[SVLS-8800] fix: hold back cold-start invocation metric until durable tag is known by lym953 · Pull Request #1301 · DataDog/datadog-lambda-extension

lym953 · 2026-07-11T00:57:33Z

Problem

For durable functions, the aws.lambda.enhanced.invocations metric for the first invocation (after cold start) doesn't have the tag durable_function:true as expected.

This is because the metric is emitted in on_invoke_event(), which, from what we observed, always happens before the extension receives platform.initStart event, which carries the information of whether the function is a durable function. As a result, the durable_function:true tag is only set on the metric for subsequent invocations but not for the first one.

Fix

Delay the flushing of aws.lambda.enhanced.invocations metric for the first invocation, if the extension doesn't yet know whether the function is a durable function.

Use an enum to track the state of the first invocation:

enum FirstInvocationMetricStatus {
    Pending,
    InitStartSeen,
    HeldBack(i64),
    Resolved,
}

This covers both cases about the order of on_invoke_event() event and platform.initStart event:

Common case: on_invoke_event() comes first, then platform.initStart:
- To start with: Pending
- on_invoke_event() comes: hold metric and move to HeldBack
- platform.initStart comes: flush metric and move to Resolved
Rare case: platform.initStart comes before on_invoke_event()
- To start with: Pending
- platform.initStart comes: InitStartSeen
- on_invoke_event() comes: flush metric immediately and move to HeldBack

Also, flush the metric at shutdown, in case platform.initStart is never received for whatever reason.

Testing

Manual testing

For a durable function, all data points for the aws.lambda.enhanced.invocations metric have the tag durable_function:true set.

Automated testing

Passed the added integration test.

…g is known In On-Demand mode, the Extensions API's `/next` poll (driving on_invoke_event) consistently beats the buffered Telemetry API's platform.initStart (which sets the durable_function tag), so the first invocation's aws.lambda.enhanced.invocations metric almost always missed durable_function:true. Hold the sandbox's first invocation metric in a pending field and flush it once platform.initStart has set the durable tag, with fallbacks on the next invoke or shutdown so the metric is never dropped if initStart never arrives.

datadog-datadog-prod-us1-2 · 2026-07-11T00:59:18Z

Tests

🔄 Datadog auto-retried 3 jobs - 2 passed on retry

_{This comment will be updated automatically if new data arrives.

🔗 Commit SHA: cc19dee | Docs | Datadog PR Page | Give us feedback!}

Deploys a real durable-configured Lambda alongside a plain guard function and asserts the cold-start aws.lambda.enhanced.invocations metric carries durable_function:true only for the durable one, exercising the fix in a905347 end-to-end. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

…field platform.initStart not arriving before the next invocation is not a realistic scenario, so drop that fallback flush and the test covering it. Renamed pending_first_invocation_metric to pending_first_invocation_metric_timestamp to reflect that it stores a timestamp, and clarified "cold-start invocation metric in On-Demand mode" comments. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

If the Invoke event reaches the processor after platform.initStart (the reverse of the usual On-Demand race), on_invoke_event would still see an empty context_buffer and hold the metric back a second time, with nothing left to flush it until shutdown. Track whether platform.initStart already ran and skip the hold-back in that case. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

…already ran" This reverts commit 35ee072.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Invoke both functions a second time after the cold start and check the tag in a separate time window, to confirm the tag isn't only set on invocation #1 but persists across warm invocations. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Replaces the two-window/two-5-min-indexing-wait design with one window and one wait, comparing tagged vs. total invocation counts (new getMetricCountWithTag helper) instead of checking existence per window. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

…e checks Each function is invoked exactly twice, so assert toBe(2) instead of a loose lower bound. Also add a guideline to AGENTS.md against unnecessary defensive checks/tests. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Copilot

Pull request overview

This PR addresses a tagging gap in Bottlecap’s enhanced invocation metric for Durable Functions by delaying the first aws.lambda.enhanced.invocations emission until platform.initStart has been processed (so the durable_function:true tag can be applied), with a shutdown fallback. It also adds an integration test and the associated CDK stack to validate the tag is present for all durable invocations and absent for non-durable ones.

Changes:

Hold back the first On-Demand invocation metric until platform.initStart (or flush at shutdown as fallback).
Add an integration test plus a new CDK stack deploying durable vs non-durable Lambdas to validate tagging.
Extend Datadog test utilities with a helper to query metric counts filtered by a tag.

Reviewed changes

Copilot reviewed 6 out of 7 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
`bottlecap/src/lifecycle/invocation/processor.rs`	Holds back the first invocation metric and flushes it on `platform.initStart` or shutdown; adds a unit test for the race.
`integration-tests/tests/utils/datadog.ts`	Adds `getMetricCountWithTag` helper for tag-filtered metric count queries.
`integration-tests/tests/durable-cold-start.test.ts`	New integration test validating `durable_function:true` tagging behavior across cold + warm invocations.
`integration-tests/lib/stacks/durable-cold-start.ts`	New CDK stack deploying durable and non-durable functions for the integration test.
`integration-tests/bin/app.ts`	Registers the new `DurableColdStart` stack in the integration test app.
`.gitlab/datasources/test-suites.yaml`	Adds the `durable-cold-start` suite to CI test suite definitions.
`AGENTS.md`	Adds a small code-style guideline note for agents.

jchrostek-dd · 2026-07-14T12:51:18Z

Do you think we should add a comment to agents.md about the ordering of lambda events so agents are aware and proactively catch issues before they are bugs?

jchrostek-dd · 2026-07-14T12:57:46Z

+    /// `platform.initStart` sets the durable tag so it's not missed on invocation #1.
+    ///
+    /// This assumes `platform.initStart` always arrives after the Invoke event, which is what
+    /// we've observed in production.


Is this actually a guarantee? That platform.initStart must appear after the first invoke event? Or is is just likely to appear after the first invoke event?

It's not a guarantee. This is just the order I consistently observed from my tests, and I didn't want to over-engineer.
Since you raised this, let me make code robust by handling both cases.

…er emptiness context_buffer.is_empty() is also true between warm invocations once platform.report is processed (an unordered, separate path from on_invoke_event), so it could wrongly re-trigger the cold-start hold-back on later invocations and drop their metric. Track it explicitly instead. Also fix a test that recomputed `now` after on_invoke_event to derive the expected metric bucket, which could mismatch the actual timestamp if the call straddled a 10s boundary; capture the held-back timestamp instead. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

…dy processed Handles both orderings of the Invoke event and platform.initStart. If initStart has already resolved the durable tag by the time invocation #1 arrives, there's nothing to wait for, so emit the metric right away instead of holding it back (which previously would only be flushed at shutdown, since platform.initStart doesn't fire again). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Replaces pending_first_invocation_metric_timestamp, first_on_demand_invocation_seen, and platform_init_start_processed with one FirstInvocationMetric enum (Pending/InitStartSeen/HeldBack/Resolved), making the state machine explicit instead of spread across three separate fields. Also drops the struct_excessive_bools allow, since the struct is back to 3 bool fields. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

…tatus Field/enum name now reflects that it tracks a status, not the metric value itself. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

…omment Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

lym953 · 2026-07-14T18:49:56Z

Oh no, I see duplicate logs in Datadog. Let me fix it.

lym953 · 2026-07-15T04:23:54Z

Oh no, I see duplicate logs in Datadog. Let me fix it.

That's unrelated to this PR. Marking the PR as Ready for Review again.

lym953 and others added 10 commits July 13, 2026 15:31

test: shorten comments in durable cold-start integration test

ad2f801

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

refactor: shorten comments around pending_first_invocation_metric

1e22255

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Revert "fix(metrics): don't hold back invocation metric if initStart …

d39d15e

…already ran" This reverts commit 35ee072.

docs: note that platform.initStart is assumed to follow Invoke

418cf99

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

lym953 changed the title ~~fix(metrics): hold back cold-start invocation metric until durable tag is known~~ [SVLS-8800] fix: hold back cold-start invocation metric until durable tag is known Jul 13, 2026

lym953 requested a review from Copilot July 13, 2026 20:44

lym953 marked this pull request as ready for review July 13, 2026 20:44

lym953 requested a review from a team as a code owner July 13, 2026 20:44

lym953 requested a review from jchrostek-dd July 13, 2026 20:44

Copilot started reviewing on behalf of lym953 July 13, 2026 20:44 View session

Copilot AI reviewed Jul 13, 2026

View reviewed changes

jchrostek-dd reviewed Jul 14, 2026

View reviewed changes

lym953 and others added 6 commits July 14, 2026 13:16

refactor: rename first_invocation_metric to first_invocation_metric_s…

decffd1

…tatus Field/enum name now reflects that it tracks a status, not the metric value itself. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

docs: trim historical rationale from first_invocation_metric_status c…

1636553

…omment Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

docs: add comment conciseness guideline to AGENTS.md

cc19dee

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

lym953 marked this pull request as draft July 14, 2026 18:50

lym953 marked this pull request as ready for review July 15, 2026 04:24

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[SVLS-8800] fix: hold back cold-start invocation metric until durable tag is known#1301

[SVLS-8800] fix: hold back cold-start invocation metric until durable tag is known#1301
lym953 wants to merge 17 commits into
mainfrom
yiming.luo/durable-metric-tag-cold-start

lym953 commented Jul 11, 2026 •

edited

Loading

Uh oh!

datadog-datadog-prod-us1-2 Bot commented Jul 11, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jchrostek-dd commented Jul 14, 2026

Uh oh!

jchrostek-dd Jul 14, 2026

Uh oh!

lym953 Jul 14, 2026

Uh oh!

lym953 Jul 14, 2026

Uh oh!

lym953 commented Jul 14, 2026

Uh oh!

lym953 commented Jul 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

lym953 commented Jul 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Problem

Fix

Testing

Manual testing

Automated testing

Uh oh!

datadog-datadog-prod-us1-2 Bot commented Jul 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jchrostek-dd commented Jul 14, 2026

Uh oh!

jchrostek-dd Jul 14, 2026

Choose a reason for hiding this comment

Uh oh!

lym953 Jul 14, 2026

Choose a reason for hiding this comment

Uh oh!

lym953 Jul 14, 2026

Choose a reason for hiding this comment

Uh oh!

lym953 commented Jul 14, 2026

Uh oh!

lym953 commented Jul 15, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

lym953 commented Jul 11, 2026 •

edited

Loading

datadog-datadog-prod-us1-2 Bot commented Jul 11, 2026 •

edited

Loading