yahoo-eng-team team mailing list archive

[Chengen Du <2024258@xxxxxxxxxxxxxxxxxx>]

Noble 3:29.0.1-0ubuntu1.3 already contains the patch, so no changes are
needed.


** Changed in: nova (Ubuntu)
       Status: Incomplete => Won't Fix

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to OpenStack Compute (nova).
https://bugs.launchpad.net/bugs/2024258

Title:
  Performance degradation archiving DB with large numbers of FK related
  records

Status in OpenStack Compute (nova):
  Fix Released
Status in OpenStack Compute (nova) antelope series:
  In Progress
Status in OpenStack Compute (nova) wallaby series:
  In Progress
Status in OpenStack Compute (nova) xena series:
  In Progress
Status in OpenStack Compute (nova) yoga series:
  In Progress
Status in OpenStack Compute (nova) zed series:
  In Progress
Status in nova package in Ubuntu:
  Won't Fix
Status in nova source package in Focal:
  In Progress
Status in nova source package in Jammy:
  In Progress

Bug description:
  [Impact]
  Originally, Nova archives deleted rows in batches consisting of a maximum number of parent rows (max_rows) plus their child rows, all within a single database transaction.
  This approach limits the maximum value of max_rows that can be specified by the caller due to the potential size of the database transaction it could generate.
  Additionally, this behavior can cause the cleanup process to frequently encounter the following error:
  oslo_db.exception.DBError: (pymysql.err.InternalError) (3100, "Error on observer while running replication hook 'before_commit'.")

  The error arises when the transaction exceeds the group replication transaction size limit, a safeguard implemented to prevent potential MySQL crashes [1].
  The default value for this limit is approximately 143MB.

  [Fix]
  An upstream commit has changed the logic to archive one parent row and its related child rows in a single database transaction.
  This change allows operators to choose more predictable values for max_rows and achieve more progress with each invocation of archive_deleted_rows.
  Additionally, this commit reduces the chances of encountering the issue where the transaction size exceeds the group replication transaction size limit.

  commit 697fa3c000696da559e52b664c04cbd8d261c037
  Author:     melanie witt <melwittt@xxxxxxxxx>
  CommitDate: Tue Jun 20 20:04:46 2023 +0000

      database: Archive parent and child rows "trees" one at a time

  [Test Plan]
  1. Create an instance and delete it in OpenStack.
  2. Log in to the Nova database and confirm that there is an entry with a deleted_at value that is not NULL.
  select display_name, deleted_at from instances where deleted_at <> 0;
  3. Execute the following command, ensuring that the timestamp specified in --before is later than the deleted_at value:
  nova-manage db archive_deleted_rows --before "XXX-XX-XX XX:XX:XX" --verbose --until-complete
  4. Log in to the Nova database again and confirm that the entry has been archived and removed.
  select display_name, deleted_at from instances where deleted_at <> 0;

  [Where problems could occur]
  The commit changes the logic for archiving deleted entries to reduce the size of transactions generated during the operation.
  If the patch contains errors, it will only impact the archiving of deleted entries and will not affect other functionalities.

  [1] https://bugs.mysql.com/bug.php?id=84785

  [Original Bug Description]

  Observed downstream in a large scale cluster with constant create/delete
  server activity and hundreds of thousands of deleted instances rows.

  Currently, we archive deleted rows in batches of max_rows parents +
  their child rows in a single database transaction. Doing it that way
  limits how high a value of max_rows can be specified by the caller
  because of the size of the database transaction it could generate.

  For example, in a large scale deployment with hundreds of thousands of
  deleted rows and constant server creation and deletion activity, a
  value of max_rows=1000 might exceed the database's configured maximum
  packet size or timeout due to a database deadlock, forcing the operator
  to use a much lower max_rows value like 100 or 50.

  And when the operator has e.g. 500,000 deleted instances rows (and
  millions of deleted rows total) they are trying to archive, being
  forced to use a max_rows value several orders of magnitude lower than
  the number of rows they need to archive is a poor user experience and
  makes it unclear if archive progress is actually being made.

To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/2024258/+subscriptions