fix(delete-user-data): kit/extension parity – kit silently skips deletions under load (concurrency 80 OOM) #3250

delete-user-data kit/extension parity. Priority: critical. User data is left undeleted, and no error is reported.

Difference

The kit's functions use the 2nd gen defaults, so one instance takes up to 80 events at once. The extension's 1st gen functions take one event per instance. When auto-discovery fans out, many handleSearch events land on the same kit instance and it goes over 256 MiB. Every event in flight on that instance is then lost: retries are off, and Cloud Run still logs the requests as 200.

Reproduced

Extension and kit deployed side by side with the same config: auto-discovery depth 1, 20 root collections × 2,500 docs, and one user-owned doc in each collection. Then one user was deleted.

• Extension: 20/20 user docs deleted.
• Kit: 1/20 deleted. Logs show Memory limit of 256 MiB exceeded and 18 × The request failed because ... connection to the instance had an error.
• Kit redeployed with concurrency: 1, cpu: "gcf_gen1": 20/20 deleted, no errors.

Code

• Extension: 1st gen resources with no runtime overrides (extension.yaml#L59-L86); triggers at index.ts#L59, #L117, #L203.
• Kit: no runtime options on clearData, handleSearch or handleDeletion (index.ts#L112-L135).

Resolution

Declare concurrency: 1 and cpu: "gcf_gen1" on all three functions, as #3209 (firestore-bigquery-export) and #3233 (firestore-send-email) do. #3206 already sets concurrency: 1 (and internal-only ingress) for this kit, but it is open with conflicts. Either land the delete-user-data part on its own with cpu added, or add cpu to #3206.

Intentional difference

Ingress is always ALLOW_INTERNAL_ONLY, the extension's default. The extension's install-time ingress setting, which ext:migrate carries over with setGlobalOptions, doesn't apply to the kit. Recorded under Intentional Differences in #3249.