List:Falcon Storage Engine« Previous MessageNext Message »
From:Vladislav Vaintroub Date:November 5 2008 9:23pm
Subject:RE: deadlock between Falcon logging and server logging
View as plain text  
Ann , you can get much more insight into what actually happens and get rid
of unresolved hexadecimals and have readable function names from C runtime
and ntdll, if you setup the symbols server to download PDBs from Microsofts
repository as  described here:

http://msdn.microsoft.com/en-us/library/b8ttk8zy(VS.80).aspx



> -----Original Message-----
> From: Ann.Harrison@stripped [mailto:Ann.Harrison@stripped] On Behalf Of
> Ann W. Harrison
> Sent: Wednesday, November 05, 2008 10:22 PM
> To: FalconDev
> Subject: deadlock between Falcon logging and server logging
> 
> The attached file (assuming it doesn't get stripped) is the "deadlock"
> I found last week between the Falcon logging system and the server
> logging.  The situation was completely trivial.  I had two connections,
> each with an active transaction.  Both were using a single table with
> three or four simple records.  One attempted to commit, and the server
> froze.
> 
> The problem is between threads 1504, 3640, and 784.
> 
> 784 is in Log::logMessage, trying to write out its commit message. It
> can't get the critical section because
> 
> 3640 has the critical section and is trying to execute an _unlock after
> having written out scavenge statistics.  It can't move because
> 
> 1504 is finishing an fflush and trying to _unlock_file after a server
> write.
> 
> The attached file may have better formatting, given the wrap induced by
> Thunderbird.
> 
> What I think we need to do is figure out how InnoDB writes its messages
> and use that.
> 
> Best,
> 
> Ann
> 


Thread
deadlock between Falcon logging and server loggingAnn W. Harrison5 Nov
  • RE: deadlock between Falcon logging and server loggingVladislav Vaintroub5 Nov