fix(debugger): retry agent /info on 5xx

[tylfin]

Description

Adds a private _info_with_retry helper in ddtrace/internal/agent.py that wraps the /info HTTP call with the same fibonacci backoff already used by the Remote Config client. AgentCheckPeriodicService._agent_check now calls the retrying helper directly, so a single transient 5xx from the local trace agent no longer flips subclasses (most visibly the Dynamic Instrumentation signal uploader) back to _agent_check state until the next periodic interval. Snapshots/diagnostics keep flushing and the instance stops looking "not heartbeating" in the UI.

The public agent.info() helper is intentionally left unchanged (single attempt, returns None and logs a warning on non-2xx). It is shared with Remote Config polling, CI Visibility, LLM Obs and the LLM Obs writer, which all have their own scheduling and timing expectations; adding implicit retries there would change those callers' behavior.

Context: observed on a single staging instance whose local agent intermittently returned 500 on /info. RC polling itself was healthy end-to-end; the DI uploader on that host was the blast radius.

Testing

• tests/tracer/test_agent.py::test_info_does_not_retry_on_5xx pins that info() keeps its single-attempt semantics.
• tests/tracer/test_agent.py::test_agent_check_retries_info_on_5xx exercises the retry through AgentCheckPeriodicService._agent_check (3 total attempts before giving up, info_check receives None).
• Existing test_info parametrization (incl. 500 case) still passes the contract: persistent 5xx returns None.
• Both new tests verified locally on py3.10.

Risks

Low. Worst case is up to ~0.5s of added latency on the periodic agent health check when the agent is persistently failing 5xx. Other callers of agent.info() (Remote Config, CI Visibility, LLM Obs) are unaffected because their code path is untouched.

Additional Notes

None.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9f02b954be

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[cit-pr-commenter-54b7da]

Codeowners resolved as

ddtrace/internal/agent.py                                               @DataDog/apm-core-python
releasenotes/notes/agent-info-retry-0a8acdda8d13a505.yaml               @DataDog/apm-python
tests/tracer/test_agent.py                                              @DataDog/apm-sdk-capabilities-python @DataDog/apm-core-python

[pr-commenter]

Benchmarks

Benchmark execution time: 2026-05-19 18:18:52

Comparing candidate commit db4e231 in PR branch tyler.finethy/agent-info-resilience with baseline commit 2166252 in branch main.

Found 0 performance improvements and 4 performance regressions! Performance is the same for 578 metrics, 10 unstable metrics.

scenario:iastaspects-lstrip_aspect

• 🟥 execution_time [+55.977µs; +61.418µs] or [+21.118%; +23.171%]

scenario:iastaspectsospath-ospathbasename_aspect

• 🟥 execution_time [+106.723µs; +113.994µs] or [+26.351%; +28.146%]

scenario:span-start

• 🟥 execution_time [+1.417ms; +1.574ms] or [+9.117%; +10.125%]

scenario:telemetryaddmetric-1-count-metric-1-times

• 🟥 execution_time [+267.777ns; +309.471ns] or [+12.999%; +15.023%]

[datadog-prod-us1-4]

🎉 All green!

🧪 All tests passed
❄️ No new flaky tests detected

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: db4e231 | Docs | Datadog PR Page | Give us feedback!}

[tylfin]

cc @P403n1x87 did we want to move forward with this or were you going to improve the resiliency alongside the UDS timeout?