[camiekim]

Description

Fixes b/574556578

Checklist

Testing

• I have tested this change on a live environment and verified it works as intended.

Compliance & Style

• I have followed the Sample Guidelines.
• The README is updated with all relevant information (if applicable).
• If this PR adds a new sample directory: I updated the CODEOWNERS file.

---

Post-Approval Actions

• Please merge this PR for me once it is approved

[snippet-bot]

Here is the summary of changes.

You are about to add 1 region tag.

• cloud_scheduler/snippets/deduplicate_job_requests.py:19, tag cloudscheduler_deduplicate_job_requests

---

This comment is generated by snippet-bot.
If you find problems with this result, please file an issue at:
https://github.com/googleapis/repo-automation-bots/issues.
To update this comment, add snippet-bot:force-run label or use the checkbox below:

• Refresh this comment

[gemini-code-assist]

Code Review

This pull request introduces a Python snippet and corresponding unit tests demonstrating how to deduplicate retried Cloud Scheduler HTTP requests using Firestore. The feedback highlights a contradiction where manual runs (lacking the schedule time header) are rejected with a 400 error despite docstring claims, and suggests allowing them to run directly. Additionally, the reviewer recommends handling expired locks to prevent crashed jobs from blocking retries due to Firestore's asynchronous TTL deletion, along with adding unit tests to cover both scenarios.

[iennae]

Thanks @camiekim for this sample and for the docs page it supports. I've opened #14772 as a rewrite built on your work. You're credited as co-author, and it keeps the cloudscheduler_deduplicate_job_requests region tag, so your docs page shouldn't need changes.

The main changes:

• Lease for crashed attempts: If an instance crashes before deleting its IN_PROGRESS record, retries currently get a 409 until TTL cleanup, which can take more than 24 hours. The rewrite claims runs in a Firestore transaction with a lease, so an expired claim can be taken over without the get/set race you pointed out in review.
• Own directory: It's a functions-framework function in cloud_scheduler/deduplicate_job_requests/ with its own requirements, instead of adding Flask and Firestore to snippets/, which no longer uses Flask after feat(cloud_scheduler): modernize samples and cover documented job tasks #14767.
• Hashed key: The job name and schedule time are hashed into the document ID instead of validated and normalized. The header stays the same across retries.
• Tests without credentials: Unit tests use a fake Firestore, and the integration tests are opt-in.

I also tested what the headers look like on each target type, which might be useful for the page:

• HTTP targets get X-CloudScheduler-JobName, which is the job ID, and X-CloudScheduler-ScheduleTime, which is RFC 3339 with an offset and microseconds. The API reference documents the same headers for App Engine targets.
• Pub/Sub messages carry no run identifier, only the job's attributes, a message ID, and a publish time. Pub/Sub targets need different guidance, such as Pub/Sub exactly-once delivery.
• A forced run (gcloud scheduler jobs run) sends the next scheduled time. Header-based deduplication then skips that scheduled run.
• A groc-schedule job also sent ScheduleTime in testing, but the reference only promises it for unix-cron schedules.

If #14772 works for you, would you be open to closing this PR in favor of it? Feedback on the rewrite is very welcome.

[camiekim]

You are awesome, @iennae . I really appreciate you taking the time to rewrite the code properly. I will close this PR and pull in your new sample instead, and make the necessary updates in my CL (for example, state that the code uses a Firestore transaction with a lease to claim runs and recover from crashed attempts). Many thanks, again!