Subject | Re: firebird freeze all connections (again) |
---|---|
Author | hvlad |
Post date | 2010-03-06T19:59:57Z |
Stephane, why you ignore my messages ?
http://www.firebirdsql.org/download/rabbits/hvlad/Firebird-2.1.4.18274_Win32.zip
http://www.ibphoenix.com/main.nfs?a=ibphoenix&s=1254244067:355713&l=;PAGES;NAME=%27ibp_pdb_win32%27
to produce crash dump for us.
Regards,
Vlad
> sorry to come back all the time with this problem, but it's look like a serious bug in firebird (and for my life too) and still now i have no idea of how to resolve it !You can try this snapshot of 2.1.4 with patch for index corruptions :
>
>
> 1/ Firebird 2.1.3 classic server
http://www.firebirdsql.org/download/rabbits/hvlad/Firebird-2.1.4.18274_Win32.zip
> 2/ Windows 2008 R2 64 bit + 32 go of memoryToo much for Classic
> 3/ 8 Processors
> 4/ default page buffer 2048
> 5/ around 100 - 500 simultaneous connexionSome application did not closed TCP connection gracefully. One of the reason - crash of application. Another one is broken network connection.
> 6/ in the firebird.log only some row like
> SERVER123 Sat Mar 06 15:05:21 2010
> INET/inet_error: read errno = 10054
> 7/ 99.99% transaction was read only when it's crash todayFollow instructions at
http://www.ibphoenix.com/main.nfs?a=ibphoenix&s=1254244067:355713&l=;PAGES;NAME=%27ibp_pdb_win32%27
to produce crash dump for us.
> 8/ in the taskmanager, 22 GO of memory used and 10 GO freeI see no crash in this case. If Classic listener process is frozen you can kill it using drWatson (see instructions above) and produce crash dump for us.
>
>
> every 2 - 3 days our server crash in the way that it's simply impossible to connect to the main database (even with ISQL). in isql for exemple, after the instruction connect database ... the command prompt is freeze. if we try to connect to another database it's ok.
> It look like also than when we reduce the number of page buffer then the crash is not so often (for exemple with a page buffer of 75)Output of fb_lock_print -h could help to answer
>
> as i know, every page need to be "lock", so more page buffer mean more lock and in classic server they are multiply by the number of client, so around 2048 page buffer * 500 connections = 1 000 000 page to lock... is it possible that it's was a lock problem ??
> any other idea ? or what to do to avoid such problem ?First, follow suggestions above
> do you thing i can try to install FB2.5 in our production server (it's can not be more bad that what is it now)
Regards,
Vlad