OK

Looking at a heap dump that Yourkit took when an out of memory condition was reached:

Name: org.opencastproject.serviceregistry.impl.JobJpaImpl
Objects: 21 117
shallow size: 4 899 144
Reatained: 1 691 868 032

Now I haven't done much profiling on matterhorn but this looks like a problem ...

D

On 01/18/2012 10:55 PM, Tobias Wunden wrote:
Hi David,

On our recently upgraded system (1.3.x) I'm seing repeated failures to distribute, the 
workflow just stalls on the "Distributing media to progressive downloads"
where is matterhorn trying to copy the files to? Is it a network share or a 
local disk? Does ist start copying the files, and if so, do you see the file 
size increase?

I've so far tried:
1) Not having the workers talk to the admin via mod_proxy
2) Doubled the memory on the admin.
I am wondering why you are thinking that it could be a memory issue? The 
distribution service first copies the files to the local workspace and then 
copy the files to the final destination using either file channels (uses the 
native os for copying) or in chunks of 1 MB.

Before I did 1. I did seem messages that "distributing file x timed out. retrying in 
yms" - but never saw any evidence of a retry. The last job after I made change 1. 
just failed after putting all 3 nodes in the cluster under heavy load.
That is *really* strange. I tried to find that log message (or similar ones) in 
the codebase without any luck. Do you think you could possibly get the correct 
message from the logs?

The media packages in question are in the region of 1.2Gb and contain just over 
an hour of recordings. Smaller media packages have no issues, this leads me to 
suspect that there is a regression in this service that is doing something like 
reading the files into memory.
A timeout seems likely. What could be possible is that we are hitting a bug 
with downloading large files from the working file repository. Do you have the 
working file repository root configured to the same share as the workspace root 
(i. e. is hard linking enabled)?

Thanks,
Tobias
_______________________________________________
Matterhorn mailing list
[email protected]
http://lists.opencastproject.org/mailman/listinfo/matterhorn


To unsubscribe please email
[email protected]
_______________________________________________

_______________________________________________
Matterhorn mailing list
[email protected]
http://lists.opencastproject.org/mailman/listinfo/matterhorn


To unsubscribe please email
[email protected]
_______________________________________________

Reply via email to