fixes deb#439384
From: martin f krafft
Subject: race condition in Maildir writing
The offlineimap Maildir code checks for file existence and then
opens a file. That's open to a race condition. It's better to open
the file and fail if it already exists. The following patch does
this. It catches OSError 17 (file exists) and re-raises all others.
I'll leave it up to you to decide whether this is appropriate.
fixes deb#433732
Date: Sun, 30 Sep 2007 13:54:56 -0400
From: Daniel Jacobowitz <drow@false.org>
To: offlineimap@complete.org
Subject: Assorted patches
Here's the result of a lazy Sunday hacking on offlineimap. Sorry for
not breaking this into multiple patches. They're mostly logically
independent so just ask if that would make a difference.
First, a new -q (quick) option. The quick option means to only update
folders that seem to have had significant changes. For Maildir, any
change to any message UID or flags is significant, because checking
the flags doesn't add a significant cost. For IMAP, only a change to
the total number of messages or a change in the UID of the most recent
message is significant. This should catch everything except for
flags changes.
The difference in bandwidth is astonishing: a quick sync takes 80K
instead of 5.3MB, and 28 seconds instead of 90.
There's a configuration variable that lets you say every tenth sync
should update flags, but let all the intervening ones be lighter.
Second, a fix to the UID validity problems many people have been
reporting with Courier. As discussed in Debian bug #433732, I changed
the UID validity check to use SELECT unless the server complains that
the folder is read-only. This avoids the Courier bug (see the Debian
log for more details). This won't fix existing validity errors, you
need to remove the local status and validity files by hand and resync.
Third, some speedups in Maildir checking. It's still pretty slow
due to a combination of poor performance in os.listdir (never reads
more than 4K of directory entries at a time) and some semaphore that
leads to lots of futex wake operations, but at least this saves
20% or so of the CPU time running offlineimap on a single folder:
Time with quick refresh and md5 in loop: 4.75s user 0.46s system 12%
cpu 41.751 total
Time with quick refresh and md5 out of loop: 4.38s user 0.50s system
14% cpu 34.799 total
Time using string compare to check folder: 4.11s user 0.47s system 13%
cpu 34.788 total
And fourth, some display fixes for Curses.Blinkenlights. I made
warnings more visible, made the new quick sync message cyan, and
made all not explicitly colored messages grey. That last one was
really bugging me. Any time OfflineIMAP printed a warning in
this UI, it had even odds of coming out black on black!
Anyway, I hope these are useful. I'm happy to revise them if you see
a problem.
--
Daniel Jacobowitz
CodeSourcery
patch from Mike Gerber
Two times today I have found my offlineimap to have died with this same
situation. It appears as if certain types of messages (both spam in my
situation), cause offlineimap to choke. When it does it cannot proceed.
This means that when I run offlineimap, it pulls in messages from some
folders, then it hits the folder with the bad message and dies, leaving
undownloaded mail on the server. The only fix to this problem is to find
the problem message on the server and remove it by hand. This isn't such
a huge deal for me, since I run the server, but other people have to
come to me to ask me to delete these messages, and until I do they
cannot download their email.
I have captured the output by running script during one of these
incidents, this has been attached. Additionally, I have also attach the
problematic message.
The patch seems to work for me, might need some Python wizard and better
testing, though.
fixes deb#396443
r[1] is a list. In case it contains more than one 'str' they are concatenated
using '. '. In processmessagesflags the join is tested, for savemessagesflags I
assume that it is also needed.
I have tested this and Dovecot no longer beats offlineimap
to the punch. (-:
I achieved that by keeping the renames in tmp/ until it finally does
one last rename in new/ or cur/.
fsync the Maildir file, its final directory when writing a new message.
fsync the localstatus file and its final directory when writing the
local status cache.
This should reduce duplication in the event of hardware trouble.
fixes#8
see thread at http://lists.complete.org/offlineimap@complete.org/2007/03/threads.html.gz
The attached patch adds syncing the INTERNALDATE of IMAP folders with
the mtime of messages in maildir folders.
I want this to happen, because I'm running a dovecot over the maildirs
synced by offlineimap, and that uses the mtime as the INTERNALDATE.
When using mutt to view messages I generally sort based on the received
date, which for IMAP folders is the INTERNALDATE.
Since this is the first real coding I've done in Python the patch may
need to be cleaned up some, but it's working pretty well for me. I've
added new messages to each side, and the received date has been
preserved going both ways.
Patch from Nikita V. Youshchenko
From: "Nikita V. Youshchenko"
To: Debian Bug Tracking System
Subject: offlineimap: exception on mail with broken headers (+fix)
Date: Wed, 24 Aug 2005 13:41:08 +0400
Package: offlineimap
Version: 4.0.10
Severity: normal
Tags: patch
Recently I've got an exception (see below) while using offlineimap.
Exception was probably caused by invalid Date header of (likely spam)
message:
Date: Sat, 20 Aug 2005 4294967295:43:18 -0700
From: "Nikita V. Youshchenko"
I trued to use offlineimap and found that while being quite fast on
small folders, it takes up to several minutes (of 100% busy CPU and
almost no network traffic) to sync a folder with 2000+ messages.
While looking into the code, I found why this happens.
In folder/Base.py, in method BaseFolder.syncmessagesto_copy(),
dest.getmessagelist() is called inside a loop, while being a loop
invariant. Similar thing happens in BaseFolder.syncmessagesto_delete()
for self.getmessagelist().
This causes quadratic complexity over folder size.
Moving these calls out of loops make large folder sync fast (several
seconds instead of several minutes for folder with 2000 messages on
700MHz P3).
