Project

General

Profile

Actions
Copy link

Bug #45093

closed

cephadm: mgrs transiently getting co-located (one node gets two when only one was asked for)

Added by Nathan Cutler about 4 years ago. Updated over 3 years ago.

Status:
Resolved
Priority:
Urgent
Assignee:
Sebastian Wagner
Category:
cephadm/scheduler
Target version:
-
% Done:

0%

Source:
Tags:
Backport:
Regression:
No
Severity:
3 - minor
Reviewed:
Affected Versions:
ceph-qa-suite:
Pull request ID:
37135
Crash signature (v1):
Crash signature (v2):

Description

After "ceph orch apply mgr node1,node2,node3", cluster has four MGRs

This started happening in master very recently, but I have not pinpointed which PR/commit did it:

After bootstrapping a cluster, I ask for three MGRs (total), as usual:

    master: ++ ceph orch apply mgr node1,node2,node3
    master: Scheduled mgr update...

Over the past few months there has been some semantic back-and-forth whether "ceph orch apply mgr node1,node2,node3" means to add three more MGRs to those already running, or whether it means I want three MGRs total. I thought the latter had finally won out, but apparently not?

master:~ # ceph-s
...
  services:
    mon: 3 daemons, quorum node1,node2,node3 (age 4h)
    mgr: node1.bwokiv(active, since 4h), standbys: node1.amxnnt, node2.hghrym, node3.vyanht
...

The Ceph version is:

master:~ # ceph --version
ceph version 16.0.0-704-g38ae96e1c9 (38ae96e1c9a4f8ad3095626c71951a122bdc8fe7) pacific (dev)
master:~ # ceph versions
{
    "mon": {
        "ceph version 16.0.0-704-g38ae96e1c9 (38ae96e1c9a4f8ad3095626c71951a122bdc8fe7) pacific (dev)": 3
    },
    "mgr": {
        "ceph version 16.0.0-704-g38ae96e1c9 (38ae96e1c9a4f8ad3095626c71951a122bdc8fe7) pacific (dev)": 4
    },
    "osd": {
        "ceph version 16.0.0-704-g38ae96e1c9 (38ae96e1c9a4f8ad3095626c71951a122bdc8fe7) pacific (dev)": 6
    },
    "mds": {},
    "overall": {
        "ceph version 16.0.0-704-g38ae96e1c9 (38ae96e1c9a4f8ad3095626c71951a122bdc8fe7) pacific (dev)": 13
    }
}

Related issues 1 (1 open — 0 closed)

Related to Orchestrator - Feature #45876: cephadm: handle port conflicts gracefullyNew

Actions
Actions
Copy link

Also available in: Atom PDF