RepoMirrors/ceph
1
0
Fork 0
mirror of https://github.com/ceph/ceph synced 2025-02-24 19:47:44 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge PR #37707 into master

Browse Source 
* refs/pull/37707/head:
	doc: document the mds_autoscaler mgr plugin

Reviewed-by: Patrick Donnelly <pdonnell@redhat.com>
This commit is contained in:
Patrick Donnelly 2020-10-20 08:21:22 -07:00
commit e5f917aa1c
No known key found for this signature in database
GPG Key ID: 3A2A7E25BEA8AADB
2 changed files with 32 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
doc/mgr/index.rst
View File

@ -45,3 +45,4 @@ sensible.
Insights module <insights>
Orchestrator module <orchestrator>
Rook module <rook>
MDS Autoscaler module <mds_autoscaler>

31
doc/mgr/mds_autoscaler.rst Normal file
View File

@ -0,0 +1,31 @@
MDS Autoscaler Module
=====================
The MDS Autoscaler Module monitors ``fsmap`` update notifications from the mgr
daemon and takes action to spawn or kill MDS daemons for a file-system as per
changes to the:
- ``max_mds`` config value
- ``standby_count_wanted`` config value
- standby promotions to active MDS state in case of active MDS rank death
Bumping up the ``max_mds`` config option value causes a standby mds to be promoted
to hold an active rank. This leads to a drop in standby mds count. The MDS
Autoscaler module detects this deficit and the orchestrator module is notified
about the required MDS count. The orchestrator back-end then takes necessary
measures to spawn standby MDSs.
Dropping the ``max_mds`` config option causes the orchestrator back-end to kill
standby mds to achieve the new reduced count. Preferably standby mds are chosen
to be killed when the ``max_mds`` count is dropped.
An increment and decrement of the ``standby_count_wanted`` config option value
has a similar effect on the total MDS count. The orchestrator is notified about
the change and necessary action to spawn or kill standby MDSs is taken.
A death of an active MDS rank also causes promotion of a standby mds to occupy
the required active rank. The MDS Autoscaler notices the change in the standby
mds count and a message is passed to the orchestrator to maintain the necessary
MDS count.
NOTE: There is no CLI associated with the MDS Autoscaler Module.