[01:26:17] --- lars.malinowsky has become available [05:58:51] --- haba has become available [06:23:39] > I reproduced the bonnie++ hang that GAWollman reported on backtrace? [07:00:47] --- reuteras has left [07:20:39] --- lars.malinowsky has left [07:46:00] --- deason has become available [08:05:07] > backtrace bonnie++ is hung in ObtainDCacheForWriting on afs_WaitForCacheDrain, the cache is full, and nothing is pushing it out to the wire. I was going to send mail, but got sleepy. [08:05:44] suggests a wakeup is missed. [08:06:35] where is the afs_CacheTruncateDaemon task at? (backtrace-wise) [08:06:41] where is cachetruncdaemon? [08:06:46] er, yeah :) [08:07:17] It's in its 100ms sleep in case something is waiting for the glock. [08:07:49] i wonder if afs_osi_Wait is broken [08:08:04] any idea if it's been in there ~forever? [08:09:35] can you see how many ticks in the msleep call? [08:37:22] --- lama has become available [08:53:09] --- haba has left [08:54:41] --- Simon Wilkinson has become available [08:55:51] The dead IRIX builder is currently stopping buildbot from reporting any results. Any chance of bringing it back to life, or removing it from the list of hosts the buildbot waits for results from? [08:57:33] > how many ticks in the msleep call Maybe; it would require mucking around in the sleepq internals. the msleep call certainly seems to look to be correct. I want to say I've looked at this before and concluded that the cachetruncatedaemon is actually continuting to run, but I don't remember enough to actually be confident about that. [09:01:14] --- Simon Wilkinson has left [09:02:42] --- Simon Wilkinson has become available [09:03:48] The dead IRIX builder is currently stopping buildbot from reporting any results. Any chance of bringing it back to life, or removing it from the list of hosts the buildbot waits for results from? [09:11:13] --- Russ has become available [09:19:07] --- Simon Wilkinson has left [09:24:40] Apparently the sleepqueue is implemented as a callout structure which is stored in per-cpu or per-thread storage (not sure which, yet). Getting to this from the core dump could be exciting ... [09:25:20] yay! if you can reproduce, debug with printf? [09:26:15] That's probably what I'm looking at. Are we interested in anything other than whether cachetruncatedaemon is actually running? [09:26:55] whether it's doing any work? whether it's trying to wake up waiters? [09:31:01] --- mfelliott has become available [09:34:48] --- Russ has left: Disconnected [09:49:54] --- Simon Wilkinson has become available [10:51:05] --- rra has become available [10:52:01] Having your build fail because dkms can't use echo to write a small amount of data to a file when your root partition is full is not a bug in openafs-modules-dkms, even if you have 300MB reserved space for root free and don't understand why it's not being used.... [10:59:32] --- Simon Wilkinson has left [12:10:48] --- haba has become available [12:19:24] --- haba has left [14:10:31] The tag for 1.6.0pre4 isn't pushed yet. Derrick, could you push that? [14:16:09] Thanks! [14:20:09] np [14:56:54] --- lama has left [15:31:20] --- deason has left [16:05:57] 1.6.0pre4 uploaded to unstable. [16:06:09] With the 2.6.39 patches that came in after it was tagged. [18:34:11] --- rra has left: Disconnected [18:55:43] --- deason has become available [18:59:54] --- Russ has become available [22:21:01] --- reuteras has become available [22:26:18] --- deason has left [23:02:39] --- Russ has left: Disconnected [23:09:45] --- lars.malinowsky has become available [23:18:26] --- lars.malinowsky has left