new FGRP app version 1.03
10 Jan 2013 5:00:52 UTC
Topic 196729
(moderation:
New Gamma-ray Pulsar WUs downloaded in the past few hours have been styled:
"Gamma-ray Pulsar Search #2 1.03" whereas for many days they have been
"Gamma-ray Pulsar Search #2 0.01".
Is there anything interesting about the difference?
new FGRP app version 1.03
)
Yes, it's broken, at least for Mac OS X. Deprecated for now until further investigation.
This release was meant to fix a bug introduced in the BOINC API in September (se thread "Einstein Gamma-ray pulsar search #2 Linux app constantly restarting with no heartbeat" on boinc_dev).
The Linux version seems to work reasonably well, it might even actually run with newer (currently Alpha) Clients.
BM
BM
RE: Yes, it's broken, at
)
Those with Mac OS X hosts will notice that there was (very approximately) a 10 hour window (20:00 UTC last night to 06:00 UTC this morning) over which FGRP2 1.03 tasks were distributed before the app was deprecated. Since that time the 'task branding' for new tasks (sent after the window) has reverted to what it was previously (0.01). I don't know for sure, but I assume BOINC will simply ignore the 1.03 branding, when those tasks get to the top of the work cache queue, and use the 0.01 app to crunch them. I don't expect the user will need to do anything.
On my Linux hosts, 1.03 tasks are continuing to arrive.
0.02 works fine with BOINC 7.0.42 (one host with a Hd7770 GPU) - no more exit 0s there. I've still got that host on 7.0.42 despite the fact that the BRP4 app continues to get occasional exit 0s. I'll promote some new FGRP2 1.03 tasks on that host and see how they go.
Cheers,
Gary.
RE: RE: Yes, it's broken,
)
No, actually. Macs will still have a local copy of the v1.03 application, and will use that to process any tasks branded 1.03
Whether a user chooses to do anything rather depends on the nature of the 'breakage'. If the tasks fail quickly, it's probably best to grin and bear it. But if they run to full term but fail to generate valid results, it would be better simply to abort the tasks and fetch v0.01 replacements.
RE: No, actually. Macs will
)
And this is exactly why I posted :-). I didn't know what would happen - I wondered whether there might be a 'deprecation procedure' built into BOINC which would (in a sched_reply message, perhaps) tell the client to substitute a replacement app for an 'officially deprecated' app. Failing that, perhaps the server could do a 'server-side abort' of the deprecated tasks. I had no clue so I posted, looking for guidance, and it was delivered :-) thanks very much Richard. I have a number of iMac hosts with 3 day caches so I'll (hopefully) be able to abort all the offending tasks tomorrow if necessary. Hopefully Bernd will have given some guidance by then.
Cheers,
Gary.
RE: Whether a user chooses
)
All failures I've seen report zero CPU time, these tasks should fail after a few seconds without doing any actual computation and don't waste any CPU time.
BM
BM
RE: The Linux version seems
)
I have one host running a newer client (7.0.42). It has been running FGRP2 0.02 tasks and has now received some 1.03 tasks. I promoted one and it ran and completed normally. It's currently 'pending'.
Cheers,
Gary.
The Linux 0.02 version has
)
The Linux 0.02 version has been working well on my systems with a very recent post 7.0.42 BOINC build from GIT repository. I have some 1.03 tasks in the queue waiting to run once the 0.02 tasks complete.