HAOS 18.0 regression: intermittent TCP connection failures to LAN devices (confirmed via curl, affects ONVIF/PTZ)

Describe the issue you are experiencing

The problem

After upgrading to HAOS 18.0, outbound TCP connections from the host to LAN devices fail intermittently, causing timeouts in any application that communicates with local network devices. Confirmed and reproduced using plain curl completely outside of Home Assistant Core or any integration code.

This was initially reported as an ONVIF integration issue in home-assistant/core#174707, where the code owner @jterrace confirmed it is a HAOS issue and suggested opening this ticket.

Environment

• Home Assistant OS: 18.0 (broken)
• Last working HAOS: 17.3
• Home Assistant Core: 2026.6.4 (confirmed NOT the cause)
• Supervisor: 2026.06.3
• Installation: Home Assistant OS (Proxmox VM, amd64)
• Affected device: Asecam PTZ 4K camera (XiongMai/XMEye chipset) at 192.168.1.5:80

Isolation testing

I did the following isolated tests to confirm the cause is HAOS 18.0 specifically, not Core or any integration:

1. Core 2026.6.2 + OS 17.3 → working normally
2. Core 2026.6.4 + OS 17.3 (OS not yet updated) → tested extensively, working normally, no timeouts
3. Core 2026.6.4 + OS 18.0 (after updating only the OS) → bug appears immediately

Reproduction with plain curl (no HA/ONVIF code involved)

Running the following command repeatedly from the HAOS host shell:

curl -v --max-time 5 -o /dev/null -s "http://192.168.1.5:80/onvif/device_service"

Result with zero delay between attempts:

• Attempt 1: ✅ SUCCESS — Established connection... HTTP 200 OK
• Attempt 2: ✅ SUCCESS — Established connection... HTTP 200 OK
• Attempt 3: ❌ FAILED — Connection timed out after 5002 milliseconds

This pattern (roughly 1 in 3 attempts failing) repeats consistently, proving the issue is at the OS network layer, not in any application code.

Additional network diagnostics (all normal)

# Ping: 0% packet loss, 1-4ms latency (20/20 packets)
ping -c 20 192.168.1.5
# Result: 20 packets transmitted, 20 received, 0% packet loss
# round-trip min/avg/max = 1.123/2.279/4.415 ms

# MTU: standard 1500
ip addr show docker0
# Result: mtu 1500

# Conntrack timeout: 432000s (5 days, not the cause)
cat /proc/sys/net/netfilter/nf_conntrack_tcp_timeout_established
# Result: 432000

# Camera responds correctly with Connection: close
# (no keep-alive involved, rules out that hypothesis)

Symptom in Home Assistant (triggered by the OS bug)

When the TCP failure hits an ONVIF PTZ command:

• PTZ commands queue up silently
• When a connection finally succeeds, all queued commands fire at once
• Camera moves uncontrollably until physical limit
• Eventually errors out with: TimeoutError: Request to http://192.168.1.5:80/onvif/ptz timed out aiohttp.client_exceptions.ConnectionTimeoutError: Connection timeout to host http://192.168.1.5/onvif/ptz

Suspected cause

HAOS 18.0 introduced two major networking-adjacent changes compared to 17.3:

• Linux kernel 6.12 → 6.18
• Docker v29.5.3 + containerd v2.2.4

The intermittent nature (~1 in 3 attempts failing) and the fact that ICMP/ping works perfectly while TCP connection establishment fails suggests a race condition or timing issue in either:

• The new kernel's TCP/bridge networking stack
• Docker 29.5.3's NAT/bridge configuration
• An interaction between the two

A tcpdump capture during failure would likely show the SYN packet being dropped or not returning an ACK, but I don't have the tools available in the HAOS shell to run this (ss, tcpdump not available).

Reference

• Original Core issue (closed, redirected here): home-assistant/core#174707
• Debug log from HA showing the ONVIF symptom: attached to original issue

What operating system image do you use?

generic-x86-64 (Generic UEFI capable x86-64 systems)

What version of Home Assistant Operating System is installed?

18.0

Did the problem occur after upgrading the Operating System?

Yes

Hardware details

Proxmox VE VM (amd64), generic UEFI x86-64. Virtualized via KVM/QEMU. No USB devices attached. Host hardware: Intel N100 Mini PC

Steps to reproduce the issue

1. Install HAOS 17.3 with Home Assistant Core 2026.6.4 on Proxmox VM (generic-x86-64)
2. Confirm ONVIF PTZ camera working normally at 192.168.1.5:80
3. Upgrade HAOS from 17.3 to 18.0 (Core unchanged at 2026.6.4)
4. Observe ONVIF PTZ commands timing out intermittently
5. From HAOS host shell, run curl repeatedly: curl -v --max-time 5 -o /dev/null -s "http://192.168.1.5:80/onvif/device_service"
6. Observe ~1 in 3 attempts timing out with no delay between requests
7. Downgrade HAOS back to 17.3 - issue disappears immediately

...

Anything in the Supervisor logs that might be useful for us?

No relevant Supervisor logs. Issue reproduces with plain curl outside of HA entirely.
See home-assistant/core#174707 for full debug logs.

Anything in the Host logs that might be useful for us?

No relevant Host logs captured. Tools ss and tcpdump not available in HAOS shell.
Issue confirmed via curl: intermittent TCP connection timeout to LAN device.

System information

No response

Additional information

No response