Project

General

Profile

Actions
Copy link

Feature #55715

closed

pybind/mgr/cephadm/upgrade: allow upgrades without reducing max_mds

Added by Patrick Donnelly almost 2 years ago. Updated over 1 year ago.

Status:
Resolved
Priority:
Urgent
Assignee:
Dhairya Parmar
Category:
-
Target version:
Ceph - v18.0.0
% Done:

0%

Source:
Development
Tags:
Backport:
Reviewed:
Affected Versions:
Component(FS):
Labels (FS):
Pull request ID:
47756

Description

See mailing list post "Alternate Multi-MDS Upgrade Procedure".

cephadm should have a configurable upgrade option to perform cephfs upgrades without reducing max_mds.

Right now, cephadm reduces max_mds to 1 to avoid having two active MDS modifying on-disk structures with new versions, communicating cross-version-incompatible messages, or other potential incompatibilities. This can be disruptive for large scale cephfs deployments because the cluster cannot easily reduce actives to 1.

Until we have true and correct rolling upgrades for CephFS, another option is to just fail the file system ("ceph fs fail ...") for every fs, upgrade all MDS servers, and then unfail the file systems ("ceph fs set foo joinable true"). This procedure should not be the default until well tested for minor/major upgrades (with qa tests!). Perhaps for Ceph S release it can be a default.

Actions
Copy link

Also available in: Atom PDF