Table of Contents

A failed Home Assistant Core 2026.11 update rolled my Yellow CM4 back to 2026.10.4. The Yellow remained reachable on the network but stopped serving its dashboard. The rollback left an HTTP settings file at storage version 3.1, while Core 2026.10.4 supports version 2.2. I preserved the file, converted a copy under strict checks, restored HTTP access, and then updated Core to 2026.11.1.

Key takeaways

  • Confirm the mismatch. Match the Core log error against the installed release and HTTP storage version.
  • Preserve the file. Keep an unchanged copy and an off-device backup before recovery.
  • Verify the result. Check Core’s state, the saved listener, the dashboard, and the installed version.

Fix for this exact case: Preserve the HTTP file , build and validate a version 2.2 candidate , install it after stopping Core , and verify the update . The conversion checks the original fields and stops when they differ. Do not apply it to an uninspected HTTP file.

Before you begin

You need physical access to a Yellow CM4, a USB-C data cable, a working serial console, and a usable backup or preserved configuration. The manual path below requires host-shell access and careful JSON inspection. Plan at least 30 minutes for diagnosis and repair. Core startup and backup restoration take longer on large installations.

Recognize the failure

The decisive error appeared in the Core log:

Storage file http has version 3 which is newer than the max supported version 2

The older Core release failed to load http. The frontend and other components depending on HTTP also failed. A router lease and healthy Observer page did not establish dashboard access. Home Assistant Core issue #190583 documents the same rollback failure across several installations, including Yellow. My comment on the issue records this Yellow’s tested recovery and final Core state.

Cause of the web outage: Core 2026.11 wrote .storage/http in format 3.1. The rollback restored Core 2026.10.4 but did not restore a compatible HTTP file. Core 2026.10.4 rejects storage version 3. This Yellow’s saved intermediate_v6 profile also fails the older schema. The reason the first update failed and triggered the rollback is unknown. As of 2026-11-15, the linked issue is open, and its discussion contains no maintainer-confirmed fix.

Check the installed Core release and the HTTP store before changing files:

ha core info --raw-json | jq -r '.data.version'

storage=/mnt/data/supervisor/homeassistant/.storage
jq '{version, minor_version, pending_present: (.data.pending != null), profile: .data.stable.ssl_profile}' "$storage/http"
ha core logs | grep -F 'Storage file http has version 3'

The output below reflects this case with timestamps and private details removed:

2026.10.4
{
  "version": 3,
  "minor_version": 1,
  "pending_present": false,
  "profile": "intermediate_v6"
}
Storage file http has version 3 which is newer than the max supported version 2

The recovery commands below compare Supervisor’s saved listener with the HTTP file without printing either value. A mismatch calls for checking Core’s actual listener before changing files.

SignalMeaning
Core supports storage version 2, file reports version 3The older release cannot load its saved HTTP settings
HTTP file and Supervisor list different listenersCheck Core’s actual listener before attributing the failure
Front page responds, Core state is still STARTINGThe frontend response alone does not prove a completed startup

Preserve recovery options

Keep a usable backup outside the Yellow before another update. Home Assistant’s backup guide describes cloud and off-device locations. An encrypted backup needs its recovery key. Do not wipe the NVMe data disk while diagnosing this storage mismatch.

Copy the HTTP file before editing it:

storage=/mnt/data/supervisor/homeassistant/.storage
test ! -e "$storage/http.before-recovery" &&
  cp -p "$storage/http" "$storage/http.before-recovery" &&
  cmp -s "$storage/http" "$storage/http.before-recovery"

Stop if the command fails. An earlier recovery copy with the same name stays untouched. Choose a new name and repeat the check before continuing.

When the dashboard is unavailable, use the Yellow serial console. The Nabu Casa guide for Linux and macOS shows the USB-C connection and JP1 UART position. Follow the CM4 instructions for your hardware.

Apply the tested fix for version 3.1

The preferred path is a compatible HTTP file from before the update. Restore the backup through Home Assistant’s documented process when a complete restore is appropriate. If you extract one file, confirm its storage version and saved listener before replacing the active file. The commands below are the direct fix tested on this Yellow when no compatible HTTP file was available.

The issue’s original workaround moves .storage/http aside and lets the older Core create a new file. A later Yellow report shows a different listener in the regenerated file. Check the new file and ha core info before concluding the web interface is fixed.

My installation had no http: YAML block, no pending HTTP configuration, and no TLS certificate or key in the stable settings. The saved profile was intermediate_v6. In this narrow case, I built a separate version 2.2 candidate and retained the existing listener:

storage=/mnt/data/supervisor/homeassistant/.storage

expected=$(ha core info --raw-json | jq -r '.data.port') &&
stored=$(jq -r '.data.stable.server_port' "$storage/http") &&
test -n "$expected" &&
test "$expected" != null &&
test "$expected" = "$stored" &&
test ! -e "$storage/http.v2-candidate" &&
jq -e '
  if .version == 3 and .minor_version == 1 and
     .data.pending == null and
     .data.stable.ssl_profile == "intermediate_v6" and
     (.data.stable | has("ssl_certificate") | not) and
     (.data.stable | has("ssl_peer_certificate") | not) and
     (.data.stable | has("ssl_key") | not) and
     (.data.stable | has("trusted_proxies") | not) and
     (.data.stable | has("use_x_forwarded_for") | not)
  then .version = 2 | .minor_version = 2 | .data.stable.ssl_profile = "intermediate"
  else error("Unexpected HTTP config. Stop and inspect")
  end
' \
  "$storage/http" > "$storage/http.v2-candidate" &&

jq -e '.version == 2 and .minor_version == 2 and .data.pending == null and .data.stable.ssl_profile == "intermediate"' \
  "$storage/http.v2-candidate" >/dev/null &&

test "$(jq -r '.data.stable.server_port' "$storage/http")" = \
  "$(jq -r '.data.stable.server_port' "$storage/http.v2-candidate")"

Every command above must succeed before replacing the active file. If jq fails, inspect the candidate file before removing it and retrying. Core 2026.10.4’s HTTP source defines storage version 2.2 and accepts the intermediate profile. Core 2026.11.1’s source uses version 3.1 and versioned profile names. I also validated the candidate against the installed older Core’s HTTP_STORAGE_SCHEMA before installation.

Run the installed older Core’s schema check while its container is still running:

docker exec homeassistant python3 -c '
import json
from homeassistant.components.http.config import HTTP_STORAGE_SCHEMA
with open("/config/.storage/http.v2-candidate") as source:
    data = json.load(source)
stable = {
    key: value for key, value in data["data"]["stable"].items()
    if key not in ("created_at", "error", "error_message")
}
HTTP_STORAGE_SCHEMA(stable)
print("Schema accepted")
'

Stop if the schema rejects the candidate. This step does not validate every integration or an untested HTTP store shape.

The schema check printed Schema accepted for this candidate. Any other result stops this recovery path.

This manual conversion was tested for the stated fields. A file with TLS settings, trusted proxies, or a pending configuration needs its own review. Changing only the top-level version number leaves a profile the older Core does not accept.

Supervisor had an active Core job and rejected ha core stop in my case. I stopped the Core container from the host console, installed the validated candidate, and started it:

storage=/mnt/data/supervisor/homeassistant/.storage
jq -e '.version == 2 and .minor_version == 2 and .data.pending == null and .data.stable.ssl_profile == "intermediate"' \
  "$storage/http.v2-candidate" >/dev/null &&
  docker stop homeassistant &&
  test "$(docker inspect -f '{{.State.Running}}' homeassistant)" = false &&
  cmp -s "$storage/http" "$storage/http.before-recovery" &&
  mv "$storage/http.v2-candidate" "$storage/http" &&
  docker start homeassistant

If a command fails after the stop, inspect the active HTTP file before running docker start homeassistant. A changed file needs a new candidate built from its latest contents. The HTTP page returned after this step. An HTTP response alone did not prove Core had finished startup, so I checked Supervisor’s state before treating the system as recovered.

On this Yellow, docker inspect -f '{{.State.Status}}' homeassistant then printed running. Check ha core info and the dashboard before requesting another update.

Verify the update outcome

After HTTP access returned, I updated Core to 2026.11.1. To request that release:

ha core update --version 2026.11.1

The command reported a failure after a long startup. Supervisor then logged Core as RUNNING and Successfully started Home Assistant 2026.11.1. A status read showed the installed release was current, the HTTP file had migrated back to version 3.1, and the container had zero restarts during observation. The authenticated dashboard loaded. Check the resulting state even when the update command reports failure.

For your system, check the release, Supervisor health, and container state after any long update:

ha core info --raw-json | jq '{version: .data.version, update_available: .data.update_available}'
ha supervisor info --raw-json | jq '{healthy: .data.healthy, supported: .data.supported}'
docker inspect -f '{{.State.Status}} restarts={{.RestartCount}}' homeassistant

The first and third commands produced these sanitized results in this case:

{
  "version": "2026.11.1",
  "update_available": false
}
running restarts=0

Then open the dashboard through your normal local address and confirm live states load. Keep the saved HTTP file until a subsequent backup and normal operation are verified.

The first update’s failure was not established by the evidence from my Yellow. The later job’s failure message also did not identify a cause. An installed release and a healthy dashboard are stronger evidence of the resulting state than the CLI message alone, but neither explains why Supervisor marked the job failed.

Next steps

If the HTTP storage error remains, inspect the preserved file and restore a compatible backup or run a Core release able to read its storage version. Restoring the preserved version 3 file under the older Core repeats the outage. If Core reaches RUNNING but individual devices remain unavailable, investigate those integrations separately. Keep the recovery copy and cloud backup until the next backup completes and the dashboard stays responsive.