World Community Grid - View Thread - Something going on with WU ex379

Okay, thanks for the info. So if there'll be another error this WU will be rejected.

The error occured just when the benchmarks started to run:

2006-06-11 21:23:39 [---] Suspending computation - running CPU benchmarks
2006-06-11 21:23:39 [World Community Grid] Pausing task ex379_2A_3 (removed from memory)
2006-06-11 21:23:39 [---] Suspending network activity - running CPU benchmarks
2006-06-11 21:23:40 [World Community Grid] Unrecoverable error for result ex379_2A_3 ( - exit code -1073741819 (0xc0000005))

We see these errors from time to time in BOINC. The app crashes when BOINC suspends it (removes it from memory). In my investigations (and opinion), there is a race condition in the BOINC code which is causing these crashes. We've seen these since we first launched BOINC with the 5.2 client. Looks like now we're still seeing them with 5.4. Unlike the "bad" batches a couple of weeks ago, this problem can hit any workunit.

When a crash like this occurs, we are sent a stack trace. In all cases the stack trace leads to code in BOINC. The problem with race conditions is that they are notoriously hard to reproduce. In fact, in our internal testing we almost never see them. But let 100,000 machines at it in the wild and it starts to show up.

Also know that not all WU errors are due to this problem. Only in the cases where the app ends after it gets suspended.

FYI, if you leave the process suspended in memory, this problem should not occur.