Ceph

h1. -1E - RADOS Gateway refactor into library internal APIs

<pre>

scuttlemonkey	ok, I just refreshed the feed	10:23

scuttlemonkey	it may reload once more	10:23

fghaas	joshd: yes indeed	10:23

*** xiaoxi has quit IRC	10:24

joshd	fghaas: lmb: I'm curious what async-replicated rbd would solve beyond sending rbd differentials between snapshots to another cluster	10:25

joshd	fghaas: lmb: if you send diffs frequently they start looking pretty similar	10:26

fghaas	joshd: well now that you mention it, shipping snapshot diffs might actually be better than streaming replication	10:28

scuttlemonkey	both video feeds should be live now	10:28

fghaas	because it would allow people to have a backup site trail x hours behind the live site on purpose	10:28

scuttlemonkey	we'll get the sessions started up again in another 2 mins or so	10:28

fghaas	joshd: is snapshot diff shipping coming for dumpling?	10:28

gregaf	wido: if you want to talk, the Google+ app supports hangouts and should open automatically if you go to the page in the browser; if you're just watching you can open it on the youtube site	10:29

wido	gregaf: I'm in already :)	10:29

joshd	fghaas: I've got good news for you then, diff import/export are already there in cuttlefish	10:29

gregaf	good good	10:30

fghaas	joshd: I do know that, but is there a daemon that will actually ship offsite or do people still have to handcraft that?	10:30

rturk	Hey all - #ceph-summit2 is for the second track	10:30

rturk	head over there if you're in track 2 :)	10:30

joshd	fghaas: shipping isn't handled by anything right now	10:31

rturk	and if you want to participate in the discussion, send scuttlemonkey a ping	10:31

fghaas	(joshd, in case you're wondering why lmb isn't responding, he stepped out for a bit, will probably throw his comments in when he returns)	10:31

*** nwat has joined #ceph-summit1	10:31

*** maclin has quit IRC	10:32

joshd	fghaas: perhaps the rgw-geo agent could do similar things for rbd, I haven't thought much about it yet	10:32

*** nwat has quit IRC	10:32

fghaas	joshd: ok. so what people would have to do is write a cron job that exports a diff, ships it, and then applies it on the other end? that's ok for testing, but people might dislike it for the real deal	10:32

paravoid	that's not accurate	10:33

paravoid	rgw isn't CGI, it's FastCGI	10:33

paravoid	and FastCGI isn't an old, deprecated way of running apps	10:33

paravoid	and fairly high performant, better than Python for sure	10:33

*** sjustlaptop has quit IRC	10:34

joshd	fghaas: I agree the management of diffs could certainly be made easier, it sounds like you might have some suggestions. I haven't thought about it much yet	10:34

joshd	fghaas: I just wanted the basic tools in place first for cuttlefish	10:35

*** Ryan_Lane has joined #ceph-summit1	10:36

fghaas	sounds a bit like the old Postgres WAL shipping. Which isn't quite as elegant as streaming, low-level RADOS replication, but probably a far better option for Glance/Cinder than an alternative where Glance images are stored as Swift objects via rgw (no objection to rgw at all, but in this case I'd greatly prefer native RBD)	10:37

fghaas	at any rate, joshd, our discussion is going OT for this channel, better switch this to #ceph :)	10:37

loicd	there is a project to move the swift API out of the swift implementation. That was discussed during the openstack summit but I don't have a link handy.	10:42

*** cdl has joined #ceph-summit1	10:42

*** jtran has left #ceph-summit1	10:47

gregaf	we already have so many logs for RGW, and coordinating consumers can be pretty annoying	10:47

scuttlemonkey	need an aggregator log tool like juju's debug-log or something :P	10:48

gregaf	sagewk was talking about keeping an event log and I don't think that's much less appealing to external services than a librgw that you can embed in your app to do all the real work ;)	10:49

loicd	joshd: do you remember about it ? ( the swift library ). Although it's already implemented, it might save time on upgrading and chasing the changes. But I don't really know anything there ;-)	10:52

*** Ryan_Lane has quit IRC	10:53

joshd	loicd: yeah, I talked to some swift folks about it, and first plan is pluggable fs backends for gluster/zfs on the backend servers, and later maybe pluggable higher-level backends (like ceph) beneath the proxy server later	10:54

paravoid	a low-level librgw library that doesn't include the HTTP bits sounds useful in general	10:54

gregaf	sagewk: not sure how behind this is your discussion, but the problem with get and put is all the streaming of those requests, right?	10:54

paravoid	hooking the logic directly into applications	10:54

sagewk	that's the devil in the details, yeah	10:54

paravoid	while also having radosgw on top of that	10:54

scuttlemonkey	(the youtube feed is about 1m behind the hangout)	10:54

paravoid	for example, we could use librgw PHP bindings for mediawiki to access directly the Ceph cluster without a gateway	10:55

joshd	loicd: it might be useful for keeping up with the apis, like new keystone etc, but probably a lot easier with librgw	10:55

loicd	it looks like it could leverage librgw indeed :-)	10:56

joshd	loicd: but the swift guys basically aren't going to do the work to make the proxy backends pluggable (and certainly not soon), so we'd need to do that as well if we wanted to use it	10:56

loicd	ah	10:56

paravoid	youtube is flaky dammit	10:58

scuttlemonkey	yeah, a bit	10:58

loicd	youtube	10:59

loicd	darn	10:59

paravoid	don't assume a MQ	11:01

paravoid	maybe a fork/exec or a unix socket (for performance)	11:01

gregaf	I'd feel better about a unix socket than a blind fork, yikes	11:04

paravoid	could the georeplication logs be reused for this?	11:04

gregaf	reused for which?	11:06

paravoid	the swift API has object expiration already	11:07

paravoid	http://docs.openstack.org/developer/swift/overview_expiring_objects.html	11:07

paravoid	X-Delete-At & X-Delete-After	11:07

paravoid	but this isn't LRU unfortunately	11:08

paravoid	I wouldn't know about S3, sorry :)	11:08

*** Aaron|away has joined #ceph-summit1	11:10

gregaf	wow, watching the etherpad edits in real-time and then hearing the conversation a minute later is very confusing	11:10

Aaron|away	swift temp urls :)	11:10

*** Aaron|away is now known as AaronSchulz	11:10

* loicd lost connectivity, changing location	11:10

paravoid	AaronSchulz: faster than you!	11:11

*** loicd has quit IRC	11:11

sagewk	what i stoken server?	11:11

michael_dreamhost	with s3 you pull acess tokens for mobile apps say	11:12

AaronSchulz	paravoid: I was listing but not on IRC	11:12

paravoid	gregaf: better than having the stream be interrupted every third minute or so	11:13

gregaf	so we're talking about token servers in terms of something like Keystone, right?	11:13

paravoid	at least I can keep up with the pad :)	11:13

joao	I propose waiting for everyone on the hangout to wait until able to sync IRC with video :p	11:13

</pre>