[feature] Bounce stalled emails after a timeout.

* Resolves: #7998

3 files changed, 41 insertions, 3 deletions

diff --git a/README.md b/README.md
index d0b5018..c78b14b 100644
--- a/README.md
+++ b/README.md
@@ -33,6 +33,14 @@ $ python setup.py install
 $ twistd -ny pkg/mx.tac
 ~~~
 
+### Stalled emails
+
+In case of problems with couchdb and other unknown sources emails can get
+stalled in the spool. There is a bouncing mechanism for long stalled emails,
+after 5 days the email will get bounced. The timestamp of stalled emails is
+hold in memory, restarting leap-mx will erase all timestamps and the stalled
+timeout will be reset.
+
 ## Hacking
 
 Please see the doc/DESIGN docs.
diff --git a/changes/next-changelog.txt b/changes/next-changelog.txt
index f34adbd..65ed19c 100644
--- a/changes/next-changelog.txt
+++ b/changes/next-changelog.txt
@@ -12,6 +12,8 @@ Features
 ~~~~~~~~
 - `#4285 <https://leap.se/code/issues/4285>`_: Add postfix lookup against couchdb for client smtp fingerprint
 - `#5959 <https://leap.se/code/issues/5959>`_: Make alias resolver to return *uuid@deliver.local*
+- `#7998 <https://leap.se/code/issues/7998>`_: Bounce stalled emails after a timeout.
+
 - `#1234 <https://leap.se/code/issues/1234>`_: Description of the new feature corresponding with issue #1234.
 - New feature without related issue number.
 
diff --git a/src/leap/mx/mail_receiver.py b/src/leap/mx/mail_receiver.py
index 69079be..7c5a368 100644
--- a/src/leap/mx/mail_receiver.py
+++ b/src/leap/mx/mail_receiver.py
@@ -40,6 +40,7 @@ import signal
 import json
 import email.utils
 
+from datetime import datetime, timedelta
 from email import message_from_string
 
 from twisted.application.service import Service, IService
@@ -75,6 +76,11 @@ class MailReceiver(Service):
     """
     RETRY_DIR_WATCH_DELAY = 60 * 5  # 5 minutes
 
+    """
+    Time delta to keep stalled emails
+    """
+    MAX_BOUNCE_DELTA = timedelta(days=5)
+
     def __init__(self, users_cdb, directories, bounce_from,
                  bounce_subject):
         """
@@ -98,6 +104,7 @@ class MailReceiver(Service):
         self._directories = directories
         self._bounce_from = bounce_from
         self._bounce_subject = bounce_subject
+        self._bounce_timestamp = {}
         self._processing_skipped = False
 
     def startService(self):
@@ -378,10 +385,31 @@ class MailReceiver(Service):
                 defer.returnValue(None)
 
             log.msg("Encrypting message to %s's pubkey" % (uuid,))
-            doc = yield self._encrypt_message(pubkey, mail_data)
+            try:
+                doc = yield self._encrypt_message(pubkey, mail_data)
+
+                yield self._export_message(uuid, doc)
+                yield self._remove(filepath)
+            except Exception as e:
+                yield self._bounce_with_timeout(filepath, msg, e)
+
+    @defer.inlineCallbacks
+    def _bounce_with_timeout(self, filepath, msg, error):
+        if filepath not in self._bounce_timestamp:
+            self._bounce_timestamp[filepath] = datetime.now()
+            log.msg("New stalled email {0!r}: {1!r}".format(filepath, error))
+            defer.returnValue(None)
 
-            yield self._export_message(uuid, doc)
-            yield self._remove(filepath)
+        current_delta = datetime.now() - self._bounce_timestamp[filepath]
+        if current_delta > self.MAX_BOUNCE_DELTA:
+            log.msg("Bouncing stalled email {0!r}: {1!r}"
+                    .format(filepath, error))
+            bounce_reason = "There was a problem in the server and the " \
+                            "email could not be delivered."
+            yield self._bounce_message(msg, filepath, bounce_reason)
+        else:
+            log.msg("Still stalled email {0!r} for the last {1}: {2!r}"
+                    .format(filepath, str(current_delta), error))
 
     @defer.inlineCallbacks
     def _process_incoming_email(self, otherself, filepath, mask):