Обсуждение: Queue order of backends waiting on WALWriteLock

Hello,


I am not sure if pgsql-novice is place for this but this sounds like a
basic locking understanding question so here:

While studying group commit implementation (commit_delay=0 in this
case) in PostgreSQL, I am having some trouble understanding how
sometimes a particular backend with earlier WAL flush location has to
wait for WALWriteLock.

In XLogFlush()

when a backend waits on WALWriteLock by a call to
LWLockAcquireOrWait(), is it a case that it is queued in the
increasing order of flush location.

For example, if 3 backends with flush location A, B, C are waiting on
WALWriteLock with A < B < C. Is it possible for backend 2 (with B) to
get lock earlier than backend 1, causing 1 to have to wait.

How does it happen that a backend with flush location which is greater
than some of the waiting backends gets WALWriteLock earlier thus
causing the flush/fsync piggybacking to happen?

--

Amit Langote