954789 - unable to establish any network connection until a restart of the application

Status: RESOLVED FIXED

(Chat Core :: Eventloop, defect)

Description

[Instantbot]

*** Original post on bio 1355 by Peter SZTAN <sztanpet AT gmail.com> at 2012-03-28 08:30:00 UTC ***

version 1.2a1pre (20120326041658)

after about a day of uptime (probably less, but I always notice after waking up)
MSN stops working, gets disconnected and reconnecting just goes for hours and the status in the accounts dialogue says something about a handshake. After this I try hitting disconnect and then reconnect, now the status says connecting for hours.

Checking for updates also stops working, the dialogue simply says Looking for newer versions of Instantbird... and nothing happens after ten minutes.
Google talk stays connected and working properly.

After restarting Instantbird, everything works properly again.

Comment 1

[Instantbot]

*** Original post on bio 1355 by Peter SZTAN <sztanpet AT gmail.com> at 2012-03-30 09:34:35 UTC ***

I have captured a process monitor (kind of like strace for windows i guess) session of instantbird if anyone is interested: http://dl.dropbox.com/u/134488/instantbird.zip
The session starts with me going to bed and leaving it there for the night, coming back at 9:11:06 (msn disconnecting at 6:20)
at 9:11:25 i try reconnecting to msn, nothing happens in the ui itself, and i try checking for updates soon after too but nothing shows up.
after this i restart instantbird and everything is working again as it should

don't know if its useful or not, but I thought I would try it.

Comment 2

[Instantbot]

Attachment: screenshot of the error console for version 1.2a1pre (20120409042941)

*** Original post on bio 1355 as attmnt 1328 by sztanpet AT gmail.com at 2012-04-11 05:31:00 UTC was without comment, so any subsequent comment numbers will be shifted ***

Comment 3

[Instantbot]

*** Original post on bio 1355 by Peter SZTAN <sztanpet AT gmail.com> at 2012-04-11 05:34:29 UTC ***

the problem is still present with version 1.2a1pre (20120409042941), with the difference that MSN now hangs with "Connecting..." and I cannot reconnect.
Seemingly it does not reset after an error?
The error console wrote the following messages right after disconnecting:

Timestamp: 2012.04.10. 23:55:19
Error: servconn 000 read error, len: -1, errno: 10054, error: Remote host closed connection.

Source File: http://hg.instantbird.org/instantbird/raw-file/ceb11e06a1fc/purple/libpurple/protocols/msn/servconn.c
Line: 433
Source Code:
msn: read_cb

-----

Timestamp: 2012.04.10. 23:55:19
Error: Connection error from Notification server (65.54.61.178): Reading error

Source File: http://hg.instantbird.org/instantbird/raw-file/ceb11e06a1fc/purple/libpurple/protocols/msn/servconn.c
Line: 154
Source Code:
msn: msn_servconn_got_error

Comment 4

[Florian Quèze [:florian]]

*** Original post on bio 1355 at 2012-04-11 09:37:41 UTC ***

Do you have a firewall installed or some similar software that could mess with network connections?

If not, the only other idea I have for something that could cause this behavior is one of the protocol plugin in use opening lots of sockets and not closing them, thus exhausting the number of sockets a process is allowed to have.

Comment 5

[Instantbot]

*** Original post on bio 1355 by Peter SZTAN <sztanpet AT gmail.com> at 2012-04-11 09:53:51 UTC ***

no I don't have any firewall or other software that might interfere
The number of sockets open for instantbird wasn't high when I looked at it with TCPView, but will check as soon as it happens again

Comment 6

[Instantbot]

Attachment: TCPView session of bug for version 1.2a1pre (20120411041356)

*** Original post on bio 1355 as attmnt 1331 by sztanpet AT gmail.com at 2012-04-12 07:25:00 UTC was without comment, so any subsequent comment numbers will be shifted ***

Comment 7

[Instantbot]

Attachment: TCPView session for normal operation for version 1.2a1pre (20120412041655)

*** Original post on bio 1355 as attmnt 1332 by sztanpet AT gmail.com at 2012-04-12 07:34:00 UTC ***

just as a reference, to see how things look when everything is working as expected

Comment 8

[Florian Quèze [:florian]]

*** Original post on bio 1355 at 2012-05-03 23:03:08 UTC ***

Although I don't have good steps to reproduce, I've seen (and been annoyed by) this several times.

I'm marking this as 1.2-blocking for now as I think it may be a regression from bug 954777 (bio 1344) and I think we need to investigate, but it's getting a clearer idea of what's going on that's blocking, not actually fixing this if it turns out to be complicated.

Comment 9

[Instantbot]

*** Original post on bio 1355 by Peter SZTAN <sztanpet AT gmail.com> at 2012-05-04 04:58:29 UTC ***

just mentioning that google talk continues to work properly for me (and can reconnect also)

Comment 10

[Benedikt Pfeifer]

*** Original post on bio 1355 at 2012-05-05 17:08:58 UTC ***

I saw the problem mentioned in comment 3, i.e. that accounts are stuck with "Connecting...". It happened with all accounts that I tried to connect at this moment (Facebook, StudiVZ (XMPP based); Yahoo!) while already connected accounts (IRC, ICQ) seemed to work.

Putting the computer to standby and waking it up again made IB recover from this without a 'real restart'. Maybe this information can help somehow?

Comment 11

[Florian Quèze [:florian]]

*** Original post on bio 1355 at 2012-05-06 08:26:20 UTC ***

(In reply to comment #10)

> Putting the computer to standby and waking it up again made IB recover from
> this without a 'real restart'. Maybe this information can help somehow?

This tends to confirm my suspicion that something is messing up the socket transport service. When you put your computer to sleep, necko is switched to offline, which has the effect of killing the network thread and clearing all pending sockets that were being watched.

Comment 12

[Benedikt Pfeifer]

*** Original post on bio 1355 at 2012-05-08 11:07:11 UTC ***

The problem occured again and this time I tried the following on the error console:

var ioService = Components.classes["@mozilla.org/network/io-service;1"].getService(Components.interfaces.nsIIOService); ioService.offline = true;

followed by 

ioService.offline = false;

It made all my accounts disconnect and even the stuck ones reconnected after the second command.

Comment 13

[aleth [:aleth]]

*** Original post on bio 1355 at 2012-05-16 17:54:46 UTC ***

Long shot: related to bug 953778 (bio 335)?

Comment 14

[Florian Quèze [:florian]]

Attachment: Patch

*** Original post on bio 1355 as attmnt 1572 at 2012-06-07 11:24:00 UTC ***

With NSPR_LOG_MODULES=purpleSockets:5,nsSocketTransport:5 on a debug build, I've observed that purpleSockets are very rarely detached from the socket transport service. The attached patch fixes this issue.

I believe that this bug is caused by idle sockets that should have been detached exhausting the maximum number of sockets we can have at once in the socket transport service, so the patch should fix the bug.

Comment 15

[Patrick Cloke [:clokep]]

Comment on attachment 8353327 [details] [diff] [review]
Patch

*** Original change on bio 1355 attmnt 1572 at 2012-06-07 11:42:18 UTC ***

Well it definitely cleans it up a lot. :) These changes look OK to me.

Comment 16

[Florian Quèze [:florian]]

*** Original post on bio 1355 at 2012-06-07 13:21:13 UTC ***

Just to be clear: this was definitely a regression from bug 954777 (bio 1344).

Pushed the patch as https://hg.instantbird.org/instantbird/rev/2ed5407c4a6a

Resolving as fixed, but it's possible the issue I fixed was just the most obvious one and was hiding other similar but less visible issues, in which case we will reopen this bug or file a new one.