Project

General

Profile

Actions

Fix #4712

open

recovery from partially partitioned networks

Added by Anonymous about 11 years ago.

Status:
New
Priority:
Normal
Assignee:
-
Category:
Monitor
Target version:
-
% Done:

0%

Source:
Development
Tags:
Backport:
Reviewed:
Affected Versions:
ceph-qa-suite:
Pull request ID:
Crash signature (v1):
Crash signature (v2):

Description

This is a hypothetical but plausible problem.

If a group of monitors and OSDs are partitioned out, they will be detected and fenced out.
But consider ...
two large sites (A and B) interconnected by a fat pipe.
a third site (C) interconnected with A & B by thin pipes for quorum.
the fat pipe is severed (which does not sound too pathological).

the monitors would probably consider all of the OSDs to be up and in,
even though many cannot talk to one another. This seems to be a situation
in which we would behave very badly. Detecting arbitrary partitionings is
probably hard, but topologically obvious partitions might be detectable ...
after which we could choose a winner and a loser and get on with life.

No data to display

Actions

Also available in: Atom PDF