fix(solis): queue entity events for the component loop

[mgazza]

Problem

Solis entity callbacks perform real API reads and writes inline — solis.py:2650, :2880, :3056.

But those callbacks run on the HA component's event loop, not Solis's own. Standalone Predbat runs every component in its own thread with its own asyncio.run() loop (hass.py:217, :223), and the HA interface is one of those components — its websocket loop awaits trigger_callback() directly (ha.py:619).

The ClientSession is bound to the Solis loop. Issuing a request from the callback loop raises inside aiohttp, the handler catches it and returns normally, and the user is told a write succeeded that never reached the inverter.

Why Solis specifically

Ohme and Octopus already avoid this — their callbacks only append to a queue and do the work in their own run():

component	callback does	exposed?
Octopus	appends a command (octopus.py:591), executed in run() (:641)	no
Ohme	queues; explicitly commented as stubs (ohme.py:689), drained in run() (:239)	no
Solis	performs the read/write inline	yes

Change

Solis follows the same pattern. select_event, number_event and switch_event become thin stubs that queue; the existing bodies become *_event_handler; run() drains the queue on the Solis loop before polling. A failing handler is logged and the rest of the queue still drains — the queue is in memory only, so anything dropped is lost outright.

This deliberately does not touch session lifetime. One session stays on one loop, so connection reuse, socket lifetime and teardown are all unchanged, and injected test sessions keep working.

Tests

Existing event tests now call the handlers directly — that is what they were always exercising.

Two new tests cover the dispatch itself: that a callback queues rather than executing on the calling loop, and that one failing event does not strand the rest of the queue. MockSolisAPI gains queued_events, since it hand-rolls the state the real __init__ sets.

Verified with the repository harness — unit_test.py --test solis, which --quick skips:

• Solis suite passes.
• The two pre-existing Unclosed client session warnings and the multi_car_iog failure are present on a clean tree too, so neither is introduced here.
• Mutation-checked: making select_event call its handler inline again fails the new test with callback executed API work on the calling loop.

Note

This supersedes #4874, which tried to solve it at the session layer by rebinding the ClientSession per loop. That was the wrong level — it broke injected test sessions, leaked a session per rebind, and destroyed connection reuse in exactly the alternating-loop case it targeted. Closing that in favour of this.

[Copilot]

🟡 Changes recommended

run() currently drains queued events before first-cycle session creation/inverter discovery, which can drop or mis-handle early queued writes during startup.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

This PR changes the Solis component so Home Assistant entity callbacks (select_event, number_event, switch_event) no longer perform API I/O inline on the HA loop, and instead queue work for the Solis component loop to execute (avoiding cross-event-loop aiohttp.ClientSession usage).

Changes:

• Add an in-memory queued_events queue to Solis, convert entity callbacks into stubs that enqueue handler calls, and drain the queue from run() on the Solis loop.
• Rename the previous callback bodies into *_event_handler methods and update Solis tests to call handlers directly.
• Add tests asserting callbacks queue (don’t execute on the calling loop) and that one failing queued event doesn’t block subsequent events.

File summaries

File	Description
apps/predbat/solis.py	Introduces queued event stubs + handler methods and drains queued events in run() to keep API work on the Solis loop.
apps/predbat/tests/test_solis.py	Updates existing tests to target handlers and adds coverage for queuing/dispatch behavior.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 1
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🟡 Changes recommended

The current queue-draining loop can starve polling indefinitely if callbacks append events while draining, which is a reliability risk and can be fixed with a small, local change.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 1
• Review effort level: Lite