I'm seeing long-running transactions (pg_dump) canceled on the standby when there are a lot of inserts happening on the master. This despite my having set max_standby_streaming_delay to -1 on the standby.
Do you have hot_standby_feedback set to "on" ?
What is the parameter max_standby_archive_delay configured to ? This will pause WAL archives from being applied when queries are executed on the standby database.
Why might that happen?
This is pg 9.3.12. When it happens I see:
pg_dump: Dumping the contents of table "TABLE" failed: PQgetResult() failed. pg_dump: Error message from server: ERROR: canceling statement due to conflict with recovery DETAIL: User query might have needed to see row versions that must be removed. pg_dump: The command was: COPY public.TABLE (COLUMNS) TO stdout;
I suspect this is due to the clean up by VACUUM on primary.
