Camelia, the Perl 6 bug

IRC log for #gluster-dev, 2013-01-28

| Channels | #gluster-dev index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
03:09 hagarth joined #gluster-dev
03:27 sgowda joined #gluster-dev
03:32 bharata joined #gluster-dev
03:35 bulde joined #gluster-dev
04:17 sripathi joined #gluster-dev
04:19 overclk joined #gluster-dev
04:22 sahina joined #gluster-dev
04:31 quillo joined #gluster-dev
04:41 sripathi1 joined #gluster-dev
04:41 sripathi joined #gluster-dev
05:01 vpshastry joined #gluster-dev
05:15 raghu joined #gluster-dev
05:33 bharata joined #gluster-dev
05:33 sripathi joined #gluster-dev
05:52 sripathi joined #gluster-dev
06:16 sripathi joined #gluster-dev
06:29 sgowda joined #gluster-dev
06:32 sripathi joined #gluster-dev
06:51 sripathi joined #gluster-dev
06:51 sgowda joined #gluster-dev
06:56 deepakcs joined #gluster-dev
07:05 sripathi1 joined #gluster-dev
07:39 vpshastry joined #gluster-dev
07:52 sgowda joined #gluster-dev
07:58 sripathi joined #gluster-dev
07:59 vpshastry1 joined #gluster-dev
08:09 puebele joined #gluster-dev
08:17 sripathi joined #gluster-dev
08:19 mohankumar joined #gluster-dev
08:25 bulde joined #gluster-dev
08:27 puebele joined #gluster-dev
08:50 sripathi1 joined #gluster-dev
09:16 gbrand_ joined #gluster-dev
09:42 sgowda joined #gluster-dev
09:45 vpshastry joined #gluster-dev
09:49 sripathi joined #gluster-dev
10:07 sripathi joined #gluster-dev
10:12 sgowda joined #gluster-dev
10:14 sripathi1 joined #gluster-dev
10:32 badone joined #gluster-dev
10:37 sripathi joined #gluster-dev
11:05 sahina joined #gluster-dev
11:26 sripathi joined #gluster-dev
11:28 bala1 joined #gluster-dev
12:01 joaquim__ joined #gluster-dev
12:06 edward1 joined #gluster-dev
12:33 bala1 joined #gluster-dev
13:25 jdarcy http://memegenerator.net/instance/33977326
13:32 bfoster :)
14:08 kkeithley joined #gluster-dev
15:05 hagarth joined #gluster-dev
15:13 wushudoin joined #gluster-dev
15:45 kr4d10 joined #gluster-dev
15:51 inodb joined #gluster-dev
15:58 puebele3 joined #gluster-dev
16:16 puebele joined #gluster-dev
16:36 puebele joined #gluster-dev
16:42 jbrooks joined #gluster-dev
17:58 vpshastry joined #gluster-dev
18:00 vpshastry1 joined #gluster-dev
18:40 vpshastry1 left #gluster-dev
18:50 portante joined #gluster-dev
21:50 a2 bfoster, ping
21:54 foster a2: pong
21:55 wushudoin joined #gluster-dev
21:59 a2 foster, abt http://review.gluster.org/#change,4441
22:00 a2 there are reasons to prefer readdirplus in dht other than for caching purposes
22:01 a2 dht_readdirplus_cbk's filtering of entries (which is based on attributes) is usually more "accurate" than dht_readdir_cbk's (which is based on entry names and hash values)
22:01 a2 in a stable state, both are accurate
22:02 foster a2: ok, I'm not terribly tied to those changes...
22:02 foster I was using it to debug/hack and figured I'd throw it out there
22:02 a2 but if there are unhashed files which do not have linkfiles (which will get self-healed on lookup), then those entries do not get listed in readdir()
22:02 a2 but readdirplus is more robust in such situations as it does not care about name/hashes
22:05 a2 foster, ok.. did you just use them as debug hooks, or anything more?
22:06 foster just debug, so I could turn it on and off easily
22:07 a2 would accepting just #4440 (md-cache patch) and abandoning dht patch still be "useful" as far as what you were acheiving for debugging?
22:08 foster hmmm, i suppose it couldn't hurt
22:08 foster i could just disable dht from said tests and get the same use out of it
22:08 a2 was it perf testing?
22:09 foster yeah, just comparing latency and whatnot
22:09 foster (details are in that bug the change references)
22:10 a2 (reading bug comments)
22:10 foster i was originally wondering if it would be useful to users, but where I'm at now I don't think that is the case
22:11 foster I'm seeing an issue where the iatt set from the initial readdirp in md-cache is ineffective
22:12 foster (so I think that might invalidate some assumptions in that bug, btw)
22:14 a2 foster, correct. md-cache's readdirplus effectiveness is only in conjunction with http://review.gluster.org/#change,3483
22:14 a2 without #3483 (i.e, now) md-cache is only useful in the second ls -l
22:15 foster hmm, ok. so why did that never go in?
22:15 a2 while working on #3483 i kind of figured FUSE doing readdirplus is "cleaner" and took that branch
22:15 foster ah, ok. well that saves me some debugging then :)
22:15 a2 that way our inode table need not keep track of two kinds of LRUs
22:16 a2 and FUSE's readdirplus abilities kind of makes md-cache useless in the fuse client too
22:17 foster right... but even with that, wouldn't we want internal readdirps to work correctly too?
22:17 a2 meaning?
22:17 foster e.g., older fuse, or no fuse at all
22:17 a2 you mean with old FUSE?
22:17 a2 ah
22:17 a2 yes
22:17 foster wouldn't we still want to fix this?
22:18 a2 sure, md-cache going uselss with new fuse was just an isolated comment
22:19 foster so should we pull amar's fix in?
22:20 a2 which amar's fix?
22:20 foster err, nm. amar is a reviewer... misread the page
22:20 foster 3483
22:20 foster should we pull that in anyways?
22:21 a2 it needs some minor work IIRC
22:21 a2 i was planning to work on that and the other FUSE_READDIRPLUS patch sometime soon
22:21 foster ok, cool. what's the other fuse path?
22:22 foster patch
22:22 a2 the one to utilize the kernel readdirplus support
22:22 foster ah, gluster support for the fuse patch
22:22 a2 http://review.gluster.org/#change,3905
22:22 a2 yes
22:22 foster cool
22:24 a2 is md-cache re-readdir slower than no-md-cache?
22:24 a2 (sorry haven't read the full bz yet, in case the answer is there already)
22:25 foster no, but it was only barely faster
22:25 foster it might not be worth reading through all the bug :)
22:26 foster whereas re-readdirp was fast, and that lead me to see that the first lookup after a readdirp wasn't handled via the cache
22:26 a2 ah, i meant to ask re-readdirp (typo on my part) and wondered why the answer was "no" :)
22:27 foster hehe
23:34 polfilm joined #gluster-dev

| Channels | #gluster-dev index | Today | | Search | Google Search | Plain-Text | summary