On Jan 19, 2012, at 8:56 PM, David Anderson wrote:
> 
> I don't understand.  What is the problem, and what did you try?
> 
The problem is the client returned stderr is missing or truncated on occasion. 
The stderr.txt file isn't necessarily written to disk when the client reads it 
to send it back, so you end up with nothing useful. It's necessary to sync it 
to be sure it's not stuck in some system write cache. Here is also an updated 
one; there was some assertion failed on windows if the redirection isn't used. 
However this should only solve the problem in the case where everything 
finishes normally and boinc_finish() is called. It could still potentially 
happen if there is a crash of some sort.

> This is probably not an issue; the window is about a microsecond.

It shouldn't really be there no matter how short it might be. Between the 
handle_finished_apps poll and the next check to write the state file there is 
the potential for disk or network activity from scheduler_rpc_poll so you can't 
expect that it will finish very quickly. I couldn't decide whether to fix it by 
forcing a write before wiping the slot, or just not deleting everything right 
when it finishes. I don't see why it should be so eager to delete everything in 
the slot.

Attachment: BoincFlush.diff
Description: Binary data

_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.

Reply via email to