10:23:35 <bulde> #startmeeting
10:23:35 <zodbot> Meeting started Fri Jul 13 10:23:35 2012 UTC.  The chair is bulde. Information about MeetBot at http://wiki.debian.org/MeetBot.
10:23:35 <zodbot> Useful Commands: #action #agreed #halp #info #idea #link #topic.
10:23:43 <avati> bulde: need some changes for libgfapi
10:23:49 <bulde> sure
10:23:51 <avati> 1. we need syncop_flush()
10:24:26 <bulde> ...
10:25:24 <avati> 2. finish the remaining fops in glfs.h
10:25:51 <avati> 3. syncop does not have xdata params for all calls
10:25:56 <avati> (3. is not urgent)
10:26:23 <bulde> for 3, we need both xdata_req, xdata_rsp :-)
10:26:44 <avati> 4. protocol/client waits 3 seconds (reconnect timeout) between portmap disconnect and brick connect
10:27:06 <avati> this is too expensive for libgfapi
10:27:20 <bulde> hmm, want it configurable?
10:27:22 <avati> i.e, delay is very visible.. for mounts it is a one time mount-time operation
10:28:01 <avati> configurable alla, there should not be reconnect thread based connection to the brick if the recent disconnection was from portmap
10:28:26 <avati> in disconnect handler we should somehow detect this can call connect immediately
10:28:52 <avati> the problem is, we do rpc_transport_disconnect() ; rpc_clnt_reconnect();
10:29:15 <avati> rpc_transport_disconnect(0 only does the shutdown and completely disconnect happens in pollerr
10:29:46 <avati> but before that rpc_clnt_reconnect() is attempted and it cannot initiate reconnection right there as cleanup is not complete.. so it just sets up timer for 3 secs later
10:30:24 <avati> so that part needs to be sorted out
10:30:37 <bulde> hmm
10:31:01 <avati> probably in pollerr handler in rpc-clnt, we can detect the 'recent portmap initiated disconnect' and call reconnect once more
10:31:18 <avati> (and clear the 'recent portmap disconnect' flag)
10:31:37 <bulde> ok
10:31:54 <bulde> is #4 urgent? (ie, before 20th thing?)
10:32:50 <avati> it will be nice to have.. the delay is annoying, but it is not urgent.. everything still works without breakage
10:33:13 <bulde> good to know, i need to prioritize with other tasks, so asked
10:34:34 <avati> that's about it for now
10:34:58 <bulde> cool, #1 can be done right away
10:35:10 <avati> i've finished the basic gfapi.. the example code works fine. there are still more changes required, will be doing that over the next few days
10:35:33 <avati> with 1 and 2 we should be mostly good for qemu consumption
10:35:54 <bharata> avati, if you have anything in sharable state, I am willing to try out
10:36:12 <avati> bharata: look at the recent patches on gerrit
10:36:33 <avati> all the fops you require are not there, but you can have a look at api/example/*
10:36:44 <avati> and api/src/glfs.h
10:37:11 <bharata> avati, ok so these patches are for reviving libglusterfsclient ?
10:37:19 <bulde> avati: #2 will be taken out only next week
10:37:21 <avati> yeah, it is now called libgfapi
10:37:51 <bharata> avati, ok
10:37:59 <bulde> #3 will be started parallelly next week, but completion time not promised...
10:38:01 <avati> bharata: your feedback on glfs.h and example/* will be appreciated
10:38:21 <bharata> avati, ok, I will take a look and let you know
10:38:36 <bulde> #4, needs one more round of review, and can be done
10:39:00 <avati> bulde: that should be ok
10:39:11 <bulde> cool
10:39:25 <bulde> #endmeeting