Jump to: navigation, search

Difference between revisions of "CinderQueensPTGSummary"

(Wednesday 9/13/2017)
(Thursday 9/14/2017)
 
(One intermediate revision by the same user not shown)
Line 91: Line 91:
 
==Thursday 9/14/2017==
 
==Thursday 9/14/2017==
 
[https://etherpad.openstack.org/p/cinder-ptg-queens-thursday-notes Etherpad with Detailed Notes]
 
[https://etherpad.openstack.org/p/cinder-ptg-queens-thursday-notes Etherpad with Detailed Notes]
 +
 +
VIDEO LINK
 +
 +
===Terminate Connections on the Source Host During Shelve Offload Instance===
 +
*'''Action:'''  (jgriffith)  Going to make it so they can request another attachment (at least reserve) before the first one is deleted.
 +
*'''Action:'''  (mriedem)  Work to get the following patch in  https://review.openstack.org/#/c/257275/
 +
*'''Action:'''  (mriedem)  Make sure that the new style attachment patch (https://review.openstack.org/#/c/330285/) is creating an 'empty' attachment to keep the volume in-use before detaching the existing attachment from the old host.
 +
 +
===Cinder <--> Nova API Update===
 +
*'''Info'''  Very close to supporting multi-attach.
 +
*'''Info'''  Old APIs:  reserve, os-initialize_connection, os-attach, begin_detaching, os-terminate_connection, os-detach
 +
*'''Info:'''  New APIs:  attachment_create, attachment_update, attachment_complete, attachment_delete
 +
 +
VIDEO LINK
 +
 +
===Multi-Attach Impacts on Cinder===
 +
*'''Action:'''  (jgriffith)  Will propose a patch that uses something like the host and attachment to create a lock when working with shared connections.
 +
*'''Agreed:'''  Policy controls are needed for:  Multi-attach, Read-only, Read/Write.
 +
*'''Action:''' (jgriffith)  New microversion will be written for the new attachment APIs/multi-attach patches.
 +
*'''Action:''' (ildikov)  Spec will be updated with regards to what we have agreed upon in the session.
 +
 +
===API to refresh volume connection info===
 +
*'''Agreed:'''  Cold migrate/resize, stop.start, suspend/resume and rebuild should refresh the connection information, but they should.  They will do this but not backport.  Add a release note and add a flag so that if the refresh fails the operation doesn't fail.
 +
*'''Action:''' (mriedem)  To own implementation.
 +
 +
===Cinder Ephemeral Storage===
 +
*'''Action''':  (mriedem)  To wruite up the spec he planned to write up for the function.
 +
*'''Action:'''  (cburgess)  To look into writing a Cinder image backend.
 +
 +
VIDEO LINK
 +
 +
===Volume Encryption===
 +
*'''Agreed:'''  Though Huawei would like to be able to specify encryption keys we have talked about doing this in the past and decided not to do so.
 +
*'''Action:'''  (eharney)  To write spec proposing how to migrate volumes encrypted using the old approach to using the new approach.
 +
 +
===Use Alembic for DB Migration in Cinder===
 +
*'''Agreed:'''  Team wasn't particularly interested in taking on this change as it is counter to what Nova had done.
 +
*'''Action:'''  (tommylikehu)  To work with Huawei to see if there is way to avoid carrying custom code or to determine a way to handle their own DB migrations.
 +
 +
===Improve Quota Management Performance===
 +
*'''Agreed:'''  That we would like to get the performance improvements into Cinder.
 +
*'''Action:'''  (tommylikehu)  Will start to implement the improvements for Cinder.  https://review.openstack.org/#/c/496562/
 +
 +
===Apply Encryption and QoS per Volume===
 +
*'''Info:'''  Another topic that has been brought up repeatedly.  Seems like customers need better education on how to use QoS and volume types.
 +
*'''Action:'''  (tommylikehu)  To go back to customer and better understand the need.  Hopefully providing some education will alleviate the need for this change.
 +
 +
VIDEO LINK
 +
 +
===Generic Backup Implementation===
 +
*'''Agreed:'''  Need to add knowledge of backups to the scheduler to make the generic implementation possible.  Don't want to write a new scheduler, the existing scheduler can just be made aware of the backup providers.
 +
*'''Action:'''  (e0ne)  Will add support for backup services to the scheduler.
 +
*'''Action:'''  (e0ne)  To set up a 3rd Party CI for the NFS backup target.  Documentation on the process will be written as it is completed for others to follow.
 +
*'''Action:'''  (jungleboyj)  Future discussion on what our 3rd Party CI requirement is for backup drivers after we know what work is required to get the 3rd Party CI working.
 +
 +
===Fibre Channel Zone Manager===
 +
*'''Agreed:'''  Need to mark Cisco driver as unsupported due to lack of 3rd Party CI.
 +
*'''Action:'''  (jungleboyj)  Mark it unsupported.
 +
*'''Action:'''  (jungleboyj)  Send e-mail to openstack-dev and the operators mailing list nothing that we will remove the driver in Rocky unless someone can pick up running 3rd Party CI for it.
 +
 +
===Create Volume from Backup===
 +
*'''Agreed:'''  We will allow users to specify a size and error if it is smaller than required for restoration of the backup.  The size value will not be required.
 +
*'''Agreed:'''  We need to update OSC so that it doesn't require you to indicate a size when restoring data from snapshot, backup, clone, etc.
 +
*'''Agreed:'''  We do not want to yet deprecate the existing backup restore commands.  There are still slightly different use cases.  We will readdress this in the future.
 +
*'''Action:'''  (tommylikehu)  To implement the new create volume from backup as an additional option to the existing 'cinder create' command.
 +
 +
===Oslo Deprecation/Cleanup Impact===
 +
*'''Action:'''  (smcginnis)  To follow up on notes from Oslo team and work on implementing any changes to Cinder where necessary.
 +
 +
===Active/Active HA Testing===
 +
*'''Action:'''  (geguileo)  To look into the RBD driver and provide some guidance as to what are the kinds of testing that he would like to see done there with the idea that it can serve as an example for others.
 +
*'''Action:''' (geguileo)  To continue working on test injection framework.
 +
 +
===Review Queens Development Priorities===
 +
*'''Info:'''  Priorities set in the [https://etherpad.openstack.org/p/cinder-spec-review-tracking Spec Review Tracking etherpad] .
 +
<br>
 +
<br>
 +
<br>
 +
==Friday 9/15/2017==
 +
[https://etherpad.openstack.org/p/cinder-ptg-queens-friday-notes Etherpad with Detailed Notes]
 +
 +
===Spend Time Looking at/verifying the Install Guide===
 +
*'''Action:'''  (jungleboyj)  To work with Hagen to run through the installation process for RedHat and Ubuntu
 +
*'''Action:'''  (jungleboyj)  Work on reducing duplication of content in the installation guide.
 +
 +
=== Discuss Bulk Volume Create===
 +
*'''Action:'''  (walshh and gman-tx)  To talk to the Nova team and see how agreeable they are to the idea of doing bulk attach.
 +
*'''Action:'''  (walshh and gman-tx)  To write a Cinder spec proposing how bulk attach could work.  It is likely that it will be more complicated than first anticipated.

Latest revision as of 22:05, 20 September 2017

Introduction

This page contains a summary of the subjects covered during the Queens PTG in Denver, Colorado Wednesday September 13th, through Friday September 15th, 2017.

The full etherpad and all associated notes may be found here.


Wednesday 9/13/2017

Etherpad with Detailed Notes

VIDEO LINK

Cinder T-Shirt Discussion

  • Decision: Team does want to do a Cinder team T-Shirt with just our new Logo on the front.
  • Action: Jay to work on making this happen before the next PTG.

Replication Failback

  • Action: (jbernard) Going to propose a spec that describes a two step process for promoting a backend to be primary for failback.
  • Action: (jungleboyj) Add the spec, when available, to the priority review list so that this get attention early in the release.
  • Action: (jungleboyj) Follow up with Xing to find out how this impacts group replication.

Using Constants for Microversions

  • Agreement: We want to move to using constants to track microversions.
  • Agreement: Reviewers in the future should enforce using constants for microversions.
  • Action: (smcginnis) A patch will be posted with a proposal on how to do this. Core team needs to review and use it as the example going forward.
  • Action: (smcginnis) Share the review with Ben so that Manila can do something similar.

Review of Base Required Features

  • Agreement: Nothing needs to be changed right now. As features continue to evolve we can readdress again in the future.

VIDEO LINK

Default Policy in Code

  • Agreement: We don't want to mix changing our policy granularity, defaults, naming, etc. at the same time as moving the default in code.
  • Agreement: Do patches as a dependency chain.
  • Action: (TommyLikeHu) Existing patches will be updated soon based on the agreed upon items above.
  • Action: Core team to work to merge these as soon as possible to avoid merge conflicts, etc.

Multi-Attach Discussion

  • Info: Concerns about each driver needing changes to support multi-attach. Could be handled by having a capability of multi_attach = False by default unless drivers enable it.
  • Agreement: Need some way for drivers to indicate when a connection is shared. This can then be used to help Brick handle things properly.
  • Agreement: Need good documentation on how this works so that the driver maintainers can properly implement it.

VIDEO LINK

Documentation

  • Action: (jungleboyj) Talk to the documentation team about pointing to the 'latest' documentation for Developer Guide, Contributor Guide and API-REF.
  • Action: (smcginnis) Sean will look at moving the microversion information to the API ref when working on the constants.
  • Info: Release Notes work differently than other code changes. If you need to change a release note in a stable branch it shouldn't go into master first.
  • Agreement: Don't need to be opening bugs for every change to the documentation. Just fix it.
  • Agreement: Do not need to make documentation build a part of our default tox runs. People need to run 'tox -e docs' on their own if they have touch documentation or strings.

Grouping Release Note Changes

  • Agreement: We would like developers to group together release note changes. I.E. all the notes on the VMAX driver go into one release note that is updated during the development cycle.
  • Action: (jungleboyj) Bring up at the next weekly team meeting.
  • Action: (jungleboyj) Send a note to the mailing list.
  • Action: (jungleboyj) Update devref on release notes.

Add the ability to collect resource's count

  • Info: Issues were already addressed and further discussion was not needed.

Appropriately Handling Microversions and the CLI

  • Agreement: For each microversion bump on the server we need to have a change that bumps the max microversion, at a minimum, in the client.
  • Action: Core reviewers need to make sure that for a microversion change in the server we find a corresponding change in the client.

Discuss cut-off Deadlines for Queens

  • Agreement: We will follow the same schedule as far as deadlines as was followed in Pike.

Automatic Configuration Generation

  • Action: (jungleboyj) To look at spinxext plugin and see how we can use that to better generate tables for the configuration reference.

Condense/Standardize on Base Set of Config Options

  • Agreement: We want to go forward with the consolidation. We will do it for drivers all at once. Kendall will do it for the drivers rather than trying to get each owner to do it.
  • Agreement: Merge the code regardless of CI state and then look at marking the driver unsupported as we find more without CI running.
  • Action: (diablo_rojo) Start implementing the consolidation.
  • Action: (diablo_rojo) Update the 'How to Contribute a Driver documentation to point to common config options that should be used.
  • Action: Reviewers need to watch for proliferation of configuration options in new drivers and discourage use of the addition of new ones.

VIDEO LINK

Support Reverting Volume to any Snapshot

  • Agreement: We are not going to change the functionality at this point in time. There isn't a good solution to make this possible for all drivers right now.

Installation Guide Validation

  • Action (jungleboyj) Common content at the start of the install guide needs to be replaced with a reference to the openstack-manuals version of the content.
  • Action (jungleboyj) Rather than having 3 different installation guides it would be good to collapse common content down into individual files as much as possible.




Thursday 9/14/2017

Etherpad with Detailed Notes

VIDEO LINK

Terminate Connections on the Source Host During Shelve Offload Instance

  • Action: (jgriffith) Going to make it so they can request another attachment (at least reserve) before the first one is deleted.
  • Action: (mriedem) Work to get the following patch in https://review.openstack.org/#/c/257275/
  • Action: (mriedem) Make sure that the new style attachment patch (https://review.openstack.org/#/c/330285/) is creating an 'empty' attachment to keep the volume in-use before detaching the existing attachment from the old host.

Cinder <--> Nova API Update

  • Info Very close to supporting multi-attach.
  • Info Old APIs: reserve, os-initialize_connection, os-attach, begin_detaching, os-terminate_connection, os-detach
  • Info: New APIs: attachment_create, attachment_update, attachment_complete, attachment_delete

VIDEO LINK

Multi-Attach Impacts on Cinder

  • Action: (jgriffith) Will propose a patch that uses something like the host and attachment to create a lock when working with shared connections.
  • Agreed: Policy controls are needed for: Multi-attach, Read-only, Read/Write.
  • Action: (jgriffith) New microversion will be written for the new attachment APIs/multi-attach patches.
  • Action: (ildikov) Spec will be updated with regards to what we have agreed upon in the session.

API to refresh volume connection info

  • Agreed: Cold migrate/resize, stop.start, suspend/resume and rebuild should refresh the connection information, but they should. They will do this but not backport. Add a release note and add a flag so that if the refresh fails the operation doesn't fail.
  • Action: (mriedem) To own implementation.

Cinder Ephemeral Storage

  • Action: (mriedem) To wruite up the spec he planned to write up for the function.
  • Action: (cburgess) To look into writing a Cinder image backend.

VIDEO LINK

Volume Encryption

  • Agreed: Though Huawei would like to be able to specify encryption keys we have talked about doing this in the past and decided not to do so.
  • Action: (eharney) To write spec proposing how to migrate volumes encrypted using the old approach to using the new approach.

Use Alembic for DB Migration in Cinder

  • Agreed: Team wasn't particularly interested in taking on this change as it is counter to what Nova had done.
  • Action: (tommylikehu) To work with Huawei to see if there is way to avoid carrying custom code or to determine a way to handle their own DB migrations.

Improve Quota Management Performance

Apply Encryption and QoS per Volume

  • Info: Another topic that has been brought up repeatedly. Seems like customers need better education on how to use QoS and volume types.
  • Action: (tommylikehu) To go back to customer and better understand the need. Hopefully providing some education will alleviate the need for this change.

VIDEO LINK

Generic Backup Implementation

  • Agreed: Need to add knowledge of backups to the scheduler to make the generic implementation possible. Don't want to write a new scheduler, the existing scheduler can just be made aware of the backup providers.
  • Action: (e0ne) Will add support for backup services to the scheduler.
  • Action: (e0ne) To set up a 3rd Party CI for the NFS backup target. Documentation on the process will be written as it is completed for others to follow.
  • Action: (jungleboyj) Future discussion on what our 3rd Party CI requirement is for backup drivers after we know what work is required to get the 3rd Party CI working.

Fibre Channel Zone Manager

  • Agreed: Need to mark Cisco driver as unsupported due to lack of 3rd Party CI.
  • Action: (jungleboyj) Mark it unsupported.
  • Action: (jungleboyj) Send e-mail to openstack-dev and the operators mailing list nothing that we will remove the driver in Rocky unless someone can pick up running 3rd Party CI for it.

Create Volume from Backup

  • Agreed: We will allow users to specify a size and error if it is smaller than required for restoration of the backup. The size value will not be required.
  • Agreed: We need to update OSC so that it doesn't require you to indicate a size when restoring data from snapshot, backup, clone, etc.
  • Agreed: We do not want to yet deprecate the existing backup restore commands. There are still slightly different use cases. We will readdress this in the future.
  • Action: (tommylikehu) To implement the new create volume from backup as an additional option to the existing 'cinder create' command.

Oslo Deprecation/Cleanup Impact

  • Action: (smcginnis) To follow up on notes from Oslo team and work on implementing any changes to Cinder where necessary.

Active/Active HA Testing

  • Action: (geguileo) To look into the RBD driver and provide some guidance as to what are the kinds of testing that he would like to see done there with the idea that it can serve as an example for others.
  • Action: (geguileo) To continue working on test injection framework.

Review Queens Development Priorities




Friday 9/15/2017

Etherpad with Detailed Notes

Spend Time Looking at/verifying the Install Guide

  • Action: (jungleboyj) To work with Hagen to run through the installation process for RedHat and Ubuntu
  • Action: (jungleboyj) Work on reducing duplication of content in the installation guide.

Discuss Bulk Volume Create

  • Action: (walshh and gman-tx) To talk to the Nova team and see how agreeable they are to the idea of doing bulk attach.
  • Action: (walshh and gman-tx) To write a Cinder spec proposing how bulk attach could work. It is likely that it will be more complicated than first anticipated.