sentry-sdk 2.61.0

on Python PyPI

New Features ✨

• Add server.address to transformed spans when stream_gen_ai_spans=True by @alexander-alderman-webb in #6307

• Allow integrations to define control flow exceptions by @sentrivana in #6425

• Disable string truncation for events by default by @alexander-alderman-webb in #6290

  Following a previous significant increase of the string truncation limit, we've now completely removed the limit by default.
  In case you have large strings in your events, you should now be able to see them.

  In rare cases, if you have really long strings (or a lot of them), you might see envelopes being dropped because of their size.
  If that happens, you can set the max_value_length init option to the previous value of 100_000:

  sentry_sdk.init(
      ...,
      max_value_length=100_000,
  )

Bug Fixes 🐛

Langchain

• Stop setting transaction status when child span fails by @alexander-alderman-webb in #6301
• Catch TypeError on langchain.agents import by @alexander-alderman-webb in #6268

Openai Agents

• Handle starting_agent keyword argument in runner patches by @ericapisani in #6428
• Remove hosted MCP tool spans by @alexander-alderman-webb in #6391
• Use name, not description in start_span by @sentrivana in #6323
• Stop setting transaction status when child span fails by @alexander-alderman-webb in #6303

Pydantic AI

• Stop setting tokens on Invoke Agent spans by @alexander-alderman-webb in #6320
• Stop setting transaction status when child span fails by @alexander-alderman-webb in #6302
• Remove Agent.run_stream_events() patch by @alexander-alderman-webb in #6281

Strawberry

• Wrap yields in try-except to ensure span cleanup by @ericapisani in #6381
• Fix AttributeError on graphql_span in resolve by @sentrivana in #6289

Other

• (anthropic) Do not set gen_ai.response.model to None by @alexander-alderman-webb in #6312

• (asyncpg) Use Sentry span attribute name conventions by @ericapisani in #6306

• (boto3) Guard setting method by @sentrivana in #6288

• (cohere) Stop setting transaction status when child span fails by @alexander-alderman-webb in #6300

• (google-genai) Guard against None response ID and response model by @alexander-alderman-webb in #6314

• (huey) Fix group and chord handling in enqueue by @ericapisani in #6392

• (integrations) Auto-wrap root gen_ai spans for openai, cohere, langgraph, huggingface_hub by @constantinius in #6285

• (serializer) Don't call __iter__ on arbitrary sequences by @sentrivana in #6304

  Previously, we'd attempt to serialize any Sequence by walking through it by calling its __iter__ function.
  We've now changed the serializer to only serialize built-in sequences (like lists, tuples, and sets) to avoid
  triggering side-effects from custom __iter__ implementations.

  This might mean some objects might be serialized differently. If you want to continue serializing a specific
  custom sequence class the old way, you can register it via sentry_sdk.serializer.add_repr_sequence_type (see
  here).

• Memory leak in SentrySpanProcessor by @volodkindv in #6271

Documentation 📚

• (celery) Remove duplicated "is" in beat.py docstring by @quyentonndbs in #6266

Internal Changes 🔧

Langchain

• Deduplicate by removing node.callspec.id matching by @alexander-alderman-webb in #6426
• Remove WatchedSpan class by @alexander-alderman-webb in #6407

Openai Agents

• Deduplicate by removing node.callspec.id matching by @alexander-alderman-webb in #6424
• Fix asyncio loop missing for sync tests by @sl0thentr0py in #6412

Pydantic Ai

• Support Agent.run_stream_events() returning a context manager by @alexander-alderman-webb in #6322
• Remove test without assertions by @alexander-alderman-webb in #6321

Other

• (openai) Deduplicate by removing node.callspec.id matching by @alexander-alderman-webb in #6427
• Respect context manager lifecycles in fake_record_sql_queries by @alexander-alderman-webb in #6295