Tom Lane wrote:
> Jeff Mcdowell <Jeff.Mcdowell@panerabread.com> writes:
> > 95% of the time, the delay is only microseconds. But we have discovered that whenever the master does an auto
vacuumof a large table, the transaction replay delay can climb is high as 1 hour. These delays don�t seem to correlate
withany particular queries that are running against the master or the standby, and the delay only subsides when the
vacuumcompletes.
>
> What PG version might this be?
This sounds related to
commit 3e4b7d87988f0835f137f15f5c1a40598dd21f3d
Author: Simon Riggs <simon@2ndQuadrant.com>
AuthorDate: Sun Apr 3 17:46:09 2016 +0100
CommitDate: Sun Apr 3 17:46:09 2016 +0100
Avoid pin scan for replay of XLOG_BTREE_VACUUM in all cases
--
Álvaro Herrera http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services