Symptoms
The vpxd service on vCenter Server crashes, producing a core.vpxd-worker file.Messages in /var/log/vmware/vpxd/vpxd.log similar to the following immediately precede the crash:
2022-10-16T23:33:33.233-04:00 info vpxd[50262] [Originator@6876 sub=vpxLro opID=lro-76175637-7d814bb5] [VpxLRO] -- BEGIN lro-76175637 -- -- DrmExecute --2022-10-16T23:33:33.233-04:00 info vpxd[50262] [Originator@6876 sub=drsExec opID=lro-76175637-7d814bb5] Executing DRS recommendation 1234 with 1 actions2022-10-16T23:33:33.235-04:00 error vpxd[50262] [Originator@6876 sub=vpxLro opID=lro-76175637-7d814bb5] [VpxLRO] Unexpected Exception: N5Vmomi5Fault11SystemError9ExceptionE(Fault cause: vmodl.fault.SystemError2022-10-16T23:33:33.236-04:00 info vpxd[50262] [Originator@6876 sub=Default opID=lro-76175637-7d814bb5] [VpxLRO] -- ERROR lro-76175637 -- -- DrmExecute: vmodl.fault.SystemError:--> Result:--> (vmodl.fault.SystemError) {--> faultCause = (vmodl.MethodFault) null,--> faultMessage = <unset>,--> reason = "Too many outstanding operations"--> msg = ""
Cause
This issue manifests when a specific routine task (vScheduleCheckVsanConfigLro) is started during a time when the vCenter's LRO job queue is already full of queued tasks.
Resolution
This is resolved in vCenter Server 7.0 Update 3i (build number 20845200).
Workaround
To work around this issue, use the vpxd log to identify and remedy the cause of the excessively queued LRO tasks.While the root cause of this issue is the specific task causing vpxd to crash instead of handling "Too many outstanding operations" properly, the fact that vCenter has so many queued tasks is also something that can be investigated. The source of the queued tasks might come from an ESXi host which has partially stopped responding to commands sent to it by vCenter.For example, if before the crash, there are many vmodl.fault.HostCommunication messages in vpxd.log pertaining to a specific host, that host might need to be set into maintenance mode for further troubleshooting. It's also possible that a solution external to vCenter is sending too many requests for vCenter to process in time, and eventually filling up the LRO job queue.
