for you.

What @sdudoladov is referring to is a major upgrade after creating a new cluster from basebackup.
Downtime in this case is significant and depends on the size of your cluster.

[ReSearchITEng]

@CyberDem0n , can you please confirm:
while it is possible to modify the version under the postgresql crd, the upgrade won't take place, right?

The operator shows:

time="2019-02-27T13:07:36Z" level=warning msg="postgresql version change(\"9.6\" -> \"10\") has no effect" cluster-name=default/mycluster2 pkg=cluster worker=3

but unfortunately the cluster gets destabilized, trying to do something it cannot:

2019-02-27 13:10:11 UTC [83]: [1-1] 5c77ddb3.53 0     FATAL:  database files are incompatible with server
2019-02-27 13:10:11 UTC [83]: [2-1] 5c77ddb3.53 0     DETAIL:  The data directory was initialized by PostgreSQL version 9.6, which is not compatible with this version 10.6 (Ubuntu 10.6-1.pgdg18.04+1).
/var/run/postgresql:5432 - no response
2019-02-27 13:10:11,982 INFO reaped unknown pid 83
2019-02-27 13:10:12,983 ERROR: postmaster is not running
2019-02-27 13:10:12,999 INFO: Lock owner: mycluster2-0; I am mycluster2-1
2019-02-27 13:10:12,999 INFO: failed to start postgres
2019-02-27 13:10:21,670 WARNING: Postgresql is not running.

[CyberDem0n]

It will not do an in-place major upgrade.

[ReSearchITEng]

Thanks for confirmation on major versions change @CyberDem0n .
While the version is actually the folder name inside the spilo image, and the spilo image does not have minor versions for either 10 or 11, I understand that maximum possible are upgrades are between 9.3->9.6 minor versions (unless you confirm those are not supported either).

[Jan-M]

Nope, 9.3 to 9.6 is a Postgres Major version upgrade.

We do major version upgrades currently as part of clone/restore. So testable and usable and then for most apps this is fine as in and out is buffered.

[dedsm]

My current strategy launches a secondary cluster with the updated version cloning the one to be updated, it seems to work but haven't done it in production

[FxKu]

Questions seems to be answered. Feel free to reopen.