immich: fix slow context search — prewarm clip_index + latency alert/healthcheck

Context (smart) search latency was caused by the 665MB vchord clip_index
decaying out of PG shared_buffers (~33% resident -> ~1.8s cold ANN reads vs
~4ms warm), NOT by yesterday's ML MODEL_TTL/clip-keepalive change (CLIP textual
is warm ~15ms on GPU). The postStart prewarm runs once at pod start and
pg_prewarm.autoprewarm only re-warms at startup, so the index decays under job
buffer-pressure over days.

- clip-index-prewarm CronJob (immich, */5): pg_prewarm('clip_index') keeps the
  whole index resident -> searches stay ~4ms.
- immich-search-probe CronJob (immich, */5): times a random-vector ANN query +
  reads clip_index residency, pushes gauges to the Pushgateway.
- Prometheus alerts ImmichSmartSearchSlow / ImmichClipIndexColdCache /
  ImmichSearchProbeStale (+ inhibition when the probe is stale).
- cluster_healthcheck.sh check #46 check_immich_search (TOTAL_CHECKS 45->46).
- Docs: infra CLAUDE.md immich note, monitoring.md, cluster-health skill.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

.claude/CLAUDE.md

@@ -153,7 +153,7 @@ Repo IDs: infra=1, Website=2, finance=3, health=4, travel_blog=5, webhook-handle
 | Service | Key Operational Knowledge |
 |---------|--------------------------|
 | Nextcloud | MaxRequestWorkers=150, needs 8Gi limit (Apache transient memory spikes, see commit eb94144), very generous startup probe |
-| Immich | ML on SSD (CUDA), disable ModSecurity (breaks streaming), frequent upgrades. **`immich-machine-learning` MUST run with `MACHINE_LEARNING_MODEL_TTL > 0`** (set to `600` in `stacks/immich/main.tf`, env on the `immich-machine-learning` deployment). At `0`, no model ever unloads and onnxruntime's CUDA arena (OCR's dynamic input shapes inflate it to ~10 GB) is held forever on the **time-sliced T4 it shares with llama-swap/frigate/immich-server** — which has no VRAM isolation, so immich-ml starved llama-swap (qwen3-8b) and silently broke recruiter-responder triage for ~5 h on 2026-06-02 (post-mortem `docs/post-mortems/2026-06-02-immich-ml-ttl-gpu-oom-recruiter.md`). TTL>0 lets idle models (OCR, face — AND CLIP) free VRAM. The TTL is a single GLOBAL knob (no per-model pin), so CLIP would also unload after 600s idle; the `clip-keepalive` CronJob (`*/5 * * * *`, same stack) pings the CLIP textual encoder so smart-search stays warm without pinning the ad-hoc models. **Video transcoding is GPU-accelerated**: `immich-server` is pinned to GPU node1 (nodeSelector `nvidia.com/gpu.present` + NoSchedule toleration + `gpu-workload` priority) with a time-sliced `nvidia.com/gpu=1` slice — the stock immich-server image's ffmpeg already ships h264/hevc_nvenc + NVDEC. Activated via `ffmpeg.accel=nvenc` + `accelDecode=true` in the **DB** system-config (`system_metadata` table, key `system-config`, JSONB — NOT Terraform; app config is DB-managed here like oauth/smtp). Direct DB edits need a pod **recreate** to reload (config is cached at boot; only API-driven changes broadcast a reload). **Streaming bitrate is capped** to keep 4K playback smooth on the contended HDD and over remote uplinks: `ffmpeg.maxBitrate=20000k` + `preset=medium` + `transcode=bitrate` (set 2026-06-01 — was uncapped `maxBitrate=0` + `ultrafast` + `targetResolution=original`, which produced 77–264 Mbps 4K transcodes that stuttered for every client, local and remote, since even a single stream needs ~10–13.5 MB/s off the shared `sdc` spindle). 4K resolution is preserved (`targetResolution=original`); originals are NEVER modified — only the `encoded-video/` streaming copy. To re-apply transcode settings to EXISTING videos (config changes only affect new/missing ones): delete the offenders' `asset_file` rows `WHERE type='encoded_video'` (derived/regenerable — never touches originals) then run videoConversion `force=false` (admin Jobs API → "Missing"); it regenerates them to the deterministic `<assetId>.mp4` path at concurrency 1 (gentle on sdc). See `docs/runbooks/immich-transcode-bitrate.md`. If Immich is ever reinstalled fresh (not restored), re-set these keys (accel, accelDecode, **maxBitrate=20000k, preset=medium, transcode=bitrate**). Thumbnails/previews live on SSD NFS (sdb) — do NOT move to block storage (HDD sdc = slower + the contended IO domain). **Background-job concurrency is capped to protect sdc** (DB-managed system-config, `system_metadata` key `system-config`, JSONB `job.*.concurrency`; re-set on fresh install): `thumbnailGeneration=2`, `metadataExtraction=2`, `library=2` — these jobs read ORIGINALS off the HDD library. Left uncapped (were 8/4/4) a library-wide job (e.g. Duplicate Detection on 2026-06-01) fans the ML/thumbnail backfill out into a read storm that saturates sdc and starves etcd → apiserver down. `sidecar`/`smartSearch`/`faceDetection` stay at Immich defaults (small `.xmp` / SSD previews). Apply via Job Settings UI or the `system-config` API; **direct DB edits need an `immich-server` pod recreate to reload** (config cached at boot). See `docs/post-mortems/2026-05-25-immich-anca-elements-io-storm.md`. |
+| Immich | ML on SSD (CUDA), disable ModSecurity (breaks streaming), frequent upgrades. **`immich-machine-learning` MUST run with `MACHINE_LEARNING_MODEL_TTL > 0`** (set to `600` in `stacks/immich/main.tf`, env on the `immich-machine-learning` deployment). At `0`, no model ever unloads and onnxruntime's CUDA arena (OCR's dynamic input shapes inflate it to ~10 GB) is held forever on the **time-sliced T4 it shares with llama-swap/frigate/immich-server** — which has no VRAM isolation, so immich-ml starved llama-swap (qwen3-8b) and silently broke recruiter-responder triage for ~5 h on 2026-06-02 (post-mortem `docs/post-mortems/2026-06-02-immich-ml-ttl-gpu-oom-recruiter.md`). TTL>0 lets idle models (OCR, face — AND CLIP) free VRAM. The TTL is a single GLOBAL knob (no per-model pin), so CLIP would also unload after 600s idle; the `clip-keepalive` CronJob (`*/5 * * * *`, same stack) pings the CLIP textual encoder so smart-search stays warm without pinning the ad-hoc models. **Smart search has a SECOND warmth layer in Postgres** (don't conflate it with the ML model): the ~665MB vchord `clip_index` must stay resident in PG `shared_buffers`, else an ANN probe that lands on an evicted list pays a ~1.8s cold storage read vs ~4ms warm. The `postStart` hook prewarms it ONCE at pod start and `pg_prewarm.autoprewarm` only re-warms at *startup*, so the index decays out of cache over days under job buffer-pressure (observed ~33% resident after 9d uptime → slow context search, easily misattributed to the ML model). The `clip-index-prewarm` CronJob (`*/5`, same stack) re-runs `pg_prewarm('clip_index')` to pin it hot; `immich-search-probe` (`*/5`) measures live latency + residency → Pushgateway gauges (`immich_smart_search_db_seconds`, `immich_clip_index_cached_pct`) → alerts `ImmichSmartSearchSlow`/`ImmichClipIndexColdCache`/`ImmichSearchProbeStale` + cluster-health check #46 (`check_immich_search`). immich PG role is a superuser so the CronJobs can run `pg_prewarm`/`pg_buffercache`. **Video transcoding is GPU-accelerated**: `immich-server` is pinned to GPU node1 (nodeSelector `nvidia.com/gpu.present` + NoSchedule toleration + `gpu-workload` priority) with a time-sliced `nvidia.com/gpu=1` slice — the stock immich-server image's ffmpeg already ships h264/hevc_nvenc + NVDEC. Activated via `ffmpeg.accel=nvenc` + `accelDecode=true` in the **DB** system-config (`system_metadata` table, key `system-config`, JSONB — NOT Terraform; app config is DB-managed here like oauth/smtp). Direct DB edits need a pod **recreate** to reload (config is cached at boot; only API-driven changes broadcast a reload). **Streaming bitrate is capped** to keep 4K playback smooth on the contended HDD and over remote uplinks: `ffmpeg.maxBitrate=20000k` + `preset=medium` + `transcode=bitrate` (set 2026-06-01 — was uncapped `maxBitrate=0` + `ultrafast` + `targetResolution=original`, which produced 77–264 Mbps 4K transcodes that stuttered for every client, local and remote, since even a single stream needs ~10–13.5 MB/s off the shared `sdc` spindle). 4K resolution is preserved (`targetResolution=original`); originals are NEVER modified — only the `encoded-video/` streaming copy. To re-apply transcode settings to EXISTING videos (config changes only affect new/missing ones): delete the offenders' `asset_file` rows `WHERE type='encoded_video'` (derived/regenerable — never touches originals) then run videoConversion `force=false` (admin Jobs API → "Missing"); it regenerates them to the deterministic `<assetId>.mp4` path at concurrency 1 (gentle on sdc). See `docs/runbooks/immich-transcode-bitrate.md`. If Immich is ever reinstalled fresh (not restored), re-set these keys (accel, accelDecode, **maxBitrate=20000k, preset=medium, transcode=bitrate**). Thumbnails/previews live on SSD NFS (sdb) — do NOT move to block storage (HDD sdc = slower + the contended IO domain). **Background-job concurrency is capped to protect sdc** (DB-managed system-config, `system_metadata` key `system-config`, JSONB `job.*.concurrency`; re-set on fresh install): `thumbnailGeneration=2`, `metadataExtraction=2`, `library=2` — these jobs read ORIGINALS off the HDD library. Left uncapped (were 8/4/4) a library-wide job (e.g. Duplicate Detection on 2026-06-01) fans the ML/thumbnail backfill out into a read storm that saturates sdc and starves etcd → apiserver down. `sidecar`/`smartSearch`/`faceDetection` stay at Immich defaults (small `.xmp` / SSD previews). Apply via Job Settings UI or the `system-config` API; **direct DB edits need an `immich-server` pod recreate to reload** (config cached at boot). See `docs/post-mortems/2026-05-25-immich-anca-elements-io-storm.md`. |
 | CrowdSec | Pin version, disable Metabase when not needed (CPU hog), LAPI scaled to 3, **DB on PostgreSQL** (migrated from MySQL), flush config: max_items=10000/max_age=7d/agents_autodelete=30d, DECISION_DURATION=168h in blocklist CronJob |
 | Frigate | GPU stall detection in liveness probe (inference speed check), high CPU |
 | Authentik | 3 replicas, PgBouncer in front of PostgreSQL, strip auth headers before forwarding |
@@ -166,7 +166,7 @@ Repo IDs: infra=1, Website=2, finance=3, health=4, travel_blog=5, webhook-handle
 - Exclude completed CronJob pods from "pod not ready" alerts.
 - Every new service gets Prometheus scrape config + Uptime Kuma monitor. External monitors auto-created for Cloudflare-proxied services by `external-monitor-sync` CronJob (10min, uptime-kuma ns). Mechanism: `ingress_factory` auto-adds `uptime.viktorbarzin.me/external-monitor=true` whenever `dns_type != "none"` (see `modules/kubernetes/ingress_factory/main.tf`) — no manual action needed on new services. The `cloudflare_proxied_names` list in `config.tfvars` is a legacy fallback for the 17 hostnames not yet migrated to `ingress_factory` `dns_type`; don't check that list when debugging "is this monitored?" questions.
 - **External monitoring**: `[External] <service>` monitors in Uptime Kuma test full external path (DNS → Cloudflare → Tunnel → Traefik). Divergence metric `external_internal_divergence_count` → alert `ExternalAccessDivergence` (15min). Config: `stacks/uptime-kuma/`, targets from `cloudflare_proxied_names` in `config.tfvars` (17 remaining centrally-managed hostnames; most DNS records now auto-created by `ingress_factory` `dns_type` param).
-- Key alerts: OOMKill, pod replica mismatch, 4xx/5xx error rates, UPS battery, CPU temp, SSD writes, NFS responsiveness, ClusterMemoryRequestsHigh (>85%), ContainerNearOOM (>85% limit), PodUnschedulable, ExternalAccessDivergence.
+- Key alerts: OOMKill, pod replica mismatch, 4xx/5xx error rates, UPS battery, CPU temp, SSD writes, NFS responsiveness, ClusterMemoryRequestsHigh (>85%), ContainerNearOOM (>85% limit), PodUnschedulable, ExternalAccessDivergence, ImmichSmartSearchSlow (context-search latency / clip_index cache eviction).
 - **E2E email monitoring**: CronJob `email-roundtrip-monitor` (every 20 min) sends test email via Brevo HTTP API to `smoke-test@viktorbarzin.me` (catch-all → `spam@`), verifies IMAP delivery, deletes test email, pushes metrics to Pushgateway + Uptime Kuma. Alerts: `EmailRoundtripFailing` (60m), `EmailRoundtripStale` (60m), `EmailRoundtripNeverRun` (60m). Outbound relay: Brevo EU (`smtp-relay.brevo.com:587`, 300/day free — migrated from Mailgun). Inbound external traffic enters via pfSense HAProxy on `10.0.20.1:{25,465,587,993}`, which forwards to k8s `mailserver-proxy` NodePort (30125-30128) with `send-proxy-v2`. Mailserver pod runs alt PROXY-speaking listeners (2525/4465/5587/10993) alongside stock PROXY-free ones (25/465/587/993) for intra-cluster clients. Real client IPs recovered from PROXY v2 header despite kube-proxy SNAT (replaces pre-2026-04-19 MetalLB `10.0.20.202` ETP:Local scheme; see bd code-yiu + `docs/runbooks/mailserver-pfsense-haproxy.md`). Vault: `brevo_api_key` in `secret/viktor` (probe + relay).
 - **Authentik walling-off guard**: `blackbox-exporter` (monitoring ns, `stacks/monitoring/modules/monitoring/authentik_walloff_probe.tf`) probes each must-stay-public `auth = "none"` carve-out URL with `no_follow_redirects` and FAILS (`fail_if_header_matches` on `Location`) iff it 302s to Authentik. Catches a carve-out regressing (TF revert / deploy / `ingress_factory` `auth` default flipping back to `"required"`). Scrape job `blackbox-authentik-walloff` (1m) → alert `AuthentikWallingOffPublicPath` (`probe_failed_due_to_regex == 1`, for 10m, `lane=security` → `#security` Slack). **To guard a new carve-out: add one line to `local.authentik_walloff_targets`** (a `service → URL` map; `valid_status_codes` includes 301/302 so legit redirects/404s stay green — only the Authentik `Location` fails the probe). `curl -sI '<url>'` must NOT show a Location to `authentik.viktorbarzin.me` before adding.