Jump to: navigation, search


Revision as of 15:27, 6 April 2016 by Duncan Thomas (talk | contribs) (Next meeting)

Weekly Cinder team meeting


If you're interested in Cinder or Block Storage in general for OpenStack, we have a weekly meetings in #openstack-meeting, on Wednesdays at 16:00 UTC. Please feel free to add items to the agenda below. NOTE: When adding topics please include your IRC name so we know who's topic it is and how to get more info.

Add your IRC nick to this list to be pinged at the start of the meeting: dulek duncant eharney geguileo winston-d e0ne jungleboyj jgriffith thingee smcginnis hemna xyang tbarron scottda erlon rhedlind jbernard _alastor_ vincent_hou kmartin patrickeast sheel dongwenjuan JaniceLee cFouts Thelo vivekd adrianofr mtanino yuriy_n17 karlamrhein diablo_rojo jay.xu jgregor baumann

Next meeting

NOTE: Include your IRC nickname with your agenda items so that you can be called upon in the meeting and arrive at the meeting promptly if placing items in agenda. Please have any supporting links ready. You might want to put this on your calendar if you are adding items.

April 6, 2016 16:00 UTC

Previous meetings

March 30, 2016 16:00 UTC

March 23, 2016 16:00 UTC

  • Some operation passes even if volume service is disabled(sheel) - postoponing this discussion, will discuss later after release

March 16, 2016 16:00 UTC

March 9, 2016 16:00 UTC

March 2, 2016 16:00 UTC

February 24, 2016 16:00 UTC

February 17, 2016 16:00 UTC

February 10, 2016 16:00 UTC

  • Release Status
  • Transition to LIO (jbernard)
    • Does anyone oppose this?
    • What exactly must be done to make this transition?
      • The only remaining hurdle here AFAIK is the question of how to migrate existing installs smoothly. Perhaps do something like nas_secure_file_operations that defaults to LIO only on deployments that have no existing volumes? (eharney)
  • I'd like to propose bit-sized projects for Outreachy (jbernard)
    • If anyone happens to have project ideas, please let me know.
  • Support for 4-byte unicode for naming volume, snapshot, CG, volume_type etc.(sheel)
    • MySQL's character set which is utf8 does not support 4 byte characters.
    • Should we prevent this issue(by prohibiting user for not using 4 byte characters) or fix it?
      • We have to fix it anyway to not blow up for deployments not using utf8mb4. This approach does not prevent use on databases that support it, but also cleanly handles the errors. Why not just fix it this way? https://review.openstack.org/#/c/266036/ . (eharney)
    • Kindly refer: https://bugs.launchpad.net/cinder/+bug/1542413
  • New behavior of DocImpact tag (smcginnis)
  • Deprecating Cinder client (thingee)
  • Status of api-microversions patch set (scottda)
  • Job distribution for HA A/A (geguileo)
    • Do we really need/want to be able to disable specific nodes of a host group (cluster)?
    • Should this disabling have the same effect as disabling has now (only prevent scheduler RPC calls)? Or should it also prevent the node from processing RPC calls coming from the API nodes?
    • Updated specs with proposed disabling mechanism (stop all incoming RPC calls to the node): https://review.openstack.org/232595

February 3, 2016 16:00 UTC

January 20, 2016 16:00 UTC

January 13, 2016 16:00 UTC

  • Nested Quota enforcement issue (mc_nair)
    • While looking into https://bugs.launchpad.net/cinder/+bug/1531502, noticed that we can't actually enforce child default quota values because we can't get to the parent_project_id as part of the create volume flow (just one example)
    • Talked with raildo and we were thinking of listening for the keystone project create event notification and explicitly creating a cinder quota for the new project at that point
    • Anyfeedback on the issue / suggested fix would be appreciated
  • Summary Messages to be displayed for success/failure in cinder operations(sheel)

January 6, 2016 16:00 UTC

December 23, 2015 16:00 UTC

December 16, 2015 16:00 UTC

  • Storing dicts on driver 'provider_location' (erlon, dulek)
    • Currently every driver has its own way to encode provider_location info. Some also use volume_metadata to store info. Is it possible to have a standard way to store that info?
  • python-cinderclient: tempest vs functional tests (e0ne)
    • we run tempest-dsvm-neutron-src-python-cinderclient job on each commit
    • only few tests from ~1500 uses cinderclient
      • need to check what projects except nova uses cinderclient
      • we don't run heat tests with tempest-dsvm-neutron-src-python-cinderclient job
    • move such tests to functional tests job will save 30 for getting CI results

December 9, 2015 16:00 UTC

December 2, 2015 16:00 UTC

November 25, 2015 16:00 UTC

  • CI Documentation (smcginnis)

November 18, 2015 16:00 UTC

  • OSLO incubator going away (DuncanT)
    • I think this is great news, BUT we need to make sure people are aware, stop rejecting fixes, and probably rename that bit of the code tree
  • Adding extra data to event notifications for searchlight to consume (DuncanT)
    • Currently the only way to get the data searchlight needs is to hit the API for an update upon every event
      • This doubles the load on the API and the db, for no really good reason
    • Anybody see a good reason not to include more info in the event? Either the full volume/snap/etc record, or at least what changed
    • Also need to add an event for metadata edits so searchlight can index them
  • Looping on failed driver init (DuncanT)
    • No point continuing as we do now
    • Failure often seems to be being unable to talk to the array
    • Sleep for a while and retry? Nothing better to do in that time, and it means no manual intervention is required if you happen to reboot the array controller and restart c-vol at the same time (or they race on power on), since the retry will succeed

November 11, 2015 16:00 UTC

November 4, 2015 16:00 UTC

October 21st, 2015 16:00 UTC

  • Rolling upgrades (dulek)
    • All the cool kids have it, what about us? - let's have initial discussion on approach in Mitaka.
    • http://lists.openstack.org/pipermail/openstack-dev/2015-October/077411.html
      • Are we going to adopt online schema changes? Or simply disallow "contracting" migrations?
      • Should we go on and demand RPC backwards compatibility on reviews?
      • thangp's Volume Object patches are in review. Are we going to merge the first big one or should we insist on splitting it?
      • Should we move dicts passed over RPC (request_spec, filter_properties, capabilities) to versioned objects?

October 14th, 2015 16:00 UTC

  • RemoteFS drivers refactoring (dguryanov)
  • Need Liberty release notes (smcginnis)
  • Finalize Summit planning (smcginnis)
  • Mitaka Midcycle planning (smcginnis)

October 7th, 2015 16:00 UTC

September 30'th, 2015 16:00 UTC

September 23rd, 2015 16:00 UTC

September 16'th, 2015 16:00 UTC

  • Nova summit session for Cinder issues (scottda)

September 9'th, 2015 16:00 UTC

  • No Agenda

September 2'nd, 2015 16:00 UTC

August 26th, 2015 16:00 UTC

  • Get Volume Driver Capabilities patch (mtanino)
    • Feature: https://review.openstack.org/#/c/201243/24
    • Merged SPEC: https://review.openstack.org/#/c/183947/
    • Based on the discussion at IRC with jgriffith and thingee, we changed policy of prefix for properties.
      • Cinder well defined key should be "bare"(without prefix).
      • Vendor unique properties should have vendor name as a prefix(like Hitachi_minIOPS).
    • I got a proposal from Gorka to use `vendor_name` in current volume_stats for vendor name prefix of vendor unique keys.
      • Proposal: Use `vendor_name` in volume_stats OR use anything the driver wants as a prefix.
      • In both cases, if vendor name includes colon, we should replace colon with _ (because colon is separator.)
      • Separator is colon. Hitachi:minIOPS, Hewlett-Packard:minIOPS or hp3par:minIOPS, SolidFire Inc:minIOPS or SolidFire:minIOPS
      • Here is vendor_name list from drivers. http://paste.openstack.org/show/427150/
    • SPEC should be updated via https://review.openstack.org/#/c/216394/2
  • Feature freeze status for Liberty (jgriffith)
  • Migration progress report (vincent_hou)
    • https://review.openstack.org/#/c/204953/
    • Concerns: Adding complexities to the Base Volume class. A driver is not supposed to deal with local processes and fetching process IDs and so on. Question: Where can be a more appropriate place? Is the code movement of copy_volume_data in https://review.openstack.org/#/c/187270/ the direction we need to head for?
    • How the progress report is implemented:
      • For host assisted migration, getting the process id for "dd" command and checking the process progress.
      • For driver specific migration, the driver should report the progress.
      • For file IO migration, being implemented in https://review.openstack.org/#/c/187270/, getting relative handle position and comparing to the volume size.

August 19th, 2015 16:00 UTC

August 12th, 2015 16:00 UTC

  • (mriedem) How to translate specific exceptions in the client? https://review.openstack.org/#/c/209150/
    • Nova needs to be able to distinguish the error when there are no volume attachments from other 400s during instance terminate.
    • Can we key off a regex of the error message from Cinder? How do we keep that error message from changing in Cinder and breaking the client translation?
  • (Thingee) Modify volume image metadata

August 5th, 2015 16:00 UTC Cancelled due to Cinder Midcycle Sprint. July 29th, 2015 16:00 UTC

  • Dinner Thursday Night ... 6:30 at Austin's (jungleboyj)
    • Does the time seem ok?
    • Need to get a feel for the number of people coming ... want to check given it is the last night.
    • Depending on number may need to do a reduced menu. Any special needs?
  • Cinder FFE request(s?) procedure (kaisers)
    • Is there a FFE decision date planned so far?
    • I'm asking because I'd like to know how to proceed on the FFE for re-add of the Quobyte driver (Original FFE and follow ups)
    • QuobyteCI has been reporting for some time now, the bugs it hit have been handled. Current logs are available showing a range of successes and a few issues it found (here, here and here).
    • Current CI state is updated at the Third party CI wiki page.
  • Unapproved BP queue
    • From last meeting: AGREED: July 31st spec/bp freeze and code feature freeze first week of september for Liberty (thingee, 17:00:16)
    • We have a number of unapproved BPs, would like to avoid last minute drama (ameade, tbarron)
  • Unit tests for John's replication patch (jungleboyj)
    • https://review.openstack.org/#/c/200269/
    • Given our interest in helping to move forward the base replication patch wonder if Vincent Hou could help add in unit tests?
    • If you are open to the idea how should we tackle? Dependent patch with unit tests?

July 22nd, 2015 16:00 UTC

  • Replication: (jungleboyj)
    • What are the expected deadlines for base functionality and subsequently drivers?
    • IBM currently has a patch out for Storwize: https://review.openstack.org/202940
      • Trying to work the process of implementing this in the community to uncover anything that might be missing in the base functionality.
      • Some review comments in John's code. Need to get feedback from John--- TaoBai
  • Config Generator updates: (jungleboyj, kjnelson, jgregor)
  • Liberty Deadlines
    • Led by: thingee

July 15th, 2015 16:00 UTC

July 1st, 2015 16:00 UTC

June 24th, 2015 16:00 UTC

June 17th, 2015 16:00 UTC

June 10th, 2015 16:00 UTC

June 3rd, 2015 16:00 UTC

May 27th, 2015 16:00 UTC

  • Live-Migration changes
  • Discuss Cinder internal tenant
    • Led by: patrickeast
    • Some info about them proposed in this spec: https://review.openstack.org/#/c/182520/
    • Looking for some consensus that we should move forward with it for the various use cases (temp volumes, temp snapshots, image cache volumes, etc)
    • This patch is introducing a hidden flag: https://review.openstack.org/#/c/185857/ . Let's get consensus on whether to use a hidden flag or a special Cinder tenant. (xyang)
  • Thin provisioning volume capacity consideration in filter scheduler
    • Led by: xyang
    • https://review.openstack.org/#/c/185764/2/cinder/scheduler/filters/capacity_filter.py
    • Currently it assumes the capacity of a new thinly provisioned volume will be consumed right away (the conservative approach).
    • The proposed patch added a flag to allow user to decide whether to assume the new thinly provisioned lun will be consumed. If True, we deduct the capacity (existing behavior); If False, we don't deduct the capacity.
    • There were many requests (after the patch was merged) not to deduct the capacity. So we need to take care of this.
    • Any one is interested in preserving the existing behavior?
  • Downgrades in new db migrations (DuncanT)
    • Several new reviews seem to be missing them - intentional or not?
    • Are we requiring make_compatible cinder-object methods for all migrations yet?

May 13th, 2015 16:00 UTC

May 6th, 2015 16:00 UTC No Agenda, no meeting.

April 29th, 2015 16:00 UTC

April 22th, 2015 16:00 UTC

April 15th, 2015 16:00 UTC

  • active-active-[active...] cinder volume services
    • Led by: DuncanT
    • What bugs are people aware of?

April 8th, 2015 16:00 UTC

April 1st, 2015 16:00 UTC

March 25th, 2015 16:00 UTC

March 18th, 2015 16:00 UTC

March 11th, 2015 16:00 UTC

  • Cinder supports several target objects now for the lvm driver. Where are those different targets getting CI'd ?
    • Submitted by: Walt (hemna)
    • Interested Partiesanish
    • when new targets get submitted, how can we ensure that they actually work. We need to think about integrating the different target objects with lvm for tempest tests

March 4th, 2015 16:00 UTC

telnet 29418

You should get this output if everything works:


Connected to

Escape character is '^]'.

  • Reminder about daylight savings time (in the USA) and how it will affect next week's meeting.
  • Move to oslo_log is coming!
    • Led by: jungleboyj, DuncanT, e0ne
    • https://review.openstack.org/#/c/157441
    • Need to get the patch in for Kilo. Planning to do it as soon as the March 10th deadline passes.
    • Unfortunately will cause may merge conflicts.
  • Replication follow up from last meeting
    • Led by: jungleboyj
    • After further internal discussion we have decided to hold off on the GPFS and Storwize patches for replication.
    • Want to use the time to focus on getting the V2 design put together and moved forward as early as possible in L.

Feb 25th, 2015 16:00 UTC

Feb 18th, 2015 16:00 UTC

Feb 11th, 2015 16:00 UTC

Feb 4th, 2015 16:00 UTC

Jan 21st, 2015 16:00 UTC

  • Sharing volume snapshots
  • Additional requirements with Third Party CI
    • Led by: Mike Perez (thingee)
    • How long can a system be down? How many warnings to the maintainer before the driver is considered inactive?
    • What response time do we expect?
  • Target drivers - DRBD transport layer Nova
    • Led by: Philipp Marek (flip214)
    • Currently the Nova node that shall use the storage isn't available on create_export. Should it be? Is that information available at this point?
    • There's already some keys in the volume parameter: "'attach_time': None, 'attached_host': None, "
    • provider_location is too small to store all storage nodes and other data; enlarge to 4kB?
    • cinder/brick/initiator/connector.py factory() used, AND cinder/volume/target/drbd.py ??
    • Only provider_location as a string gets passed around, no structured data? (Passing to Nova, too)
  • Persistent TaskFlow workflows

Jan 14th, 2015 16:00 UTC

Jan 7th, 2015 16:00 UTC

Dec 17th, 2014 16:00 UTC

Dec 10th, 2014 16:00 UTC

Dec 3rd, 2014 16:00 UTC

Nov 26th, 2014 16:00 UTC

Nov 19th, 2014 16:00 UTC

Nov 12th, 2014 16:00 UTC

  • Cancelled since the summit just happened last week. We'll meet again November 19th.

Nov 5th, 2014

  • No meeting - Openstack Summit in Paris

Oct 29th, 2014 16:00 UTC

Oct 22nd, 2014 16:00 UTC

  • Discuss kilo summit design session proposals for the cinder track
  • Led by: thingee

Oct 14th, 2014 16:00 UTC

Oct 8th, 2014 16:00 UTC

Oct 1st, 2014 16:00 UTC

Sep 24th, 2014 16:00 UTC

  • Remove XenAPI driver (https://review.openstack.org/#/c/123162/) (DuncanT)
    • Should we go through deprecation?
    • There's an HP MSA driver that is unmaintained, unCIed and as far as we know unused, so we need a policy
      • Might have found a maintainer (DuncanT)
  • Please remember we're coming up to RC so lots of testing and bug fixing please, not just getting stuff ready for Kilo opening!

Sep 17th, 2014 16:00 UTC

Sep 10th, 2014 16:00 UTC

Sep 3rd, 2014 16:00 UTC

Aug 27th, 2014 16:00 UTC (Some problems with the bot, so minutes are incomplete)

  • https://review.openstack.org/111659 - DuncanT
    • Should cinder be doing this, or should cinder remain an API that does exactly what it says on the tin and let neutron/nova be 'smarter' if it wants?
    • Should we be changing API behaviour? Should we be changing it just before milestone 3?

Aug 20th, 2014 16:00 UTC

Aug 13th, 2014 16:00 UTC


Aug 6th, 2014 16:00 UTC

July 30'th, 2014 16:00 UTC

July 23th, 2014 16:00 UTC

July 16th, 2014 16:00 UTC

  • Putting the fun back into cinder development.
    • There's been a mailing list thread recently about how nit-picky reviews are getting about typos, white space and the like, and how it is a motivation killer. I'm inclinded to agree - the formatting of the doc strings, fullstops at the end of comments etc doesn't actually improve the code much at all, and getting a -1 for it is a buzz kill of the highest order. Should we leave that sort of thing to the gate, and say that if there is no hacking check for it then it isn't important in general? (DuncanT)
  • Log translations https://review.openstack.org/#/c/105665/ is still stuck - any thoughts? Options I can see: (DuncanT)
    • A better technical solution - should be possible where the message format is not expanded outside the logging call i.e.:


    LOG.warning("The flubigar id %d exploded messily", flu_id)

Not ok:

    msg = _("The flubigar id %d exploded messily") % flu_id
    • We don't break up our message categories
      • This makes life harder for the translation team, makes us inconsistent with Openstack in general but keeps the code from descending into ugliness
    • Related discussion on enabling translation (jungleboyj):
  • 3rd Party CI (jungleboyj):
    • Clarification on when drivers are going to be removed.

Previous meetings

July 9, 2014 16:00 UTC

July 2nd, 2014 16:00 UTC

June 25th, 2014 16:00 UTC

June 18th, 2014 16:00 UTC

  • It's review day !?! [jdg]
  • Mid cycle meetup plans/updates [jdg]
  • Separation of Connectors from Driver/Device Interface (status update) [jdg]
  • Updates on 3'rd party CI [jdg]
  • Things we need to decide upon (not today, but do your homework for next week)
    • Software Define Storage layers/drivers
    • Pools implementation

June 11th, 2014 16:00 UTC

June 4th, 2014 16:00 UTC

May 28th, 2014 16:00 UTC

  • 3rd Party CI (jungleboyj)
    • What tempest test cases to run?
    • iSCSI only? What about for FC only drivers then?
    • Progress on where to record results?
  • SSH host keys (jungleboyj)
  • Dynamic multi-pools (navneet)
  • cinder-specs (jgriffith)
    • Specs repo is live
    • Process
    • Reviews

May 21st, 2014 16:00 UTC

  • Consistency Groups (xyang)
    • A few people have concerns on the restriction of one volume type per CG. Should we allow one CG to have multiple volume types on the same backend? Let's discuss about it.
  • Third-Party CI (jgriffith)
    • Who's started, who's planning to and how can we help support each other to get this going smoothly
  • Moving GlusterFS snapshot code into the NFS RemoteFs driver (mberlin)
    • The GlusterFS snapshot code using qcow2 snapshots is useful for all file based storage systems. I would volunteer to move the GlusterFS snapshot code into the general RemoteFs driver - making it easier to get our driver accepted ;-)
    • Eric Harney is fine with this and planned to do this for Juno anyway (see his blueprint). I've put it on the agenda to make sure others also agree with this approach.

May 7th, 2014 16:00 UTC

April 16th, 2014 16:00 UTC

  • Release Status
  • Summit Session Updates
  • Next Stop ATL!!!
  • Cinder resource status - thingee

April 9th, 2014 16:00 UTC (Agenda entered retrospectively)

  • Cinder Spec (jgriffith)
    • Just a heads up that cinder blueprints will move to a gerrit based process shortly, a la nova. Details and wiki entry to follow.
  • RC2 status (jgriffith)
    • Just after cutting RC2, a bunch of bugs
  • Testing RC code (jgriffith)
    • Get on it, folks!
    • Looks like theres some serious, intermittent performance issues in the API somewhere...

April 2, 2014 16:00 UTC _Meeting cancelled and summary discussion held on #openstack-cinder_

  • Release status and bugs
  • -2s left on reviews from before Junos opened - please check if they are still valid
   - https://review.openstack.org/#/c/73446/ (JGriffith)
   - https://review.openstack.org/#/c/80550/ (JBryant)
   - https://review.openstack.org/#/c/82100/ (Avishay)
   - https://review.openstack.org/#/c/74158/ (Avishay)
   - + a whole bunch of stable branch stuff

Previous meetings

Mar 26, 2014 16:00 UTC

  • RC1 updates (jgriffith)
  • Design Summit Sessions (jgriffith)


Mar 19, 2014 16:00 UTC

  • ProphetStor Driver Exception request for Icehouse (jgriffith)
  • Bug status/updates (jgriffith)
  • What we should be punting to Juno (aka immediate -2 in Gerrit) (jgriffith)
  • Continuous Integration for Cinder Certification (jungleboyj)

Mar 12, 2014 16:00 UTC

  • Cancelled due to nothing on the agenda. Ad-hoc discussion on #openstack-cinder instead

Mar 5, 2014 16:00 UTC

Feb 19, 2014 16:00 UTC

Feb 5, 2014 16:00 UTC

  • I3 Status check/updates
  • Cert test
  • Multiple pools per backend (bswartz)

Jan 8, 2014 16:00 UTC

  • I2 is just around the corner, blueprint updates
  • Alternating meeting time proposal, results on feedback
  • Driver cert test, it's there... use it
  • Prioritizing patches and reviews

December 18, 2013 16:00 UTC

December 11, 2013 16:00 UTC

December 4, 2013 16:00 UTC

November 27, 2013 16:00 UTC

November 20, 2013 16:00 UTC

  • I-1 scheduling - JGriffith

November 13, 2013 16:00 UTC

October 30, 2013 16:00 UTC

October 23, 2013 16:00 UTC

October 2, 2013 16:00 UTC

  • What's still broken in Havana
  • Backups and multibackend (https://code.launchpad.net/bugs/1228223): Fix committed
  • Configuration - Global CONF settings in brick don't belong, and a number of them break multi-backend (Bug #1230066): Fix committed
  • TaskFlow retry mechanism - The majority felt this should be left as a white list, but no work has been done to fix it so we still have ugly failures/roll-backs (3 bugs logged here): ???
  • Quotas - Don't know that anybody has gotten to the bottom of the quota syncing issue (Bug #1202896): Still open
  • iSCSI Target creation failures - This was thought to have been fixed but showed up last night (Bug #1223469): Probable fix committed
  • Summary of gate issue pertaining to Cinder can be viewed here: http://paste.openstack.org/show/47798/
  • Moving to taskflow - avishay

Sept 25, 2013, 16:00 UTC

  • PTL nomination process is open until the 26'th, if you want to run send your nomination proposal out to the dev ML
  • What's broken in Havana
  • Backups (specifically when configured with multi-backend volumes)
  • Configuration - Global CONF settings in brick don't belong, and a number of them break multi-backend (Bug #1230066)
  • TaskFlow retry mechanism - The majority felt this should be left as a white list, but no work has been done to fix it so we still have ugly failures/roll-backs (3 bugs logged here)
  • Quotas - Don't know that anybody has gotten to the bottom of the quota syncing issue (Bug #1202896)
  • iSCSI Target creation failures - This was thought to have been fixed but showed up last night (Bug #1223469)
  •  ????
  • Cinderclient release plans/status? (Eharney)
  • OSLO imports (DuncanT)
  • bp/cinder-backup-improvements (dosaboy)
  • bp/multi-attach (zhiyan)

Aug 21, 2013, 16:00 UTC

  1. No agenda, no meeting.

Aug 14, 2013, 16:00 UTC

  1. Volume migration status - avishay
  2. API extensions using metadata. This comes from the readonly volume attach support. - thingee


Aug 7, 2013, 16:00 UTC

  1. RFC - make all rbd clones copy-on-write -- Dosaboy
  2. V1 API removal issues, plans and timescales - DuncanT

Meeting Minutes