Project

General

Profile

Fix #58023

mds: do not evict clients if OSDs are laggy

Added by Patrick Donnelly 3 months ago. Updated 3 months ago.

Status:
New
Priority:
High
Assignee:
Dhairya Parmar
Category:
Correctness/Safety
Target version:
Ceph - v18.0.0
% Done:

0%

Source:
Development
Tags:
Backport:
quincy,pacific
Reviewed:
Affected Versions:
ceph-qa-suite:
Component(FS):
MDS
Labels (FS):
Pull request ID:
Crash signature (v1):
Crash signature (v2):

Description

Monitoring perf dumps from the MDS can sometimes show that OSDs are laggy, "objecter.op_laggy" and "objecter.osd_laggy". This can indicate the cause for laggy clients (cannot flush dirty data for cap revokes).

In Server::find_idle_sessions, if the MDS detects laggy OSDs, defer evicting/removing "laggy" client sessions until OSDs are no longer laggy. Log a warning to the cluster indicating the situation.

History

#1 Updated by Venky Shankar 3 months ago

  • Category set to Correctness/Safety
  • Assignee set to Dhairya Parmar

Also available in: Atom PDF