feat(traces): span export queue with retries

[turnipdabeets]

💡 Motivation and Context

Adds SpanExporter, the in-memory queue ended spans wait in until they are batched and sent, separate from the events queue.

• A timer flushes every flush_interval; a full batch flushes at once; only one flush runs at a time.
• A full queue drops the incoming span, never a queued parent.
• Failures back off exponentially with jitter, floored by Retry-After (clamped to 30 s, extended by a later deadline but never shortened). Automatic sends pause while backing off.
• A batch refused across 8 backoff windows is dropped. A 413 halves the batch and ramps back. Other 4xx drop the batch.
• flush(timeout) always sends the first batch, so a serverless handler with no budget left still ships spans.

Not reachable from the client yet.

Stack (PR 6 of 9, based on traces/05-pipeline):

1. feat(traces): W3C trace context ids and traceparent parsing #949 traces/01-ids-traceparent
2. feat(traces): OTLP span encoding and client-side validity #950 traces/02-otlp-encoding
3. feat(traces): span handles #951 traces/03-span-handles
4. feat(traces): span batch transport #952 traces/04-transport
5. feat(traces): span pipeline #953 traces/05-pipeline
6. feat(traces): span export queue with retries #954 traces/06-export ← this PR
7. feat(traces): per-span limits and exception stacktraces #955 traces/07-span-limits
8. feat(traces): before_span_send hook #956 traces/08-before-span-send
9. feat(traces): wire tracing into the client #957 traces/09-client-wiring

💚 How did you test it?

Unit tests in posthog/test/tracing/test_export.py cover timer and size flushes, backoff and Retry-After, 413 halving, drop rules and the bounded flush(timeout).

📝 Checklist

• I reviewed the submitted code.
• I added tests to verify the changes.
• I updated the docs if needed.
• No breaking change or entry added to the changelog.

If releasing new changes

• Ran sampo add to generate a changeset file

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

Implemented with Claude Code (Claude Opus 5) against the traces spec, one commit per slice so each PR reviews on its own. Rebased onto main and opened as a stacked draft in a later Claude Code session (Claude Fable 5.1).

🤖 Generated with Claude Code

https://claude.ai/code/session_012o7CtHLfcypjmXL7g9ZGRC

[github-actions]

posthog-python Compliance Report

Date: 2026-09-17 03:42:47 UTC
Duration: 256431ms

✅ All Tests Passed!

111/111 tests passed

---

Capture_V1 Tests

✅ 94/94 tests passed

View Details

Test	Status	Duration
Endpoint And Method.Targets V1 Endpoint	✅	518ms
Endpoint And Method.Does Not Use Legacy Endpoints	✅	511ms
Required Headers.Has Authorization Bearer Header	✅	510ms
Required Headers.Has Content Type Json	✅	511ms
Required Headers.Has Posthog Sdk Info Format	✅	510ms
Required Headers.Has Posthog Attempt Header	✅	510ms
Required Headers.Has Posthog Request Id	✅	510ms
Required Headers.Has Posthog Request Timestamp	✅	511ms
Required Headers.Has User Agent	✅	510ms
Body Format.Body Has Created At And Batch	✅	511ms
Body Format.No Api Key In Body	✅	511ms
Body Format.No Sent At In Body	✅	510ms
Event Format.Event Has Required Root Fields	✅	511ms
Event Format.Event Uuid Is Valid	✅	510ms
Event Format.Event Timestamp Is Rfc3339	✅	510ms
Event Format.Distinct Id Is String	✅	510ms
Event Format.Distinct Id At Root Not Properties	✅	510ms
Event Format.Custom Properties Preserved	✅	510ms
Event Format.Set Properties Preserved	✅	511ms
Event Format.Set Once Properties Preserved	✅	510ms
Event Format.Groups Properties Preserved	✅	511ms
Event Format.Sdk Generates Uuid If Not Provided	✅	510ms
Event Format.Event Has Required Root Fields Batch	✅	514ms
Event Format.Event Uuid Is Valid Batch	✅	514ms
Event Format.Event Timestamp Is Rfc3339 Batch	✅	513ms
Event Format.Distinct Id Is String Batch	✅	514ms
Event Format.Distinct Id At Root Not Properties Batch	✅	513ms
Event Format.Custom Properties Preserved Batch	✅	514ms
Event Format.Set Properties Preserved Batch	✅	514ms
Event Format.Set Once Properties Preserved Batch	✅	514ms
Event Format.Groups Properties Preserved Batch	✅	514ms
Event Format.Sdk Generates Uuid If Not Provided Batch	✅	514ms
Batch Behavior.Multiple Events In Single Batch	✅	518ms
Batch Behavior.Batch Envelope Smoke	✅	516ms
Batch Behavior.Flush With No Events Sends Nothing	✅	506ms
Batch Behavior.Flush At Triggers Batch	✅	1011ms
Batch Behavior.Created At Reflects Batch Creation Time	✅	511ms
Deduplication.Generates Unique Uuids	✅	518ms
Deduplication.Different Events Same Content Different Uuids	✅	513ms
Deduplication.Preserves Uuid On Retry	✅	6518ms
Deduplication.Preserves Timestamp On Retry	✅	6518ms
Deduplication.Preserves Uuid And Timestamp On Batch Retry	✅	6524ms
Deduplication.No Duplicate Events In Batch	✅	518ms
Header Behavior On Retry.Attempt Header Starts At One	✅	511ms
Header Behavior On Retry.Attempt Header Increments On Retry	✅	13523ms
Header Behavior On Retry.Request Id Preserved On Retry	✅	6520ms
Header Behavior On Retry.Different Requests Have Different Request Ids	✅	3020ms
Header Behavior On Retry.Request Timestamp Changes On Retry	✅	6520ms
Response Format Validation.Success Response Has Uuid Keyed Results	✅	511ms
Response Format Validation.Success Response Has Ok For Each Event	✅	514ms
Response Format Validation.Success No Retry After When All Ok	✅	515ms
Response Format Validation.Success Retry After Present When Retry Events	✅	1516ms
Response Format Validation.Success No Retry After When Drop Only	✅	513ms
Response Format Validation.Response Echoes Request Id	✅	510ms
Retry Behavior.Retries On 408	✅	6519ms
Retry Behavior.Retries On 500	✅	6520ms
Retry Behavior.Retries On 503	✅	8523ms
Retry Behavior.Retries On 504	✅	6520ms
Retry Behavior.Retryable Errors Have Retry After	✅	3516ms
Retry Behavior.Respects Retry After On Retryable Error	✅	11521ms
Retry Behavior.Does Not Retry On 400	✅	2514ms
Retry Behavior.Does Not Retry On 401	✅	2514ms
Retry Behavior.Does Not Retry On 402	✅	2514ms
Retry Behavior.Does Not Retry On 413	✅	2514ms
Retry Behavior.Does Not Retry On 415	✅	2513ms
Retry Behavior.Non Retryable Errors Have No Retry After	✅	2514ms
Retry Behavior.Implements Backoff	✅	22526ms
Retry Behavior.Max Retries Respected	✅	22536ms
Partial Batch Handling.Handles 200 Full Success	✅	2513ms
Partial Batch Handling.Handles 200 With All Ok	✅	3516ms
Partial Batch Handling.Does Not Retry Dropped Events	✅	3516ms
Partial Batch Handling.Does Not Retry Limited Events	✅	3514ms
Partial Batch Handling.Prunes Ok Events On Partial Retry	✅	6522ms
Partial Batch Handling.Prunes Dropped Events On Partial Retry	✅	6521ms
Partial Batch Handling.Retries Only Retry Events From Partial	✅	6523ms
Partial Batch Handling.Partial Retry Preserves Uuids	✅	6518ms
Partial Batch Handling.Partial Retry Attempt Header Increments	✅	6522ms
Partial Batch Handling.Partial Retry Request Id Preserved	✅	6522ms
Partial Batch Handling.Respects Retry After On Partial	✅	8522ms
Partial Batch Handling.Unknown Result Treated As Terminal	✅	3515ms
Partial Batch Handling.Mixed Ok Drop Limited No Retry	✅	3517ms
Compression.Sends Gzip Content Encoding	✅	512ms
Compression.No Content Encoding When Disabled	✅	510ms
Compression.Compressed Body Is Decompressible	✅	511ms
Error Handling.Does Not Retry On Unknown 4Xx	✅	2513ms
Event Options.Cookieless Mode Override	✅	510ms
Event Options.Disable Skew Correction Override	✅	511ms
Event Options.Process Person Profile Override	✅	510ms
Event Options.Product Tour Id Override	✅	511ms
Event Options.Unset Options Omitted	✅	510ms
Event Options.Options Override In Batch	✅	515ms
Geoip And Historical Migration.Geoip Disable Injected Into Properties	✅	510ms
Geoip And Historical Migration.Historical Migration Set In Body	✅	510ms
Geoip And Historical Migration.Historical Migration Absent By Default	✅	510ms

Feature_Flags Tests

✅ 17/17 tests passed

View Details

Test	Status	Duration
Request Payload.Request With Person Properties Device Id	✅	11ms
Request Payload.Flags Request Uses V2 Query Param	✅	9ms
Request Payload.Flags Request Hits Flags Path Not Decide	✅	10ms
Request Payload.Flags Request Omits Authorization Header	✅	9ms
Request Payload.Token In Flags Body Matches Init	✅	9ms
Request Payload.Groups Round Trip	✅	10ms
Request Payload.Groups Default To Empty Object	✅	9ms
Request Payload.Disable Geoip False Propagates As Geoip Disable False	✅	9ms
Request Payload.Disable Geoip Omitted Defaults To False	✅	9ms
Request Payload.Flag Keys To Evaluate Contains Only Requested Key	✅	10ms
Request Lifecycle.No Flags Request On Init Alone	✅	3ms
Request Lifecycle.No Flags Request On Normal Capture	✅	510ms
Request Lifecycle.Two Flag Calls Produce Two Remote Requests	✅	14ms
Request Lifecycle.Mock Response Value Is Returned To Caller	✅	9ms
Retry Behavior.Retries Flags On 502	✅	313ms
Retry Behavior.Retries Flags On 504	✅	313ms
Side Effect Events.Get Feature Flag Captures Feature Flag Called Event	✅	512ms

[greptile-apps]

Important Files Changed

Filename	Overview
posthog/tracing/_export.py	Adds the span queue, batching, retries, and timers. The timer replacement can lose a zero-delay flush.
posthog/test/tracing/test_export.py	Adds broad tests for batching, retries, drops, deadlines, and timers. The fake timer does not cover a callback that runs during start().
posthog/test/tracing/helpers.py	Adds tracing test helpers and timer fakes.

Prompt To Fix All With AI

### Issue 1
posthog/tracing/_export.py:449-452
`timer.start()` can run the callback before `_flush_timer` points to the new timer. A zero-delay flush can then look stale and return. The old timer is canceled, and the dead timer is stored. Publish the replacement before starting it, and restore the old timer if `start()` fails.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

_{Reviews (2): Last reviewed commit: "fix(traces): keep the scheduled flush wh..." | Re-trigger Greptile}

[dustinbyrne]

Looks good for the export-queue slice, with one non-blocking note: _export.py:182 calls warn_if_due(force=True) after every flush, including automatic flushes. Repeated rejected batches can therefore emit a warning per flush rather than at most once per configured interval. Consider reserving forced reporting for shutdown/exit and updating the fixed-clock test that currently expects two warnings in one interval.

The transport and pipeline feedback belongs to their respective stack PRs; this approval does not cover the complete public tracing release.

AI-assisted review using source, tests and existing CI; no new tests were run.

[turnipdabeets]

Thanks. We warn after every flush on purpose, to match Node. It calls _warnAboutDrops() at the end of every flush and has a test called "warns once per flush with the total". The once-per-interval rule is for drops that happen outside a flush, like a full queue during an outage.

[dustinbyrne]

Thanks, confirmed against Node: the flush path calls _warnAboutDrops(), with an explicit once-per-flush test. I withdraw my warning-cadence suggestion. The spec wording should be reconciled separately, not by changing Python alone. Approval stands.

AI-assisted follow-up source review, including Node parity and the rebased transport change; no new tests run locally.

[jzhu13]

Reviewed against traces/05-pipeline. Tests pass at the head. Lock ordering is consistent, _send runs with no lock held, the 413 halving cannot reach zero or loop, and the Retry-After window handles units, clamp, and negatives correctly. Two items I would fix before merge; both surface as user-visible data loss once #957 wires this in.

Blocking

1. posthog/tracing/_export.py:348 a flush(timeout) with budget remaining never retries a retry-later batch: _apply_outcome_locked returns stop=True, _drain propagates it, flush skips the follow-up pass, and nothing sleeps. Reproduced: one 503 then ok; flush(30.0) made one attempt in 0.000 s and left the queue intact. #957's shutdown then calls close() and discards the whole backlog with a "Discarding N span(s)" warning, while the events lane retries up to ten times with sleeps. Suggest: when timeout is not None and the pass ended in retry-later, time.sleep(min(self._next_flush_delay_locked(), remaining)) and loop _drain while budget remains, charging normally. Timer-driven flushes (_timer is not None) keep the current no-sleep behavior.
2. posthog/tracing/_export.py:155 the deadline is computed before _flush_lock.acquire(timeout=...), and acquire failure returns silently with zero requests and no log. That contradicts the docstring's "no request starts once it is spent, except the first", which #957 repeats to users in Client.flush. Reproduced: a timer flush parked in _send, flush(0.2) returned at 0.20 s with 0 send attempts and 2 spans still queued. Either start the budget after the lock is acquired, or narrow the contract and debug-log the give-up path.

Non-blocking

3. posthog/tracing/_export.py:317 a server 413 caused by one oversized span collapses _max_export_batch_size to 1 and the +1 ramp keeps it small for a long time. Reproduced: 1 giant + 511 small spans at batch 512 took 42 requests (sizes 512, 256, ..., 1, 1, 2, 3, ...) where 2 would do, and left the size at 33 so the depth trigger fired every ~35 spans afterwards. Restore the pre-halving size once the size-1 413 isolates the culprit, and ramp multiplicatively.
4. posthog/tracing/_export.py:350 dropping a head batch after 8 windows resets _consecutive_failures, so the exporter immediately sends the next batch to an endpoint it just saw fail 8 times and re-enables the depth trigger. Keep the count on a retry-later drop.
5. posthog/tracing/_export.py:29 duplicates the 30 s ceiling that capture_v1._MAX_BACKOFF_SECONDS defines and AGENTS.md calls "the single ceiling for both the exponential backoff and the Retry-After clamp". Import it.
6. Nits: ten fields track failure and timer state where one _next_attempt_at deadline would replace _jitter and _head_batch_chargeable_at; FakeTimer.fire() ignores cancelled, so a test can pass by firing a timer the code already cancelled; shared fixtures are imported from helpers.py and re-exported via __all__ rather than a conftest.py.

Reviewed with Claude Code (Claude Fable 5.1). Behaviors above were reproduced against this branch head.

[veria-ai]

PR overview

All previously flagged issues have been addressed. No open security concerns remain on this pull request.

Security review

No open security issues remain on this pull request.

Fixed/addressed: 1 · PR risk: 0/10

[turnipdabeets]

Thanks. Done in a451080: item 4 (a dropped batch keeps the backoff), item 5 (ceiling imported from capture_v1) and the FakeTimer nit.
Items 1 to 3 (flush retry within budget, budget after the lock, batch size after a 413) are still open. I'd like to do them together in one change.

[turnipdabeets]

Follow-up, 79869b9:

• 1: a caller-driven timed flush waits out the backoff and retries while budget remains, with one last attempt at the deadline. Timer flushes and untimed flushes are unchanged. close() wakes a waiting flush.
• 2: the budget starts once the flush lock is held, and a flush that never gets it says so at debug. Docstring updated.
• 3: a size-1 413 restores the batch size it halved from, and the ramp doubles instead of adding one.
  Ready for another look.