Recently I migrated my Zigbee network to a new controller. From my old Sonoff ZBDongle-P (Z-Stack) to a Sonoff Dongle Plus MG24 (Ember) I got for about 13€ from AliExpress. I was following this guide and everything went well. But a few days later, suddenly in the night, the new dongle crashed. These were the related Z2M logs:
[2026-11-21 02:20:13] [31merror[39m: zh:ember:ezsp: ERROR Transaction failure; status=ASH_ERROR_TIMEOUTS. Last Frame: [FRAME: ID=52:"SEND_UNICAST" Seq=53 Len=7].
[2026-11-21 02:20:13] [31merror[39m: zh:ember:ezsp: ERROR Transaction failure; status=NOT_CONNECTED. Last Frame: [FRAME: ID=52:"SEND_UNICAST" Seq=53 Len=7].
[2026-11-21 02:20:13] [31merror[39m: zh:ember:ezsp: ERROR Transaction failure; status=NOT_CONNECTED. Last Frame: [FRAME: ID=52:"SEND_UNICAST" Seq=53 Len=7].
[2026-11-21 02:20:13] [31merror[39m: z2m: Adapter disconnected, stopping
WATCHDOG: Waiting 0.03333333333333333min before next start try.
Starting Zigbee2MQTT with watchdog (2000,60000,300000,900000,1800000,3600000).
[2026-11-21 02:20:41] [31merror[39m: z2m: Error while starting zigbee-herdsman
[2026-11-21 02:20:41] [31merror[39m: z2m: Failed to start zigbee-herdsman
[2026-11-21 02:20:41] [31merror[39m: z2m: Check https://www.zigbee2mqtt.io/guide/installation/20_zigbee2mqtt-fails-to-start_crashes-runtime.html for possible solutions
[2026-11-21 02:20:41] [31merror[39m: z2m: Exiting...
[2026-11-21 02:20:41] [31merror[39m: z2m: Error: Failed to start EZSP layer with status=HOST_FATAL_ERROR.
at EmberAdapter.initEzsp (/app/node_modules/.pnpm/zigbee-herdsman@11.2.0_supports-color@7.2.0/node_modules/zigbee-herdsman/src/adapter/ember/adapter/emberAdapter.ts:677:19)
at async EmberAdapter.start (/app/node_modules/.pnpm/zigbee-herdsman@11.2.0_supports-color@7.2.0/node_modules/zigbee-herdsman/src/adapter/ember/adapter/emberAdapter.ts:1558:24)
at async Controller.start (/app/node_modules/.pnpm/zigbee-herdsman@11.2.0_supports-color@7.2.0/node_modules/zigbee-herdsman/src/controller/controller.ts:146:29)
at async Zigbee.start (/app/lib/zigbee.ts:70:27)
at async Controller.start (/app/lib/controller.ts:117:29)
at async start (/app/index.js:156:5)
at async triggerWatchdog (/app/index.js:36:9)
at async Controller.exit [as exitCallback] (/app/index.js:52:13)
at async Controller.exit (/app/lib/controller.ts:379:16)
at async Controller.stop (/app/lib/controller.ts:374:16)
WATCHDOG: Waiting 1min before next start try.
Turned out, the reason was a backup job on my Proxmox host on which my Home Assistant VM is running.
journalctl
--since "2026-11-21 02:15:00"
--until "2026-11-21 02:25:00"
--no-pager
| grep -Ei 'vzdump|backup|snapshot|freeze|thaw|qemu|usb|reset'
Nov 21 02:15:01 pve2 CRON: (root) CMD (which zfs-auto-snapshot > /dev/null || exit 0 ; zfs-auto-snapshot --quiet --syslog --label=frequent --keep=4 //)
Nov 21 02:16:33 pve2 pvescheduler: INFO: Finished Backup of VM 195 (00:04:28)
Nov 21 02:16:33 pve2 pvescheduler: INFO: Starting Backup of VM 196 (lxc)
Nov 21 02:18:40 pve2 pvescheduler: INFO: Finished Backup of VM 196 (00:02:07)
Nov 21 02:18:40 pve2 pvescheduler: INFO: Starting Backup of VM 204 (qemu)
VM 204 backup started at 02:18:40, which is my HA VM. A few minutes later, the MG24 dongle crashed. By doing another manual backup, I could reproduce the controller crash.
My backup job was using the following settings:
--mode snapshot
--storage TrueNAS
--compress zstd
--fleecing 0
AI suggested different solutions, one of it was to enable fleecing in the job configuration. This reduces guest-write delays caused by copying old blocks directly to the remote backup target. To be honest, I had never heard of this setting before, but after enabling it and running another backup, the MG24 dongle stopped crashing. I have no idea why my previous Dongle-P didn’t have this problem, but since it was a simple solution, I’ll just take it.






