[Pvfs2-users] MD server crash

[Sam Lang]

I think this is probably related to some of the changes we added  
recently that are supposed to allow in-progress operations at the  
server to complete when it receives a SIGINT.  Simply allowing them  
to complete appears to have expressed some bugs esp. if the client is  
gone.
Thanks for the detailed report, I'm assuming the same behavior occurs  
with tcp as well, but I guess you've only tried with ib?

-sam

On Sep 6, 2006, at 10:31 AM, Kyle Schochenmaier wrote:

> I was wondering if this is an expected operation:
>
> I did pvfs2-cp from client to server, ctrl-c'd the client  
> operation, and it killed the server.
> The client log looks like it handled everything correctly, and de- 
> registered all the memory and etc. that it should have done.  Not  
> so lucky on the server I think, this is the log from the MD server,  
> which is the only one that went down, the Data servers are all  
> still up.
>
> -- using CVS head from last week, and openIB, client is ppc64,  
> servers are x86_64 --
>
> [D 10:27:55.956330] PVFS2 Server version  
> 1.5.1pre1-2006-08-30-144459 starting.
> [E 10:28:22.852931] src/proto/PINT-le-bytefield.c line 573:  
> lebf_decode_req: imp
> roper input buffer size: 0x63fc8c != 0x63fc58 + 80
> [E 10:28:22.853531]     [bt] /usr/local/sbin/pvfs2-server [0x429895]
> [E 10:28:22.853541]     [bt] /usr/local/sbin/pvfs2-server 
> (PINT_decode+0x1ac) [0x
> 4277ec]
> [E 10:28:22.853549]     [bt] /usr/local/sbin/pvfs2-server [0x40edfc]
> [E 10:28:22.853561]     [bt] /usr/local/sbin/pvfs2-server(main 
> +0x11d4) [0x4101e4
> ]
> [E 10:28:22.853568]     [bt] /lib/libc.so.6(__libc_start_main+0xda)  
> [0x2aeb3b668
> 4ca]
> [E 10:28:22.853576]     [bt] /usr/local/sbin/pvfs2-server [0x40d5ba]
> (END)
> Is this the expected operation when a client 'dies'?  I've seen in  
> the past that simply killing the pvfs2-cp process hasnt caused  
> server's to go down, but maybe I'm wrong?
>
>    -- Kyle
>
> -- 
> Kyle Schochenmaier
> kschoche at scl.ameslab.gov
> Research Assistant, Dr. Brett Bode
> AmesLab - US Dept.Energy
> Scalable Computing Laboratory
> _______________________________________________
> Pvfs2-users mailing list
> Pvfs2-users at beowulf-underground.org
> http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
>