O1Spot1TLo 1.02 computation errors
19 May 2017 7:40:03 UTC
Topic 207728
(moderation:
All running instances of einstein_O1Spot1TLo_1.02_x86_64-pc-linux-gnu__AVX are killed with "finish file present too long" when BOINC is restarted. This is on two hosts that were running 1.00 or 1.01 before. Debian jessie + BOINC 7.6.33 from backports.
I had also one (Hi freq
)
I had also one (Hi freq task).
https://einsteinathome.org/task/646634804
That happened yesterday after I had reboot that host. It had been running only 1 of those tuning searches at a time. That specific task had been running for almost 7 hours. I shut down Boinc and even gave about ten seconds for whatever write operations to complete before I hit reboot.
After reboot, I let Boinc start working again and it said something like files of that task do not exist... and a fresh task begun from the beginning.
I looked at your tasks and
)
I looked at your tasks and the problem seems to be that when you restart BOINC the science app gets send a SIGTERM which aborts the computation. Your Client shouldn't send a SIGTERM in the first place. We didn't change the science app on how it reacts to this kind of signal between 1.01 and 1.02.
A possible explanation is that it takes too long for the app to finish normally and that the Client tries to expedite this by sending a SIGTERM.
We are investigating a possible build system related change between 1.01 and 1.02.
There is going to be a
)
There is going to be a version 1.03 for Linux that will address this problem soon.
I now see version 1.04 on
)
I now see version 1.04 on Linux.
Everything is back to normal
)
Everything is back to normal with 1.04.
I see a couple of these
)
I see a couple of these errors on v1.05 - see here after a daily restart.
That's different app though
)
That's a different app though (Gamma-ray pulsar binary search) ...
Good point! I'll start a
)
Good point! I have started a separate thread, hopefully the fix will be exactly the same.