Troubleshooting and Repairing
Where to look first
Troubleshooting follows layers: management, compute, network, storage, guest. If vCenter is unreachable, check the VCSA services and VAMI, then certificate expiry and NTP. If hosts are disconnected, check hostd, vpxd, DNS, and the management network. If VMs are slow, check CPU ready, memory balloon/swap, datastore latency, and network drops before changing settings. Know log locations: vpxd.log for vCenter, hostd.log and vmkernel.log on ESXi, fdm.log for HA, and vobd.log for events. Use esxcli for storage paths and vmkping for VMkernel connectivity.
Common failure patterns
APD (all paths down) means the host has lost all access to a datastore but may recover; PDL (permanent device loss) means the device is gone and requires action. HA failover may not occur if admission control has reserved too little or if the isolation response is misconfigured. vMotion can fail from EVC/CPU incompatibility, network misconfiguration, or insufficient resources. Snapshot consolidation failures usually mean a delta file is locked or the datastore is full. Certificate expiry can break vCenter, SSO, and HA simultaneously-correlate timestamps.
Traps
Do not reboot everything before capturing logs. Do not confuse APD with PDL. Do not assume a disconnected host is powered off. Do not treat a vMotion error as a storage error without checking EVC and VMkernel networks.
Worked example: After a planned certificate replacement, all hosts show disconnected and HA reports errors. Check vpxd.log for certificate validation failures, verify the machine certificate chain and time sync, then restart vpxd and the HA services in order. Capture a support bundle before changes.
Rule of thumb: Capture evidence, isolate the layer, change one variable at a time, and verify recovery before closing the incident.