Hi,
I'm not a member of this list, so please CC me on responses and discussion.

After a system crash PostgreSQL startup is slow as the database recovers.  So the db_connect() call from pg_autovacuum terminates as soon as it tries to connect to "template1".

Looking at the README file, I find this note:
    pg_autovacuum does not get started automatically by either the
    postmaster or by pg_ctl.  Similarly, when the postmaster exits, no one
    tells pg_autovacuum.  The result of that is that at the start of the
    next loop, pg_autovacuum will fail to connect to the server and
    exit().  Any time it fails to connect pg_autovacuum exit()s.

So the failure we're experiencing is an unintended result of an intended solution.   Any suggestions on how I can work-around this problem?

Would it make sense to put the first db_connect() call in the init_db_list() routine inside a [configurable repeatition] loop, sleeping after disappointed attempt to connect, and breaking out on success?   That way, I think, when pg_autovacuum is initiated, we assume the postmaster is up, but when the VacuumLoop connection fails, we assume the postmaster went away, and take our exit().

Thanks,
Jonathan