Project

General

Profile

Fix #11187

ceph-qa-suite: reduce required machine numbers

Added by Greg Farnum about 9 years ago. Updated over 8 years ago.

Status:
Resolved
Priority:
High
Assignee:
Category:
Testing
Target version:
-
% Done:

0%

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

Description

We've got a lot of tests that frequently get stuck because they can't lock their 4 or 5 nodes needed to run. Many of these probably don't actually require that many nodes, since we should be able to mount multiple clients on the same node, etc. Run through the suites and try to reduce them.

Eg,

roles:
- [mon.a, mon.b, mon.c, mds.a, osd.0, osd.1, osd.2]
- [client.2] 
- [client.1]
- [client.0]

could probably just become

roles:
- [mon.a, mon.b, mon.c, mds.a, osd.0, osd.1, osd.2]
- [client.2, client.1, client.0] 

without breaking anything? Or we can put a client on the first node (and optionally spread out the other daemons) and get it down to three nodes, which is the important bit. This just requires actually testing that it doesn't break anything.

History

#1 Updated by Greg Farnum about 9 years ago

That example is from suites/fs/multiclient/cluster/three_clients.yaml

We've also got suites/knfs/basic/clusters/extra-client.yaml taking up 4 nodes that could probably be 2 (although maybe not, since we have to be careful about loopback mounting).

#2 Updated by Greg Farnum almost 9 years ago

  • Status changed from New to In Progress
  • Assignee set to Greg Farnum

Okay, there are some specialized ones in suites/fs that are pretty easy (done locally).

kcephfs has 2-client tests which is hard because they can't be colocated with the OSDs. :( I guess we can fall back to one node's worth of OSDs and get it down to 3 nodes? Or is multiple mounts on a single node sufficient for those?
(suites/kcephfs/clusters/mixed-clients/clusters-2-clients.yaml)

I created a clusters/fixed-2-ucephfs.yaml for use by the userspace tests that previously grabbed clusters/fixed-3-cephfs.yaml. (The kernel ones need separate nodes for mounting.)

I did not try and do anything with the tests in marginal or experimental; not sure if those are ever used.

#3 Updated by Greg Farnum almost 9 years ago

We can consolidate kernel mounts on the same machine as long as we use the -o noshare option when mounting.

#4 Updated by Greg Farnum almost 9 years ago

  • Status changed from In Progress to 7

#5 Updated by Greg Farnum over 8 years ago

  • Status changed from 7 to Resolved

This is done for the regular suite, at least, and has not been provoking new problems that I can tell.

Also available in: Atom PDF