sendmail starts too early?

[Gregory Neil Shapiro]

On Sun, Aug 08, 2004 at 09:53:44AM +0400, Andrey Chernov wrote:
> There is a problem using sendmail with milters, which usually lives in 
> /usr/local/* and starts from /usr/local/etc/rc.d/*
> 
> Currently, sendmail starts before them, which produce this example 
> diagnostic:
> 
> Starting sendmail.
> WARNING: Xsentinel: local socket name /var/spool/sentinel/sentinel.socket missing
> WARNING: Xantivir-milter: local socket name /var/spool/avmilter/avmilter.sock missing
> 
> When they starts afterwards, situation normalizes, but there is a gap 
> between, where whole system can reject or do something potentially wrong 
> with messages.
> 
> Since converting each milter startup script under rc.subr technology is
> painful task, what about starting sendmail _after_ /etc/rc.d/localpkg 
> starts? I see no drawback in that case.

I personally don't see a problem in that unless some of the
/usr/local/etc/rc.d/ scripts require an MTA to be running by the time
they start.

If things are going to change in this area, I would also suggest moving
virecover to after sendmail starts (currently it is before) as that
script's purpose is to mail users telling them that recovery files
exist.

Likewise, the resolv and nsswitch scripts should run before sendmail
starts (currently they are run after) since they will both impact
sendmail's operation.

This should however be passed by the rc.d folks as well (freebsd-rc
Cc'ed).