Re: Logical replication monitoring

On 13/08/2018 13:16, AYahorau(at)ibagroup(dot)eu wrote:
> Hello!
>
> Thank you for your suggestion.
> I  afraid this approach is not suitable for me. As a rule my postgresql log  on subscriber side contains a  bunch of the following entries:
>
> *ERROR:  terminating logical replication worker due to timeout*
> *00000 LOG:  worker process: logical replication worker for subscription 24578 (PID 6217) exited with exit code 1*
> *
> *How should I handle this situation? *
> *As I understand this is quite normal situation. But why is severity for it  an ERROR ?
In my short experience with logical replication, those ERRORs should be avoided, you better increase the relevant timeouts: wal_receiver_timeout , wal_sender_timeout . One hand tool we use here is
tail_n_mail , it is tailored specifically for parsing PostgreSQL logs and sending emails with alerts.
>
> I have another assumption. Could you correct me if I am wrong.
> I found out in the source code that logical replication worker termination depends on *wal_receiver_timeout *paramer.
> So I propose setting *wal_receiver_timeout *to 0.
> In this case I think that monitoring of the following views *pg_stat_subscription, pg_publication *and *pg_stat_replication * is enough.
> In case if there  is some problem with connection or with replication *pg_stat_replication * will show nothing because wal sender will not be working otherwise it will give some information.
> Am I right? Are there any vulnerabilities in this approach ?
>
> Best regards,
> Andrei Yahorau
>
>
>
> From: Andrei Yahorau/IBA
> To: pgsql-admin(at)postgresql(dot)org,
> Cc: Mikalai Keida/IBA(at)IBA
> Date: 10/08/2018 13:05
> Subject: Logical replication monitoring
> --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
>
>
> Hello PostgreSQL Community!
>
> I configured logical replication for PostgreSQL 10.4 on 2 machines, set wal_level to logical, created a publication on master node and created a subscription on standby node according to the
> PostgreSQL documentation.
> Could you please suggest an approach for replication state monitoring.
>
> According to my experience the monitoring of *pg_stat_subscription *and *pg_publication, pg_replication_slots *unfortunately is not enough for this aim. Moreover standby database does not prohibit
> write operations by default and it can lead to some inconsistency between these databases.
>
> For example a chain of queries as
> *SELECT pg_is_is_recovery()*,
> *SELECT * FROM pg_stat_replication* and
> *SELECT * FROM pg_stat_wal_receiver*
> provide insight into replication state for hot_standby replication.
>
> So is there a reliable way of replication state monitoring for logical replication?
>
> Best regards,
> Andrei Yahorau

--
Achilleas Mantzios
IT DEV Lead
IT DEPT
Dynacom Tankers Mgmt