[BUG] Reacquired upstream history transport can bind to a closed aiohttp connector

[CaneTLOTW]

Follow-up/blocker discovered during the live runtime validation of Issue #53 on exact merged develop SHA 3f458b3906ce2c1db13ea982f2488418978ef0d3.

Runtime result from Issue #53:

initial:
server_history_ready=false
server_history_source=local_fallback
server_history_reason=upstream_vehicle_unavailable

later reacquisition:
server_history_ready=false
server_history_source=local_fallback
server_history_reason=sync_failed
server_history_capability=authenticated_trips_transport
server_history_error=Connector is closed.

The reacquisition worker therefore did what #53 intended: it later resolved a VIN-matching loaded upstream client/vehicle and reached the authenticated historical transport. The remaining blocker is that the resolved upstream HTTP transport was not usable at request time.

Relevant upstream lifecycle code at inspected upstream commit 950cc93074c7a43974956a3134579e2cbc81228d:

• StellantisBase.start_session() creates a new aiohttp.ClientSession() only when self._session is falsey;
• close_session() returns immediately when self._session.closed is already true and in that early-return path does not reset self._session = None;
• make_http_request() catches request failures and calls close_session();
• normal upstream config-entry shutdown also closes the shared session.

This makes a stranded closed-session/closed-connector reference a plausible cause, but it is NOT yet proven. A concurrent upstream request closing the shared session, a stale/shutting-down runtime client, or another lifecycle race must also be ruled out.

Required audit first

On the current HA runtime, without modifying source or stores, reproduce/inspect the failure and capture for the exact client object that SV resolves:

1. upstream ConfigEntry id, loaded state and runtime_data object identity;
2. exact VIN and coordinator identity returned by async_get_coordinator_by_vin(vin);
3. client._shutting_down if present;
4. client._session identity and whether it is None;
5. client._session.closed when present;
6. connector identity and connector.closed when present;
7. whether a normal upstream coordinator/API refresh immediately before/after the SV history attempt succeeds or fails;
8. whether the resolved runtime_data object changes across reload/reacquisition;
9. exact exception type/traceback for Connector is closed. rather than only the message.

Do not expose tokens/credentials in the report.

Fix constraints

After the root cause is proven, implement the smallest SV-side compatibility fix. Keep the Stellantis Vehicles repository unchanged.

Architecture constraints:

• exact VIN matching remains mandatory;
• no independent OAuth/login/credential stack;
• do not monkey-patch upstream methods/classes;
• do not silently bind to a stale or shutting-down client;
• prefer the currently loaded ConfigEntry.runtime_data object;
• preserve local/archive fallback until an authenticated /trips sync actually succeeds;
• transient transport/session lifecycle failures must remain recoverable in the same HA uptime and must not permanently strand server history in sync_failed;
• do not clear/archive raw history on failure.

If the proven cause is a closed/stale upstream session object, the implementation must explicitly document why the chosen SV compatibility action is safe and narrowly scoped. Do not introduce a second long-lived HTTP session owned by SV without an explicit architecture review.

Regression tests

Cover the proven failure mode, including:

• resolved client with unusable/closed transport;
• later recovery to a usable authenticated transport;
• strict VIN matching;
• no duplicate retry loops;
• unload cancellation;
• archive/local fallback preservation;
• readiness only after successful sync.

Issue #53 remains open until this blocker is fixed and the same runtime acceptance reaches server_history_ready=true, source=server, reason=sync_succeeded with a fresh last_sync.

No main promotion or release before runtime validation.

[CaneTLOTW]

ChatGPT → Codex Handoff

Repository: CaneTLOTW/sv_dashboard

Base SHA: 3f458b3906ce2c1db13ea982f2488418978ef0d3

Work branch already exists: fix/server-history-closed-connector

Task: Audit and fix Issue #55 in sv_dashboard only. Do not modify or patch the Stellantis Vehicles upstream repository.

Phase 1 — prove the transport-lifecycle root cause

The previous exact runtime candidate resolved the upstream VIN later, then failed the historical /trips call with Connector is closed.. Before changing source, inspect the live upstream runtime object and available HA logs/traceback. Current HA is rolled back to beta.13, so use existing logs first; if the failure cannot be proven from them, temporarily reinstall exact merged SHA 3f458b3906ce2c1db13ea982f2488418978ef0d3 using the same backup-first discipline, reproduce once, collect the diagnostics below, then proceed. Do not alter HA stores/config/entities manually.

For the exact client object selected by SV, capture without secrets:

1. Stellantis ConfigEntry id and loaded state;
2. runtime_data object identity;
3. configured VIN, coordinator identity and exact VIN match;
4. _shutting_down value if present;
5. _session identity / None;
6. _session.closed if present;
7. connector identity and connector.closed if present;
8. exact exception class + traceback for Connector is closed.;
9. whether the same upstream client can perform its normal coordinator refresh immediately before/after the failure;
10. whether runtime_data or session identity changes between initial miss, later reacquisition and any upstream reload.

Explicitly test the hypothesis from upstream StellantisBase lifecycle code: a closed session/connector may remain referenced because start_session() only creates a session when _session is falsey, while close_session() returns early when _session.closed is already true. This is a hypothesis, not permission to mutate upstream internals before proving it.

Phase 2 — smallest SV-side recovery

After the cause is proven, implement the narrowest compatibility fix with these constraints:

• keep upstream source untouched;
• no independent login/OAuth/credential stack;
• no monkey-patching methods/classes;
• exact VIN match remains mandatory;
• never bind a _shutting_down or otherwise demonstrably stale client;
• continue to use the upstream authenticated request machinery where viable;
• preserve archive/local fallback while transport is unavailable;
• a transient transport/session lifecycle failure must remain retryable in the same HA uptime instead of permanently ending in sync_failed;
• do not create duplicate retry/sync loops;
• cancellation on SV unload must remain clean.

If the proven fix requires touching a private upstream runtime field such as a closed _session reference, stop before implementing that mutation and report the evidence + proposed action back in #55 for architecture approval. Prefer a public/upstream-owned recovery path if one exists. Do not introduce an SV-owned long-lived HTTP session without explicit approval.

Tests and delivery

Add targeted regression coverage for the proven failure mode and later successful recovery, plus existing VIN/archive/unload/no-duplicate guarantees. Run the full AGENTS validation. If the implementation stays within the approved constraints, open a PR to develop and report exact HEAD/tests/CI/PR under ## Codex → ChatGPT Ergebnis in #55.

Do not merge, release, touch main, or leave the HA runtime on an unvalidated candidate.