Tuesday, 2021-06-01

*** martinkennelly has quit IRC00:09
*** zenkuro has quit IRC00:42
*** whoami-rajat has quit IRC01:48
opendevreviewXuan Yandong proposed openstack/cinder master: Add timeout when execute ssh cmd.  https://review.opendev.org/c/openstack/cinder/+/79326402:57
opendevreviewXuan Yandong proposed openstack/cinder master: Add timeout when execute ssh cmd.  https://review.opendev.org/c/openstack/cinder/+/79326403:03
*** carloss has quit IRC03:21
*** ricolin_ has joined #openstack-cinder04:32
*** vishalmanchanda has joined #openstack-cinder04:35
*** ricolin has quit IRC04:36
*** ricolin_ is now known as ricolin04:36
*** abhishekk has joined #openstack-cinder04:42
opendevreviewTushar Trambak Gite proposed openstack/cinderlib master: Moving to OFTC  https://review.opendev.org/c/openstack/cinderlib/+/79388104:45
opendevreviewAdrian Smith proposed openstack/cinder master: allow project_id/user_id in create_snapshot filter  https://review.opendev.org/c/openstack/cinder/+/79388505:32
*** josecastroleon has joined #openstack-cinder06:26
*** hemna has quit IRC07:01
*** hemna has joined #openstack-cinder07:01
*** tosky has joined #openstack-cinder07:48
*** tosky has quit IRC07:48
*** martinkennelly has joined #openstack-cinder07:48
*** jrosser has joined #openstack-cinder07:52
*** tosky has joined #openstack-cinder08:08
*** abhishekk has quit IRC08:35
*** abhishekk has joined #openstack-cinder08:35
*** stephenfin has quit IRC08:49
*** martinkennelly has quit IRC08:50
*** martinkennelly has joined #openstack-cinder08:50
*** abhishekk has quit IRC08:56
opendevreviewGirish Chilukuri proposed openstack/cinder master: [SVF]:Fix error message for volume with snapshot  https://review.opendev.org/c/openstack/cinder/+/78726209:12
*** josecastroleon has quit IRC09:14
*** josecastroleon has joined #openstack-cinder09:15
*** stephenfin has joined #openstack-cinder09:15
opendevreviewGirish Chilukuri proposed openstack/cinder master: [SVF]:Bulk create Hyperswap volume is failing.  https://review.opendev.org/c/openstack/cinder/+/77919409:24
opendevreviewkatari manoj kumar proposed openstack/cinder master: [SVF]:Fix add volumes to GMCV group  https://review.opendev.org/c/openstack/cinder/+/78409209:27
opendevreviewsaikrishnagk proposed openstack/cinder master: [SVF] Fixed add volumes to clone group issue  https://review.opendev.org/c/openstack/cinder/+/78477909:30
*** amotoki has joined #openstack-cinder09:36
opendevreviewAmar Rajeev Singuluri proposed openstack/cinder master: [SVF]: Fix mkvdisk volume name length issue  https://review.opendev.org/c/openstack/cinder/+/79091009:46
*** zenkuro has joined #openstack-cinder10:00
opendevreviewAmar Rajeev Singuluri proposed openstack/cinder master: [SVF]: Fix extend issue for a clone of rep-volume  https://review.opendev.org/c/openstack/cinder/+/78240510:04
opendevreviewHelen Walsh proposed openstack/cinder master: PowerMax Driver - Re-use existing initiator group/host.  https://review.opendev.org/c/openstack/cinder/+/78776410:19
*** abhishekk has joined #openstack-cinder10:45
*** carloss has joined #openstack-cinder10:54
*** Luzi has joined #openstack-cinder10:57
hemnagood morning11:29
opendevreviewMerged openstack/cinderlib master: Moving to OFTC  https://review.opendev.org/c/openstack/cinderlib/+/79388111:38
opendevreviewVictoria Martinez de la Cruz proposed openstack/devstack-plugin-ceph master: DNM  https://review.opendev.org/c/openstack/devstack-plugin-ceph/+/79395211:39
*** whoami-rajat has joined #openstack-cinder11:55
whoami-rajatgeguileo: hey i did the suggested changes in same patch, can you take another look? https://review.opendev.org/c/openstack/os-brick/+/78928911:57
opendevreviewRajat Dhasmana proposed openstack/cinder master: Add user messages for volume operations  https://review.opendev.org/c/openstack/cinder/+/79351511:57
*** dviroel has joined #openstack-cinder11:57
*** fabiooliveira has joined #openstack-cinder12:03
*** abhishekk has quit IRC12:04
opendevreviewGirish Chilukuri proposed openstack/cinder master: [SVF]:Storwize HyperSwap snapshot clone is failing  https://review.opendev.org/c/openstack/cinder/+/77630712:05
opendevreviewGirish Chilukuri proposed openstack/cinder master: [SVF]:Fix error message for volume with snapshot  https://review.opendev.org/c/openstack/cinder/+/78726212:12
opendevreviewGirish Chilukuri proposed openstack/cinder master: [SVF]:Retype in-use hyperswap volume  https://review.opendev.org/c/openstack/cinder/+/78239012:29
*** rosmaita has joined #openstack-cinder12:33
*** tbarron has quit IRC12:52
*** eharney has joined #openstack-cinder12:53
*** enriquetaso has joined #openstack-cinder12:53
*** tbarron has joined #openstack-cinder13:07
*** Luzi has quit IRC13:21
*** abhishekk has joined #openstack-cinder13:35
*** hemna has quit IRC13:53
*** hemna has joined #openstack-cinder13:54
rosmaitahemna: thanks for keeping an eye on that ceph-iscsi-CI patch ... seems like everything *except* the ceph-iscsi job has been failing13:59
hemnahoping to get that to merge13:59
*** _pewp_ has joined #openstack-cinder14:01
_pewp_hemna (✧∇✧)╯14:04
hemnarosmaita, geguileo,  so the discussion the other day about the initialize_connection rpc timeouts/failures.   The idea was to add a microversion and make initialize_connection async.   The problem is, cinder doesn't know who to call back.   Since the cinderclient can call initialize_connection (for the cinderclient os brick extension), cinder14:07
hemnavolume manager can't assume that we should call nova back after the driver is done w/ initialize_connection.14:07
geguileohemna: we can make the callback optional (only working for nova) and when there is no callback the client would do polling and then get information from the attachment14:08
rosmaitahemna: what geguileo said14:09
hemnaso I think making it a callback is going to be really hard on the nova side of things14:09
hemnaas nova would have to somehow 'pause' the attachment process and wait14:10
geguileohemna: yes, that is the problem14:13
geguileothe Nova side is the hard part14:13
rosmaitaguess we need to talk to the nova team about what ideas they have to address this14:14
hemnamy nova guy suggested polling from nova's side14:15
hemnabut I'm not sure really what to poll14:15
geguileohemna: alternatively we could do the first Nova implementation using polling, which simplifies the nova implementation14:15
geguileohemna: the attachment id14:15
hemnalike maybe repeatedly call initialize_connection and if cinder responds with 'not done yet', keep polling14:15
geguileohemna: it should have no connection info if it's still doing the export/map14:15
rosmaitaso, basically same thing someone would do using the cinderclient14:15
geguileohemna: before the initialize_connection call Nova already has an attachment id14:16
hemnathe reserve call creates the attachment and replies with the attachment id ?14:16
geguileohemna: and then calls cinder with that attachment id and the connector properties to do the attachment14:16
geguileohemna: I believe there is no longer a reserve call on the new attach mechanism14:16
geguileowe have a create attachment without the connector properties14:17
hemnaI think there is, to get cinder to put the volume in attaching state early on14:17
hemnato prevent someone else requesting an attach14:17
geguileowhich makes the volume go to reserved or whatever state its used for locking14:17
geguileohemna: there is no reserve, there is create attachment without connector properties14:17
geguileothen you get an attachment id14:17
geguileoand then there is an update to that attachment id with the connector properties14:18
geguileothat call would be the one that should be changed to support async14:18
geguileoand then nova would do the polling14:18
hemnaattachment_create (from nova side looks like).  ok I see the nova/api.py call now, that superceeded the reserve_volume w/ the newer microversion14:19
geguileoor if we want to make it even better/harder the async is only triggered if it takes longer than 45 seconds or whatever14:19
hemnawell the problem is on initialize_connection on cinder's side.   That rpc call is what times out14:22
hemnadoes attachment_update return anything?14:22
hemnacould we make initialize_connection async and then return immediately14:23
hemnaand the nova could call attachment_get as the polling mechanism14:23
hemnacinder volume manager would stuff the connection_info from the driver's initialize_connection into the attachment entry14:23
hemnathe next call to attachment_get by nova could get it14:24
*** gouthamr_ has joined #openstack-cinder14:34
*** gouthamr has joined #openstack-cinder14:40
hemnawould just have to modify the volume manager to stuff the driver's connection info into the attachment db record......but that also includes the iscsi CHAP creds as well14:41
*** abishop has quit IRC14:43
*** opendevreview has quit IRC14:44
*** abishop has joined #openstack-cinder14:51
hemnaok looking at this a bit more, it seems that the volume manager's attachment_update sort of does this, but it also calls the driver's initialize_connection every time too14:54
hemnaand I see a bug in there.14:54
hemnarequire_driver_initialized is called after _connection_create() which calls the driver's initialize_connection14:55
hemnashould we talk about removing the legacy volume attach process in nova?15:02
hemnanova shouldn't be calling initialize_connection directly with the process we implemented ages ago now15:02
*** dviroel is now known as dviroel|lunch15:04
hemnanova's legacy_volume_attach was created by john back in 201715:04
*** luksky has joined #openstack-cinder15:09
*** gouthamr_ has quit IRC15:12
*** opendevreview has joined #openstack-cinder15:13
opendevreviewRajat Dhasmana proposed openstack/cinder master: Add user messages for volume operations  https://review.opendev.org/c/openstack/cinder/+/79351515:13
*** gouthamr has quit IRC15:13
*** gouthamr has joined #openstack-cinder15:14
hemnaso if attachment_create is called w/ a connector (which happens during nova's live migration), then initialize_connection is called as well.15:15
*** gouthamr is now known as identify15:18
*** gouthamr has joined #openstack-cinder15:19
*** identify has quit IRC15:20
hemnaok so the connection_info is saved in the db record currently by cinder api after the attachment_update rpc into the volume manager returns15:26
*** abishop has quit IRC15:28
*** abishop has joined #openstack-cinder15:30
opendevreviewWalt proposed openstack/cinder master: Ensure driver is initialized before attach  https://review.opendev.org/c/openstack/cinder/+/79400115:56
hemna^ moved require_driver_initialized prior to allow a call to initialize_connection15:57
*** dviroel|lunch is now known as dviroel16:11
*** vishalmanchanda has quit IRC17:10
*** abhishekk has quit IRC17:22
*** zenkuro has quit IRC17:28
hemnaother weirdness with attachment_update......17:38
hemnalooks like we are doing multiple DB update calls instead of just 117:41
hemnatwice we write to the db for the attachment record17:41
hemnacinder api does it here: https://github.com/openstack/cinder/blob/master/cinder/volume/api.py#L2246-L225217:41
hemnabut the volume manager also does it here:  https://github.com/openstack/cinder/blob/39762e408ed2aaafb873cdcdbde90c3700792bbf/cinder/volume/manager.py#L468917:42
hemnathe volume manager has the connection_info, but doesn't write it to the db entry.  it waits for cinder api to do it later ?17:42
opendevreviewFernando Ferraz Silva proposed openstack/cinder master: [DNM] testing NetApp CI  https://review.opendev.org/c/openstack/cinder/+/79402118:09
*** sfernand has joined #openstack-cinder18:11
*** _pewp_ has quit IRC18:11
*** luksky has quit IRC21:59
opendevreviewMerged openstack/cinder master: Update IRC info  https://review.opendev.org/c/openstack/cinder/+/79323022:25
*** tosky has quit IRC23:11
*** martinkennelly has quit IRC23:52

Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!