Viktor Dukhovni:
> On Mon, Jan 21, 2019 at 07:04:56PM -0500, Wietse Venema wrote:
>
> > postfix-3.4-20190121-nonprod-logger has lightly-tested code for
> > logging to file without using syslogd.
>
> I see that the postlogd(8) service correctly uses a unix-dgram
> service to avoid message reordering, so the usual attention to
> detail has not deserted you. :-) Thanks!
A lot of attention to detail went into this, but I have no control
over how the kernel's unix-domain datagram implementation buffers
messages until the postlogd service is ready. That's why messages
are time-stamped by the sender, not receiver.
> For log rotation instructions, I would add that the old file should
> not be modified, compressed or otherwise assumed complete, until a
> new file is observed to have been written by postlogd(8) after
> processing the "reload" command. If one is not sure whether Postfix
> is running, one could use postlog(1) (as root) to cause a new file
> to be written, one way or the other.
The Postfix master immediately signals postlogd to terminate. This
being a single-instance event-driven service, it should terminate
immediately unless it is blocked on file writes.
> The only other nit that comes to mind, is that some syslog servers
> needlessly delete and re-create their unix-dgram sockets on reload,
> which leads to needless opportunities for brief loss of messages
> written to the previous incarnation of the socket.
The Postfix master does not close a socket, unless the service is
removed from master.cf, or master_service_disable is in effect.
Wietse
