[Mauiusers] maui crashes with segv (version 3.2.6.p16)

wightman wightman at clusterresources.com
Fri Nov 3 09:07:48 MST 2006


Answers are inline:

> > I thought now would be a good time for Cluster Resources Incorporated to
> > chime in and give an official position.
> > 
> > First off, let me say that Maui is still being maintained.  There aren't
> > any new features planned but bug fixes and enhancements from the
> > community are always welcome and are actively applied.  
> > 
> That is good to hear.
> 
> Several people have posted patches or enhancements to the list with no 
> reaction from the maintainers. In the past i have also submitted 
> patches/enhancements for maui and i got a reaction if the patch will be 
> applied or rejected.  Some patches also fixed issues in moab ;-)
> 
> To give an example. I have posted a patch that fixes an issue with MAXPS 
> for our site. I submitted a bug report and asked several times to the 
> list if the patch will be applied, rejected or must i change the patch 
> so it would be applied. Till now i have not had a reaction at all ;-(

My apologies.  I thought another developer had gotten back to you about
this patch.  The only tricky part about this patch, and the reason that
it hasn't yet been applied, is that it changes default behavior.
Because Maui is in use at so many locations, we try not to change the
default behavior ever.  Normally, in a situation like this, it's best to
create a new parameter, like NODEALLOCMAXPS {TRUE|FALSE}.  When set to
TRUE then you get the behavior you desire.  But default, however, it is
set to FALSE so as not to disrupt the standard behavior that others have
come to expect.  The developers here were split as to whether this was a
"bug" or not.

> > As for the specific segv bug that this thread centers on, I have not yet
> > seen a patch from the community that the community agrees should be
> > applied.  As soon the community agrees on a patch it will be applied and
> > a new snapshot will be released.  
> 
> Somebody has to make the decision to apply a patch. Who makes the 
> decision? The community is a bit vague ;-).

Most times we've found that a fix is a fix and as long as it fixes the
bug without introducing any new ones we'll put it in.  Obviously there
are some patches that are so obvious we would be stupid not to apply
them :)

> > Many bugs, such as this specific segv, cannot be reproduced in our own
> > labs simply because we don't locally maintain the large clusters that
> > the community maintains.  However, should a remote site ask us to log in
> > remotely to track down an issue we are always happy to do so.
> > 
> To trigger the segv, you do not need a big cluster at all. I run it on a 
> test cluster with 2 nodes and adjusted MAX_MJOB to 32768 and submit a 
> lot of jobs and see if maui could cope with it. So it is a simple test.

Sounds good, we'll give it a try and see what we can find.

> > Cluster Resources has no plans to drop Maui.  If there is a member of
> > the OSS community that believes they can act as an efficient liaison
> > between Cluster Resources and the rest of the community we are happy to
> > give that person admin rights to the Maui bugzilla and also enable
> > direct code access.
> 
> Could this also be a team instead of one person?

Possibly.  What do you recommend?


- Douglas



More information about the mauiusers mailing list