ESXi hosts across two clusters are experiencing intermittent "not responding" states in vCenter

Products

• VMware vCenter Server

Issue/Introduction

 Disconnections are tied to resource exhaustion on the ESXi management agents (hostd), occurring specifically during weekend and evening hours when Veeam backups are scheduled.

Environment

vSphere vCenter 8.0u3

Cause

Depending on Veeam's configuration, it utilizes either NBD mode (pulling data over the management network) or HotAdd mode (hot-adding virtual disks to a proxy VM). The logs confirm HotAdd is being used and is overwhelming the host.

In vmkernel.log 

at 2026-04-28T08:25:18.376Z, we see ESXi querying a virtual disk for UNMAP (VAAI Space Reclamation) support, which the disk rejects, causing ESXi to treat it as a thick disk. Exactly three seconds prior (08:25:15.928Z), hostd.log records a massive 46-second timeout: N7Vmacore16TimeoutExceptionE(Operation timed out... duration: 00:00:46.368165).

This perfectly links the Veeam HotAdd disk mount to the ESXi management agent freezing. Because hostd was locked up for over 46 seconds waiting on storage API resources, it failed to send its required keep-alive heartbeats to vCenter, resulting in the host disconnect.

Resolution

To mitigate this issue, the following workarounds are recommended:

1. Reduce Concurrent Tasks: In the Veeam console, navigate to Backup Infrastructure > Backup Proxies. Right-click the proxy handling these clusters, select Properties, and lower the Max concurrent tasks (e.g., reduce from 8-10 down to 4). This spreads the API load over a longer period.

2. Stagger Backup Jobs: Avoid having multiple Veeam jobs targeting the same cluster start at the exact same time. Stagger start times by 15-30 minutes to allow the ESXi API time to process snapshot creations and consolidations.

3. Check Storage Array Latency: Engage the storage vendor to ensure the underlying SAN/NAS is not struggling to process the snapshot I/O during these backup windows.

Additional Information

After applying recommendation 1 (lowering concurrent tasks on backup proxies) and Recommendation 2 (staggering jobs). Subsequent monitoring over the following weekend confirmed that backups completed successfully without causing hostd timeouts or vCenter disconnects.