commit 22d6fe98a3f437709901fac4e4ec65fec414f7d0
Author: Goutham Pacha Ravi <email address hidden>
Date: Fri Aug 7 12:38:05 2020 -0700
Fix capacity calculations in the CephFS driver
The driver inflated total and available capacity
due to an incorrect calculation. The driver was
also ignoring the configuration option
"reserved_share_percentage" that allows
deployers to set aside space from scheduling
to prevent oversubscription.
While this bugfix may have an upgrade impact,
some things must be clarified:
- Inflating the total, free space will allow
manila to schedule workloads that may run
out of space - this may cause end user
downtime and frustration, because shares are
created (empty subvolumes on ceph occupy no
space) easily, but they could get throttled
as they start to fill up.
- CephFS shares are always thinly provisioned
but, the driver does not support oversubscription
via manila. So, real free space is what
determines capacity based scheduler decisions.
Users however expect share sizes to be honored,
and manila will allow provisioning as long
as there is free space on the cluster. This
means that Ceph cluster administrators
must manage oversubscription outside of manila
to prevent misbehavior.
Depends-On: Ic96b65d2caab788afca8bfc45575f3c05dc88008
Change-Id: I6ab157d6d099fe910ec1d90193783b55053ce8f6
Closes-Bug: #1890833
Signed-off-by: Goutham Pacha Ravi <email address hidden>
Reviewed: https:/ /review. opendev. org/745402 /git.openstack. org/cgit/ openstack/ manila/ commit/ ?id=22d6fe98a3f 437709901fac4e4 ec65fec414f7d0
Committed: https:/
Submitter: Zuul
Branch: master
commit 22d6fe98a3f4377 09901fac4e4ec65 fec414f7d0
Author: Goutham Pacha Ravi <email address hidden>
Date: Fri Aug 7 12:38:05 2020 -0700
Fix capacity calculations in the CephFS driver
The driver inflated total and available capacity share_percentag e" that allows
due to an incorrect calculation. The driver was
also ignoring the configuration option
"reserved_
deployers to set aside space from scheduling
to prevent oversubscription.
While this bugfix may have an upgrade impact,
some things must be clarified:
- Inflating the total, free space will allow
manila to schedule workloads that may run
out of space - this may cause end user
downtime and frustration, because shares are
created (empty subvolumes on ceph occupy no
space) easily, but they could get throttled
as they start to fill up.
- CephFS shares are always thinly provisioned
but, the driver does not support oversubscription
via manila. So, real free space is what
determines capacity based scheduler decisions.
Users however expect share sizes to be honored,
and manila will allow provisioning as long
as there is free space on the cluster. This
means that Ceph cluster administrators
must manage oversubscription outside of manila
to prevent misbehavior.
Depends-On: Ic96b65d2caab78 8afca8bfc45575f 3c05dc88008 910ec1d90193783 b55053ce8f6
Change-Id: I6ab157d6d099fe
Closes-Bug: #1890833
Signed-off-by: Goutham Pacha Ravi <email address hidden>