Three proposals addressing the 2026-02-25 trace pipeline OOM crash (114K CIDs):
1. trace-events-memory-triage: fetchmany iterator (read_sql_df_slow_iter),
admission control (50K CID limit for non-MSD), cache skip for large queries,
early memory release with gc.collect()
2. trace-async-job-queue: RQ-based async jobs for queries >20K CIDs,
separate worker process with isolated memory, frontend polling via
useTraceProgress composable, systemd service + deploy scripts
3. trace-streaming-response: chunked Redis storage (TRACE_STREAM_BATCH_SIZE=5000),
NDJSON stream endpoint (GET /api/trace/job/<id>/stream), frontend
ReadableStream consumer for progressive rendering, backward-compatible
with legacy single-key storage
All three proposals archived. 1101 tests pass, frontend builds clean.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Two changes combined:
1. historical-query-slow-connection: Migrate all historical query pages
to read_sql_df_slow with semaphore concurrency control (max 3),
raise DB slow timeout to 300s, gunicorn timeout to 360s, and
unify frontend timeouts to 360s for all historical pages.
2. hold-resource-history-dataset-cache: Convert hold-history and
resource-history from multi-query to single-query + dataset cache
pattern (L1 ProcessLevelCache + L2 Redis parquet/base64, TTL=900s).
Replace old GET endpoints with POST /query + GET /view two-phase
API. Frontend auto-retries on 410 cache_expired.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Reject History:
- Compute dimension pareto (package/type/workflow/workcenter/equipment) from
cached DataFrame instead of re-querying Oracle per dimension change
- Propagate supplementary filters and trend date selection to dimension pareto
- Add staleness tracking to prevent race conditions on rapid dimension switches
- Add WORKFLOWNAME to detail and export outputs
- Fix button hover visibility with CSS specificity
MSD (製程不良追溯分析):
- Separate raw events caching from aggregation computation so changing
loss_reasons uses EventFetcher per-domain cache (fast) and recomputes
aggregation with current filters instead of returning stale cached results
- Exclude loss_reasons from MSD seed cache key since seed resolution does
not use it, avoiding unnecessary Oracle re-queries
- Add suspect context panel, analysis summary, upstream station/spec filters
- Add machine bar click drill-down and filtered attribution charts
Query Tool:
- Support batch container_ids in lot CSV export (history/materials/rejects/holds)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Transform /mid-section-defect from TMTT-only backward analysis into a full-line
bidirectional defect traceability center supporting all detection stations.
Key changes:
- Parameterized station detection: any workcenter group as detection station
- Bidirectional tracing: backward (upstream attribution) + forward (downstream reject rates)
- Dual query mode: date range OR LOT/工單/WAFER container-based seed resolution
- Multi-select filters for upstream station, equipment model (RESOURCEFAMILYNAME), and loss reasons
- Progressive 3-stage trace pipeline (seed-resolve → lineage → events) with streaming UI
- Equipment model lookup via resource cache instead of SPECNAME
- Session caching, auto-refresh, searchable MultiSelect with fuzzy matching
- Remove legacy tmtt-defect module (fully superseded)
- Archive openspec change artifacts
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
