Wietse Venema:
> Viktor Dukhovni:
> > On Thu, Sep 29, 2022 at 05:14:12PM +0200, michae...@rocketmail.com wrote:
> >
> > > systemd's start command is
> > >
> > > ExecStart=/usr/sbin/postfix start
> > >
> > > .. and when trying to "postfix -vvv start" at shell, postfix starts!
> > >
> > > But takes very, very long. So I'll search for increasing timeouts for
> > > systemd service units, next.
> >
> > How many (queue) files are (recursively) in:
> >
> > /var/spool/postfix/{deferred,incoming,active,hold}?
>
> find /var/spool/postfix/{deferred,incoming,active,hold} -type f | wc
>
> > The "postfix start" command executs "postfix-script" which runs mostly
> > cheap commands, but also "postsuper -s" to repair the queue directories.
> > If you have millions of queue files on hold or similar, this could take
> > a while. If your queue directory is in NFS or other slow file system,
> > that could compound the issue.
>
> We tried to run some of the checks in the background at some point.
>
> There was some discussion 17 years ago that Postfix startup was
> affecting system start-up time. So we moved some checks into the
> background:
>
> 20050117
>
> Performance: reduced start-up delay by moving warning-only
> startup checks into the background; they now start after
> one minute to allow the system to finish booting. File:
> conf/postfix-script.
>
> After a few weeks we had to move those checks back into the foreground
> (and we removed the 1-minute delay), because the change had introduced
> new problems.
>
> 20050203
>
> Cleanup: "postfix start" background checks moved back to
> the foreground so they can be stopped more easily. File:
> conf/postfix-script.
>
> Not documented, not even in the HISTORY change log, is the
> quick-start feature that skips the start-up checks.
>
> We could make the 20050117 behavior configurable, but then it
> would have to be on by default, because no-one would turn it on.
Some results from email archive digging:
The system start-up latency concern was brought up by someone from
SUSE. I was unable to get confirmation that my change made things
better, and some other people didn't like it because it would delay
the detection of some errors. All that seems to be the reason move
these Postfix start-up checks back into the foreground.
There was a conflict early on with the patch for multi-instance
support, but that may not have been the reason to move these Postfix
start-up checks back into the foreground.
Wietse
