Dashboard cruft cleanup

Daily cron that prunes stale index-only dashboard rows for users inactive ≥ 1 year. Complements lazy me() eviction for active users and one-off ops scripts (prune-stale-recent-completed, purge-usergame-orphans).

Schedule
Handler src/functions/dashboard-cruft-cleanup.ts
Schedule Daily 03:00 UTC (prod only)
Timeout / memory 900 s / 1024 MB
Layer No

What it cleans

For each eligible user (live confirm after S3 dump candidate scan):

Partition Action
RECENTCOMPLETED# Delete rows not dashboard-eligible (merged USERGAME# overlays)
USERGAME# Delete overlays for pruned recent rows and orphan overlays not on CURRENTGAMES# ∪ eligible RECENTCOMPLETED#

Does not touch USER.lastSeen, USERS.lastSeen, or legacy USER.games[] (retired in Phase 5).

Candidate discovery (no DynamoDB scan)

  1. List abstractplay-db-dump and pick the latest manifest-summary.json export uid (same pattern as Records pipeline).
  2. Stream ION USER items from that export.
  3. Collect sk (user id) when lastSeen < now - 1y and cleaned != true.

Dump may be up to ~24h stale; each candidate is re-validated with live GetItem before cleanup.

Live processing

For up to DASHBOARD_CRUFT_BATCH_SIZE candidates (default 75):

  1. GetItem USER — skip if active since dump, already cleaned, or missing.
  2. Skip bots (GetItem BOT).
  3. Run cleanupUserDashboardCruft (Query + Delete only).
  4. If cruft was removed, SET cleaned = true on USER (not USERS).

me() in node-backend clears cleaned on login so a future long absence can be cleaned again.

Environment

Variable Default Purpose
ABSTRACT_PLAY_TABLE abstract-play-{stage} DynamoDB table
DASHBOARD_CRUFT_BATCH_SIZE 75 Max users processed per run
ABANDONED_ACCOUNT_INACTIVE_MS 31536000000 (365 days) Inactivity threshold

Manual invoke

serverless invoke -f dashboard-cruft-cleanup --stage prod

Use prod with care — this deletes dashboard index rows for inactive users.

Related