Automated Release PR

0.11.8 (2026-06-01)

Full Changelog: v0.11.7...v0.11.8

Features

• cli: add Temporal + LangGraph agent template and example (#383) (bbc9e02)
• tracing: OTel span queue and export telemetry (SGPINF-1863) (#373) (6669012)

This pull request is managed by Stainless's GitHub App.

The semver version number is based on included commit messages. Alternatively, you can manually set the version number in the title of this pull request.

For a better experience, it is recommended to use either rebase-merge or squash-merge when merging this pull request.

🔗 Stainless website
📚 Read the docs
🙋 Reach out for help or questions

Greptile Summary

This release (0.11.8) ships two features: a new Temporal + LangGraph agent template/example, and OTel span queue + export telemetry instrumentation.

• Temporal + LangGraph template: Adds a full CLI template (temporal-langgraph) and a matching tutorial example. The template includes human-in-the-loop interrupt handling via Temporal signals, live introspection queries, and per-turn SGP tracing (correctly guarded by SGP_API_KEY / SGP_ACCOUNT_ID). A new emit_langgraph_messages ADK helper converts LangGraph message objects to Agentex content types.
• OTel span queue telemetry: Adds tracing_metrics.py (instrument definitions) and tracing_metrics_recording.py (best-effort recording helpers) with bounded tag cardinality throughout. The recording path is disabled via AGENTEX_TRACING_METRICS=0 and swallows all exceptions so it is safe on hot paths. enqueued_at timestamps are correctly preserved through re-enqueues for accurate lag histograms.

Confidence Score: 5/5

Safe to merge — changes are additive, all tag values are bounded, and the recording helpers are exception-safe on hot paths.

Both features are purely additive. The tracing metrics are well-isolated behind a feature flag and best-effort exception handling. The LangGraph template correctly guards credential-dependent setup. No behavioral regressions are introduced in existing code paths.

src/agentex/lib/core/tracing/span_queue.py — the retry path silently re-enqueues failed items without emitting a failure metric, making transient errors invisible until retries are exhausted.

Important Files Changed

Sequence Diagram

sequenceDiagram
    participant Caller
    participant AsyncSpanQueue
    participant Metrics as tracing_metrics_recording
    participant Processor as SGPAsyncTracingProcessor
    participant SGP as SGP HTTP API

    Caller->>AsyncSpanQueue: enqueue(event_type, span, processors)
    AsyncSpanQueue->>Metrics: record_span_enqueued(event_type)
    Note over AsyncSpanQueue: stores enqueued_at timestamp

    AsyncSpanQueue->>AsyncSpanQueue: drain loop (linger_ms)
    AsyncSpanQueue->>Metrics: record_batch_coalesced(queue_depth, batch_items)

    AsyncSpanQueue->>Processor: on_spans_start / on_spans_end
    Processor->>SGP: upsert_batch(...)

    alt HTTP success
        SGP-->>Processor: 200 OK
        Processor->>Metrics: record_export_success(event_type, span_count, sgp)
    else Retriable failure (retries remaining)
        SGP-->>Processor: 5xx / network error
        Processor-->>AsyncSpanQueue: raises exception
        Note over AsyncSpanQueue: re-enqueues item, no metric emitted
    else Retries exhausted
        AsyncSpanQueue->>Metrics: record_export_failure(exhausted_count)
    else Permanent non-retriable failure
        AsyncSpanQueue->>Metrics: record_export_failure(all items)
    end

    AsyncSpanQueue->>Metrics: record_batch_phase(phase, size, duration_ms)

    Note over AsyncSpanQueue: on shutdown timeout
    AsyncSpanQueue->>Metrics: record_shutdown_timeout(remaining_items)

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
src/agentex/lib/core/tracing/span_queue.py:349-361
**Transient export failures are invisible in metrics**

`record_export_failure` is only emitted for spans whose retry budget is exhausted (or non-retriable permanent failures). Items that fail but are re-enqueued for another attempt emit no metric. In practice, a span that fails 4 times and succeeds on the 5th retry shows only a single `record_export_success` with zero failure events recorded. The counter names `export_batch_failures` / `export_span_failures` (described as "failed HTTP export batches") imply every failure attempt, but the current implementation counts only permanent losses, making transient error rates unobservable. Consider calling `record_export_failure` (with the retriable span count) for intermediate failures as well, alongside the re-enqueue path.

Reviews (3): Last reviewed commit: "release: 0.11.8" | Re-trigger Greptile