chore(octopus): log repeated GraphQL responses and slot lists briefly instead of in full

[chalfontchubby]

Opened by Claude on Rik's behalf.

Summary

On a live system with an Octopus account, about 40% of predbat.log was the same text logged again and again. The biggest single item was the saving sessions GraphQL response, about 22 KB, logged in full on every poll and then a second time as "Fetched saving sessions data". This logs a repeat as one short line instead, without losing anything.

• GraphQL requests and responses: a request or response already logged in full within the last hour is logged as one line saying when it was logged in full (and, for a response, its size). It is logged in full again at least hourly, so the full text is never only in a rotated-out log. An error response is always logged in full, with the request that caused it, including on the 401/403, WAF-block and timeout paths. The duplicate "Fetched saving sessions data" line is dropped.
• Intelligent slot list: about 16 lines every plan, now logged in full only when it changes (or hourly), otherwise as one line. An empty list is not logged, as before.
• Weekend Happy Hour: "Not offering ... it cannot be joined" is logged hourly per event, not on every poll.

The decision is one small class, RepeatLogGate (utils.py). For each key it keeps a hash of each text logged in full, not the text itself, so:

• several texts can alternate under one key without forcing each other out: one request context serving several Intelligent devices, or tariff comparison runs whose slot prices differ;
• entries older than the interval are dropped, so memory stays bounded;
• expiry runs on a monotonic clock, so a clock change cannot stretch it, and the time shown is local wall-clock time, matching the log's own timestamps.

The JWT stays redacted on every line.

Not changed

• write_and_poll ... No write needed, which shows the control is being checked.
• The per-run "Inverter does not support reserve" note, about 0.5% of the log.
• The short repeat lines name the request context, not the device. With two Intelligent devices, the brief lines for each look alike. The full lines still differ.

Changes

• apps/predbat/utils.py: RepeatLogGate.
• apps/predbat/const.py: REPEAT_FULL_LOG_SECONDS (3600).
• apps/predbat/octopus.py: the GraphQL logging, the slot list and the Happy Hour line go through the gate.
• apps/predbat/predbat.py: the slot-list gate in reset().
• docs/components.md: the get_log example line no longer quotes the dropped line.
• Tests: test_octopus_logging (repeat, change, hourly re-log, two queries per context, errors, the gate itself, token redaction), test_rate_add_io_slots (slot list), test_octopus_saving_event_type (Happy Hour, including an event with no code), and test_web_mcp (fixture line).

Test plan

• Full suite and pre-commit pass on each commit.
• Measure the log size on a live system after deploying.

🤖 Generated with Claude Code