Bound CHASM scheduler request IDs

[chaptersix]

Summary

Generate request IDs as sched-<auto|backfill-id>-<UUIDv5>. The UUID covers the namespace ID, schedule ID, conflict token, nominal time, and actual time.

This keeps generated IDs within the 255-byte SQL request-ID column while retaining the useful auto/backfill marker.

Validation

• go test -tags test_dep ./chasm/lib/scheduler/...
• make lint-code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 964f336470

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chaptersix]

Why this request-ID representation change is safe:

• A request ID is generated when a BufferedStart is created and is persisted with that entry. The invoker retries the stored value; it does not regenerate it.
• Existing buffered starts therefore retain their current request IDs. The UUIDv5 representation applies only to actions buffered after this change is deployed.
• UUIDv5 is deterministic: reprocessing the same logical action under the new code produces the same schedule-ID component, while the namespace, backfill ID, conflict token, and nominal/actual times continue to distinguish actions.
• Generator and backfiller state advance together with buffered starts, so a completed action is not normally generated again across a rollout. If a task is retried before that state is committed, deterministic request IDs preserve idempotency.

The change is consequently a bounded representation change for future scheduler-generated IDs, not a rewrite of existing buffered state or a source of duplicate starts.

[davidporter-id-au]

Not super sure I understand, wouldn't hashing the entire thing be the goal? What it if the NS is really long?

I am not sure if we need the requestID to be human-readable, but if we don't, then what about:

return uuid.NewSHA1(uuid.Nil, []byte(fmt.Sprintf(
		"sched-%s-%s-%s-%d-%d-%d",
		backfillID,
		namespaceID,
		scheduleID,
		conflictToken,
		nominal.UnixMilli(),
		actual.UnixMilli(),
	)))

[chaptersix]

I think the only places we'd see this is in tdbg execution dumps of the chasms schedule or workflows started by a schedule.

I'd also like to keep auto vs backfiller id since the boolean marking a buffered start as manual is not displayed in tdbg output when false. Having the backfiller is also help with identifying which backfiller a buffered start came from.

[davidporter-id-au]

if you want to be completionist, it might be worth capping the backfill ID as well at a max length so that user-defined inputs can't exceed the field size.