Hello,
In our experience the initial dump of the tables we wanted to publish
took >1 hr which was the idle_in_transaction_session_timeout initial
value. It is our understanding that while the initial dump of tables
is happening, the wal is not consumed, so we think the value of this
setting (for the postgres user that initiates the replication) has to
be higher than the length of the original dump. Otherwise *it seemed*
the slot was reclaimed right after the initial dump.
Now, it could very well might have been other factors but setting:
>ALTER USER debezium SET idle_in_transaction_session_timeout TO 14400000; -- 4 hrs
seemed to have removed the disconnects that happened soon after the
initial dump.
>We certainly could have the connector automatically restart if it encounters this particular exception (see the PostgresErrorHandler class)
I think is fine having the connector work as it does now (re-starting
itself after retriable exceptions) but if others run into this issue
it might be worth documenting as a setting worth having in RDS.
Thanks,
Nuria
P.S a helpful post:
https://www.cybertec-postgresql.com/en/idle_in_transaction_session_timeout-terminating-idle-transactions-in-postgresql/
> --
> You received this message because you are subscribed to a topic in the Google Groups "debezium" group.
> To unsubscribe from this topic, visit
https://groups.google.com/d/topic/debezium/ZY3E0t_2FLo/unsubscribe.
> To unsubscribe from this group and all its topics, send an email to
debezium+u...@googlegroups.com.
> To view this discussion on the web visit
https://groups.google.com/d/msgid/debezium/efe3080b-1512-4fff-ad6d-d12976da90f4n%40googlegroups.com.