Project

General

Profile

Fix #41782

mds: allow stray directories to fragment and switch from 10 stray directories to 1

Added by Patrick Donnelly over 1 year ago. Updated 2 months ago.

Status:
Resolved
Priority:
High
Assignee:
-
Category:
-
Target version:
Ceph - v16.0.0
% Done:

0%

Source:
Development
Tags:
Backport:
Reviewed:
Affected Versions:
ceph-qa-suite:
Component(FS):
MDS
Labels (FS):
Pull request ID:
33479
Crash signature:

Description

Stray directories can become too full which can result in unexpected ENOSPC errors. See for example, #41778.

Evaluate whether it's feasible to allow fragmentation on strays. It is probably a conservative barrier erected for when fragmentation wasn't considered stable (pre-luminous). The use of 10 stray directories was a workaround. Using only 1 with fragmentation is much simpler.

out.tar.bz - vstart cluster logs (733 KB) Milind Changire, 10/31/2019 02:24 PM

Related issues

Related to CephFS - Bug #41778: 'No space left on device' due to snapshots New 09/11/2019

History

#1 Updated by Patrick Donnelly over 1 year ago

  • Related to Bug #41778: 'No space left on device' due to snapshots added

#2 Updated by Patrick Donnelly over 1 year ago

  • Assignee set to Milind Changire

#3 Updated by Milind Changire about 1 year ago

please see attachment out.tar.bz which includes ceph.conf as to why `ceph status` command hangs on Fedora 30 laptop.

#4 Updated by Patrick Donnelly about 1 year ago

Milind Changire wrote:

please see attachment out.tar.bz which includes ceph.conf as to why `ceph status` command hangs on Fedora 30 laptop.

perhaps it's because you aren't running "bin/ceph status"? Is ceph also installed on your system using Fedora packages?

#5 Updated by Milind Changire about 1 year ago

Patrick Donnelly wrote:

Milind Changire wrote:

please see attachment out.tar.bz which includes ceph.conf as to why `ceph status` command hangs on Fedora 30 laptop.

perhaps it's because you aren't running "bin/ceph status"? Is ceph also installed on your system using Fedora packages?

this wasn't the case
I had a /etc/ceph/ceph.conf along with build/ceph.conf
Zheng pointed this case to me and it indeed being the case, my vstart cluster started working normally after I removed /etc/ceph/ceph.conf

#6 Updated by Milind Changire about 1 year ago

  • Status changed from New to Fix Under Review
  • Pull request ID set to 31521

Update:
Stray dirs are not being dropped from 10 to 1. Zheng recommended having more stray dirs.
Only fragmentation on stray dirs is being enabled with this PR.

#7 Updated by Patrick Donnelly 12 months ago

  • Target version changed from v15.0.0 to v16.0.0

We'll look at merging this at the beginning of Pacific release cycle.

#8 Updated by Patrick Donnelly 2 months ago

  • Status changed from Fix Under Review to Resolved
  • Assignee deleted (Milind Changire)
  • Pull request ID changed from 31521 to 33479

Also available in: Atom PDF