Wietse Venema via Postfix-users wrote in
 <4rvkgl4lfyzj...@spike.porcupine.org>:
 |Steffen Nurpmeso via Postfix-users:
 |> "But" postfix's sendmail reads the postfix configuration, it will
 |> not work otherwise, at least once i tried last.
 |> Going over SMTP (submission that is) can share a single postfix
 |> instance in between many containers that do not have access to
 |> the actual configuration, filesystem-wise.
 |> 
 |> Often i have wished postfix's sendmail(1) would work otherwise;
 |> i also used dma (DragonFly Mail Agent) as is now used by default
 |> on FreeBSD for that purpose (which was quite messy installation-
 |> order wise), ie, simple sendmail that relays to real postfix.
 |> 
 |> Yes, a compile-time configuration option, and a postdrop that
 |> listens via UNIX domain socket on a path that can be mounted into
 |> containers, and a postfix sendmail that only works on that path,
 |> that would be a nice thing container-wise, wouldn't it?
 |> Just an idea...
 |
 |Alls you need is to 'mount' the maildtop directory into a container
 |with read/write permission, and install the Postfix sendmail and
 |postdrop programs insalled in the container. As long as there is a
 |Postfix pickup daemon running somewhere, it will pick up new messages
 |in at most 60 seconds.

"The problem" (i have given up and did not try it for long) is the
configuration directory.  Does this work without configuration
directory?  I had to try again.

So last i tried.
If you do not compile custom, but still want a custom
configuration (directory), you need command line options. 

Yes, one could write a shell wrapper around /usr/sbin/sendmail
which applies the necessary hints (and needs to be reinstalled as
a replacement after each update).  But this works only for root.
Unless the path all through to the configuration etc is +[r]x for
all users -- and this counteracts the host-specific configuration
as i it use for the other daemons i drive: these live under
/root/hosts/$HOSTNAME/.

  Actually, you know, this is all git(1) driven; some branches
  (bin, linux.kconfig, ..., $HOSTNAME), and then you join only
  those branches into $HOSTNAME that are needed, and then you "git
  push $HOSTNAME" to that host.  Each host only has its own
  branch.  On each $HOSTNAME you can use bin/backup.sh (which
  reads /root/hosts/$HOSTNAME/backup) to backup some host-specific
  files (mostly resolv.conf, hosts, and whatever is only there),
  commit that, and then you can "git fetch" that back to the
  "super server" which has all the git branches.  This sounds
  messy but does not require any external software like puppet or
  how it is called and such (Ansible?), and for hosts with
  identical hardware you effectively can share branches with just
  a few lines of "git diff HOST1 HOST2".

 |Originally. Postfix was designed to support NFS file sharing, where
 |the maildrop direcfory is shared read/write, the server runs a full
 |Postfix install, and the clients have Postfix sendmil and postdrop
 |programs.

You mean, clients have a minimal configuration just to find the
maildrop directory, and nothing else?  This is sufficient?

--steffen
|
|Der Kragenbaer,                The moon bear,
|der holt sich munter           he cheerfully and one by one
|einen nach dem anderen runter  wa.ks himself off
|(By Robert Gernhardt)
_______________________________________________
Postfix-users mailing list -- postfix-users@postfix.org
To unsubscribe send an email to postfix-users-le...@postfix.org

Reply via email to