All three scripts have now been changed to always quickly queue their

messages to the qfiles directory.  This once and for all avoids the
possibility that we hit the MTA's command time limit.  The mailing
list objects are never locked so we can't time out there.  They don't
need to be locked for message queuing.

The penalty is that we do more disk i/o for every message destined to
the list, the list-owner or list-request, and messages are not
delivered immediately.  Both are probably worthy tradeoffs for
absolutely guaranteeing that messages never get lost.

1 files changed, 13 insertions, 18 deletions

diff --git a/scripts/mailcmd b/scripts/mailcmd
index 979030d4a..4201e18c8 100755
--- a/scripts/mailcmd
+++ b/scripts/mailcmd
@@ -28,12 +28,9 @@ Errors are redirected to logs/errors.
 import sys
 
 import paths
-from Mailman import mm_cfg
 from Mailman import MailList
-from Mailman import Utils
 from Mailman import Message
 from Mailman import Errors
-from Mailman.pythonlib.StringIO import StringIO
 from Mailman.Logging.Utils import LogStdErr
 
 LogStdErr("error", "mailcmd")
@@ -52,21 +49,19 @@ def main():
         sys.stderr.write('Mailman error: mailcmd got bad listname: %s\n%s' %
                          (listname, e))
         sys.exit(1)
-    # Try to obtain the mailing list lock.  If that fails, enqueue the message
-    # for delivery by the qrunner cron job.
-    try:
-        mlist.Lock(timeout=mm_cfg.LIST_LOCK_TIMEOUT)
-    except LockFile.TimeOutError:
-        msg.Enqueue(mlist)
-        return
-    try:
-        s = StringIO(sys.stdin.read())
-        msg = Message.Message(s)
-        # Moved the setting of 'torequest' into MailCommandHandler
-        mlist.ParseMailCommands(msg)
-    finally:
-        mlist.Save()
-        mlist.Unlock()
+    # Create the message object
+    msg = Message.Message(sys.stdin)
+    # Immediately queue the message for disposition by qrunner, most likely in
+    # about a minute from now.  The advantage to this approach is that
+    # messages should never get lost -- some MTAs have a hard limit to the
+    # time a filter prog can run.  Postfix is a good example; if the limit is
+    # hit, the proc is SIGKILL'd giving us no chance to save the message.  It
+    # could take a long time to acquire the lock.  This way we're fairly safe
+    # against catastrophe at the expense of more disk I/O.
+    #
+    # The `torequest' flag is a message to qrunner that an alternative route
+    # should be taken for this message.
+    msg.Enqueue(mlist, torequest=1)