[flagsmithsmith]

Contributes to Flagsmith/sdk-spec#23

[coderabbitai]

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

Run configuration

• Source mode: Auto
• Repository: Unavailable
• Branch: Unavailable
• File: No configuration file from this repository applied
• Commit: Unavailable
• Layers: Organization UI
• Review profile: ASSERTIVE
• Plan: Advanced
• Run ID: 44265a80-dee7-4de0-a47e-b85cfcfc6d8e

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

• X
• Mastodon
• Reddit
• LinkedIn

[codspeed]

Merging this PR will not alter performance

✅ 15 untouched benchmarks

---

_{Comparing openhands/sdk-spec-v0.8.0 (ca330ed) with main (b480743)}

[flagsmithsmith]

Implemented SDK spec v0.8.0 in d019e76. The spec job failed because the SDK was missing four steps, and the event delivery behaviour behind them.

SDK changes

• Retries: event batches are retried on transient failures (5xx, 408, 429, connection errors, timeouts). There are up to 3 attempts with a 3 s request timeout and a full-jitter backoff starting at 1 s and doubling. Only one batch is in flight at a time.
• After the last attempt: if it failed transiently, the batch is kept for the next flush and sent ahead of newer events. If it got a 500, the batch is dropped. A 400-class rejection drops the batch. A 401/403 stops event tracking.
• Full buffer: the oldest event is now dropped. Before, re-queued events were truncated from the newest.
• New Flagsmith.flush_events(): sends the tracked events synchronously and returns once they are delivered, dropped or kept. If another batch was in flight, events it already attempted are not sent again. This matters for the full-buffer scenario, where reaching max_buffer_items starts a background flush.
• New Flagsmith.close(timeout_seconds=5): stops polling and streaming, then sends the buffered events, returning within the timeout. __del__ and the atexit hook use the same path.

Spec wiring (sdk-spec/steps.py): added Create the SDK client with local evaluation and events enabled, tracks events "event_{first}" to "event_{last}", flushes events and closes the client with a timeout of {timeout} seconds. Each one only calls the SDK.

Verification: I ran the spec v0.8.0 locally with tags core or events or local-evaluation: 24 passed. Unit tests (392) pass, as do mypy --strict . and pre-commit. The new delivery tests mock the events API with responses.

[flagsmithsmith]

Unit tests failed on 3.10/3.12/3.13 (the spec job passed). test_stop__sends_buffered_events checked that the timer's thread was no longer alive, but Timer.cancel() only signals it, so the thread can outlive stop() briefly. The test now checks that the timer is cancelled. Looking into it also showed a real race: a timer flush running while the client closed would schedule another timer, so the client kept flushing after close(). Stopping now prevents any further timer flushes, with a test. Fixed in ca330ed.

[flagsmithsmith]

Superseded by #283, which carries this pull request's work over to the SDK spec's latest version.