Re: [Openais] cpg behavior on transitional membership change
Hi Jiaju, 03.09.2011 19:52, Jiaju Zhang wrote: > On Fri, Sep 02, 2011 at 10:12:11PM +0300, Vladislav Bogdanov wrote: >> 02.09.2011 20:55, David Teigland wrote: >> [snip] >>> >>> I really can't make any sense of the report, sorry. Maybe reproduce it >>> without pacemaker, and then describe the specific steps to create the >>> issue and resulting symptoms. After that we can determine what logs, if >>> any, would be useful. >>> >> >> I just tried to ask a question about cluster components logic based on >> information I discovered from both logs and code analysis. I'm sorry if >> I was unclear in that, probably some language barrier still exists. >> >> Please see my previous mail, I tried to add some explanations why I >> think current logic is not complete. > > Hi Vladislav, I guess I have known the problem what you described;) > I'd like to give a example to make the things more clear. > > 3-node cluster, for whatever reason, especially on heavy workload, (my case) > corosync may detect one node disappear and reappear again. So the BTW, could this be prevented (at least for majority of cases) by some corosync timeout params? Steve? > membership information changes are as follows: > membership 1: nodeA, nodeB, nodeC > membership 2: nodeB, nodeC > membership 3: nodeA, nodeB, nodeC Exactly. > > From the membership change 1 -> 2, dlm_controld konws nodeA is down, > and have many things to do, like check_fs_done, check_fencing_done ... > The key point here is dlm need to wait the fencing is really done > before it proceed. If we employ a cluster filesystem here, like ocfs2, > it also needs the fencing is really done. I believe in the normal > cases, pacemaker will fence nodeA and then everything should be OK. > > However, there is a possibility here that pacemaker won't fence nodeA. > Say nodeA is the original DC of the cluster, when nodeA is down, the > cluster should elect a new DC. But if the time window where membership > change 2 -> 3 is too small, node A is up again and attend the election > too, then node A is elected to be the DC again and it won't fence > itself. Ahm, I think you are perfectly right in what exactly happens. This addresses my case I think, just because that node which left cluster for a short moment is usually 1. Started earlier on a cold boot (it is a VM, others are bare-metal which boot diskless via PXE, and there is 1 minute timeout before that bare-metal systems get their boot image, just because of cisco implementation of ether-channel) 2. Upgraded (and rebooted if needed) first, before bare-metal nodes are booted with new image. So that node is usually a DC. > Andrew, correct me if my understanding on pacemaker is wrong;) > > So I think the membership change should be like a transaction in > database or filesystem field, that is, for the membership change > 1 -> 2, every thing should be done (e.g. fencing nodeA), no matter the > following change 2 -> 3 will happen or not. For the situation where a > node magically disappear and reappear, and the situation where a node > normally down and then up, ocfs2 and dlm should not be able to see any > difference between them, what they can do is just waiting the fencing > to be done. > > Any comments? thoughts? I'd have protection at as many layers as possible. So, if you are right about pacemaker, then it is great if it could be fixed in pacemaker. But, I'd prefer to be safe and have DLM et all schedule fencing as well if they notice unrecoverable problem which leads to cluster subsystem freeze if fencing is not done for whatever reason. I suppose that current behavior is just a artifact from cluster2, where groupd was on duty for that event (I may be wrong, because I didn't look at groupd code closely, but comments in other daemons code make me think so). Thank you for your comments, Vladislav ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] cpg behavior on transitional membership change
On Fri, Sep 02, 2011 at 10:12:11PM +0300, Vladislav Bogdanov wrote: > 02.09.2011 20:55, David Teigland wrote: > [snip] > > > > I really can't make any sense of the report, sorry. Maybe reproduce it > > without pacemaker, and then describe the specific steps to create the > > issue and resulting symptoms. After that we can determine what logs, if > > any, would be useful. > > > > I just tried to ask a question about cluster components logic based on > information I discovered from both logs and code analysis. I'm sorry if > I was unclear in that, probably some language barrier still exists. > > Please see my previous mail, I tried to add some explanations why I > think current logic is not complete. Hi Vladislav, I guess I have known the problem what you described;) I'd like to give a example to make the things more clear. 3-node cluster, for whatever reason, especially on heavy workload, corosync may detect one node disappear and reappear again. So the membership information changes are as follows: membership 1: nodeA, nodeB, nodeC membership 2: nodeB, nodeC membership 3: nodeA, nodeB, nodeC >From the membership change 1 -> 2, dlm_controld konws nodeA is down, and have many things to do, like check_fs_done, check_fencing_done ... The key point here is dlm need to wait the fencing is really done before it proceed. If we employ a cluster filesystem here, like ocfs2, it also needs the fencing is really done. I believe in the normal cases, pacemaker will fence nodeA and then everything should be OK. However, there is a possibility here that pacemaker won't fence nodeA. Say nodeA is the original DC of the cluster, when nodeA is down, the cluster should elect a new DC. But if the time window where membership change 2 -> 3 is too small, node A is up again and attend the election too, then node A is elected to be the DC again and it won't fence itself. Andrew, correct me if my understanding on pacemaker is wrong;) So I think the membership change should be like a transaction in database or filesystem field, that is, for the membership change 1 -> 2, every thing should be done (e.g. fencing nodeA), no matter the following change 2 -> 3 will happen or not. For the situation where a node magically disappear and reappear, and the situation where a node normally down and then up, ocfs2 and dlm should not be able to see any difference between them, what they can do is just waiting the fencing to be done. Any comments? thoughts? Thanks, Jiaju ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] cpg behavior on transitional membership change
02.09.2011 20:55, David Teigland wrote: [snip] > > I really can't make any sense of the report, sorry. Maybe reproduce it > without pacemaker, and then describe the specific steps to create the > issue and resulting symptoms. After that we can determine what logs, if > any, would be useful. > I just tried to ask a question about cluster components logic based on information I discovered from both logs and code analysis. I'm sorry if I was unclear in that, probably some language barrier still exists. Please see my previous mail, I tried to add some explanations why I think current logic is not complete. Thank you, Vladislav ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] cpg behavior on transitional membership change
Hi Steve, 02.09.2011 20:30, Steven Dake wrote: > On 09/02/2011 12:59 AM, Vladislav Bogdanov wrote: ... >> I'm trying to further investigate problem I described at >> https://www.redhat.com/archives/cluster-devel/2011-August/msg00133.html >> >> The main problem for me there is that pacemaker first sees transitional >> membership with left nodes, then it sees stable membership with that >> nodes returned back, and does nothing about that. On the other hand, >> dlm_controld sees CPG_REASON_NODEDOWN events on CPGs related to all its >> lockspaces (at the same time with transitional membership change) and >> stops kernel part of each lockspace until whole cluster is rebooted (or >> until some other recovery procedure which unfortunately does not happen > > I believe fenced should reboot the node, but only if there is quorum. > It is possible your cluster has lost quorum during this series of > events. I have copied Dave for his feedback on this point. Aha. I think so too. But fenced doesn't do that as well as all other daemons from cluster3, this part of code is identical among them, that's why I think this does not depend on whether cman or pacemaker stack is used: fence/fenced/cpg.c around line 1440 (as for 3.1.1) if (left_list[i].reason == CPG_REASON_NODEDOWN || left_list[i].reason == CPG_REASON_PROCDOWN) { memb->failed = 1; cg->failed_count++; } ... if (left_list[i].reason == CPG_REASON_PROCDOWN) kick_node_from_cluster(memb->nodeid); probably last lines should be: if (left_list[i].reason == CPG_REASON_NODEDOWN || left_list[i].reason == CPG_REASON_PROCDOWN) kick_node_from_cluster(memb->nodeid); at least in one of daemons (fenced is a good candidate, but I prefer dlm_controld)? About quorum: 3 node cluster was split to two partitions, 2 bare-metal and 1 VM nodes. When I found that, two metal ones were in 'kern_stop' state, transitioning via 'kern_stop,fencing' state I suppose. VM did not have quorum, so it was left in 'kern_stop,fencing' state. dlm dump says: 1313579105 clvmd add_change cg 4 remove nodeid 1543767306 reason 3 That means CPG_REASON_NODEDOWN event. Then: 1313579105 Node 1543767306/mgmt01 has not been shot yet 1313579105 clvmd check_fencing 1543767306 wait add 1313562825 fail 1313579105 last 0 1313579107 Node 1543767306/mgmt01 was last shot 'now' This is not true, there is no line about actual fencing scheduling (and it is clear from code why). This could be a deficiency of .pcmk dlm_controld variant, but that is not important here I think. 1313579107 clvmd check_fencing 1543767306 done add 1313562825 fail 1313579105 last 1313579107 1313579107 clvmd check_fencing done > >> :( ). It neither requests to fence left node nor recovers when node is >> returned on next stable membership. >> >> Could anyone please help me to understand, what is a correct CPG >> behavior on membership change? >> From what I see, CPG emits CPG_REASON_NODEDOWN event on both >> transitional and stable membership if there is node which left the >> cluster. Am I correct here? And is that a right thing if I am? Ah, I should be mixed something, it was quite long ago. Actually, yes, that was transitional one. There was only one such event. >> > > Line #'s where this happens? I just saw that in pacemaker plugin logs and in dlm_tool dump logs. Their timestamps are identical. > >> If yes, is there a way do detect membership change type (transitional pr >> stable) through CPG API? >> > > A transitional membership will always contain a subset of the previous > regular membership. This means it will always contains 0 or more left > members. A transitional membership means "The membership of nodes > transitioning from previous regular membership to new regular mebmership". > > A regular configuration is where members are added to the configuration > when detected. A transitional membership never has nodes added to it. Thank you for clarification very much. Shouldn't pacemaker then schedule fencing itself (from the partition with quorum) if there are left nodes? BTW, actually there was only second or two between transitional and regular membership. I probably need to ask Andrew for pacemaker logic details. Unfortunately I lost that logs and hardly can reproduce that :( That was a VM which left the cluster, and it probably just suffered from insufficient host CPU time. And... Just wandering, what could be a reason to recalculate membership if there are 0 left or added members? > >> Hoping for answer, >> > > It would be nice if cpg and totem had a direct relationship in how their > transitional and regular configurations were generated, but this doesn't > happen currently. I am not sure if there is a good reason for this. Pacemaker uses totem? At least it doesn't use cpg. May be that is the reason of not-fenc
Re: [Openais] cpg behavior on transitional membership change
On Fri, Sep 02, 2011 at 10:30:53AM -0700, Steven Dake wrote: > On 09/02/2011 12:59 AM, Vladislav Bogdanov wrote: > > Hi all, > > > > I'm trying to further investigate problem I described at > > https://www.redhat.com/archives/cluster-devel/2011-August/msg00133.html > > > > The main problem for me there is that pacemaker first sees transitional > > membership with left nodes, then it sees stable membership with that > > nodes returned back, and does nothing about that. On the other hand, > > dlm_controld sees CPG_REASON_NODEDOWN events on CPGs related to all its > > lockspaces (at the same time with transitional membership change) and > > stops kernel part of each lockspace until whole cluster is rebooted (or > > until some other recovery procedure which unfortunately does not happen > > I believe fenced should reboot the node, but only if there is quorum. > It is possible your cluster has lost quorum during this series of > events. I have copied Dave for his feedback on this point. I really can't make any sense of the report, sorry. Maybe reproduce it without pacemaker, and then describe the specific steps to create the issue and resulting symptoms. After that we can determine what logs, if any, would be useful. ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
Re: [Openais] cpg behavior on transitional membership change
On 09/02/2011 12:59 AM, Vladislav Bogdanov wrote: > Hi all, > > I'm trying to further investigate problem I described at > https://www.redhat.com/archives/cluster-devel/2011-August/msg00133.html > > The main problem for me there is that pacemaker first sees transitional > membership with left nodes, then it sees stable membership with that > nodes returned back, and does nothing about that. On the other hand, > dlm_controld sees CPG_REASON_NODEDOWN events on CPGs related to all its > lockspaces (at the same time with transitional membership change) and > stops kernel part of each lockspace until whole cluster is rebooted (or > until some other recovery procedure which unfortunately does not happen I believe fenced should reboot the node, but only if there is quorum. It is possible your cluster has lost quorum during this series of events. I have copied Dave for his feedback on this point. > :( ). It neither requests to fence left node nor recovers when node is > returned on next stable membership. > > Could anyone please help me to understand, what is a correct CPG > behavior on membership change? > From what I see, CPG emits CPG_REASON_NODEDOWN event on both > transitional and stable membership if there is node which left the > cluster. Am I correct here? And is that a right thing if I am? > Line #'s where this happens? > If yes, is there a way do detect membership change type (transitional pr > stable) through CPG API? > A transitional membership will always contain a subset of the previous regular membership. This means it will always contains 0 or more left members. A transitional membership means "The membership of nodes transitioning from previous regular membership to new regular mebmership". A regular configuration is where members are added to the configuration when detected. A transitional membership never has nodes added to it. > Hoping for answer, > It would be nice if cpg and totem had a direct relationship in how their transitional and regular configurations were generated, but this doesn't happen currently. I am not sure if there is a good reason for this. Regards -steve > Best regards, > Vladislav > ___ > Openais mailing list > Openais@lists.linux-foundation.org > https://lists.linux-foundation.org/mailman/listinfo/openais ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais
[Openais] cpg behavior on transitional membership change
Hi all, I'm trying to further investigate problem I described at https://www.redhat.com/archives/cluster-devel/2011-August/msg00133.html The main problem for me there is that pacemaker first sees transitional membership with left nodes, then it sees stable membership with that nodes returned back, and does nothing about that. On the other hand, dlm_controld sees CPG_REASON_NODEDOWN events on CPGs related to all its lockspaces (at the same time with transitional membership change) and stops kernel part of each lockspace until whole cluster is rebooted (or until some other recovery procedure which unfortunately does not happen :( ). It neither requests to fence left node nor recovers when node is returned on next stable membership. Could anyone please help me to understand, what is a correct CPG behavior on membership change? >From what I see, CPG emits CPG_REASON_NODEDOWN event on both transitional and stable membership if there is node which left the cluster. Am I correct here? And is that a right thing if I am? If yes, is there a way do detect membership change type (transitional pr stable) through CPG API? Hoping for answer, Best regards, Vladislav ___ Openais mailing list Openais@lists.linux-foundation.org https://lists.linux-foundation.org/mailman/listinfo/openais