Client issues 'empty' RPC requesting zero work

[Richard Haselgrove]

Description If a client has a CUDA work shortfall, but no CUDA project is fetchable, it may issue a CPU work fetch instead. If there is no CPU shortfall, the work fetch is issued for 0.00 seconds.

For example:

17/04/2009 11:21:51 [wfd] CPU: shortfall 0.00 nidle 0.00 est. delay 0.00
17/04/2009 11:21:51 [wfd] CUDA: shortfall 21420.16 nidle 0.00 est. delay 0.00
17/04/2009 11:21:51 Einstein@Home chosen: CPU minor shortfall
17/04/2009 11:21:51 Einstein@Home [wfd] request: CPU (0.00 sec, 0) CUDA (0.00 sec, 0)

Reproducibility Difficult. Only observed during SETI server crashes, but observed twice during different outages - once with client v6.6.23, and several times with client v6.6.28

Suspect code area Function WORK_FETCH::choose_project() in work_fetch.cpp

There is no "none of the above" escape path to cancel the work fetch if no project with the desired resource is fetchable.

Reporting history On boinc_alpha:

17 Apr 2009, "Work fetch bug in v6.6.23" (with log)
23 Apr 2009, "Fw: Work fetch bug in v6.6.23" (analysis with JM7)
10 May 2009, "Work fetch bug continues in v6.6.28" (7 instances logged)

[Nicolas]

Anything a project could do to intentionally cause this problem, and help reproduce?

[Richard Haselgrove]

Nobody noticed (or if they noticed, they were too polite to mention) that I typed v6.6.28 wrongly.

Observed again today, with v6.6.31, and no SETI server crash (though the feeder/scheduler is under some extra stress because they've squandered all the Astropulse jobs).

[Nicolas]

Fixed version number.