Track API tokens per device instead of globally

[simons-plugins]

Problem

The plugin tracks a single _token_remaining counter in NetroAPIClient that gets overwritten by every API response's meta.token_remaining. Since Netro confirmed that each device has its own independent 2000/day token limit, the last device polled determines the displayed count.

This causes the proactive pause threshold (100 tokens) to trigger based on one device's count, pausing ALL polling — including devices with plenty of tokens remaining.

Context from Netro support

The 2000 limit applies per API key as well as per device. The API key is generated based on the serial number, so the two limits are equivalent.
Each device — Spark, Whisper, and Pixie — is independent and has its own 2000 limit.
All devices perform a poll every 5 minutes, which should theoretically be sufficient: 60 × 24 / 5 = 288 times per day.

Expected token usage per device

• Sprinkler controller: ~4 calls/poll × 288 polls/day = 1,152/day (within 2000)
• Whisperer sensor: ~1 call/poll × 288 polls/day = 288/day (within 2000)

No single device should approach the 2000 limit at normal polling intervals.

Proposed solution

• Track token_remaining and token_reset per device serial number in NetroAPIClient
• Apply pause/warning thresholds per device, not globally
• Update the token_remaining device state per device
• Consider whether proactive pause is still needed given the maths above

Files involved

• api_client.py — _update_token_budget(), should_pause_polling, _token_remaining
• plugin.py — runConcurrentThread() pause check, device state updates

[simons-plugins]

Finding: Token budget appears to be per-account, not per-device

Despite Netro support stating each device has an independent 2000/day limit, historical data from the SQL Logger shows all devices drawing from the same pool.

Evidence

Token history for today (from Indigo SQL Logger via Domio plugin):

Veg Plant Sensor (Whisperer, 1 API call/poll):

02:00  tokens=1988  (reset)
03:00  tokens=1894  (-94 in 1hr)
05:05  tokens=1703
07:05  tokens=1515
09:00  tokens=1323
11:00  tokens=1047
13:00  tokens=768
15:00  tokens=489
15:55  tokens=357
16:55  tokens=205

A whisperer makes 1 API call per poll. At 5-min intervals that's 12 calls/hour = 288/day. Yet tokens dropped by ~95/hour — meaning other devices' API calls (different API keys) are draining the same budget.

Irrigation Controller (Sprinkler, ~4 calls/poll + weather/forecast):

16:25  tokens=289
16:30  tokens=277  (-12 per 5 min)
16:35  tokens=265
...
17:40  tokens=94

~12 tokens per 5-min cycle matches the sprinkler's actual API calls (4 polling + weather + forecast), but the absolute count (289 at 16:25) is much lower than it should be for a device with its own 2000 budget that only started polling today after a rate-limit reset.

Front Garden Window Bed (Sprinkler): current=1139
Fern Bed Plant Sensor (Whisperer): current=1711 (stale — was paused by old global tracking)

Interpretation

All devices report different token_remaining values because each API response reflects the account-wide budget at the moment that specific key was used. A whisperer polled at 13:00 saw 768. The controller polled at 16:25 saw 289. They're reading from the same counter at different times.

If tokens were truly per-device:

• Whisperer (1 call/poll × 12 polls/hr) would use ~288 tokens/day — should still show ~1700+
• Instead it shows 205, consistent with account-wide drain from all 9 devices

Impact on PR #47

The per-device tracking in PR #47 is still correct — it accurately records what each device's API response reports. But the proactive pause logic needs to account for a shared budget. Currently a whisperer might appear to have plenty of tokens while the shared pool is nearly empty.

Action items

• Follow up with Netro support with this evidence
• Consider whether to pause all devices when any device reports low tokens (closer to the old global behavior, but now intentional)
• Monitor token consumption over 24h with the new per-device tracking to get clean data