Symptom
After upgrading to Proxmox VE 9.2.20 (September 24, from 9.2.4), LXC/VM consoles in the web interface sometimes failed to start (termproxy ... failed: exit code 1, failed waiting for client: timed out), and overall interaction (console use, text highlighting/copy-paste, even SSH connection setup at times) was noticeably slower than before the upgrade.
Confirmed contributing causes
Open loose end (unexplained, doesn't appear to be blocking)
VM 705 keeps logging qga command failed - unable to open monitor socket roughly every 60 seconds, despite agent: 0 in its config and a full stop/start plus host reboot. However, an strace at the moment of the call showed a fast, successful connect() — so this looks more like cosmetic log noise than an actual hang.
Conclusion
Two concrete, confirmed contributing factors were resolved (pbs1 reachability, nftables firewall off), with noticeable improvement. The remaining slowness likely matches a broader "sluggish webUI after upgrade to PVE 9" pattern
After upgrading to Proxmox VE 9.2.20 (September 24, from 9.2.4), LXC/VM consoles in the web interface sometimes failed to start (termproxy ... failed: exit code 1, failed waiting for client: timed out), and overall interaction (console use, text highlighting/copy-paste, even SSH connection setup at times) was noticeably slower than before the upgrade.
Confirmed contributing causes
- PBS VM (705, Proxmox Backup Server,aka pbs1) temporarily unreachable: pvestatd/pvedaemon were periodically checking pbs1's datastores, and those checks timed out (No route to host / Connection timed out), tying up worker processes — so other API tasks (like console requests) had to wait for a free worker. Resolved by fully stopping/starting VM 705.
- Proxmox firewall (nftables) active at node level: deliberately enabled a week earlier for a strict rule on one LXC, with ACCEPT/ACCEPT policy at datacenter level. Disabling it gave a noticeable improvement, though not fully back to the pre-upgrade level.
- Orphaned dtach sockets — present but not the core issue.
- DNS/reverse-DNS resolution — fast, not the cause.
- GSSAPI/SSH auth delay — disabled, not the cause.
- Cluster leftovers (old corosync config, /etc/pve/nodes/) — clean; node is correctly single-node.
- QEMU monitor socket lock on VM 705 — no other processes holding the socket.
- Cron jobs / vzdump backup config — VM 705 is not in the backup job's VMID list.
- PCIe correctable errors (r8169 NIC) — present since January, unrelated to this upgrade.
- CIFS shares (TNAS, ProtonDrive) — now responding in milliseconds, no hang.
- Ballooning setting on VM 705 — disabling had no effect.
Open loose end (unexplained, doesn't appear to be blocking)
VM 705 keeps logging qga command failed - unable to open monitor socket roughly every 60 seconds, despite agent: 0 in its config and a full stop/start plus host reboot. However, an strace at the moment of the call showed a fast, successful connect() — so this looks more like cosmetic log noise than an actual hang.
Conclusion
Two concrete, confirmed contributing factors were resolved (pbs1 reachability, nftables firewall off), with noticeable improvement. The remaining slowness likely matches a broader "sluggish webUI after upgrade to PVE 9" pattern
Last edited:
