On 07/31/2018 02:32 PM, Savvas Karagiannidis wrote:
Hi Nic,
as far as I know there are currently a couple of nasty open
issues when upgrading old mailboxes, created from earlier
versions of cyrus, that have not been upgraded to v13 index
using reconstruct. These issues are:
The main problem is frequent crashes with error message
"assertion failed" when accessing these mailboxes.
In issue #2208 a reconstruct command was proposed as a
workaround:
reconstruct -G -V max [mailboxname]
Note that -G parameter seems necessary, otherwise the
problem persists...
Note also, that this should be done while still in version
2.5.x! The behavior of version 3.x reconstruct is not the same
and you will not have the same results.
As for the cyrdump you mention, it would be better if you
compared the results before running reconstruct and after
that.
I guess what you would see is that the uidlist before
would probably be in the range 6 - 9 (or you'd see a crash).
cyrus 3.0.x considers that part of the index corrupt so simply
cannot access those uids (1-5). Reconstruct detects this issue
and what it does is that reappends those "newly
found" messages, giving them a new uid, so that's what uids
10-14 probably really are. Note that this causes these
messages to appear unread, since they have no flags (not
something you want for your clients' mailboxes).
Bron mentioned something about this bug in the cyrus devel
list a few days ago:
but I didn't see any related commit or anything mentioned
in the above issues, so I'm not sure if there is a fix at the
moment...
[Cross posting to cyrus-devel, as am now dumping core]
Savvas,
Thanks for your response. I think, however, that we've already
tackled the issue of assert failures in reconstruct. If you follow
the entire thread of the message you linked to, Bron's message from
the other day, he was actually writing to me, and his patch,
referenced with a commit, within that message, worked just fine for
me, as addressed in my follow up:
https://lists.andrew.cmu.edu/pipermail/cyrus-devel/2018-July/004315.html
But I can see that you're correct about the cyrdump issue. That's a
real mess. Here's the top of a cyrdump from the old host, running
2.5.10:
onlight@mail:~$ sudo su cyrus -c "/usr/lib/cyrus/bin/cyrdump user.onlight" | head -12
Content-Type: multipart/related; boundary="dump-10426-1533070392-23406623"
--dump-10426-1533070392-23406623
Content-Type: text/xml
IMAP-Dump-Version: 0
<imapdump uniqueid="710a47ca47ebc676">
<mailbox-url>imap://mail.ahwi.us/user.onlight</mailbox-url>
<incremental-uid>0</incremental-uid>
<nextuid>10</nextuid>
<uidlist>1 2 3 4 5 6 7 9 </uidlist>
which comports with the existing mailbox directory:
onlight@mail:~$ sudo ls /var/spool/cyrus/mail/I/user/onlight/
1. 2. 3. 4. 5. 6. 7. 8. 9. cyrus.annotations cyrus.cache cyrus.header cyrus.index Drafts Sent Spam Trash
And here's what it looks like after reconstructing on the new
server:
root@newmail:~# ls /var/spool/cyrus/mail/I/user/onlight/
10. 11. 12. 13. 14. 6. 7. 8. 9. cyrus.annotations cyrus.cache cyrus.header cyrus.index Drafts Sent Spam Trash
But I cannot even run cyrdump on the new mailbox now, I keep dumping
core with a Floating Point error:
root@newmail:~# /usr/lib/cyrus/bin/cyrdump user/onlight
Content-Type: multipart/related; boundary="dump-27702-1533070740-1079351705"
--dump-27702-1533070740-1079351705
Content-Type: text/xml
IMAP-Dump-Version: 0
<imapdump uniqueid="710a47ca47ebc676">
<mailbox-url>imap://newmail.ahwi.us/user.onlight</mailbox-url>
<incremental-uid>0</incremental-uid>
<nextuid>15</nextuid>
<uidlist>6 7 9 10 11 12 13 14 </uidlist>
<flags>
Floating point exception (core dumped)
Here's a backtrace from gdb:
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Core was generated by `/usr/lib/cyrus/bin/cyrdump user/onlight'.
Program terminated with signal SIGFPE, Arithmetic exception.
#0 0x00007fcbc1057955 in hash_lookup (key=key@entry=0x7ffd75ae4610 "systemflags",
table=table@entry=0x7fcbc1894e30 <attrs_by_name>) at lib/hash.c:159
159 lib/hash.c: No such file or directory.
(gdb) bt
#0 0x00007fcbc1057955 in hash_lookup (key=key@entry=0x7ffd75ae4610 "systemflags",
table=table@entry=0x7fcbc1894e30 <attrs_by_name>) at lib/hash.c:159
#1 0x00007fcbc1636822 in search_attr_find (name=<optimized out>) at imap/search_expr.c:2465
#2 0x000055c087d5a694 in systemflag_match (flag=1) at imap/cyrdump.c:149
#3 0x000055c087d5a9dd in dump_me (data="" out>, rock=0x7ffd75ae68e0) at imap/cyrdump.c:209
#4 0x00007fcbc16164c3 in find_cb (rockp=rockp@entry=0x7ffd75ae6820,
key=key@entry=0x7fcbc1aa1a28 <error: Cannot access memory at address 0x7fcbc1aa1a28>, keylen=keylen@entry=12,
data="" <error: Cannot access memory at address 0x7fcbc1aa1a38>, datalen=<optimized out>)
at imap/mboxlist.c:2410
#5 0x00007fcbc131f553 in myforeach (db=0x55c0894b2610, prefix=0x7ffd75ae5fb0 "user.onlight", prefixlen=0,
goodp=0x7fcbc16162c0 <find_p>, cb=0x7fcbc1616390 <find_cb>, rock=0x7ffd75ae6820, tidptr=<optimized out>)
at lib/cyrusdb_skiplist.c:1177
#6 0x00007fcbc16141cd in mboxlist_find_category (rock=rock@entry=0x7ffd75ae6820,
prefix=prefix@entry=0x7ffd75ae5fb0 "user.onlight", len=len@entry=0) at imap/mboxlist.c:2685
#7 0x00007fcbc1614b2c in mboxlist_do_find (rock=rock@entry=0x7ffd75ae6820, patterns=patterns@entry=0x55c0894b26f0)
at imap/mboxlist.c:2877
#8 0x00007fcbc1619fe6 in mboxlist_findallmulti (namespace=<optimized out>, patterns=0x55c0894b26f0, isadmin=<optimized out>,
userid=0x0, auth_state=<optimized out>, proc=<optimized out>, rock=0x7ffd75ae68e0) at imap/mboxlist.c:2942
#9 0x000055c087d5a4bd in main (argc=2, argv=<optimized out>) at imap/cyrdump.c:119
I'll note, too, that the behavior of reconstruct, with regards to
those low numbered UIDs, is totally inconsistent. Here's a re-run
on the same mailbox, after re-rsyncing (with --delete option) from
the original server:
root@newmail:~# su cyrus -c "/usr/lib/cyrus/bin/reconstruct -V max user/onlight"
user.onlight: update uniqueid from header (null) => 710a47ca47ebc676
user.onlight updating quota_mailbox_used: 10715 => 3910
user.onlight: updating exists 8 => 3
user.onlight: updating sync_crc 2250269913 => 1705867986
user/onlight
Repacked user/onlight to version 13
root@newmail:~# ls -l /var/spool/cyrus/mail/I/user/onlight/
total 76
-rw------- 1 cyrus mail 2924 Mar 27 2008 1.
-rw------- 1 cyrus mail 748 Mar 27 2008 2.
-rw------- 1 cyrus mail 804 Jul 15 2008 3.
-rw------- 1 cyrus mail 1137 Oct 2 2009 4.
-rw------- 1 cyrus mail 1192 Apr 22 2012 5.
-rw------- 1 cyrus mail 1423 Sep 19 2016 6.
-rw------- 1 cyrus mail 1743 Dec 22 2016 7.
-rw------- 2 cyrus mail 1357 Jun 1 2017 8.
-rw------- 1 cyrus mail 744 Jun 1 2017 9.
-rw------- 1 cyrus mail 336 Mar 1 2017 cyrus.annotations
-rw------- 1 cyrus mail 10276 Jul 31 15:54 cyrus.cache
-rw------- 1 cyrus mail 165 Feb 28 2017 cyrus.header
-rw------- 1 cyrus mail 1064 Jul 31 16:16 cyrus.index
drwx------ 2 cyrus mail 4096 Nov 30 2017 Drafts
drwx------ 2 cyrus mail 4096 Nov 30 2017 Sent
drwx------ 2 cyrus mail 4096 Nov 30 2017 Spam
drwx------ 2 cyrus mail 4096 Nov 30 2017 Trash
So now it's just ignoring UIDs 1-6, and claiming the mailbox has
just 3 messages.
Color me confused.
Honestly, I am not sure how anyone is managing large upgrades from
2.5 to 3.0? Not being able to reliably bring the existing message
store up to the necessary level is quite unsettling.
Cheers,
-nic
Hope this helps,
Regards,
Savvas Karagiannidis
Friends,
I'm preparing for a couple of belated 2.5.X to 3.0.X
upgrades, and have a question about how necessary it is to
run "reconstruct -V max" on the mailstore. Both systems are
currently running 2.5.10, and are already at index version
13. However, when performing "reconstruct -V max" on one,
on a new 3.0.7 (with patches) system, I see this:
root@newmail:~# /usr/lib/cyrus/bin/reconstruct -V max user/onlight
user.onlight uid 1 rediscovered - appending
user.onlight uid 2 rediscovered - appending
user.onlight uid 3 rediscovered - appending
user.onlight uid 4 rediscovered - appending
user.onlight uid 5 rediscovered - appending
user/onlight
Repacked user/onlight to version 13
The last line can be ignored, as it's really a noop. The
"rediscovered - appending" stuff is what catches my eye.
However, once the reconstruct is complete, here's what the
mailbox looks like:
root@newmail:/var/spool/cyrus/mail/I/user/onlight# /usr/lib/cyrus/bin/cyrdump user/onlight
Content-Type: multipart/related; boundary="dump-27466-1533049817-351841533"
--dump-27466-1533049817-351841533
Content-Type: text/xml
IMAP-Dump-Version: 0
<imapdump uniqueid="710a47ca47ebc676">
<mailbox-url>imap://newmail.example.com/user.onlight</mailbox-url>
<incremental-uid>0</incremental-uid>
<nextuid>15</nextuid>
<uidlist>6 7 9 10 11 12 13 14 </uidlist>
<flags>
...
Note that the <uidlist> doesn't list those low number
UIDs which were listed in the reconstruct sequence. In
other words, I think this all is harmless, but I'm not sure
how much overhead it brings to the whole process.
One of the servers has a total of 70GB of mail, so a
complete reconstruct run only takes a short while. The
other, however, has over 8TB scattered over >30
partitions. If I could avoid running reconstruct across
that whole wad, it'd be great.
Thoughts please?
-nic
--
Nic Bernstein nic@xxxxxxxxxxx
Onlight, Inc. www.onlight.com
6525 W Bluemound Road, Suite 24 v. 414.272.4477
Milwaukee, Wisconsin 53213-4073
Cyrus Home Page: http://www.cyrusimap.org/
List Archives/Info: http://lists.andrew.cmu.edu/pipermail/info-cyrus/
To Unsubscribe:
https://lists.andrew.cmu.edu/mailman/listinfo/info-cyrus
--
Nic Bernstein nic@xxxxxxxxxxx
Onlight, Inc. www.onlight.com
6525 W Bluemound Road, Suite 24 v. 414.272.4477
Milwaukee, Wisconsin 53213-4073
|
