Re: [jonas] JOnAS 4.6.6 Transaction bug

[David.Egolf@xxxxxxxxxxxxx]

Hello Roland,

I am a little concerned by your belief that the problem occurs only under 
load.  The problem is occurring
during the use of a thread local collection.  The problem I saw with the 
code is an apparent use of the
collection while an iterator was active.  The collection happens to be 
thread local.

o o o 

After consultation with several people, scouring the web for similar error 
reports, and examination of the 
code, I believe that the problem probably arises when an XAResource notes 
an error on a connection 
DURING the enlist callback.  It must decide to use the 
connectionErrorOccured() callback during the enlist 
callback.  The connectionErrorOccured method will modify the collection 
and ruin the iterator for future use.

Please note that we have no evidence that this is the case.  It's only one 
of the plausible explanations I can 
think of which might have some dependence on system load.

Another possibility is that some code neglects to call the 
connectionClosed() method in one TX instance
and the resource manager somehow figures out that when a subsequent 
begin() on the same thread 
attempts to enlist the XAResource, then an error has occurred which 
results in either an error or close call.

That being said, copying the original list prior to the enlist calls 
appears to still be the correct solution.

Regards,

David

Roland Hedayat <roland@xxxxxxxxxx> wrote on 07/06/2006 01:02:18 AM:

> Hello David,
> 
> Tomorrow is the last day of work before leaving for vacations.
> 
> However, my colleague Mikael will be back by the beginning of
> next week, and will probably be able to give testing of  the new
> code priority.
> 
> We have made some refactoring and simplifications in our code
> so unfortunately we will not be able to test on the same code
> base that provoked the error in the production environment,
> because there are slight incompatibilities in the database.
> 
> Reverting to the original code and testing it in our test environment
> will probably not be meaningful, since I think the error occurs under
> load only.
> 
> Now when there are signs that we will be scaling the application, we
> will need to create a test environment where we are able to generate
> synthetic load, but that is for later.
> 
> Anyway we are very glad that this will be addressed, because we are
> extremely keen on seeing Speedo/Jotm/JOnAS/ working in harmony.
> 
> Best regards,
> 
> Roland
> 
> 
> -- 
> ==================================
> | Roland Hedayat
> | Inherit S AB
> | Långsjöv. 8
> | 131 33 Nacka, Sweden
> | roland@xxxxxxxxxx 
> | Tel: +46 (0)8  641 64 14 
> | Mob: +46 (0)708 18 07 69 
> | Fax:  +46 (0)8  718 52 88 
>