[mapguide-internals] MapGuide 2.2 beta crashes

[Jason Birch]

How are you checking for resource existence? There's a nasty memory
corruption problem if you just try to catch a failed
getresourcecontent call.

On 2010-10-15, Gabriele Monfardini <gabrimonfa at gmail.com> wrote:
> Hi Trevor,
>
> thank you for the quick reply.
>
> I've opened a ticket
> http://trac.osgeo.org/mapguide/ticket/1469
>
> with the demangled log
>
>> Also, have you found any *easy* way to reproduce this?
> The short answer is unfortunately no.
> We have a repositories with a lot of folder and layers and maps.
> We probably have a few problematic layers or missing ones, we're
> investigating.
>
> The crashes are quite recurring (usually around 10/day but up to
> 40-50/day), but only a few of them leave some traces into the log.
> I've reviewed Error log and Trace log to find out problematic calls,
> but I've not found anything useful.
>
> mapguide-apache error log is polluted by a lot of failed ACE_ASSERT
> that make apache child abort with signal Aborted (6) or more rarely
> Segmentation Fault (11)
>
> ACE_ASSERT: file Configuration.cpp, line 1522 assertion failed for
> 'this->allocator_'.Aborting...
> 292 [Wed Oct 13 17:02:46 2010] [notice] child pid 27120 exit signal Aborted
> (6)
>
> This two lines occurs hundreds of time a day and seems to not kill
> mapguide nor mapguide-apache (but I'm not 100% sure)
>
> We're using the OGR provider with gdal library compiled to support
> PostgreSQL and multi thread.
> No missing libraries nor symbols in gdal and FDO providers.
>
> Number of crashes grows with concurrent load, at the point that we
> need to carefully balance our clients to several mapguide servers.
> The number of crashes appears also to grow with the use of features
> that involve the use of the API in our PHP code, selecting, redlining
> and so on.
> Single features are tested in a separated test environment and they
> work well. Not to say that testing server crashes very rarely even
> under load.
>
> I've carefully reviewed our scripts to catch all possible MgExceptions
> and to do basic checks (layer exists before using it and so on) to
> avoid making problematic calls to the server.
> No single action seems to trigger the crash and we're not yet able to
> reproduce reliably the problem.
>
> Sometimes one single user is able to crash mapguide several times,
> some others many customers doing about the same things are served
> without problems and without crashes.
>
>> There was a recent submission put into trunk to correct some Web Tier
>> initialization behaviour.  We can try backporting it.
> Regarding Web Tier initialization I have a question.
>
> What exactly does the InitializeWebTier() call do?
> All the PHP script appears to work also without. Is this mandatory?
>
> In the following days we probably try the 2.2 RC1, hoping for better
> stability.
>
> Regards,
>
> Gabriele Monfardini
> _______________________________________________
> mapguide-internals mailing list
> mapguide-internals at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/mapguide-internals
>