Monitor Mr

Monitor a Map/Reduce run

Companion to run-poller and reprocess-transaction. Those skills trigger; this one answers "did it work?" without anyone opening NetSuite.

The underlying tables, column contracts, log-title catalog per flow, and gotchas live in reference/mapreduce-monitoring.md. Read it before improvising queries.

When to use

• Right after run-poller or reprocess-transaction — these return fast while the MR runs async; this skill is the follow-through.
• The user asks whether a poll/reprocess/outbound run worked, why an Orderful Transaction is still Pending/Error, or what an MR script logged.
• Before re-triggering anything: confirm nothing is already QUEUED/PROCESSING.

Inputs needed

1. Customer directory with a .env (same convention as every other skill). All queries are read-only SELECTs; --env production is acceptable for monitoring a prod account.
2. The flow being watched — one of inbound-polling, inbound-processing, reprocess, outbound-consolidation, outbound-sending, outbound-status (maps to the script id; see reference doc).
3. Whatever correlation handles exist: the taskId from run-poller's response, and/or the OT record id(s) in play.

The recipe

The cardinal rule: capture markers before triggering, then watch, then verify outputs. Status alone never proves success.

A. Watching an inbound poll (you have a taskId)

1. Trigger via run-poller; keep taskId from its response.
2. Immediately (the chained-task diff only sees tasks born during the watch): node monitor-mr.mjs <dir> watch --flow inbound-polling --task <taskId>
   • Blocks until the task is COMPLETE/FAILED (default timeout 600s), streaming new log rows.
   • Reports the MR SUMMARY (mapErrors/reduceErrors) and any new taskids — polling chains the processing + simplified MRs unconditionally within seconds of completing, even when it polled nothing.
3. Verify output: new OTs created since the poll — SELECT COUNT(*) FROM customrecord_orderful_transaction WHERE created >= <t0> or ot mode on specific ids.
4. Set expectations on the chained processing run: it fires seconds after polling and normally picks up freshly-polled OTs — the nominal 10-minute freshness gate is near-vacuous, biting only ~00:00–00:10 account time (mechanism: reference/mapreduce-monitoring.md flow 2). If an OT wasn't picked up, suspect the pending-query filters (status, direction, pending_transactions), or push it through now with reprocess-transaction (the single-id path bypasses the batch query entirely).

B. Watching a reprocess (no taskId is returned)

1. Before triggering: node monitor-mr.mjs <dir> logs --flow reprocess --tail 1 → note the printed marker.
2. Trigger via reprocess-transaction.
3. node monitor-mr.mjs <dir> watch --flow reprocess --after-id <marker> --ot <otId>
   • With no --task, completion is detected by the next MR SUMMARY row past the marker or by every watched OT's lastmodified advancing (OTS_UPDATED) — the latter is what saves you on deployments whose log level (ERROR) suppresses the SUMMARY entirely.
   • Look for the log line Beginning to process NetSuite Orderful Transaction internalid: <otId> to confirm the run picked up your record.
4. The OT status printed at the end is the authoritative outcome (Success / Error + error text / Stale).

Many transactions: trigger each reprocess (they serialize — concurrency is 1), then run a single watch with --ot id1,id2,... and a longer --timeout. Judge by OT statuses, not by counting tasks.

C. Inbound processing on its own schedule

node monitor-mr.mjs <dir> watch --flow inbound-processing --ot <ids> --timeout 1200 — no trigger, just wait for the next scheduled cycle's SUMMARY beacon and check the OTs. If an OT wasn't picked up, suspect the pending-query filters (status, direction, pending_transactions null), not the near-vacuous 10-minute freshness filter.

D. Outbound consolidation / sending

There is no sanctioned remote trigger for these MRs as of v1.22 (the canonical agent-write action inventory lives in script-execution-map.md §2 — and don't reach for the testHook RESTlet). Nudge outbound via product paths (flip custbody_orderful_ready_to_process_* to re-fire the UE — see outbound-dispatch.md) or run the deployment from the NS UI. Then monitor exactly as above: watch --flow outbound-consolidation (or outbound-sending) --ot <ids>. Outbound success = OT flips toward Success with orderful_id populated; follow up in Orderful with fetch-validations if it lands Error/INVALID.

Ad-hoc forensics (no live run)

• runs --since 2h — what queued/ran recently (all scripts; identify by corroborating with logs).
• logs --flow inbound-processing --tail 30 [--errors-only] — recent history, parsed SUMMARY rows included.
• ot --ot 123,456 — current state of specific Orderful Transactions.

Behaviour rules

• Never report success from task status alone. COMPLETE + mapErrors=3 is a failed run; COMPLETE + empty mapKeys processed nothing. Always read the SUMMARY and, when record ids are known, the OT statuses.
• Never re-trigger while a task is QUEUED or PROCESSING. Check runs --since 30m first; concurrency is 1 per deployment, so re-triggering only deepens the queue.
• A QUEUED verdict that persists for minutes usually means a scheduled instance is in flight — wait, don't escalate.
• "Task not visible yet" immediately after triggering is normal lag — poll again before concluding.
• A silent execution log is not a silent failure. Deployments at log level ERROR persist nothing for a clean run (not even the SUMMARY). Judge by task verdict + OT state, and say so explicitly when reporting.
• A stale lastmodified is not proof the MR skipped the record. If a record's map stage throws, the MR runs and errors without ever saving it — retry_count and lastmodified don't move (mechanism + live evidence: mapreduce-monitoring's SUMMARY-row caveats). For a known-broken OT, trust the taskid verdict and the AUDIT SUMMARY beacon, not OT movement. A run that keeps OTS_UPDATED from firing on such a record is expected, not a tool bug.
• This skill is read-only by design. It never PATCHes records, never submits tasks, and is safe against production. Anything mutating (reprocess, flag flips) belongs to the trigger skills.
• Quote exact ERROR log rows (name + message) when reporting failures — don't paraphrase stack traces into vagueness.
• Timestamps from NetSuite are account-local; Orderful API timestamps are UTC. Reconcile before claiming latency numbers.

Reference material

• reference/mapreduce-monitoring.md — table contracts, per-flow script ids + log titles, verdict semantics, the 10-minute filter, all validated against a live account.
• reference/outbound-dispatch.md — why outbound rarely involves an MR at all.
• reference/record-types.md — OT record + join-table schema.