Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
Actually, we are not skipping 'Started' state - we just consider resource as started when beam is powered up and rabbitmq start_app/stop_app action succeeds. Such a node is considered as a good one that can be marked as 'Master' to which the nodes should connect and then all the cluster join/leave actions are handled using multi-state notification mechanism. On Wed, May 20, 2015 at 5:05 AM, Andrew Beekhof abeek...@redhat.com wrote: On 20 May 2015, at 6:05 am, Andrew Woodward xar...@gmail.com wrote: On Thu, May 7, 2015 at 5:01 PM Andrew Beekhof abeek...@redhat.com wrote: On 5 May 2015, at 1:19 pm, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: Thank you Andrew. on 2015/05/05 08:03, Andrew Beekhof wrote: On 28 Apr 2015, at 11:15 pm, Bogdan Dobrelya bdobre...@mirantis.com wrote: Hello, Hello, Zhou I using Fuel 6.0.1 and find that RabbitMQ recover time is long after power failure. I have a running HA environment, then I reset power of all the machines at the same time. I observe that after reboot it usually takes 10 minutes for RabittMQ cluster to appear running master-slave mode in pacemaker. If I power off all the 3 controllers and only start 2 of them, the downtime sometimes can be as long as 20 minutes. Yes, this is a known issue [0]. Note, there were many bugfixes, like [1],[2],[3], merged for MQ OCF script, so you may want to try to backport them as well by the following guide [4] [0] https://bugs.launchpad.net/fuel/+bug/1432603 [1] https://review.openstack.org/#/c/175460/ [2] https://review.openstack.org/#/c/175457/ [3] https://review.openstack.org/#/c/175371/ [4] https://review.openstack.org/#/c/170476/ Is there a reason you’re using a custom OCF script instead of the upstream[a] one? Please have a chat with David (the maintainer, in CC) if there is something you believe is wrong with it. [a] https://github.com/ClusterLabs/resource-agents/blob/master/heartbeat/rabbitmq-cluster I'm using the OCF script from the Fuel project, specifically from the 6.0 stable branch [alpha]. Ah, I’m still learning who is who... i thought you were part of that project :-) Comparing with upstream OCF code, the main difference is that Fuel RabbitMQ OCF is a master-slave resource. Fuel RabbitMQ OCF does more bookkeeping, for example, blocking client access when RabbitMQ cluster is not ready. I beleive the upstream OCF should be OK to use as well after I read the code, but it might not fit into the Fuel project. As far as I test, the Fuel OCF script is good except sometimes the full reassemble time is long, and as I find out, it is mostly because the Fuel MySQL Galera OCF script keeps pacemaker from promoting RabbitMQ resource, as I mentioned in the previous emails. Maybe Vladimir and Sergey can give us more insight on why Fuel needs a master-slave RabbitMQ. That would be good to know. Browsing the agent, promote seems to be a no-op if rabbit is already running. To the master / slave reason due to how the ocf script is structured to deal with rabbit's poor ability to handle its self in some scenarios. Hopefully the state transition diagram [5] is enough to clarify what's going on. [5] http://goo.gl/PPNrw7 Not really. It seems to be under the impression you can skip started and go directly from stopped to master. __ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev -- Yours Faithfully, Vladimir Kuklin, Fuel Library Tech Lead, Mirantis, Inc. +7 (495) 640-49-04 +7 (926) 702-39-68 Skype kuklinvv 35bk3, Vorontsovskaya Str. Moscow, Russia, www.mirantis.com http://www.mirantis.ru/ www.mirantis.ru vkuk...@mirantis.com __ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
On 20 May 2015, at 6:05 am, Andrew Woodward xar...@gmail.com wrote: On Thu, May 7, 2015 at 5:01 PM Andrew Beekhof abeek...@redhat.com wrote: On 5 May 2015, at 1:19 pm, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: Thank you Andrew. on 2015/05/05 08:03, Andrew Beekhof wrote: On 28 Apr 2015, at 11:15 pm, Bogdan Dobrelya bdobre...@mirantis.com wrote: Hello, Hello, Zhou I using Fuel 6.0.1 and find that RabbitMQ recover time is long after power failure. I have a running HA environment, then I reset power of all the machines at the same time. I observe that after reboot it usually takes 10 minutes for RabittMQ cluster to appear running master-slave mode in pacemaker. If I power off all the 3 controllers and only start 2 of them, the downtime sometimes can be as long as 20 minutes. Yes, this is a known issue [0]. Note, there were many bugfixes, like [1],[2],[3], merged for MQ OCF script, so you may want to try to backport them as well by the following guide [4] [0] https://bugs.launchpad.net/fuel/+bug/1432603 [1] https://review.openstack.org/#/c/175460/ [2] https://review.openstack.org/#/c/175457/ [3] https://review.openstack.org/#/c/175371/ [4] https://review.openstack.org/#/c/170476/ Is there a reason you’re using a custom OCF script instead of the upstream[a] one? Please have a chat with David (the maintainer, in CC) if there is something you believe is wrong with it. [a] https://github.com/ClusterLabs/resource-agents/blob/master/heartbeat/rabbitmq-cluster I'm using the OCF script from the Fuel project, specifically from the 6.0 stable branch [alpha]. Ah, I’m still learning who is who... i thought you were part of that project :-) Comparing with upstream OCF code, the main difference is that Fuel RabbitMQ OCF is a master-slave resource. Fuel RabbitMQ OCF does more bookkeeping, for example, blocking client access when RabbitMQ cluster is not ready. I beleive the upstream OCF should be OK to use as well after I read the code, but it might not fit into the Fuel project. As far as I test, the Fuel OCF script is good except sometimes the full reassemble time is long, and as I find out, it is mostly because the Fuel MySQL Galera OCF script keeps pacemaker from promoting RabbitMQ resource, as I mentioned in the previous emails. Maybe Vladimir and Sergey can give us more insight on why Fuel needs a master-slave RabbitMQ. That would be good to know. Browsing the agent, promote seems to be a no-op if rabbit is already running. To the master / slave reason due to how the ocf script is structured to deal with rabbit's poor ability to handle its self in some scenarios. Hopefully the state transition diagram [5] is enough to clarify what's going on. [5] http://goo.gl/PPNrw7 Not really. It seems to be under the impression you can skip started and go directly from stopped to master. __ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
On Thu, May 7, 2015 at 5:01 PM Andrew Beekhof abeek...@redhat.com wrote: On 5 May 2015, at 1:19 pm, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: Thank you Andrew. on 2015/05/05 08:03, Andrew Beekhof wrote: On 28 Apr 2015, at 11:15 pm, Bogdan Dobrelya bdobre...@mirantis.com wrote: Hello, Hello, Zhou I using Fuel 6.0.1 and find that RabbitMQ recover time is long after power failure. I have a running HA environment, then I reset power of all the machines at the same time. I observe that after reboot it usually takes 10 minutes for RabittMQ cluster to appear running master-slave mode in pacemaker. If I power off all the 3 controllers and only start 2 of them, the downtime sometimes can be as long as 20 minutes. Yes, this is a known issue [0]. Note, there were many bugfixes, like [1],[2],[3], merged for MQ OCF script, so you may want to try to backport them as well by the following guide [4] [0] https://bugs.launchpad.net/fuel/+bug/1432603 [1] https://review.openstack.org/#/c/175460/ [2] https://review.openstack.org/#/c/175457/ [3] https://review.openstack.org/#/c/175371/ [4] https://review.openstack.org/#/c/170476/ Is there a reason you’re using a custom OCF script instead of the upstream[a] one? Please have a chat with David (the maintainer, in CC) if there is something you believe is wrong with it. [a] https://github.com/ClusterLabs/resource-agents/blob/master/heartbeat/rabbitmq-cluster I'm using the OCF script from the Fuel project, specifically from the 6.0 stable branch [alpha]. Ah, I’m still learning who is who... i thought you were part of that project :-) Comparing with upstream OCF code, the main difference is that Fuel RabbitMQ OCF is a master-slave resource. Fuel RabbitMQ OCF does more bookkeeping, for example, blocking client access when RabbitMQ cluster is not ready. I beleive the upstream OCF should be OK to use as well after I read the code, but it might not fit into the Fuel project. As far as I test, the Fuel OCF script is good except sometimes the full reassemble time is long, and as I find out, it is mostly because the Fuel MySQL Galera OCF script keeps pacemaker from promoting RabbitMQ resource, as I mentioned in the previous emails. Maybe Vladimir and Sergey can give us more insight on why Fuel needs a master-slave RabbitMQ. That would be good to know. Browsing the agent, promote seems to be a no-op if rabbit is already running. To the master / slave reason due to how the ocf script is structured to deal with rabbit's poor ability to handle its self in some scenarios. Hopefully the state transition diagram [5] is enough to clarify what's going on. [5] http://goo.gl/PPNrw7 I see Vladimir and Sergey works on the original Fuel blueprint RabbitMQ cluster [beta]. [alpha] https://github.com/stackforge/fuel-library/blob/stable/6.0/deployment/puppet/nova/files/ocf/rabbitmq [beta] https://blueprints.launchpad.net/fuel/+spec/rabbitmq-cluster-controlled-by-pacemaker I have a little investigation and find out there are some possible causes. 1. MySQL Recovery Takes Too Long [1] and Blocking RabbitMQ Clustering in Pacemaker The pacemaker resource p_mysql start timeout is set to 475s. Sometimes MySQL-wss fails to start after power failure, and pacemaker would wait 475s before retry starting it. The problem is that pacemaker divides resource state transitions into batches. Since RabbitMQ is master-slave resource, I assume that starting all the slaves and promoting master are put into two different batches. If unfortunately starting all RabbitMQ slaves are put in the same batch as MySQL starting, even if RabbitMQ slaves and all other resources are ready, pacemaker will not continue but just wait for MySQL timeout. Could you please elaborate the what is the same/different batches for MQ and DB? Note, there is a MQ clustering logic flow charts available here [5] and we're planning to release a dedicated technical bulletin for this. [5] http://goo.gl/PPNrw7 I can re-produce this by hard powering off all the controllers and start them again. It's more likely to trigger MySQL failure in this way. Then I observe that if there is one cloned mysql instance not starting, the whole pacemaker cluster gets stuck and does not emit any log. On the host of the failed instance, I can see a mysql resource agent process calling the sleep command. If I kill that process, the pacemaker comes back alive and RabbitMQ master gets promoted. In fact this long timeout is blocking every resource from state transition in pacemaker. This maybe a known problem of pacemaker and there are some discussions in Linux-HA mailing list [2]. It might not be fixed in the near future. It seems in generally it's bad to have long timeout in state transition actions (start/stop/promote/demote). There maybe another way to implement
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
On 5 May 2015, at 1:19 pm, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: Thank you Andrew. on 2015/05/05 08:03, Andrew Beekhof wrote: On 28 Apr 2015, at 11:15 pm, Bogdan Dobrelya bdobre...@mirantis.com wrote: Hello, Hello, Zhou I using Fuel 6.0.1 and find that RabbitMQ recover time is long after power failure. I have a running HA environment, then I reset power of all the machines at the same time. I observe that after reboot it usually takes 10 minutes for RabittMQ cluster to appear running master-slave mode in pacemaker. If I power off all the 3 controllers and only start 2 of them, the downtime sometimes can be as long as 20 minutes. Yes, this is a known issue [0]. Note, there were many bugfixes, like [1],[2],[3], merged for MQ OCF script, so you may want to try to backport them as well by the following guide [4] [0] https://bugs.launchpad.net/fuel/+bug/1432603 [1] https://review.openstack.org/#/c/175460/ [2] https://review.openstack.org/#/c/175457/ [3] https://review.openstack.org/#/c/175371/ [4] https://review.openstack.org/#/c/170476/ Is there a reason you’re using a custom OCF script instead of the upstream[a] one? Please have a chat with David (the maintainer, in CC) if there is something you believe is wrong with it. [a] https://github.com/ClusterLabs/resource-agents/blob/master/heartbeat/rabbitmq-cluster I'm using the OCF script from the Fuel project, specifically from the 6.0 stable branch [alpha]. Ah, I’m still learning who is who... i thought you were part of that project :-) Comparing with upstream OCF code, the main difference is that Fuel RabbitMQ OCF is a master-slave resource. Fuel RabbitMQ OCF does more bookkeeping, for example, blocking client access when RabbitMQ cluster is not ready. I beleive the upstream OCF should be OK to use as well after I read the code, but it might not fit into the Fuel project. As far as I test, the Fuel OCF script is good except sometimes the full reassemble time is long, and as I find out, it is mostly because the Fuel MySQL Galera OCF script keeps pacemaker from promoting RabbitMQ resource, as I mentioned in the previous emails. Maybe Vladimir and Sergey can give us more insight on why Fuel needs a master-slave RabbitMQ. That would be good to know. Browsing the agent, promote seems to be a no-op if rabbit is already running. I see Vladimir and Sergey works on the original Fuel blueprint RabbitMQ cluster [beta]. [alpha] https://github.com/stackforge/fuel-library/blob/stable/6.0/deployment/puppet/nova/files/ocf/rabbitmq [beta] https://blueprints.launchpad.net/fuel/+spec/rabbitmq-cluster-controlled-by-pacemaker I have a little investigation and find out there are some possible causes. 1. MySQL Recovery Takes Too Long [1] and Blocking RabbitMQ Clustering in Pacemaker The pacemaker resource p_mysql start timeout is set to 475s. Sometimes MySQL-wss fails to start after power failure, and pacemaker would wait 475s before retry starting it. The problem is that pacemaker divides resource state transitions into batches. Since RabbitMQ is master-slave resource, I assume that starting all the slaves and promoting master are put into two different batches. If unfortunately starting all RabbitMQ slaves are put in the same batch as MySQL starting, even if RabbitMQ slaves and all other resources are ready, pacemaker will not continue but just wait for MySQL timeout. Could you please elaborate the what is the same/different batches for MQ and DB? Note, there is a MQ clustering logic flow charts available here [5] and we're planning to release a dedicated technical bulletin for this. [5] http://goo.gl/PPNrw7 I can re-produce this by hard powering off all the controllers and start them again. It's more likely to trigger MySQL failure in this way. Then I observe that if there is one cloned mysql instance not starting, the whole pacemaker cluster gets stuck and does not emit any log. On the host of the failed instance, I can see a mysql resource agent process calling the sleep command. If I kill that process, the pacemaker comes back alive and RabbitMQ master gets promoted. In fact this long timeout is blocking every resource from state transition in pacemaker. This maybe a known problem of pacemaker and there are some discussions in Linux-HA mailing list [2]. It might not be fixed in the near future. It seems in generally it's bad to have long timeout in state transition actions (start/stop/promote/demote). There maybe another way to implement MySQL-wss resource agent to use a short start timeout and monitor the wss cluster state using monitor action. This is very interesting, thank you! I believe all commands for MySQL RA OCF script should be as well wrapped with timeout -SIGTERM or -SIGKILL as we did for MQ RA OCF. And there should no be any sleep calls. I created a bug for this [6]. [6] https://bugs.launchpad.net/fuel/+bug/1449542 I also
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
On 5 May 2015, at 9:30 pm, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: Thank you Andrew. Sorry for misspell your name in the previous email. on 2015/05/05 14:25, Andrew Beekhof wrote: On 5 May 2015, at 2:31 pm, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: Thank you Bogdan for clearing the pacemaker promotion process for me. on 2015/05/05 10:32, Andrew Beekhof wrote: On 29 Apr 2015, at 5:38 pm, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: [snip] Batch is a pacemaker concept I found when I was reading its documentation and code. There is a batch-limit: 30 in the output of pcs property list --all. The pacemaker official documentation explanation is that it's The number of jobs that the TE is allowed to execute in parallel. From my understanding, pacemaker maintains cluster states, and when we start/stop/promote/demote a resource, it triggers a state transition. Pacemaker puts as many as possible transition jobs into a batch, and process them in parallel. Technically it calculates an ordered graph of actions that need to be performed for a set of related resources. You can see an example of the kinds of graphs it produces at: http://clusterlabs.org/doc/en-US/Pacemaker/1.1-pcs/html/Pacemaker_Explained/s-config-testing-changes.html There is a more complex one which includes promotion and demotion on the next page. The number of actions that can run at any one time is therefor limited by - the value of batch-limit (the total number of in-flight actions) - the number of resources that do not have ordering constraints between them (eg. rsc{1,2,3} in the above example) So in the above example, if batch-limit = 3, the monitor_0 actions will still all execute in parallel. If batch-limit == 2, one of them will be deferred until the others complete. Processing of the graph stops the moment any action returns a value that was not expected. If that happens, we wait for currently in-flight actions to complete, re-calculate a new graph based on the new information and start again. So can I infer the following statement? In a big cluster with many resources, chances are some resource agent actions return unexpected values, The size of the cluster shouldn’t increase the chance of this happening unless you’ve set the timeouts too aggressively. If there are many types of resource agents, and anyone of them is not well written, it might cause trouble, right? Yes, but really only for the things that depend on it. For example if resources B, C, D, E all depend (in some way) on A, then their startup is going to be delayed. But F, G, H and J will be able to start while we wait around for B to time out. and if any of the in-flight action timeout is long, it would block pacemaker from re-calculating a new transition graph? Yes, but its actually an argument for making the timeouts longer, not shorter. Setting the timeouts too aggressively actually increases downtime because of all the extra delays and recovery it induces. So set them to be long enough that there is unquestionably a problem if you hit them. But we absolutely recognise that starting/stopping a database can take a very long time comparatively and that it shouldn’t block recovery of other unrelated services. I would expect to see this land in Pacemaker 1.1.14 It will be great to see this in Pacemaker 1.1.14. From my experience using Pacemaker, I think customized resource agents are possibly the weakest part. This is why we encourage people wanting new agents to get involved with the upstream resource-agents project :-) This feature should improve the handling for resource action timeouts. I see the current batch-limit is 30 and I tried to increase it to 100, but did not help. Correct. It only puts an upper limit on the number of in-flight actions, actions still need to wait for all their dependants to complete before executing. I'm sure that the cloned MySQL Galera resource is not related to master-slave RabbitMQ resource. I don't find any dependency, order or rule connecting them in the cluster deployed by Fuel [1]. In general it should not have needed to wait, but if you send me a crm_report covering the period you’re talking about I’ll be able to comment specifically about the behaviour you saw. You are very nice, thank you. I uploaded the file generated by crm_report to google drive. https://drive.google.com/file/d/0B_vDkYRYHPSIZ29NdzV3NXotYU0/view?usp=sharing Hmmm... there’s no logs included here for some reason. I suspect it a bug on my part, can you apply this patch to report.collector on the machine you’re running crm_report from and retry? https://github.com/ClusterLabs/pacemaker/commit/96427ec Is there anything I can do to make sure all the resource actions return expected values in a full reassembling? In general, if we say ‘start’, do your best to start or return ‘0’ if you
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
On 5 May 2015, at 7:52 pm, Bogdan Dobrelya bdobre...@mirantis.com wrote: On 05.05.2015 04:32, Andrew Beekhof wrote: [snip] Technically it calculates an ordered graph of actions that need to be performed for a set of related resources. You can see an example of the kinds of graphs it produces at: http://clusterlabs.org/doc/en-US/Pacemaker/1.1-pcs/html/Pacemaker_Explained/s-config-testing-changes.html There is a more complex one which includes promotion and demotion on the next page. The number of actions that can run at any one time is therefor limited by - the value of batch-limit (the total number of in-flight actions) - the number of resources that do not have ordering constraints between them (eg. rsc{1,2,3} in the above example) So in the above example, if batch-limit = 3, the monitor_0 actions will still all execute in parallel. If batch-limit == 2, one of them will be deferred until the others complete. Processing of the graph stops the moment any action returns a value that was not expected. If that happens, we wait for currently in-flight actions to complete, re-calculate a new graph based on the new information and start again. First we do a non-recurring monitor (*_monitor_0) to check what state the resource is in. We can’t assume its off because a) we might have crashed, b) the admin might have accidentally configured it to start at boot or c) the admin may have asked us to re-check everything. Also important to know, the order of actions is: I should clarify something here: s/actions is/actions for each resource is/ 1. any necessary demotions 2. any necessary stops 3. any necessary starts 4. any necessary promotions Thank you for explaining this, Andrew! So, in the context of the given two example DB(MySQL) and messaging(RabbitMQ) resources: The problem is that pacemaker can only promote a resource after it detects the resource is started. During a full reassemble, in the first transition batch, pacemaker starts all the resources including MySQL and RabbitMQ. Pacemaker issues resource agent start invocation in parallel and reaps the results. For a multi-state resource agent like RabbitMQ, pacemaker needs the start result reported in the first batch, then transition engine and policy engine decide if it has to retry starting or promote, and put this new transition job into a new batch. So, for given example, it looks like we currently have: _batch start_ ... 3. DB, messaging resources start in a one batch Since there is no dependancy between them, yes. 4. messaging resource promote blocked by the step 3 completion _batch end_ Not quite, I wasn’t as clear as I could have been in my previous email. We wont promote Rabbit instances until all they have all been started. However we don’t need to wait for all the DBs to finish starting (again, because there is no dependancy between them) before we begin promoting Rabbit. So a single transition that did this is totally possible: t0. Begin transition t1. Rabbit start node1(begin) t2. DB start node 3 (begin) t3. DB start node 2 (begin) t4. Rabbit start node2(begin) t5. Rabbit start node3(begin) t6. DB start node 1 (begin) t7. Rabbit start node2(complete) t8. Rabbit start node1(complete) t9. DB start node 3 (complete) t10. Rabbit start node3(complete) t11. Rabbit promote node 1 (begin) t12. Rabbit promote node 3 (begin) t13. Rabbit promote node 2 (begin) ... etc etc ... For something like cinder however, these are some of the dependancies we define: pcs constraint order start keystone-clone then cinder-api-clone pcs constraint order start cinder-api-clone then cinder-scheduler-clone pcs constraint order start galera-master then keystone-clone So first all the galera instances must be started. Then we can begin to promote some. Once all the promotions complete, then we can start the keystone instances. Once all the keystone instances are up, then we can bring up the cinder API instances, which allows us to start the scheduler, etc etc. And assuming nothing fails, this can all happen in one transition. Bottom line: Pacemaker will do as much as it can as soon as it can. The only restrictions are ordering constraints you specify, the batch-limit, and each master/slave (or clone) resource’s _internal_ demote-stop-start-promote ordering. Am I making it better or worse? Does this mean what an artificial constraints ordering between DB and messaging could help them to get into the separate transition batches, like: ... 3. messaging multistate clone resource start 4. messaging multistate clone resource promote _batch end_ _next batch start_ ... 3. DB simple clone resource start ? -- Best regards, Bogdan Dobrelya, Skype #bogdando_at_yahoo.com Irc #bogdando __
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
Thank you Andrew. Sorry for misspell your name in the previous email. on 2015/05/05 14:25, Andrew Beekhof wrote: On 5 May 2015, at 2:31 pm, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: Thank you Bogdan for clearing the pacemaker promotion process for me. on 2015/05/05 10:32, Andrew Beekhof wrote: On 29 Apr 2015, at 5:38 pm, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: [snip] Batch is a pacemaker concept I found when I was reading its documentation and code. There is a batch-limit: 30 in the output of pcs property list --all. The pacemaker official documentation explanation is that it's The number of jobs that the TE is allowed to execute in parallel. From my understanding, pacemaker maintains cluster states, and when we start/stop/promote/demote a resource, it triggers a state transition. Pacemaker puts as many as possible transition jobs into a batch, and process them in parallel. Technically it calculates an ordered graph of actions that need to be performed for a set of related resources. You can see an example of the kinds of graphs it produces at: http://clusterlabs.org/doc/en-US/Pacemaker/1.1-pcs/html/Pacemaker_Explained/s-config-testing-changes.html There is a more complex one which includes promotion and demotion on the next page. The number of actions that can run at any one time is therefor limited by - the value of batch-limit (the total number of in-flight actions) - the number of resources that do not have ordering constraints between them (eg. rsc{1,2,3} in the above example) So in the above example, if batch-limit = 3, the monitor_0 actions will still all execute in parallel. If batch-limit == 2, one of them will be deferred until the others complete. Processing of the graph stops the moment any action returns a value that was not expected. If that happens, we wait for currently in-flight actions to complete, re-calculate a new graph based on the new information and start again. So can I infer the following statement? In a big cluster with many resources, chances are some resource agent actions return unexpected values, The size of the cluster shouldn’t increase the chance of this happening unless you’ve set the timeouts too aggressively. If there are many types of resource agents, and anyone of them is not well written, it might cause trouble, right? and if any of the in-flight action timeout is long, it would block pacemaker from re-calculating a new transition graph? Yes, but its actually an argument for making the timeouts longer, not shorter. Setting the timeouts too aggressively actually increases downtime because of all the extra delays and recovery it induces. So set them to be long enough that there is unquestionably a problem if you hit them. But we absolutely recognise that starting/stopping a database can take a very long time comparatively and that it shouldn’t block recovery of other unrelated services. I would expect to see this land in Pacemaker 1.1.14 It will be great to see this in Pacemaker 1.1.14. From my experience using Pacemaker, I think customized resource agents are possibly the weakest part. This feature should improve the handling for resource action timeouts. I see the current batch-limit is 30 and I tried to increase it to 100, but did not help. Correct. It only puts an upper limit on the number of in-flight actions, actions still need to wait for all their dependants to complete before executing. I'm sure that the cloned MySQL Galera resource is not related to master-slave RabbitMQ resource. I don't find any dependency, order or rule connecting them in the cluster deployed by Fuel [1]. In general it should not have needed to wait, but if you send me a crm_report covering the period you’re talking about I’ll be able to comment specifically about the behaviour you saw. You are very nice, thank you. I uploaded the file generated by crm_report to google drive. https://drive.google.com/file/d/0B_vDkYRYHPSIZ29NdzV3NXotYU0/view?usp=sharing Is there anything I can do to make sure all the resource actions return expected values in a full reassembling? In general, if we say ‘start’, do your best to start or return ‘0’ if you already were started. Likewise for stop. Otherwise its really specific to your agent. For example an IP resource just needs to add itself to an interface - it cant do much differently, if it times out then the system much be very very busy. The only other thing I would say is: - avoid blocking calls where possible - have empathy for the machine (do as little as is needed) +1 for the empathy :) Is it because node-1 and node-2 happen to boot faster than node-3 and form a cluster, when node-3 joins, it triggers new state transition? Or may because some resources are already started, so pacemaker needs to stop them firstly? We only stop them if they shouldn’t yet be running (ie. a colocation or ordering dependancy
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
On 5 May 2015, at 2:31 pm, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: Thank you Bogdan for clearing the pacemaker promotion process for me. on 2015/05/05 10:32, Andrew Beekhof wrote: On 29 Apr 2015, at 5:38 pm, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: [snip] Batch is a pacemaker concept I found when I was reading its documentation and code. There is a batch-limit: 30 in the output of pcs property list --all. The pacemaker official documentation explanation is that it's The number of jobs that the TE is allowed to execute in parallel. From my understanding, pacemaker maintains cluster states, and when we start/stop/promote/demote a resource, it triggers a state transition. Pacemaker puts as many as possible transition jobs into a batch, and process them in parallel. Technically it calculates an ordered graph of actions that need to be performed for a set of related resources. You can see an example of the kinds of graphs it produces at: http://clusterlabs.org/doc/en-US/Pacemaker/1.1-pcs/html/Pacemaker_Explained/s-config-testing-changes.html There is a more complex one which includes promotion and demotion on the next page. The number of actions that can run at any one time is therefor limited by - the value of batch-limit (the total number of in-flight actions) - the number of resources that do not have ordering constraints between them (eg. rsc{1,2,3} in the above example) So in the above example, if batch-limit = 3, the monitor_0 actions will still all execute in parallel. If batch-limit == 2, one of them will be deferred until the others complete. Processing of the graph stops the moment any action returns a value that was not expected. If that happens, we wait for currently in-flight actions to complete, re-calculate a new graph based on the new information and start again. So can I infer the following statement? In a big cluster with many resources, chances are some resource agent actions return unexpected values, The size of the cluster shouldn’t increase the chance of this happening unless you’ve set the timeouts too aggressively. and if any of the in-flight action timeout is long, it would block pacemaker from re-calculating a new transition graph? Yes, but its actually an argument for making the timeouts longer, not shorter. Setting the timeouts too aggressively actually increases downtime because of all the extra delays and recovery it induces. So set them to be long enough that there is unquestionably a problem if you hit them. But we absolutely recognise that starting/stopping a database can take a very long time comparatively and that it shouldn’t block recovery of other unrelated services. I would expect to see this land in Pacemaker 1.1.14 I see the current batch-limit is 30 and I tried to increase it to 100, but did not help. Correct. It only puts an upper limit on the number of in-flight actions, actions still need to wait for all their dependants to complete before executing. I'm sure that the cloned MySQL Galera resource is not related to master-slave RabbitMQ resource. I don't find any dependency, order or rule connecting them in the cluster deployed by Fuel [1]. In general it should not have needed to wait, but if you send me a crm_report covering the period you’re talking about I’ll be able to comment specifically about the behaviour you saw. Is there anything I can do to make sure all the resource actions return expected values in a full reassembling? In general, if we say ‘start’, do your best to start or return ‘0’ if you already were started. Likewise for stop. Otherwise its really specific to your agent. For example an IP resource just needs to add itself to an interface - it cant do much differently, if it times out then the system much be very very busy. The only other thing I would say is: - avoid blocking calls where possible - have empathy for the machine (do as little as is needed) Is it because node-1 and node-2 happen to boot faster than node-3 and form a cluster, when node-3 joins, it triggers new state transition? Or may because some resources are already started, so pacemaker needs to stop them firstly? We only stop them if they shouldn’t yet be running (ie. a colocation or ordering dependancy is not yet started also). Does setting default-resource-stickiness to 1 help? From 0 or INFINITY? I also tried crm history XXX commands in a live and correct cluster, I’m not familiar with that tool anymore. but didn't find much information. I can see there are many log entries like run_graph: Transition 7108 Next I'll inspect the pacemaker log to see which resource action returns the unexpected value or which thing triggers new state transition. [1] http://paste.openstack.org/show/214919/ I’d not recommend mixing the two CLI tools. The problem is that pacemaker can only promote a resource after it detects the resource is
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
On 05.05.2015 04:32, Andrew Beekhof wrote: [snip] Technically it calculates an ordered graph of actions that need to be performed for a set of related resources. You can see an example of the kinds of graphs it produces at: http://clusterlabs.org/doc/en-US/Pacemaker/1.1-pcs/html/Pacemaker_Explained/s-config-testing-changes.html There is a more complex one which includes promotion and demotion on the next page. The number of actions that can run at any one time is therefor limited by - the value of batch-limit (the total number of in-flight actions) - the number of resources that do not have ordering constraints between them (eg. rsc{1,2,3} in the above example) So in the above example, if batch-limit = 3, the monitor_0 actions will still all execute in parallel. If batch-limit == 2, one of them will be deferred until the others complete. Processing of the graph stops the moment any action returns a value that was not expected. If that happens, we wait for currently in-flight actions to complete, re-calculate a new graph based on the new information and start again. First we do a non-recurring monitor (*_monitor_0) to check what state the resource is in. We can’t assume its off because a) we might have crashed, b) the admin might have accidentally configured it to start at boot or c) the admin may have asked us to re-check everything. Also important to know, the order of actions is: 1. any necessary demotions 2. any necessary stops 3. any necessary starts 4. any necessary promotions Thank you for explaining this, Andrew! So, in the context of the given two example DB(MySQL) and messaging(RabbitMQ) resources: The problem is that pacemaker can only promote a resource after it detects the resource is started. During a full reassemble, in the first transition batch, pacemaker starts all the resources including MySQL and RabbitMQ. Pacemaker issues resource agent start invocation in parallel and reaps the results. For a multi-state resource agent like RabbitMQ, pacemaker needs the start result reported in the first batch, then transition engine and policy engine decide if it has to retry starting or promote, and put this new transition job into a new batch. So, for given example, it looks like we currently have: _batch start_ ... 3. DB, messaging resources start in a one batch 4. messaging resource promote blocked by the step 3 completion _batch end_ Does this mean what an artificial constraints ordering between DB and messaging could help them to get into the separate transition batches, like: ... 3. messaging multistate clone resource start 4. messaging multistate clone resource promote _batch end_ _next batch start_ ... 3. DB simple clone resource start ? -- Best regards, Bogdan Dobrelya, Skype #bogdando_at_yahoo.com Irc #bogdando __ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
Thank you Andrew. on 2015/05/05 08:03, Andrew Beekhof wrote: On 28 Apr 2015, at 11:15 pm, Bogdan Dobrelya bdobre...@mirantis.com wrote: Hello, Hello, Zhou I using Fuel 6.0.1 and find that RabbitMQ recover time is long after power failure. I have a running HA environment, then I reset power of all the machines at the same time. I observe that after reboot it usually takes 10 minutes for RabittMQ cluster to appear running master-slave mode in pacemaker. If I power off all the 3 controllers and only start 2 of them, the downtime sometimes can be as long as 20 minutes. Yes, this is a known issue [0]. Note, there were many bugfixes, like [1],[2],[3], merged for MQ OCF script, so you may want to try to backport them as well by the following guide [4] [0] https://bugs.launchpad.net/fuel/+bug/1432603 [1] https://review.openstack.org/#/c/175460/ [2] https://review.openstack.org/#/c/175457/ [3] https://review.openstack.org/#/c/175371/ [4] https://review.openstack.org/#/c/170476/ Is there a reason you’re using a custom OCF script instead of the upstream[a] one? Please have a chat with David (the maintainer, in CC) if there is something you believe is wrong with it. [a] https://github.com/ClusterLabs/resource-agents/blob/master/heartbeat/rabbitmq-cluster I'm using the OCF script from the Fuel project, specifically from the 6.0 stable branch [alpha]. Comparing with upstream OCF code, the main difference is that Fuel RabbitMQ OCF is a master-slave resource. Fuel RabbitMQ OCF does more bookkeeping, for example, blocking client access when RabbitMQ cluster is not ready. I beleive the upstream OCF should be OK to use as well after I read the code, but it might not fit into the Fuel project. As far as I test, the Fuel OCF script is good except sometimes the full reassemble time is long, and as I find out, it is mostly because the Fuel MySQL Galera OCF script keeps pacemaker from promoting RabbitMQ resource, as I mentioned in the previous emails. Maybe Vladimir and Sergey can give us more insight on why Fuel needs a master-slave RabbitMQ. I see Vladimir and Sergey works on the original Fuel blueprint RabbitMQ cluster [beta]. [alpha] https://github.com/stackforge/fuel-library/blob/stable/6.0/deployment/puppet/nova/files/ocf/rabbitmq [beta] https://blueprints.launchpad.net/fuel/+spec/rabbitmq-cluster-controlled-by-pacemaker I have a little investigation and find out there are some possible causes. 1. MySQL Recovery Takes Too Long [1] and Blocking RabbitMQ Clustering in Pacemaker The pacemaker resource p_mysql start timeout is set to 475s. Sometimes MySQL-wss fails to start after power failure, and pacemaker would wait 475s before retry starting it. The problem is that pacemaker divides resource state transitions into batches. Since RabbitMQ is master-slave resource, I assume that starting all the slaves and promoting master are put into two different batches. If unfortunately starting all RabbitMQ slaves are put in the same batch as MySQL starting, even if RabbitMQ slaves and all other resources are ready, pacemaker will not continue but just wait for MySQL timeout. Could you please elaborate the what is the same/different batches for MQ and DB? Note, there is a MQ clustering logic flow charts available here [5] and we're planning to release a dedicated technical bulletin for this. [5] http://goo.gl/PPNrw7 I can re-produce this by hard powering off all the controllers and start them again. It's more likely to trigger MySQL failure in this way. Then I observe that if there is one cloned mysql instance not starting, the whole pacemaker cluster gets stuck and does not emit any log. On the host of the failed instance, I can see a mysql resource agent process calling the sleep command. If I kill that process, the pacemaker comes back alive and RabbitMQ master gets promoted. In fact this long timeout is blocking every resource from state transition in pacemaker. This maybe a known problem of pacemaker and there are some discussions in Linux-HA mailing list [2]. It might not be fixed in the near future. It seems in generally it's bad to have long timeout in state transition actions (start/stop/promote/demote). There maybe another way to implement MySQL-wss resource agent to use a short start timeout and monitor the wss cluster state using monitor action. This is very interesting, thank you! I believe all commands for MySQL RA OCF script should be as well wrapped with timeout -SIGTERM or -SIGKILL as we did for MQ RA OCF. And there should no be any sleep calls. I created a bug for this [6]. [6] https://bugs.launchpad.net/fuel/+bug/1449542 I also find a fix to improve MySQL start timeout [3]. It shortens the timeout to 300s. At the time I sending this email, I can not find it in stable/6.0 branch. Maybe the maintainer needs to cherry-pick it to stable/6.0 ? [1] https://bugs.launchpad.net/fuel/+bug/1441885 [2]
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
Thank you Bogdan for clearing the pacemaker promotion process for me. on 2015/05/05 10:32, Andrew Beekhof wrote: On 29 Apr 2015, at 5:38 pm, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: [snip] Batch is a pacemaker concept I found when I was reading its documentation and code. There is a batch-limit: 30 in the output of pcs property list --all. The pacemaker official documentation explanation is that it's The number of jobs that the TE is allowed to execute in parallel. From my understanding, pacemaker maintains cluster states, and when we start/stop/promote/demote a resource, it triggers a state transition. Pacemaker puts as many as possible transition jobs into a batch, and process them in parallel. Technically it calculates an ordered graph of actions that need to be performed for a set of related resources. You can see an example of the kinds of graphs it produces at: http://clusterlabs.org/doc/en-US/Pacemaker/1.1-pcs/html/Pacemaker_Explained/s-config-testing-changes.html There is a more complex one which includes promotion and demotion on the next page. The number of actions that can run at any one time is therefor limited by - the value of batch-limit (the total number of in-flight actions) - the number of resources that do not have ordering constraints between them (eg. rsc{1,2,3} in the above example) So in the above example, if batch-limit = 3, the monitor_0 actions will still all execute in parallel. If batch-limit == 2, one of them will be deferred until the others complete. Processing of the graph stops the moment any action returns a value that was not expected. If that happens, we wait for currently in-flight actions to complete, re-calculate a new graph based on the new information and start again. So can I infer the following statement? In a big cluster with many resources, chances are some resource agent actions return unexpected values, and if any of the in-flight action timeout is long, it would block pacemaker from re-calculating a new transition graph? I see the current batch-limit is 30 and I tried to increase it to 100, but did not help. I'm sure that the cloned MySQL Galera resource is not related to master-slave RabbitMQ resource. I don't find any dependency, order or rule connecting them in the cluster deployed by Fuel [1]. Is there anything I can do to make sure all the resource actions return expected values in a full reassembling? Is it because node-1 and node-2 happen to boot faster than node-3 and form a cluster, when node-3 joins, it triggers new state transition? Or may because some resources are already started, so pacemaker needs to stop them firstly? Does setting default-resource-stickiness to 1 help? I also tried crm history XXX commands in a live and correct cluster, but didn't find much information. I can see there are many log entries like run_graph: Transition 7108 Next I'll inspect the pacemaker log to see which resource action returns the unexpected value or which thing triggers new state transition. [1] http://paste.openstack.org/show/214919/ The problem is that pacemaker can only promote a resource after it detects the resource is started. First we do a non-recurring monitor (*_monitor_0) to check what state the resource is in. We can’t assume its off because a) we might have crashed, b) the admin might have accidentally configured it to start at boot or c) the admin may have asked us to re-check everything. During a full reassemble, in the first transition batch, pacemaker starts all the resources including MySQL and RabbitMQ. Pacemaker issues resource agent start invocation in parallel and reaps the results. For a multi-state resource agent like RabbitMQ, pacemaker needs the start result reported in the first batch, then transition engine and policy engine decide if it has to retry starting or promote, and put this new transition job into a new batch. Also important to know, the order of actions is: 1. any necessary demotions 2. any necessary stops 3. any necessary starts 4. any necessary promotions __ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev -- Best wishes! Zhou Zheng Sheng / 周征晟 Software Engineer Beijing AWcloud Software Co., Ltd. __ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
On 28 Apr 2015, at 11:15 pm, Bogdan Dobrelya bdobre...@mirantis.com wrote: Hello, Hello, Zhou I using Fuel 6.0.1 and find that RabbitMQ recover time is long after power failure. I have a running HA environment, then I reset power of all the machines at the same time. I observe that after reboot it usually takes 10 minutes for RabittMQ cluster to appear running master-slave mode in pacemaker. If I power off all the 3 controllers and only start 2 of them, the downtime sometimes can be as long as 20 minutes. Yes, this is a known issue [0]. Note, there were many bugfixes, like [1],[2],[3], merged for MQ OCF script, so you may want to try to backport them as well by the following guide [4] [0] https://bugs.launchpad.net/fuel/+bug/1432603 [1] https://review.openstack.org/#/c/175460/ [2] https://review.openstack.org/#/c/175457/ [3] https://review.openstack.org/#/c/175371/ [4] https://review.openstack.org/#/c/170476/ Is there a reason you’re using a custom OCF script instead of the upstream[a] one? Please have a chat with David (the maintainer, in CC) if there is something you believe is wrong with it. [a] https://github.com/ClusterLabs/resource-agents/blob/master/heartbeat/rabbitmq-cluster I have a little investigation and find out there are some possible causes. 1. MySQL Recovery Takes Too Long [1] and Blocking RabbitMQ Clustering in Pacemaker The pacemaker resource p_mysql start timeout is set to 475s. Sometimes MySQL-wss fails to start after power failure, and pacemaker would wait 475s before retry starting it. The problem is that pacemaker divides resource state transitions into batches. Since RabbitMQ is master-slave resource, I assume that starting all the slaves and promoting master are put into two different batches. If unfortunately starting all RabbitMQ slaves are put in the same batch as MySQL starting, even if RabbitMQ slaves and all other resources are ready, pacemaker will not continue but just wait for MySQL timeout. Could you please elaborate the what is the same/different batches for MQ and DB? Note, there is a MQ clustering logic flow charts available here [5] and we're planning to release a dedicated technical bulletin for this. [5] http://goo.gl/PPNrw7 I can re-produce this by hard powering off all the controllers and start them again. It's more likely to trigger MySQL failure in this way. Then I observe that if there is one cloned mysql instance not starting, the whole pacemaker cluster gets stuck and does not emit any log. On the host of the failed instance, I can see a mysql resource agent process calling the sleep command. If I kill that process, the pacemaker comes back alive and RabbitMQ master gets promoted. In fact this long timeout is blocking every resource from state transition in pacemaker. This maybe a known problem of pacemaker and there are some discussions in Linux-HA mailing list [2]. It might not be fixed in the near future. It seems in generally it's bad to have long timeout in state transition actions (start/stop/promote/demote). There maybe another way to implement MySQL-wss resource agent to use a short start timeout and monitor the wss cluster state using monitor action. This is very interesting, thank you! I believe all commands for MySQL RA OCF script should be as well wrapped with timeout -SIGTERM or -SIGKILL as we did for MQ RA OCF. And there should no be any sleep calls. I created a bug for this [6]. [6] https://bugs.launchpad.net/fuel/+bug/1449542 I also find a fix to improve MySQL start timeout [3]. It shortens the timeout to 300s. At the time I sending this email, I can not find it in stable/6.0 branch. Maybe the maintainer needs to cherry-pick it to stable/6.0 ? [1] https://bugs.launchpad.net/fuel/+bug/1441885 [2] http://lists.linux-ha.org/pipermail/linux-ha/2014-March/047989.html [3] https://review.openstack.org/#/c/171333/ 2. RabbitMQ Resource Agent Breaks Existing Cluster Read the code of the RabbitMQ resource agent, I find it does the following to start RabbitMQ master-slave cluster. On all the controllers: (1) Start Erlang beam process (2) Start RabbitMQ App (If failed, reset mnesia DB and cluster state) (3) Stop RabbitMQ App but do not stop the beam process Then in pacemaker, all the RabbitMQ instances are in slave state. After pacemaker determines the master, it does the following. On the to-be-master host: (4) Start RabbitMQ App (If failed, reset mnesia DB and cluster state) On the slaves hosts: (5) Start RabbitMQ App (If failed, reset mnesia DB and cluster state) (6) Join RabbitMQ cluster of the master host Yes, something like that. As I mentioned, there were several bug fixes in the 6.1 dev, and you can also check the MQ clustering flow charts. As far as I can understand, this process is to make sure the master determined by pacemaker is the same as the master determined in RabbitMQ cluster. If there is no existing
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
On 29 Apr 2015, at 5:38 pm, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: [snip] Batch is a pacemaker concept I found when I was reading its documentation and code. There is a batch-limit: 30 in the output of pcs property list --all. The pacemaker official documentation explanation is that it's The number of jobs that the TE is allowed to execute in parallel. From my understanding, pacemaker maintains cluster states, and when we start/stop/promote/demote a resource, it triggers a state transition. Pacemaker puts as many as possible transition jobs into a batch, and process them in parallel. Technically it calculates an ordered graph of actions that need to be performed for a set of related resources. You can see an example of the kinds of graphs it produces at: http://clusterlabs.org/doc/en-US/Pacemaker/1.1-pcs/html/Pacemaker_Explained/s-config-testing-changes.html There is a more complex one which includes promotion and demotion on the next page. The number of actions that can run at any one time is therefor limited by - the value of batch-limit (the total number of in-flight actions) - the number of resources that do not have ordering constraints between them (eg. rsc{1,2,3} in the above example) So in the above example, if batch-limit = 3, the monitor_0 actions will still all execute in parallel. If batch-limit == 2, one of them will be deferred until the others complete. Processing of the graph stops the moment any action returns a value that was not expected. If that happens, we wait for currently in-flight actions to complete, re-calculate a new graph based on the new information and start again. The problem is that pacemaker can only promote a resource after it detects the resource is started. First we do a non-recurring monitor (*_monitor_0) to check what state the resource is in. We can’t assume its off because a) we might have crashed, b) the admin might have accidentally configured it to start at boot or c) the admin may have asked us to re-check everything. During a full reassemble, in the first transition batch, pacemaker starts all the resources including MySQL and RabbitMQ. Pacemaker issues resource agent start invocation in parallel and reaps the results. For a multi-state resource agent like RabbitMQ, pacemaker needs the start result reported in the first batch, then transition engine and policy engine decide if it has to retry starting or promote, and put this new transition job into a new batch. Also important to know, the order of actions is: 1. any necessary demotions 2. any necessary stops 3. any necessary starts 4. any necessary promotions __ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
Hello Sergii, Thank you for the great explanation on Galera OCF script. I replied your question inline. on 2015/05/03 04:49, Sergii Golovatiuk wrote: Hi Zhou, Galera OCF script is a bit special. Since MySQL keeps the most important data we should find the most recent data on all nodes across the cluster. check_if_galera_pc is specially designed for that. Every server registers the latest status from grastate.dat to CIB. Once all nodes are registered, the one with the most recent data will be selected as Primary Component. All others should join to that node. 5 minutes is a time for all nodes to appear and register position from grastate.dat to CIB. Usually, it takes much faster. Though there are cases when node is stuck on fsck or grub or power outlet or some other cases. If all nodes are registered there shouldn't be 5 minute penalty timeout. If one node is stuck (at least present in CIB), then all other nodes will be waiting for 5 minutes then will assemble cluster without it. Concerning dependencies, I agree that RabbitMQ may start in parallel to Galera cluster assemble procedure. It makes no sense to start other services as they are dependent on Galera and RabbitMQ. Also, I have a quick question to you. Shutting down all three controllers is a unique case, like whole power outage in whole datacenter (DC). In this case, 5 minute delay is very small comparing to DC recovery procedure. Reboot of one controller is more optimistic scenario. What's a special case to restart all 3-5 at once? Sorry, I am not very clear about what 3-5 refers to. Is the question about why we want to make the full reassemble time short, and why this case is important for us? We have some small customers forming a long-tail in local market. They have neither dedicated datacenter houses nor dual power supply. Some of them would even shutdown all the machines when they go home, and start all of the machines when they start to work. Considering of data privacy, they are not willing to put their virtual machines on the public cloud. Usually, this kind of customer don't have IT skills to troubleshoot a full reassemble process. We want to make this process as simple as turning on all the machines roughly at the same time and wait about several minutes, so they don't call our service team. Also, I would like to say a big thank for digging it out. It's very useful to use your findings in our next steps. -- Best regards, Sergii Golovatiuk, Skype #golserge IRC #holser On Wed, Apr 29, 2015 at 9:38 AM, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com mailto:zhengsh...@awcloud.com wrote: Hi! Thank you very much Vladimir and Bogdan! Thanks for the fast respond and rich information. I backported MySQL and RabbitMQ ocf patches from stable/6.0 and tested again. A full reassemble takes about 5mins, this improves a lot. Adding the force_load trick I mentioned in the previous email, it takes about 4mins. I get that there is not really a RabbitMQ master instance because queue masters spreads to all the RabbitMQ instances. The pacemaker master is an abstract one. However there is still an mnesia node from which other mnesia nodes sync table schema. The exception timeout_waiting_for_tables in log is actually reported by mnesia. By default, it places a mark on the last alive mnesia node, and other nodes have to sync table from it (http://www.erlang.org/doc/apps/mnesia/Mnesia_chap7.html#id78477). RabbitMQ clustering inherits the behavior, and the last RabbitMQ instance shutdown must be the first instance to start. Otherwise it produces timeout_waiting_for_tables (http://www.rabbitmq.com/clustering.html#transcript search for the last node to go down). The 1 minute difference is because without force_load, the abstract master determined by pacemaker during a promote action may not be the last RabbitMQ instance shut down in the last start action. So there is chance for rabbitmqctl start_app to wait 30s and trigger a RabbitMQ exception timeout_waiting_for_tables. We may able to see table timeout and mnesa resetting for once during a reassemble process on some of the RabbitMQ instances, but it only introduces 30s of wait, which is acceptable for me. I also inspect the RabbitMQ resource agent code in latest master branch. There are timeout wrapper and other improvements which are great. It does not change the master promotion process much, so it may still run into the problems I described. Please see the inline reply below. on 2015/04/28/ 21:15, Bogdan Dobrelya wrote: Hello, Hello, Zhou I using Fuel 6.0.1 and find that RabbitMQ recover time is long after power failure. I have a running HA environment, then I reset power of all the machines at the
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
Hi! Thank you very much Vladimir and Bogdan! Thanks for the fast respond and rich information. I backported MySQL and RabbitMQ ocf patches from stable/6.0 and tested again. A full reassemble takes about 5mins, this improves a lot. Adding the force_load trick I mentioned in the previous email, it takes about 4mins. I get that there is not really a RabbitMQ master instance because queue masters spreads to all the RabbitMQ instances. The pacemaker master is an abstract one. However there is still an mnesia node from which other mnesia nodes sync table schema. The exception timeout_waiting_for_tables in log is actually reported by mnesia. By default, it places a mark on the last alive mnesia node, and other nodes have to sync table from it (http://www.erlang.org/doc/apps/mnesia/Mnesia_chap7.html#id78477). RabbitMQ clustering inherits the behavior, and the last RabbitMQ instance shutdown must be the first instance to start. Otherwise it produces timeout_waiting_for_tables (http://www.rabbitmq.com/clustering.html#transcript search for the last node to go down). The 1 minute difference is because without force_load, the abstract master determined by pacemaker during a promote action may not be the last RabbitMQ instance shut down in the last start action. So there is chance for rabbitmqctl start_app to wait 30s and trigger a RabbitMQ exception timeout_waiting_for_tables. We may able to see table timeout and mnesa resetting for once during a reassemble process on some of the RabbitMQ instances, but it only introduces 30s of wait, which is acceptable for me. I also inspect the RabbitMQ resource agent code in latest master branch. There are timeout wrapper and other improvements which are great. It does not change the master promotion process much, so it may still run into the problems I described. Please see the inline reply below. on 2015/04/28/ 21:15, Bogdan Dobrelya wrote: Hello, Hello, Zhou I using Fuel 6.0.1 and find that RabbitMQ recover time is long after power failure. I have a running HA environment, then I reset power of all the machines at the same time. I observe that after reboot it usually takes 10 minutes for RabittMQ cluster to appear running master-slave mode in pacemaker. If I power off all the 3 controllers and only start 2 of them, the downtime sometimes can be as long as 20 minutes. Yes, this is a known issue [0]. Note, there were many bugfixes, like [1],[2],[3], merged for MQ OCF script, so you may want to try to backport them as well by the following guide [4] [0] https://bugs.launchpad.net/fuel/+bug/1432603 [1] https://review.openstack.org/#/c/175460/ [2] https://review.openstack.org/#/c/175457/ [3] https://review.openstack.org/#/c/175371/ [4] https://review.openstack.org/#/c/170476/ I have a little investigation and find out there are some possible causes. 1. MySQL Recovery Takes Too Long [1] and Blocking RabbitMQ Clustering in Pacemaker The pacemaker resource p_mysql start timeout is set to 475s. Sometimes MySQL-wss fails to start after power failure, and pacemaker would wait 475s before retry starting it. The problem is that pacemaker divides resource state transitions into batches. Since RabbitMQ is master-slave resource, I assume that starting all the slaves and promoting master are put into two different batches. If unfortunately starting all RabbitMQ slaves are put in the same batch as MySQL starting, even if RabbitMQ slaves and all other resources are ready, pacemaker will not continue but just wait for MySQL timeout. Could you please elaborate the what is the same/different batches for MQ and DB? Note, there is a MQ clustering logic flow charts available here [5] and we're planning to release a dedicated technical bulletin for this. [5] http://goo.gl/PPNrw7 Batch is a pacemaker concept I found when I was reading its documentation and code. There is a batch-limit: 30 in the output of pcs property list --all. The pacemaker official documentation explanation is that it's The number of jobs that the TE is allowed to execute in parallel. From my understanding, pacemaker maintains cluster states, and when we start/stop/promote/demote a resource, it triggers a state transition. Pacemaker puts as many as possible transition jobs into a batch, and process them in parallel. The problem is that pacemaker can only promote a resource after it detects the resource is started. During a full reassemble, in the first transition batch, pacemaker starts all the resources including MySQL and RabbitMQ. Pacemaker issues resource agent start invocation in parallel and reaps the results. For a multi-state resource agent like RabbitMQ, pacemaker needs the start result reported in the first batch, then transition engine and policy engine decide if it has to retry starting or promote, and put this new transition job into a new batch. I see improvements to put individual commands inside a timeout wrapper in RabbitMQ resource agent, and a bug created yesterday
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
On 28.04.2015 15:15, Bogdan Dobrelya wrote: Hello, Zhou Yes, this is a known issue [0]. Note, there were many bugfixes, like [1],[2],[3], merged for MQ OCF script, so you may want to try to backport them as well by the following guide [4] [0] https://bugs.launchpad.net/fuel/+bug/1432603 [1] https://review.openstack.org/#/c/175460/ [2] https://review.openstack.org/#/c/175457/ [3] https://review.openstack.org/#/c/175371/ [4] https://review.openstack.org/#/c/170476/ Could you please elaborate the what is the same/different batches for MQ and DB? Note, there is a MQ clustering logic flow charts available here [5] and we're planning to release a dedicated technical bulletin for this. [5] http://goo.gl/PPNrw7 This is very interesting, thank you! I believe all commands for MySQL RA OCF script should be as well wrapped with timeout -SIGTERM or -SIGKILL as we did for MQ RA OCF. And there should no be any sleep calls. I created a bug for this [6]. [6] https://bugs.launchpad.net/fuel/+bug/1449542 Yes, something like that. As I mentioned, there were several bug fixes in the 6.1 dev, and you can also check the MQ clustering flow charts. after Not exactly. There is no master in mirrored MQ cluster. We define the rabbit_hosts configuration option from Oslo.messaging. What ensures all queue masters will be spread around all of MQ nodes in a long run. And we use a master abstraction only for the Pacemaker RA clustering layer. Here, a master is the MQ node what joins the rest of the MQ nodes. We do erase the node master attribute in CIB for such cases. This should not bring problems into the master election logic. (Note, the RabbitMQ documentation mentions *queue* masters and slaves, which are not the case for the Pacemaker RA clustering abstraction layer.) We made an assumption what the node with the highest MQ uptime should know the most about recent cluster state, so other nodes must join it. RA OCF does not work with queue masters directly. The full MQ cluster reassemble logic is far from the perfect state, indeed. This might erase all mnesia files, hence any custom entities, like users or vhosts, would be removed as well. Note, we do not configure durable queues for Openstack so there is nothing to care about here - the full cluster downtime assumes there will be no AMQP messages stored at all. Yes, this option is only supported for newest RabbitMQ versions. But we definitely should look how this could help. Indeed, there are cases when MQ's autoheal can do nothing with existing partitions and remains partitioned for ever, for example: Masters: [ node-1 ] Slaves: [ node-2 node-3 ] root@node-1:~# rabbitmqctl cluster_status Cluster status of node 'rabbit@node-1' ... [{nodes,[{disc,['rabbit@node-1','rabbit@node-2']}]}, {running_nodes,['rabbit@node-1']}, {cluster_name,rabbit@node-2}, {partitions,[]}] ...done. root@node-2:~# rabbitmqctl cluster_status Cluster status of node 'rabbit@node-2' ... [{nodes,[{disc,['rabbit@node-2']}]}] ...done. root@node-3:~# rabbitmqctl cluster_status Cluster status of node 'rabbit@node-3' ... [{nodes,[{disc,['rabbit@node-1','rabbit@node-2','rabbit@node-3']}]}, {running_nodes,['rabbit@node-3']}, {cluster_name,rabbit@node-2}, {partitions,[]}] Sorry, here is the correct one [0] ! [0] http://pastebin.com/m3fDdMA6 So we should test the pause-minority value as well. But I strongly believe we should make MQ multi-state clone to support many masters, related bp [7] [7] https://blueprints.launchpad.net/fuel/+spec/rabbitmq-pacemaker-multimaster-clone Well, we should not mess the queue masters and multi-clone master for MQ resource in the pacemaker. As I said, pacemaker RA has nothing to do with queue masters. And we introduced this master mostly in order to support the full cluster reassemble case - there must be a node promoted and other nodes should join. This is a very good point, thank you. Thank you for a thorough feedback! This was a really great job. -- Best regards, Bogdan Dobrelya, Skype #bogdando_at_yahoo.com Irc #bogdando __ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
Hello, Hello, Zhou I using Fuel 6.0.1 and find that RabbitMQ recover time is long after power failure. I have a running HA environment, then I reset power of all the machines at the same time. I observe that after reboot it usually takes 10 minutes for RabittMQ cluster to appear running master-slave mode in pacemaker. If I power off all the 3 controllers and only start 2 of them, the downtime sometimes can be as long as 20 minutes. Yes, this is a known issue [0]. Note, there were many bugfixes, like [1],[2],[3], merged for MQ OCF script, so you may want to try to backport them as well by the following guide [4] [0] https://bugs.launchpad.net/fuel/+bug/1432603 [1] https://review.openstack.org/#/c/175460/ [2] https://review.openstack.org/#/c/175457/ [3] https://review.openstack.org/#/c/175371/ [4] https://review.openstack.org/#/c/170476/ I have a little investigation and find out there are some possible causes. 1. MySQL Recovery Takes Too Long [1] and Blocking RabbitMQ Clustering in Pacemaker The pacemaker resource p_mysql start timeout is set to 475s. Sometimes MySQL-wss fails to start after power failure, and pacemaker would wait 475s before retry starting it. The problem is that pacemaker divides resource state transitions into batches. Since RabbitMQ is master-slave resource, I assume that starting all the slaves and promoting master are put into two different batches. If unfortunately starting all RabbitMQ slaves are put in the same batch as MySQL starting, even if RabbitMQ slaves and all other resources are ready, pacemaker will not continue but just wait for MySQL timeout. Could you please elaborate the what is the same/different batches for MQ and DB? Note, there is a MQ clustering logic flow charts available here [5] and we're planning to release a dedicated technical bulletin for this. [5] http://goo.gl/PPNrw7 I can re-produce this by hard powering off all the controllers and start them again. It's more likely to trigger MySQL failure in this way. Then I observe that if there is one cloned mysql instance not starting, the whole pacemaker cluster gets stuck and does not emit any log. On the host of the failed instance, I can see a mysql resource agent process calling the sleep command. If I kill that process, the pacemaker comes back alive and RabbitMQ master gets promoted. In fact this long timeout is blocking every resource from state transition in pacemaker. This maybe a known problem of pacemaker and there are some discussions in Linux-HA mailing list [2]. It might not be fixed in the near future. It seems in generally it's bad to have long timeout in state transition actions (start/stop/promote/demote). There maybe another way to implement MySQL-wss resource agent to use a short start timeout and monitor the wss cluster state using monitor action. This is very interesting, thank you! I believe all commands for MySQL RA OCF script should be as well wrapped with timeout -SIGTERM or -SIGKILL as we did for MQ RA OCF. And there should no be any sleep calls. I created a bug for this [6]. [6] https://bugs.launchpad.net/fuel/+bug/1449542 I also find a fix to improve MySQL start timeout [3]. It shortens the timeout to 300s. At the time I sending this email, I can not find it in stable/6.0 branch. Maybe the maintainer needs to cherry-pick it to stable/6.0 ? [1] https://bugs.launchpad.net/fuel/+bug/1441885 [2] http://lists.linux-ha.org/pipermail/linux-ha/2014-March/047989.html [3] https://review.openstack.org/#/c/171333/ 2. RabbitMQ Resource Agent Breaks Existing Cluster Read the code of the RabbitMQ resource agent, I find it does the following to start RabbitMQ master-slave cluster. On all the controllers: (1) Start Erlang beam process (2) Start RabbitMQ App (If failed, reset mnesia DB and cluster state) (3) Stop RabbitMQ App but do not stop the beam process Then in pacemaker, all the RabbitMQ instances are in slave state. After pacemaker determines the master, it does the following. On the to-be-master host: (4) Start RabbitMQ App (If failed, reset mnesia DB and cluster state) On the slaves hosts: (5) Start RabbitMQ App (If failed, reset mnesia DB and cluster state) (6) Join RabbitMQ cluster of the master host Yes, something like that. As I mentioned, there were several bug fixes in the 6.1 dev, and you can also check the MQ clustering flow charts. As far as I can understand, this process is to make sure the master determined by pacemaker is the same as the master determined in RabbitMQ cluster. If there is no existing cluster, it's fine. If it is run after Not exactly. There is no master in mirrored MQ cluster. We define the rabbit_hosts configuration option from Oslo.messaging. What ensures all queue masters will be spread around all of MQ nodes in a long run. And we use a master abstraction only for the Pacemaker RA clustering layer. Here, a master is the MQ node what joins the rest of the MQ nodes. power failure
Re: [openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
Hi, Zhou Thank you for writing these awesome recommendations. We will look into them and see whether they provide significant impact. BTW, we have found a bunch of issues with our 5.1 and 6.0 RabbitMQ OCF script and fixed them in current master. Would you be so kind as to check out the newest version and say if any of issues mentioned by you are gone? On Tue, Apr 28, 2015 at 9:03 AM, Zhou Zheng Sheng / 周征晟 zhengsh...@awcloud.com wrote: Hello, I using Fuel 6.0.1 and find that RabbitMQ recover time is long after power failure. I have a running HA environment, then I reset power of all the machines at the same time. I observe that after reboot it usually takes 10 minutes for RabittMQ cluster to appear running master-slave mode in pacemaker. If I power off all the 3 controllers and only start 2 of them, the downtime sometimes can be as long as 20 minutes. I have a little investigation and find out there are some possible causes. 1. MySQL Recovery Takes Too Long [1] and Blocking RabbitMQ Clustering in Pacemaker The pacemaker resource p_mysql start timeout is set to 475s. Sometimes MySQL-wss fails to start after power failure, and pacemaker would wait 475s before retry starting it. The problem is that pacemaker divides resource state transitions into batches. Since RabbitMQ is master-slave resource, I assume that starting all the slaves and promoting master are put into two different batches. If unfortunately starting all RabbitMQ slaves are put in the same batch as MySQL starting, even if RabbitMQ slaves and all other resources are ready, pacemaker will not continue but just wait for MySQL timeout. I can re-produce this by hard powering off all the controllers and start them again. It's more likely to trigger MySQL failure in this way. Then I observe that if there is one cloned mysql instance not starting, the whole pacemaker cluster gets stuck and does not emit any log. On the host of the failed instance, I can see a mysql resource agent process calling the sleep command. If I kill that process, the pacemaker comes back alive and RabbitMQ master gets promoted. In fact this long timeout is blocking every resource from state transition in pacemaker. This maybe a known problem of pacemaker and there are some discussions in Linux-HA mailing list [2]. It might not be fixed in the near future. It seems in generally it's bad to have long timeout in state transition actions (start/stop/promote/demote). There maybe another way to implement MySQL-wss resource agent to use a short start timeout and monitor the wss cluster state using monitor action. I also find a fix to improve MySQL start timeout [3]. It shortens the timeout to 300s. At the time I sending this email, I can not find it in stable/6.0 branch. Maybe the maintainer needs to cherry-pick it to stable/6.0 ? [1] https://bugs.launchpad.net/fuel/+bug/1441885 [2] http://lists.linux-ha.org/pipermail/linux-ha/2014-March/047989.html [3] https://review.openstack.org/#/c/171333/ 2. RabbitMQ Resource Agent Breaks Existing Cluster Read the code of the RabbitMQ resource agent, I find it does the following to start RabbitMQ master-slave cluster. On all the controllers: (1) Start Erlang beam process (2) Start RabbitMQ App (If failed, reset mnesia DB and cluster state) (3) Stop RabbitMQ App but do not stop the beam process Then in pacemaker, all the RabbitMQ instances are in slave state. After pacemaker determines the master, it does the following. On the to-be-master host: (4) Start RabbitMQ App (If failed, reset mnesia DB and cluster state) On the slaves hosts: (5) Start RabbitMQ App (If failed, reset mnesia DB and cluster state) (6) Join RabbitMQ cluster of the master host As far as I can understand, this process is to make sure the master determined by pacemaker is the same as the master determined in RabbitMQ cluster. If there is no existing cluster, it's fine. If it is run after power failure and recovery, it introduces the a new problem. After power recovery, if some of the RabbitMQ instances reach step (2) roughly at the same time (within 30s which is hard coded in RabbitMQ) as the original RabbitMQ master instance, they form the original cluster again and then shutdown. The other instances would have to wait for 30s before it reports failure waiting for tables, and be reset to a standalone cluster. In RabbitMQ documentation [4], it is also mentioned that if we shutdown RabbitMQ master, a new master is elected from the rest of slaves. If we continue to shutdown nodes in step (3), we reach a point that the last node is the RabbitMQ master, and pacemaker is not aware of it. I can see there is code to bookkeeping a rabbit-start-time attribute in pacemaker to record the most long lived instance to help pacemaker determine the master, but it does not cover the case mentioned above. A recent patch [5] checks existing rabbit-master attribute but it neither cover the above case. So
[openstack-dev] [Fuel] Speed Up RabbitMQ Recovering
Hello, I using Fuel 6.0.1 and find that RabbitMQ recover time is long after power failure. I have a running HA environment, then I reset power of all the machines at the same time. I observe that after reboot it usually takes 10 minutes for RabittMQ cluster to appear running master-slave mode in pacemaker. If I power off all the 3 controllers and only start 2 of them, the downtime sometimes can be as long as 20 minutes. I have a little investigation and find out there are some possible causes. 1. MySQL Recovery Takes Too Long [1] and Blocking RabbitMQ Clustering in Pacemaker The pacemaker resource p_mysql start timeout is set to 475s. Sometimes MySQL-wss fails to start after power failure, and pacemaker would wait 475s before retry starting it. The problem is that pacemaker divides resource state transitions into batches. Since RabbitMQ is master-slave resource, I assume that starting all the slaves and promoting master are put into two different batches. If unfortunately starting all RabbitMQ slaves are put in the same batch as MySQL starting, even if RabbitMQ slaves and all other resources are ready, pacemaker will not continue but just wait for MySQL timeout. I can re-produce this by hard powering off all the controllers and start them again. It's more likely to trigger MySQL failure in this way. Then I observe that if there is one cloned mysql instance not starting, the whole pacemaker cluster gets stuck and does not emit any log. On the host of the failed instance, I can see a mysql resource agent process calling the sleep command. If I kill that process, the pacemaker comes back alive and RabbitMQ master gets promoted. In fact this long timeout is blocking every resource from state transition in pacemaker. This maybe a known problem of pacemaker and there are some discussions in Linux-HA mailing list [2]. It might not be fixed in the near future. It seems in generally it's bad to have long timeout in state transition actions (start/stop/promote/demote). There maybe another way to implement MySQL-wss resource agent to use a short start timeout and monitor the wss cluster state using monitor action. I also find a fix to improve MySQL start timeout [3]. It shortens the timeout to 300s. At the time I sending this email, I can not find it in stable/6.0 branch. Maybe the maintainer needs to cherry-pick it to stable/6.0 ? [1] https://bugs.launchpad.net/fuel/+bug/1441885 [2] http://lists.linux-ha.org/pipermail/linux-ha/2014-March/047989.html [3] https://review.openstack.org/#/c/171333/ 2. RabbitMQ Resource Agent Breaks Existing Cluster Read the code of the RabbitMQ resource agent, I find it does the following to start RabbitMQ master-slave cluster. On all the controllers: (1) Start Erlang beam process (2) Start RabbitMQ App (If failed, reset mnesia DB and cluster state) (3) Stop RabbitMQ App but do not stop the beam process Then in pacemaker, all the RabbitMQ instances are in slave state. After pacemaker determines the master, it does the following. On the to-be-master host: (4) Start RabbitMQ App (If failed, reset mnesia DB and cluster state) On the slaves hosts: (5) Start RabbitMQ App (If failed, reset mnesia DB and cluster state) (6) Join RabbitMQ cluster of the master host As far as I can understand, this process is to make sure the master determined by pacemaker is the same as the master determined in RabbitMQ cluster. If there is no existing cluster, it's fine. If it is run after power failure and recovery, it introduces the a new problem. After power recovery, if some of the RabbitMQ instances reach step (2) roughly at the same time (within 30s which is hard coded in RabbitMQ) as the original RabbitMQ master instance, they form the original cluster again and then shutdown. The other instances would have to wait for 30s before it reports failure waiting for tables, and be reset to a standalone cluster. In RabbitMQ documentation [4], it is also mentioned that if we shutdown RabbitMQ master, a new master is elected from the rest of slaves. If we continue to shutdown nodes in step (3), we reach a point that the last node is the RabbitMQ master, and pacemaker is not aware of it. I can see there is code to bookkeeping a rabbit-start-time attribute in pacemaker to record the most long lived instance to help pacemaker determine the master, but it does not cover the case mentioned above. A recent patch [5] checks existing rabbit-master attribute but it neither cover the above case. So in step (4), pacemaker determines a different master which was a RabbitMQ slave last time. It would wait for its original RabbitMQ master for 30s and fail, then it gets reset to a standalone cluster. Here we get some different clusters, so in step (5) and (6), it is likely to report error in log saying timeout waiting for tables or fail to merge mnesia database schema, then the those instances get reset. You can easily re-produce the case by hard resetting power of all the controllers. As you can see, if you