Hi Ryan and Felipe! Could you eventually tell me the case number if you remember it? I opened the case and would reference to your case ID Or shall I send you mine ?
From: gpfsug-discuss <[email protected]> On Behalf Of Ryan Novosielski Sent: Freitag, 17. Februar 2023 06:43 To: gpfsug main discussion list <[email protected]> Subject: Re: [gpfsug-discuss] Reasons for DiskLeaseThread Overloaded Thanks for this, Felipe. We’ve started seeing intermittent overdue leases in large numbers and don’t otherwise have an explanation for it, other than “look at your network,” which actually does show occasional signs of strange behavior/higher-than-normal RTO values, but we’re not necessarily seeing those things happen at the same times as the lease issues. We’ve also seen “GPFS Critical Thread Watchdog” recently. We had a case open about it, but didn’t draw any real conclusions. If any of our data might be helpful/if there’s a case we could reference to see if we’re also running into that, we could provide a gpfs.snap. FWIW, we are running 5.1.3-1 on the storage side (except one system that’s about to be upgraded that runs a combination of 5.0.3-2 and 5.0.5-1), and 5.1.6-0 (soon to be 5.1.6-1) on the remote/client cluster side. -- #BlackLivesMatter ____ || \\UTGERS<file://UTGERS>, |---------------------------*O*--------------------------- ||_// the State | Ryan Novosielski - [email protected]<mailto:[email protected]> || \\ University | Sr. Technologist - 973/972.0922 (2x0922) ~*~ RBHS Campus || \\ of NJ | Office of Advanced Research Computing - MSB C630, Newark `' On Feb 16, 2023, at 12:02, Felipe Knop <[email protected]<mailto:[email protected]>> wrote: Walter, Thanks for the detailed description. I don’t yet see anything glaringly incorrect on your configuration, but perhaps others might find something out of place. I’d encourage you to open a case, since I spoke with a colleague yesterday, and he mentioned that he is working on a problem that may cause the lease thread to “loop” for a while. That might cause the critical thread watchdog to flag the lease thread as taking too long to “check in”. Capturing gpfs.snap is important, since we’d be looking into all the [W] ------------------[GPFS Critical Thread Watchdog]------------------ instances. Thanks, Felipe ---- Felipe Knop [email protected]<mailto:[email protected]> GPFS Development and Security IBM Systems IBM Building 008 2455 South Rd, Poughkeepsie, NY 12601 From: gpfsug-discuss <[email protected]<mailto:[email protected]>> on behalf of Walter Sklenka <[email protected]<mailto:[email protected]>> Reply-To: gpfsug main discussion list <[email protected]<mailto:[email protected]>> Date: Thursday, February 16, 2023 at 9:16 AM To: gpfsug main discussion list <[email protected]<mailto:[email protected]>> Subject: [EXTERNAL] Re: [gpfsug-discuss] Reasons for DiskLeaseThread Overloaded Hi Felipe! Once again me. Thank you very much for the hint I did not open a PMR yet because I fear they will ask me/us if we are cracy ☹ I did not tell the full story yet We have a 3 node cluster, 2 NSD servers o1,o2 (same site ) and g1 (different ZjQcmQRYFpfptBannerStart This Message Is From an External Sender This message came from outside your organization. ZjQcmQRYFpfptBannerEnd Hi Felipe! Once again me. Thank you very much for the hint I did not open a PMR yet because I fear they will ask me/us if we are cracy ☹ I did not tell the full story yet We have a 3 node cluster, 2 NSD servers o1,o2 (same site ) and g1 (different site). (rhel 8.7) All of them are Vmware VMs O1 and o2 have each 4 NVME drives passed through , there is a software raid 5 made over these NVMEs , and from them made a single NSD , for a filesystem fs4vm (m,r=2 ) [root@ogpfs1 ras]# mmlscluster GPFS cluster information ======================== GPFS cluster name: edvdesign-cluster.local GPFS cluster id: 12147978822727803186 GPFS UID domain: edvdesign-cluster.local Remote shell command: /usr/bin/ssh Remote file copy command: /usr/bin/scp Repository type: CCR Node Daemon node name IP address Admin node name Designation ---------------------------------------------------------------------------- 1 ogpfs1-hs.local 10.20.30.1 ogpfs1-hs.local quorum-manager-perfmon 2 ogpfs2-hs.local 10.20.30.2 ogpfs2-hs.local quorum-manager-perfmon 3 ggpfsq.mgmt.cloudia xxxx.other.net<http://xxxx.other.net/> ggpfsq.mgmt. a quorum-perfmon [root@ogpfs1 ras]# mmlsconfig Configuration data for cluster edvdesign-cluster.local: ------------------------------------------------------- clusterName edvdesign-cluster.local clusterId 12147978822727803186 autoload yes profile gpfsProtocolRandomIO dmapiFileHandleSize 32 minReleaseLevel 5.1.6.0 tscCmdAllowRemoteConnections no ccrEnabled yes cipherList AUTHONLY sdrNotifyAuthEnabled yes maxblocksize 16M [cesNodes] maxMBpS 5000 numaMemoryInterleave yes enforceFilesetQuotaOnRoot yes workerThreads 512 [common] tscCmdPortRange 60000-61000 [srv] verbsPorts mlx5_0/1 mlx5_1/1 [common] cesSharedRoot /fs4vmware/cesSharedRoot [srv] maxFilesToCache 10000 maxStatCache 20000 [common] verbsRdma enable [ggpfsq] verbsRdma disable [common] verbsRdmaSend yes [ggpfsq] verbsRdmaSend no [common] verbsRdmaCm enable [ggpfsq] verbsRdmaCm disable [srv] pagepool 32G [common] adminMode central File systems in cluster edvdesign-cluster.local: ------------------------------------------------ /dev/fs4vm [root@ogpfs1 ras]# mmlsdisk fs4vm -L disk driver sector failure holds holds storage name type size group metadata data status availability disk id pool remarks ------------ -------- ------ ----------- -------- ----- ------------- ------------ ------- ------------ --------- ogpfs1_1 nsd 512 1 yes yes ready up 1 system desc ogpfs2_1 nsd 512 2 yes yes ready up 2 system desc ggpfsq_qdisk nsd 512 -1 no no ready up 3 system desc Number of quorum disks: 3 Read quorum value: 2 Write quorum value: 2 And the two nodes o1 and o2 export the filesystem via CES NFS functions ( for VMware) I think this isn´supported , that a NSD Server is also a CES Node? And finally the RDMA Network : The both NSD servers also have a Mellanox ConnectX-6 Lx dual port 25Gb adapter also via passthrough And these interfaces we configured for rdma (RoCE) , Last but not least: this network is not switched but direct attached ( 2x25Gb directly connected between the NSD nodes ) RDMA Connections between nodes: Fabric 0 - Device mlx5_0 Port 1 Width 1x Speed EDR lid 0 hostname idx CM state VS buff RDMA_CT(ERR) RDMA_RCV_MB RDMA_SND_MB VS_CT(ERR) VS_SND_MB VS_RCV_MB WAIT_CON_SLOT WAIT_NODE_SLOT ogpfs2-hs.local 0 Y RTS (Y)256 478202 (0 ) 12728 67024 8864789(0 ) 22776 4643 0 0 Fabric 0 - Device mlx5_1 Port 1 Width 1x Speed EDR lid 0 hostname idx CM state VS buff RDMA_CT(ERR) RDMA_RCV_MB RDMA_SND_MB VS_CT(ERR) VS_SND_MB VS_RCV_MB WAIT_CON_SLOT WAIT_NODE_SLOT ogpfs2-hs.local 1 Y RTS (Y)256 477659 (0 ) 12489 67034 8864773(0 ) 22794 4639 0 0 [root@ogpfs1 ras]# You mentioned that it might be a cpu contention : Maybe due to the VM layer (scheduling with other VMS) ? And wrong layout of VMs ( 8 vCPUs and 64GB Mem) [ esxis only single socket with 32/64 cores HT) And also the direct attached RDMA ( +DAEMON) network is also not good? Do you think IBM would say no to check such a configuration ? Best regards Walter From: gpfsug-discuss <[email protected]<mailto:[email protected]>> On Behalf Of Felipe Knop Sent: Mittwoch, 15. Februar 2023 15:59 To: gpfsug main discussion list <[email protected]<mailto:[email protected]>> Subject: Re: [gpfsug-discuss] Reasons for DiskLeaseThread Overloaded Walter, Thanks for the details. The stack trace below captures the lease thread in the middle of sending the “lease” RPC. This operation normally is not blocking, and we do not often block while sending the RPC. But the stack trace “does not show” whether there was anything blocking the thread prior to the point where the RPCs are sent. At a first glance: 2023-02-14_19:44:07.430+0100: [W] counter: 0 (mark-idle: 0 mark-active: 0 pre-work: 0 post-work: 0) sched: (nvcsw: 0 nivcsw: 10) I believe nivcsw: 10 means that the thread was scheduled out of the CPU involuntarily, possibly indicating that there is some CPU contention going on. Could you open a case to get debug data collected? If the problem can be recreated, I think we’ll need a recreate of the problem with traces enabled. Thanks, Felipe ---- Felipe Knop [email protected]<mailto:[email protected]> GPFS Development and Security IBM Systems IBM Building 008 2455 South Rd, Poughkeepsie, NY 12601 From: gpfsug-discuss <[email protected]<mailto:[email protected]>> on behalf of Walter Sklenka <[email protected]<mailto:[email protected]>> Reply-To: gpfsug main discussion list <[email protected]<mailto:[email protected]>> Date: Wednesday, February 15, 2023 at 4:23 AM To: gpfsug main discussion list <[email protected]<mailto:[email protected]>> Subject: [EXTERNAL] Re: [gpfsug-discuss] Reasons for DiskLeaseThread Overloaded Hi! This is a „full“ sequence in mmfs. log. latest Fortunately this was also the last event until now (yesterday evening) Maybe you can have a look? 2023-02-14_19: 43: 51. 474+0100: [N] Disk lease period expired 0. 030 seconds ago in cluster ZjQcmQRYFpfptBannerStart This Message Is From an External Sender This message came from outside your organization. ZjQcmQRYFpfptBannerEnd Hi! This is a „full“ sequence in mmfs.log.latest Fortunately this was also the last event until now (yesterday evening) Maybe you can have a look? 2023-02-14_19:43:51.474+0100: [N] Disk lease period expired 0.030 seconds ago in cluster xxx-cluster. Attempting to reacquire the lease. 2023-02-14_19:44:07.430+0100: [W] ------------------[GPFS Critical Thread Watchdog]------------------ 2023-02-14_19:44:07.430+0100: [W] PID: 7294 State: R (DiskLeaseThread) is overloaded for more than 8 seconds 2023-02-14_19:44:07.430+0100: [W] counter: 0 (mark-idle: 0 mark-active: 0 pre-work: 0 post-work: 0) sched: (nvcsw: 0 nivcsw: 10) 2023-02-14_19:44:07.430+0100: [W] Call Trace(PID: 7294): 2023-02-14_19:44:07.431+0100: [W] #0: 0x000055CABE4A56AB NodeConn::sendMessage(TcpConn**, iovec*, int, unsigned char, int, int, int, unsigned int, DestTag*, int*, unsigned long long*, unsigned long long*, unsi gned int*, CondvarName, vsendCallback_t*) + 0x42B at ??:0 2023-02-14_19:44:07.432+0100: [W] #1: 0x000055CABE4A595F llc_send_msg(ClusterConfiguration*, NodeAddr, iovec*, int, unsigned char, int, int, int, unsigned int, DestTag*, int*, TcpConn**, unsigned long long*, u nsigned long long*, unsigned int*, CondvarName, vsendCallback_t*, int, unsigned int) + 0xDF at ??:0 2023-02-14_19:44:07.437+0100: [W] #2: 0x000055CABE479A55 MsgRecord::send() + 0x1345 at ??:0 2023-02-14_19:44:07.438+0100: [W] #3: 0x000055CABE47A169 tscSendInternal(ClusterConfiguration*, unsigned int, unsigned char, int, int, NodeAddr*, TscReply*, TscScatteredBuff*, int, int (*)(void*, ClusterConfig uration*, int, NodeAddr*, TscReply*), void*, ChainedCallback**, __va_list_tag*) + 0x339 at ??:0 2023-02-14_19:44:07.439+0100: [W] #4: 0x000055CABE47C39A tscSendWithCallback(ClusterConfiguration*, unsigned int, unsigned char, int, NodeAddr*, TscReply*, int (*)(void*, ClusterConfiguration*, int, NodeAddr*, TscReply*), void*, void**, int, ...) + 0x1DA at ??:0 2023-02-14_19:44:07.440+0100: [W] #5: 0x000055CABE5F9853 MyLeaseState::renewLease(NodeAddr, TickTime) + 0x6E3 at ??:0 2023-02-14_19:44:07.440+0100: [W] #6: 0x000055CABE5FA682 ClusterConfiguration::checkAndRenewLease(TickTime) + 0x192 at ??:0 2023-02-14_19:44:07.441+0100: [W] #7: 0x000055CABE5FAAC6 ClusterConfiguration::RunLeaseChecks(void*) + 0x366 at ??:0 2023-02-14_19:44:07.441+0100: [W] #8: 0x000055CABDF2B662 Thread::callBody(Thread*) + 0x42 at ??:0 2023-02-14_19:44:07.441+0100: [W] #9: 0x000055CABDF18680 Thread::callBodyWrapper(Thread*) + 0xA0 at ??:0 2023-02-14_19:44:07.441+0100: [W] #10: 0x00007F3B7563D1CA start_thread + 0xEA at ??:0 2023-02-14_19:44:07.441+0100: [W] #11: 0x00007F3B7435BE73 __GI___clone + 0x43 at ??:0 2023-02-14_19:44:10.512+0100: [N] Disk lease reacquired in cluster xxx-cluster. 2023-02-14_19:44:10.512+0100: [N] Disk lease period expired 7.970 seconds ago in cluster xxx-cluster. Attempting to reacquire the lease. 2023-02-14_19:44:12.563+0100: [N] Disk lease reacquired in cluster xxx-cluster. Thank you very much! Best regards Walter From: gpfsug-discuss <[email protected]<mailto:[email protected]>> On Behalf Of Felipe Knop Sent: Mittwoch, 15. Februar 2023 00:06 To: gpfsug main discussion list <[email protected]<mailto:[email protected]>> Subject: Re: [gpfsug-discuss] Reasons for DiskLeaseThread Overloaded All, These messages like [W] ------------------[GPFS Critical Thread Watchdog]------------------ indicate that a “critical thread”, in this case the lease thread, was apparently blocked for longer than expected. This is usually not caused by delays in the network, but possibly by excessive CPU load, blockage while accessing the local file system, or possible mutex contention. Do you have other samples of the message, with a more complete stack trace? Or was the instance below the only one? Felipe ---- Felipe Knop [email protected]<mailto:[email protected]> GPFS Development and Security IBM Systems IBM Building 008 2455 South Rd, Poughkeepsie, NY 12601 From: gpfsug-discuss <[email protected]<mailto:[email protected]>> on behalf of Walter Sklenka <[email protected]<mailto:[email protected]>> Reply-To: gpfsug main discussion list <[email protected]<mailto:[email protected]>> Date: Tuesday, February 14, 2023 at 10:49 AM To: "[email protected]<mailto:[email protected]>" <[email protected]<mailto:[email protected]>> Subject: [EXTERNAL] Re: [gpfsug-discuss] Reasons for DiskLeaseThread Overloaded Hi! I started with 5. 1. 6. 0 and now am at [root@ ogpfs1 ~]# mmfsadm dump version Dump level: verbose Build branch "5. 1. 6. 1 ". the messages started from the beginning From: gpfsug-discuss <gpfsug-discuss-bounces@ gpfsug. org> On ZjQcmQRYFpfptBannerStart This Message Is From an External Sender This message came from outside your organization. ZjQcmQRYFpfptBannerEnd Hi! I started with 5.1.6.0 and now am at [root@ogpfs1 ~]# mmfsadm dump version Dump level: verbose Build branch "5.1.6.1 ". the messages started from the beginning From: gpfsug-discuss <[email protected]<mailto:[email protected]>> On Behalf Of Christian Vieser Sent: Dienstag, 14. Februar 2023 15:34 To: [email protected]<mailto:[email protected]> Subject: Re: [gpfsug-discuss] Reasons for DiskLeaseThread Overloaded What version of Spectrum Scale is running there? Do these errors appear since your last version update? Am 14.02.23 um 14:09 schrieb Walter Sklenka: Dear Collegues! May I ask if anyone has a hint what could be the reason for Critical Thread Watchdog warnings for Disk Leases Threads? Is this a “local node” Problem or a network problem ? I see these messages sometimes arriving when NSD Servers which also serve as NFS servers when they get under heavy NFS load Following is an excerpt from mmfs.log.latest 2023-02-14_12:06:53.235+0100: [N] Disk lease period expired 0.040 seconds ago in cluster xxx-cluster. Attempting to reacquire the lease. 2023-02-14_12:06:53.600+0100: [W] ------------------[GPFS Critical Thread Watchdog]------------------ 2023-02-14_12:06:53.600+0100: [W] PID: 7294 State: R (DiskLeaseThread) is overloaded for more than 8 seconds 2023-02-14_12:06:53.600+0100: [W] counter: 0 (mark-idle: 0 mark-active: 0 pre-work: 0 post-work: 0) sched: (nvcsw: 0 nivcsw: 8) 2023-02-14_12:06:53.600+0100: [W] Call Trace(PID: 7294): 2023-02-14_12:06:53.600+0100: [W] #0: 0x000055CABDF49521 BaseMutexClass::release() + 0x12 at ??:0 2023-02-14_12:06:53.600+0100: [W] #1: 0xB1557721BBABD900 _etext + 0xB154F7E646041C0E at ??:0 2023-02-14_12:07:09.554+0100: [N] Disk lease reacquired in cluster xxx-cluster. 2023-02-14_12:07:09.554+0100: [N] Disk lease period expired 5.680 seconds ago in cluster xxx-cluster. Attempting to reacquire the lease. 2023-02-14_12:07:11.605+0100: [N] Disk lease reacquired in cluster xxx-cluster. 2023-02-14_12:10:55.990+0100: [I] Command: mmlspool /dev/fs4vm all -L -Y 2023-02-14_12:10:55.990+0100: [I] Command: successful mmlspool /dev/fs4vm all -L -Y 2023-02-14_12:30:58.756+0100: [I] Command: mmlspool /dev/fs4vm all -L -Y 2023-02-14_12:30:58.756+0100: [I] Command: successful mmlspool /dev/fs4vm all -L -Y 2023-02-14_13:10:55.988+0100: [I] Command: mmlspool /dev/fs4vm all -L -Y 2023-02-14_13:10:55.989+0100: [I] Command: successful mmlspool /dev/fs4vm all -L -Y 2023-02-14_13:21:40.892+0100: [N] Node 10.20.30.2 (ogpfs2-hs.local) lease renewal is overdue. Pinging to check if it is alive 2023-02-14_13:21:40.892+0100: [I] The TCP connection to IP address 10.20.30.2 ogpfs2-hs.local <c0n1>:[1] (socket 106) state: state=1 ca_state=0 snd_cwnd=10 snd_ssthresh=2147483647 unacked=0 probes=0 backoff=0 retransmits=0 rto=201000 rcv_ssthresh=1219344 rtt=121 rttvar=69 sacked=0 retrans=0 reordering=3 lost=0 2023-02-14_13:22:00.220+0100: [N] Disk lease period expired 0.010 seconds ago in cluster xxx-cluster. Attempting to reacquire the lease. 2023-02-14_13:22:08.298+0100: [N] Disk lease reacquired in cluster xxx-cluster. 2023-02-14_13:30:58.760+0100: [I] Command: mmlspool /dev/fs4vm all -L -Y 2023-02-14_13:30:58.760+0100: [I] Command: successful mmlspool /dev/fs4vm all -L -Y Mit freundlichen Grüßen Walter Sklenka Technical Consultant _______________________________________________ gpfsug-discuss mailing list gpfsug-discuss at gpfsug.org<http://gpfsug.org/> http://gpfsug.org/mailman/listinfo/gpfsug-discuss_gpfsug.org
_______________________________________________ gpfsug-discuss mailing list gpfsug-discuss at gpfsug.org http://gpfsug.org/mailman/listinfo/gpfsug-discuss_gpfsug.org
