Lee-W opened a new pull request, #76999:
URL: https://github.com/apache/airflow/pull/76999
<!-- SPDX-License-Identifier: Apache-2.0
https://www.apache.org/licenses/LICENSE-2.0 -->
<!--
Thank you for contributing!
Please provide above a brief description of the changes made in this pull
request.
Write a good git commit message following this guide:
https://chris.beams.io/posts/git-commit/
Please make sure that your code changes are covered with tests.
And in case of new features or big changes remember to adjust the
documentation.
For user-facing UI changes, please attach before/after screenshots (or a
short
screen recording) so reviewers can assess the visual impact.
Feel free to ping (in general) for the review if you do not see reaction for
a few days
(72 Hours is the minimum reaction time you can expect from volunteers) - we
sometimes miss notifications.
In case of an existing issue, reference it using one of the following:
* closes: #ISSUE
* related: #ISSUE
-->
---
## Why
When every connection of a `fallback_conn_ids` chain fails, `LLMOperator`
and `AgentOperator` raise pydantic-ai's `FallbackExceptionGroup`. `RetryRule`
compares one exception type with `isinstance`, so a rule written for a
provider-specific exception such as `ModelHTTPError` stops matching once a
chain is configured. A rule set that retries rate limits but fails bad requests
collapses into one rule. The classifier prompt of `LLMRetryPolicy` and
`ClassifierRetryPolicy` sees only the group's message, not what each connection
raised. (This is the gap the previous docs in `retry_policies.rst` and
`recovery.rst` described.)
## What
-
`providers/common/ai/src/airflow/providers/common/ai/fallback_exceptions.py`
(new): adds `FallbackChainExhaustedError`, a subclass of
`FallbackExceptionGroup`. It exposes `member_errors` (each connection's error,
in chain order) and `replayed_steps` / `recorded_steps` (durable journal
progress of the failed attempt, `None` when the run was not durable). It lives
outside `common.ai.exceptions` so that module stays importable without
pydantic-ai.
- `.../mixins/cancellable_run.py`: `run_agent_sync` re-raises a
`FallbackExceptionGroup` as `FallbackChainExhaustedError`, `from` the original
group (kept as `__cause__`), filling the journal counts from the current
journal. An error that is already a `FallbackChainExhaustedError` is not
wrapped again.
- `.../policies/retry.py`:
- `fallback_rules` of `LLMRetryPolicy` and `ClassifierRetryPolicy` are
tried, rule by rule in the listed order, against the chain's own error and then
against the **last** connection's error. The last connection decides when
members disagree. A decision taken on it ends its reason with `(last chain
member)`.
- The classifier prompt lists each member as `chain member N of M: <Type>:
<message>`, redacted and limited to `max_exception_length` per member, so one
long message cannot push the last member out. With `durable=True` it also
states how many steps the attempt replayed and recorded. The prompt-building
code was split into a `_limit` helper to share the redact-then-truncate step.
- The `include_traceback=True` path adds no member lines (it already
formats the group).
##### Was generative AI tooling used to co-author this PR?
<!--
If generative AI tooling has been used in the process of authoring this PR,
please
change below checkbox to `[X]` followed by the name of the tool, uncomment
the "Generated-by".
-->
- [x] Yes (please specify the tool below)
Generated-by: [Claude] following [the
guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#gen-ai-assisted-contributions)
---
* Read the **[Pull Request
Guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#pull-request-guidelines)**
for more information. Note: commit author/co-author name and email in commits
become permanently public when merged.
* For fundamental code changes, an Airflow Improvement Proposal
([AIP](https://cwiki.apache.org/confluence/display/AIRFLOW/Airflow+Improvement+Proposals))
is needed.
* When adding dependency, check compliance with the [ASF 3rd Party License
Policy](https://www.apache.org/legal/resolved.html#category-x).
* For significant user-facing changes create newsfragment:
`{pr_number}.significant.rst`, in
[airflow-core/newsfragments](https://github.com/apache/airflow/tree/main/airflow-core/newsfragments).
You can add this file in a follow-up commit after the PR is created so you
know the PR number.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
