fix(#804): OTLP endpoint path, CLI retry, and flush warning

[fullsend-ai-coder]

Summary

Fixes three related regressions in the OTLP export path introduced by the OTel SDK migration (fullsend-ai#4510), each verified with the issue's live reproductions against 3c57dda9:

• Endpoint path handling: resolveEndpoint() now appends /v1/traces to the generic OTEL_EXPORTER_OTLP_ENDPOINT per the OTLP spec; the signal-specific OTEL_EXPORTER_OTLP_TRACES_ENDPOINT is used verbatim. Previously both were passed identically to WithEndpointURL, which overrides the SDK's path-join, causing path-prefixed collector endpoints (e.g. /otlp) to 404.
• CLI retry config: Exporter uses CLI-appropriate retry timing (500 ms initial, 2 s max, 4 s elapsed) instead of SDK defaults (5 s initial, 30 s max, 60 s elapsed) which cannot complete a single retry within the 5 s flush budget.
• Flush warning: tp.Shutdown errors are surfaced to stderr (fullsend: OTLP flush incomplete: ...) instead of being silently discarded, so operators can distinguish failed exports from successful ones.

Changes

• internal/telemetry/telemetry.go: Replaced endpointFromEnv() with resolveEndpoint() that differentiates generic vs signal-specific endpoint vars per OTLP spec. Added cliRetry config with short intervals. Surfaced tp.Shutdown errors to stderr.
• internal/telemetry/telemetry_test.go: Updated TestSetup_OTLPExporterSeam and TestSetup_TracesEndpointPreferred to verify endpoint argument values. Added TestResolveEndpoint_* unit tests covering generic append, signal-specific verbatim, and empty cases.
• internal/telemetry/wire_test.go (new): Wire-level integration tests with an in-process OTLP HTTP protobuf sink covering endpoint path construction (generic with path prefix, bare host:port, signal-specific verbatim), retry-after-503 delivery, header injection, and retryable-failure completion within budget.

Testing

• All telemetry package tests pass with -race and 94.2% coverage
• go vet passes
• Wire-level tests exercise real HTTP exporter against in-process OTLP sink
• golangci-lint not available in sandbox; CI will run it

---

Closes #804

Post-script verification

• Branch is not main/master (agent/804-otel-endpoint-flush-tests)
• Secret scan passed (gitleaks — ba77dbdbbd17e9bb18c47bf6efa023c98b697158..HEAD)
• PR body secret scan passed (gitleaks — no-git)
• Pre-commit hooks passed (authoritative run on runner)
• Tests ran inside sandbox

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 3:08 AM UTC · Completed 3:20 AM UTC
Commit: 8920bc0 · View workflow run →

[fullsend-ai-review]

Review

Findings

Low

• [naming-convention] internal/telemetry/telemetry.go:87 — Function renamed from endpointFromEnv to resolveEndpoint. The codebase has both *FromEnv and resolve* patterns for functions of similar complexity. The rename is defensible — the function now performs URL parsing and path construction beyond simple env-var reading, and the rationale is documented inline. protocolFromEnv retains *FromEnv because it only reads env vars, which is a reasonable distinction.

Previous run

Review

Findings

Low

• [naming-consistency] internal/telemetry/telemetry.go:100 — The rename from endpointFromEnv to resolveEndpoint breaks the established *FromEnv naming pattern used by sibling functions (protocolFromEnv). The new name is defensible since the function now does URL parsing and path appending beyond simple env-var reading, but protocolFromEnv remains as a naming divergence.
• [error-handling-idiom] internal/telemetry/telemetry.go:84 — The cleanup function logs tp.Shutdown errors to stderr but silently discards f.Close() errors on the next line (_ = f.Close()). There is a reasonable justification — shutdown failure means trace data loss (user-visible), while file close on an append-mode file that has already been flushed is rarely actionable — but a brief inline comment would clarify the intent.
• [missing-operational-detail] docs/guides/infrastructure/distributed-tracing.md:64 — The new OTLP flush incomplete stderr warning and CLI retry policy (500 ms initial, 2 s max, 4 s elapsed cap) are not mentioned in the operational details section. Follow-up suggestion — the docs file is not in this PR's changed set, and the existing docs already partially cover the fail-open contract.

---

Labels: PR fixes a bug in OTLP telemetry export (endpoint path, retry, flush warning) per issue #804

[guyoron1]

/fs-fix

[fullsend-ai-coder]

🤖 Finished Fix · ✅ Success · Started 6:48 AM UTC · Completed 6:54 AM UTC
Commit: 8920bc0 · View workflow run →

[fullsend-ai-coder]

🔧 Fix agent — iteration 1 (human-triggered)

Addressed 2 of 3 review findings with documentation comments. Finding 1 (naming-consistency): added godoc explaining the intentional *FromEnv naming divergence for resolveEndpoint. Finding 2 (error-handling-idiom): added inline comment on f.Close() explaining why the error is discarded. Finding 3 (missing-operational-detail): disagreed — docs file is not in PR scope, and reviewer flagged it as a follow-up suggestion.

Fixed (2):

1. *naming-consistency: resolveEndpoint breaks FromEnv naming pattern (internal/telemetry/telemetry.go): Added godoc comment to resolveEndpoint explaining why it diverges from the *FromEnv naming pattern — it performs URL parsing and path construction beyond simple env-var reading, whereas protocolFromEnv retains the suffix since it only reads env vars.
2. error-handling-idiom: f.Close() error silently discarded (internal/telemetry/telemetry.go): Added inline comment on the _ = f.Close() line explaining the intent: the file is already flushed by SimpleSpanProcessor and close errors are not actionable.

Disagreed (1):

1. missing-operational-detail: docs not updated with flush warning and retry policy: The docs file (docs/guides/infrastructure/distributed-tracing.md) is not in this PR's changed set. The reviewer explicitly noted this is a follow-up suggestion. Adding docs changes would expand the PR scope beyond the bug fixes it addresses.

Tests: passed

Next steps:

• /fs-review — request a re-review of the changes
• /fs-fix <your instruction> — run another fix pass with specific guidance
• Push commits directly — review re-runs automatically on push
  _{Updated by fullsend fix agent}

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 6:59 AM UTC · Completed 7:13 AM UTC
Commit: 5e5602a · View workflow run →

[fullsend-ai-review]

Note: The following inline comments could not be posted on the diff (GitHub returned 422) and are included here instead:

• internal/telemetry/telemetry.go (file-level): Line 87 · [low] naming-convention

Function renamed from 'endpointFromEnv' to 'resolveEndpoint'. The codebase has both FromEnv and resolve patterns for functions of similar complexity. The rename is defensible — the function now performs URL parsing and path construction beyond simple env-var reading, and the rationale is documented inline.

[github-actions]

This pull request has been automatically marked as stale because it has not had any activity in the last month. It will be closed in 2 weeks if no further activity occurs. Remove the stale label to reset the inactivity timer.