On Wed, Jul 16, 2008 at 11:15:50PM -0401, jared r r spiegel wrote: > On Wed, Jul 16, 2008 at 09:13:14PM -0400, jared r r spiegel wrote: > > > > on jul 11 snapshots now, have gone thru i think 2 or 3 snapshot iterations > > since ~early/mid june. > > cracked out again hardcore a bit ago (when it shits out it seems accurate > to call it a deadlock), so now am on: > > OpenBSD 4.4-beta (GENERIC.MP) #800: Mon Jul 14 20:28:22 MDT 2008
still on that snapshot; seems to be OK so far. two curious spuriousness happened today tho: Aug 1 22:12:47 iorek /bsd: uvm_mapent_alloc: out of static map entries Aug 1 23:12:49 iorek /bsd: uvm_mapent_alloc: out of static map entries those're the only instances of this message since previous email. two of those log warnings showed up; i corroborated those times against the log for the vmstat_watcher script i made up (which gets all its data from sysctl(1)): --- Aug 1 22:10:38 iorek vmstat_watcher: vmstat_watcher: name: UVM_amap memuse: 6337000 maxused: 8650000 limit: 39322000 Aug 1 22:12:40 iorek vmstat_watcher: vmstat_watcher: name: UVM_amap memuse: 6338000 maxused: 8650000 limit: 39322000 Aug 1 22:12:50 iorek vmstat_watcher: vmstat_watcher: name: UVM_amap memuse: 7014000 maxused: 8650000 limit: 39322000 Aug 1 22:13:51 iorek vmstat_watcher: vmstat_watcher: name: UVM_amap memuse: 6340000 maxused: 8650000 limit: 39322000 <...> Aug 1 23:11:56 iorek vmstat_watcher: vmstat_watcher: name: UVM_amap memuse: 6871000 maxused: 8650000 limit: 39322000 Aug 1 23:12:47 iorek vmstat_watcher: vmstat_watcher: name: UVM_amap memuse: 7951000 maxused: 8650000 limit: 39322000 Aug 1 23:12:57 iorek vmstat_watcher: vmstat_watcher: name: UVM_amap memuse: 6868000 maxused: 8899000 limit: 39322000 Aug 1 23:13:07 iorek vmstat_watcher: vmstat_watcher: name: UVM_amap memuse: 6869000 maxused: 8899000 limit: 39322000 --- no other warnings in dmesg. so... something made the yelp to the ringbuffer, but i've got no guess as to what it was... no instance of unrecoverable deadlock for sure, no instance of recoverable deadlock (comalock?) *afaik*. i'm fine with the notion that the alloc/free or whatever happens faster than any granularity than my script can catch, but what i actually find curious is that the 'maxused' i get from sysctl also doesn't reflect even coming anywhere near the roof. http://www.ice-nine.org/jrrs/UVM.png is updated, fwiw... -- jared
