Slim repo-fleet-monitor: snapshot/SQL-first state fetch + 2h cadence (target ≤3M/run)

repo-fleet-monitor's function is load-bearing (queue↔GitHub reconciliation: claim/label divergence repair, dead-in-queue alerts, PR↔queue binding — the duty that ended the duplicate-PR class of failure). Its cost is not: 7.2M input per run, hourly (108M per 15h at check-in #5), because it re-derives state from GitHub lists and the queue helper every run.

Rebuild the fetch on the fleet-watchdog template (its 2026-10-28 rewrite cut state fetch to two SQL queries): precompute a compact state snapshot (queues, claims, labels, open PRs) deterministically, let the agent judge only the divergences, and move cadence 1h → 2h. No duty may be dropped. Target ≤3M input/run measured over the following three check-ins.

CODE-1032

Progress 2026-10-29 ~01:2x CDT: the cadence half is LIVE — repo-fleet-monitor now runs every 2h (definition edited in the git copy + posted as a replacement job; scheduler verified interval@2h, next run 03:12 CDT). Remaining on this issue: the snapshot/SQL-first state-fetch rewrite (the larger half of the saving, target ≤3M/run). Issue stays open for that work.

TOP PRIORITY (Drew, 2026-10-29 01:23 CDT): Batch 3 fleet consolidation — pick this up in the next runs, ahead of routine work. Being enqueued to repo-health-code-pilot at top priority now.

🤖 Agent note · picked up

Agent: queue worker (repo-worker-code), Lane B shift
Session / ID: queue task 3164
Found via: queue claim
Task: Triage Slim repo-fleet-monitor: snapshot/SQL-first state fetch + 2h cadence (target ≤3M/run) #2134 (Batch 3, Drew-approved): verify the fleet-monitor fetch-rewrite premise against live state, produce the snapshot/SQL-first design, route implementation in the same pass, update epic epic: fleet expansion & placement program (2026-10-28 directive, LIST 1–5) #2055. No live job changes in this triage.
Expectation: m — done looks like a verdict + fetch design on Slim repo-fleet-monitor: snapshot/SQL-first state fetch + 2h cadence (target ≤3M/run) #2134, a follow-on packet or linked implementation issue, and an epic epic: fleet expansion & placement program (2026-10-28 directive, LIST 1–5) #2055 update.
Links: epic epic: fleet expansion & placement program (2026-10-28 directive, LIST 1–5) #2055

🤖 Agent note · done

Agent: queue worker (repo-worker-code), Lane B shift
Session / ID: queue task 3164
Found via: queue claim
Task: Triage Slim repo-fleet-monitor: snapshot/SQL-first state fetch + 2h cadence (target ≤3M/run) #2134: premise verified against live state, fetch design decided below, implementation routed to Implement repo-fleet-monitor snapshot/SQL-first fetch (digest-judged runs, ≤3M input/run) #2154 (queue task pending in this same pass), epic epic: fleet expansion & placement program (2026-10-28 directive, LIST 1–5) #2055 updated. No live job changes made in this triage.
Expectation: m — verdict + design in this comment; the rewrite is Implement repo-fleet-monitor snapshot/SQL-first fetch (digest-judged runs, ≤3M input/run) #2154's.
Links: implementation Implement repo-fleet-monitor snapshot/SQL-first fetch (digest-judged runs, ≤3M input/run) #2154 · epic epic: fleet expansion & placement program (2026-10-28 directive, LIST 1–5) #2055

Triage verdict — premise CONFIRMED with one correction, design decided, implementation routed

Premise check (live state, 2026-10-29):

✅ Cadence half is already done: repo-fleet-monitor is enabled at 2h in the live scheduler — the 1h → 2h move in this issue's text already happened, so the remaining work is the fetch rewrite only.
✅ The cost mechanism is as described: the monitor re-derives state from GitHub lists + the queue helper each run (7.2M input/run at check-in Bump build/buck/preludes/buck2-prelude from 3a02ea3 to 370cd4d #5).
⚠️ Correction to the "rebuild from scratch" framing: snapshot prototypes already exist in the goal workspace (hidden_files/monitor-fetch.py, monitor-query.py — PostgREST pulls of live queue tasks), but they are not wired into the monitor's run, and their saved outputs (monitor-gh-snapshot.json, monitor-tasks-snapshot.json) are stale from 2026-10-28. The implementation extends and wires these prototypes; it does not start from zero, and it must never judge from those stale files.

Design (fetch rewrite)

Deterministic precompute at run start. A script extends the prototypes above and emits ONE compact digest: live tasks from all four pilot queues (one PostgREST query per queue, live statuses only), joined against a bounded GitHub pull — only issues carrying queue/*/work/* labels and PRs bound to live tasks. Divergences are pre-flagged mechanically (claimed-but-unlabelled, labelled-but-no-live-task, expired lease still claimed, expired park, PR merged/closed but task open, issue closed but task open, dead-in-queue at max_attempts).
The agent judges only the digest. Full lists stay out of context; a flagged item gets a targeted single-item fetch. Intake enrichment stays bounded/newest-first but draws candidates from the digest.
Freshness is binding (the watchdog's rule, applied here): the digest is generated inside the run; a snapshot older than the run is not evidence.
No duty dropped — duty → digest field mapping: reconciliation/auto-repair → divergence flags; dead-in-queue alerts → attempts flag; intake enrichment → unenriched-issue list; in-progress tracker → claimed-task/label join; protocol checks → work/* items whose latest agent note is stale or missing.
Guardrail metric: success is ≤3M input/run over the following three daily check-ins and divergence-repair counts at or above the pre-rewrite run average — a cheaper run that quietly repairs less is a failure, not a saving.

Routing

Implementation filed as #2154 (instruction rewrite to the digest flow, script wiring, one evidenced full run, three-check-in measurement) and enqueued as a follow-on queue task in this same pass. Role note: repo-fleet-monitor stays Tend in the #2131 map; this rewrite is its slimming child in everything but name.

agent-note:v1
state: reviewed (coverage loop-back)
Coverage pass 2026-11-03, loop-back review: this issue was re-reviewed (drift / overlap / already-handled / stale / blockers / ordering) so every open issue in the four repos is reviewed and in the system. Verdict: still live; in the system as task 7296 (triage, queued, priority 50) in repo-health-code-pilot. Labels audited and synced. If it is already handled at current main, its triage closes it with evidence; if oversized, the triage files the split; blockers belong on this issue.
Found via: coverage loop-back 2026-11-03

agent-note:v1
event: triaged
task: 7296 (triage, repo-health-code-pilot; repo-worker-code; OpenChamber scheduled lane 02cafe76-eafd-4d0b-aaa6-2b7a737d68dc)
session: ses_e5e5b5b8e50ffeShFhVywaC0NL
state: triaged at current repository state; #2134 remains OPEN. No source or live job was changed.
evidence:

The cadence half is complete: the 2026-10-29 progress comment says the live repo-fleet-monitor schedule was verified at interval@2h; current main's doc/fleet-schedule-registry.md:161 also lists repo-fleet-monitor as enabled at interval@2h.
The fetch/snapshot half is not complete. Follow-up issue Implement repo-fleet-monitor snapshot/SQL-first fetch (digest-judged runs, ≤3M input/run) #2154 remains OPEN, has no linked PR, and currently carries queue/parked + status/parked. Its latest evidence records a >300s monitor-digest timeout and a data-quality flaw: the digest drops status/effort labels and uses last-pull-wins updatedAt, with corrective work still required.
Implement repo-fleet-monitor snapshot/SQL-first fetch (digest-judged runs, ≤3M input/run) #2154's current issue note identifies queue task 3809 as the parked implementation follow-up; the stated success condition still includes the ≤3M/run measurement over three check-ins.
decision: The “already implemented” premise is only partially true. Keep Slim repo-fleet-monitor: snapshot/SQL-first state fetch + 2h cadence (target ≤3M/run) #2134 open; do not duplicate the implementation. Resume Implement repo-fleet-monitor snapshot/SQL-first fetch (digest-judged runs, ≤3M input/run) #2154/task 3809 for digest correctness, successful full runs, and the three-check-in measurement, then update Slim repo-fleet-monitor: snapshot/SQL-first state fetch + 2h cadence (target ≤3M/run) #2134 and epic epic: fleet expansion & placement program (2026-10-28 directive, LIST 1–5) #2055.

Navigation Menu

Repository navigation

Slim repo-fleet-monitor: snapshot/SQL-first state fetch + 2h cadence (target ≤3M/run) #2134

Triage verdict — premise CONFIRMED with one correction, design decided, implementation routed

Design (fetch rewrite)

Routing

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Slim repo-fleet-monitor: snapshot/SQL-first state fetch + 2h cadence (target ≤3M/run) #2134

Description

Activity

developing-today commented on Oct 29, 2026

dezren39 commented on Oct 29, 2026

dezren39 commented on Oct 29, 2026

dezren39 commented on Oct 29, 2026

dezren39 commented on Oct 29, 2026

Triage verdict — premise CONFIRMED with one correction, design decided, implementation routed

Design (fetch rewrite)

Routing

developing-today commented on Nov 3, 2026

dezren39 commented on Nov 4, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions