[ClusterLabs] Pacemaker serializes start operations

[Nekrasov, Alexander]

Hello,

I have a dependency tree starting with PgsqlShared as root, and two sub trees:
1.      PgsqlShared->apache->apl->c4fastvpa
2.      PgsqlShared->iproute

With the version on Pacemaker that came with SLES11, the starting was done in parallel, such as
1.      PgsqlShared
2.      apache and iproute - parallel, as expected
3.      apl
4.      c4fastvpa

With the version of Pacemaker that comes with SLES12, the starting is serial:
1.      PgsqlShared
2.      apache
3.      apl
4.      c4fastvpa
5.      iproute - delayed until the first sub tree completes

I have searched for solutions and found a number of variables that should affect this area. I tried to set all of them in different combinations, to no effect.

20:18:40 root at BC-H1008-spb spb:~> echo $LRMD_MAX_CHILDREN
30
20:18:45 root at BC-H1008-spb spb:~> echo $PCMK_MAX_CHILDREN
30
20:05:56 root at BC-H1008-spb spb:~> crm_attribute -Q --name batch-limit
30
20:06:01 root at BC-H1008-spb spb:~> crm_attribute -Q --type nodes --node spb --name node-action-limit
30
20:06:28 root at BC-H1008-spb spb:~> crm_attribute -Q --name migration-limit
-1

Any pointers appreciated.

libpacemaker3-1.1.12-7.1.x86_64
pacemaker-1.1.12-7.1.x86_64
pacemaker-debuginfo-1.1.12-7.1.x86_64
pacemaker-cli-1.1.12-7.1.x86_64

Thanks,
Alexander


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.clusterlabs.org/pipermail/users/attachments/20150811/e0863dae/attachment-0002.html>