VCP-DCV Section 5 - Troubleshooting and Repairing¶

Diagnosing the common failures, and knowing which log to read.

Method¶

1. Establish scope: one VM, one host, one cluster, or everything.
2. Check recent change. Most faults follow a change.
3. Check dependencies bottom-up: physical, storage, network, host, VM.
4. Read the relevant log rather than guessing.
5. Verify the fix and document it.

Logs worth knowing¶

The per-VM vmware.log living beside the VM's files, not in /var/log, is a detail worth remembering.

• Syslog forwarding - ESXi logs should be sent to a central collector, because local logs are limited and may be lost.
• vm-support / vc-support - diagnostic bundle collection.

Host problems¶

• Host disconnected from vCenter - check network to the management VMkernel, then restart the management agents from the DCUI or with services.sh restart. The VMs keep running while the host is disconnected; disconnection is a management-plane problem, not a workload outage.
• Host not responding versus disconnected - "not responding" means vCenter lost contact unexpectedly; "disconnected" means an administrator removed it.
• PSOD (purple screen of death) - a VMkernel panic. Record the screen, collect the core dump, and check hardware and driver compatibility. Usually hardware, firmware, or a driver fault.
• Host isolation - the host is running but cannot reach the isolation address. HA applies the configured isolation response.
• Time drift - breaks certificate validation and SSO. Check NTP.

VM problems¶

• VM will not power on - common causes: insufficient host resources, a reservation that cannot be satisfied, a locked file from an ungraceful shutdown or an existing lock by another host, a missing or corrupt .vmdk, or a full datastore.
• Locked file - shown as "unable to access file since it is locked." Identify the host holding the lock and release it.
• VM performance poor - check %RDY, ballooning, swapping, and storage latency, in that order.
• VMware Tools not running - breaks graceful shutdown, time sync, quiescing, and paravirtual drivers.
• Snapshot consolidation needed - a warning that delta disks remain after a failed snapshot removal. Consolidate promptly; delta chains degrade performance and consume space.
• Full datastore - a very common root cause, frequently caused by thin disks growing or by forgotten snapshots.

Storage problems¶

• APD (All Paths Down) - the host has lost all paths to a device but the array has not said the device is gone. The host keeps retrying, hoping it returns.
• PDL (Permanent Device Loss) - the array has explicitly signaled the device is gone permanently.
• VMCP (VM Component Protection) - configures HA's response to APD and PDL conditions, including restarting affected VMs elsewhere.
• Path thrashing - two hosts alternately claiming ownership of a LUN on an active/passive array, usually caused by a wrong path selection policy.
• High DAVG - array or fabric latency, escalate to the storage team.
• High KAVG - host queuing, check queue depth and multipathing.
• Datastore full - clear snapshots, expand the datastore, or Storage vMotion VMs away.

APD is temporary and retried; PDL is permanent and declared by the array. That distinction is directly examined.

Network problems¶

• VM has no connectivity - check port group VLAN ID, uplink assignment, physical switch trunking, and teaming policy in that order.
• VLAN mismatch - the most common cause. The port group VLAN must match the physical switch configuration.
• IP hash without EtherChannel - configured teaming that the physical switch does not match, producing intermittent or total loss.
• vMotion fails - check the vMotion-enabled VMkernel adapter on both hosts, IP connectivity between them, MTU consistency if jumbo frames are used, and CPU compatibility (EVC).
• MTU mismatch - jumbo frames configured on some elements but not all; large frames silently dropped while small pings succeed.

Cluster problems¶

• HA configuration error - often DNS resolution, management network redundancy warnings, or insufficient heartbeat datastores.
• Insufficient failover resources - admission control preventing a power-on because the cluster could not guarantee restart capacity.
• DRS not balancing - check the automation level, migration threshold, and whether "must" affinity rules are constraining placement.
• vSAN health warnings - use the built-in vSAN health checks; component and object status indicate whether policy compliance is met.

Backup and restore¶

• vCenter file-based backup and restore - the supported path for vCenter recovery.
• VM backup - via vStorage APIs for Data Protection (VADP) using a backup product with changed block tracking. Snapshots are a mechanism inside this, not a backup themselves.
• Changed Block Tracking (CBT) - lets backup software copy only changed blocks.

Exam pointers¶

• vmware.log is in the VM's folder on the datastore; host logs are in /var/log.
• A disconnected host does not stop its VMs; restart management agents.
• APD is transient, PDL is permanent and array-declared.
• A locked file usually follows an ungraceful shutdown or a host still holding the lock.
• vMotion failures: VMkernel service enabled, network reachability, MTU, and CPU compatibility.
• Snapshots are not backups, and unconsolidated deltas cause both performance and space problems.

Official documentation¶

📖 vSphere troubleshooting - diagnostic procedures 📖 ESXi log file locations - log reference 📖 Broadcom support knowledge base - specific error resolution