yahoo-eng-team team mailing list archive

[Piotr Parczewski <2078476@xxxxxxxxxxxxxxxxxx>]

Public bug reported:

Versions affected: from current master to at least Yoga.

The documentation
(https://docs.openstack.org/glance/2024.1/configuration/configuring.html#configuring-
the-rbd-storage-backend) states that the default rbd_store_chunk_size
defaults to 4M while in reality it's 8M. This could have been 'only' a
documentation bug, but there are two concerns here:

1) Was it the original intention to have 8M chunk size (which is
different from Ceph's defaults = 4M) or was it an inadvertent effect of
other changes?

2) Cinder defaults to rbd_store_chunk_size=4M. Having volumes created
from Glance images results in an inherited chunk size of 8M (due to
snapshotting) and could have unpredicted performance consequences. It
feels like this scenario should at least be documented, if not avoided.

Steps to reproduce:
- deploy Glance with RBD backend enabled and default config;
- query stores information for the configured chunk size (/v2/info/stores/detail)
Optional:
- have an image created in Ceph pool and validate its chunk size with rbd info command.

** Affects: glance
     Importance: Undecided
         Status: New

** Summary changed:

- rbd_store_chunk_size defaults to 8192M not 4096M
+ rbd_store_chunk_size defaults to M not 4M

** Summary changed:

- rbd_store_chunk_size defaults to M not 4M
+ rbd_store_chunk_size defaults to 8M not 4M

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to Glance.
https://bugs.launchpad.net/bugs/2078476

Title:
  rbd_store_chunk_size defaults to 8M not 4M

Status in Glance:
  New

Bug description:
  Versions affected: from current master to at least Yoga.

  The documentation
  (https://docs.openstack.org/glance/2024.1/configuration/configuring.html#configuring-
  the-rbd-storage-backend) states that the default rbd_store_chunk_size
  defaults to 4M while in reality it's 8M. This could have been 'only' a
  documentation bug, but there are two concerns here:

  1) Was it the original intention to have 8M chunk size (which is
  different from Ceph's defaults = 4M) or was it an inadvertent effect
  of other changes?

  2) Cinder defaults to rbd_store_chunk_size=4M. Having volumes created
  from Glance images results in an inherited chunk size of 8M (due to
  snapshotting) and could have unpredicted performance consequences. It
  feels like this scenario should at least be documented, if not
  avoided.

  Steps to reproduce:
  - deploy Glance with RBD backend enabled and default config;
  - query stores information for the configured chunk size (/v2/info/stores/detail)
  Optional:
  - have an image created in Ceph pool and validate its chunk size with rbd info command.

To manage notifications about this bug go to:
https://bugs.launchpad.net/glance/+bug/2078476/+subscriptions