Standby pg_dump Conflict with Recovery

[Louis Battuello <louis.battuello@xxxxxxxxxxx>]

Hell All,

I’ve got a confusing issue with dumping data from a standby PostgreSQL 9.4.5 database.

At night, on a nearly completely idle server, I run a pg_dump of a database that contains numerous small tables and one 3GB table. The dump consistently fails when reaching the 3GB table with this message:

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 <table> (...) TO stdout;

I have replication slots enabled on the primary (“repmgr_slot_3" for the standby pg_dump source), and I’m using hot_standby_feedback. After getting the failure a couple times, I temporarily set max_standby_archive_delay and max_standby_streaming_delay to -1 to allow infinite delay on the standby,  just to see if I could get the dump to complete. I still encountered the above error.

postgres=# select * from pg_replication_slots ;

   slot_name   | plugin | slot_type | datoid | database | active |  xmin   | catalog_xmin | restart_lsn 

---------------+--------+-----------+--------+----------+--------+---------+--------------+-------------

 repmgr_slot_2 |        | physical  |        |          | t      |         |              | A/C6502880

 repmgr_slot_3 |        | physical  |        |          | t      | 1356283 |              | A/C6502880

(2 rows)

Is there some other configuration setting I’m forgetting?

Thanks,

Louis