Cold start latency regression: pre-release and first 30 minutes checklists

Serverless or container cold starts become slower after a dependency or image change.

The pre-release checklist maps each item to a test class from the incident page. The first-30-minutes checklist covers detection, containment, the rollback decision and customer communication.

7 pre-release items · 7 first-30-minutes items · verified 2026-10-22

Pre-release checklist

PR-01Measure P90 cold start duration over 100 invokes of a minimal function on the new runtime version and on the current one before switching the runtime. load
PR-02Run the cold start benchmark at the lowest memory and CPU setting the serverless functions use in production, not with container benchmark settings. load
PR-03Measure function init duration with and without the SDK client import on the new runtime version and fail the check when the SDK's share exceeds its share on the current version. load
PR-04Run the init benchmark with a minimal repro that imports only the runtime's http request module, to separate runtime import cost from SDK code. load
PR-05Benchmark preview releases of the next runtime major version in the serverless cold start scenario before the general release. load
PR-06Report the init duration of canary cold starts as a separate metric from warm invoke duration. canary
PR-07Halt the runtime switch when the canary's P90 init duration exceeds the baseline's by more than a set number of milliseconds. canary

First 30 minutes

FM-01Compare init duration of cold start invokes before and after the runtime or SDK update, with warm invokes excluded.
FM-02Deploy a copy of an affected function without the SDK import and compare its init duration, to split runtime cost from SDK cost.
FM-03Move the affected functions back to the previous runtime version if their cold start duration breaks the latency budget.
FM-04Check whether the SDK's http handler has a release that loads the http module lazily, and confirm the installed SDK clients are compatible with it.
FM-05Record the runtime version, SDK version and memory setting of every affected function in the incident log.
FM-06Tell owners of latency-sensitive services which runtime version raised cold start duration and by how many milliseconds.
FM-07Open an issue with the runtime or SDK maintainers that includes the minimal repro and the before-and-after init durations.

Download the checklist

Pre-release checklist (Markdown)1.6 KB · v1.0.0
First 30 minutes checklist (Markdown)1.2 KB · v1.0.0

Only in this checklist

Every test class in this checklist also appears in at least one sibling checklist in the group; the item texts differ: Measure P90 cold start duration over 100 invokes of a minimal function on the new runtime version and on the current one before switching the runtime.

Questions

What should be checked before a release to prevent Cold start latency regression?

7 pre-release items, PR-01 to PR-07. PR-01: Measure P90 cold start duration over 100 invokes of a minimal function on the new runtime version and on the current one before switching the runtime.

Is there a rollback plan for Cold start latency regression?

The first-30-minutes checklist has 7 items. FM-01: Compare init duration of cold start invokes before and after the runtime or SDK update, with warm invokes excluded.

What should a team do in the first 30 minutes after Cold start latency regression is detected?

The first-30-minutes checklist has 7 items. FM-01: Compare init duration of cold start invokes before and after the runtime or SDK update, with warm invokes excluded.

Is a Cold start latency regression checklist still needed if deployment is automated?

Yes. Automation runs the steps; the checklist names what the tests must cover: load and performance tests and canary and progressive delivery checks.

Which pipeline stages does the Cold start latency regression checklist cover?

The pre-release items map to the pre-deploy and post-deploy stages.

In what formats can the Cold start latency regression checklist be downloaded?

Both checklists download as Markdown files: /downloads/cold-start-latency-regression-pre-release.md and /downloads/cold-start-latency-regression-first-30-minutes.md, licensed CC BY 4.0.