Bug #18085: osd thrashing deadlock: copy-from vs max-backfills - Ceph - Ceph

Actions

Copy link

Bug #18085

closed

osd thrashing deadlock: copy-from vs max-backfills

Added by Sage Weil over 7 years ago. Updated over 7 years ago.

Status:

Resolved

Priority:

Immediate

Assignee:

Samuel Just

Category:

Target version:

% Done:

Source:

Tags:

Backport:

Regression:

Severity:

3 - minor

Reviewed:

Affected Versions:

ceph-qa-suite:

Pull request ID:

Crash signature (v1):

Crash signature (v2):

Description

one pg is backfilling. backfilling blocks on a copy-from operation, which blocks on the read (copy-get) because the pg is

    "state": "undersized+degraded+remapped+backfill_wait+peered",
    "snap_trimq": "[dd~1,e4~2,e8~1,eb~1,ed~7]",
    "epoch": 1401,
    "up": [
        1,
        5
    ],
    "acting": [
        5
    ],
    "backfill_targets": [
        "1" 
    ],

on the same osd. it won't backfill because osd_max_backfills=1

I think the simple fix is to just set osd_max_backfills=2 during thrashing with cache tiering?