HAOS 17.x breaks ZHA with network-based EZSP coordinator (SLZB-06M via socket://)

[Kruppes]

Describe the issue you are experiencing

After upgrading to HAOS 17.0/17.1, ZHA integration fails to connect to a network-based SLZB-06M coordinator via socket://192.168.10.48:6638. The integration enters setup_error state and cannot recover.

Rolling back to the pre-17.x HAOS backup immediately resolves the issue. The same HA Core version (2026.1.3) works on pre-17.x HAOS but fails on 17.x.

Key findings during debugging

• The SLZB-06M is fully reachable (ping, HTTP web UI, TCP port 6638 open)
• Raw EZSP/ASH protocol test succeeds: Sending an ASH RST frame to port 6638 returns a valid RSTACK response (1ac1020b0a527e)
• Despite the coordinator responding correctly at the protocol level, ZHA/bellows fails with a 500 Internal Server Error during config flow setup
• The SLZB-06M shows zb_channel: 0 (no active Zigbee network), suggesting bellows never gets far enough to initialize the coordinator
• Attempted fixes that did NOT help:
  • ZHA integration reload
  • ZHA integration delete and re-add
  • SLZB-06M software reboot (Zigbee chip + core)
  • SLZB-06M physical power cycle
  • SLZB-06M Zigbee firmware re-flash (same version)
  • Full HA VM reboot via Proxmox
  • Trying both flow_control: software and flow_control: none
• Rolling back to pre-HAOS 17.x backup: immediately fixes the issue

What operating system image do you use?

generic-x86-64 (Proxmox VM)

What version of Home Assistant Operating System is installed?

17.1 (also reproduced on 17.0)

Did the problem occur after upgrading the Operating System?

Yes

Hardware details

• Coordinator: SMLIGHT SLZB-06M (EFR32MG21, firmware v20250220, SLZB-OS v3.2.4)
• Connection: Network/Ethernet (PoE), socket://192.168.10.48:6638
• Host: Proxmox VM (generic-x86-64)
• HA Core: 2026.1.3 (same version before and after HAOS upgrade)

Suspected cause

HAOS 17.0 includes major updates to Docker (v29.1.3), Containerd (v2.2.0), and the Linux kernel (6.12.63→6.12.67). One of these changes likely affects how HA Core containers handle TCP socket connections to network-based Zigbee coordinators. This is similar to #4494 (HAOS 17.0 broke Z2M with USB-based ZBT-2) but specifically affects network/socket-based coordinators using the EZSP protocol.

Steps to reproduce

1. Have a working ZHA setup with a network-based SLZB-06M coordinator on HAOS 16.x
2. Upgrade to HAOS 17.0 or 17.1
3. ZHA enters setup_error and cannot reconnect
4. Attempting to delete and re-add ZHA results in 500 Internal Server Error during config flow

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

The 500 error occurs during the ZHA config flow at the manual_port_config step when submitting the socket path.

[kp-bit]

Seems to be the same problem when using SkyConnect dongle

[sairon]

If you want to get help here, please fill in the issue form as requested and provide logs and diagnostics data, instead of pasting pre-processed AI summary full of misunderstood things. The SLZB-06M is just a serial-to-TCP bridge - it passes the EZSP/ASH traffic through without interpreting it, so zb_channel: 0 doesn't indicate anything.

I have two SLZB coordinators working without issues on 17.1, one talks to a Proxmox VM, the other to a Raspberry Pi 5, and since there are no further reports, given how popular they are, it all points to something wrong in your setup.