v202604240135longchunk image shipped with only 1 file in the baked
corpus (NEXT-SPRINT.md) because the corpus tar was accidentally built
from the Intranet.Web working directory instead of the Notes repo
root. Rebuilt from the right cwd; new image has the expected 370
*.md + *.html files at /srv/flowercore-notes/docs/.
Same long-chunk handling code as v202604240135longchunk; just a clean
rebuild.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Image bump v202604240108gpu -> v202604240135longchunk, rebuilt from
FlowerCore.Intranet.Web@feat/shared-indexing-search HEAD which transitively
picks up FlowerCore.Common@feat/shared-indexing@105af75:
- MarkdownChunker hard-caps oversized heading-bounded sections at
ChunkSizeTokens × 4 chars and splits with overlap (same pattern as
JsonArticleChunker). Stops the indexer from producing chunks above
nomic-embed-text's 8192-token input limit at the source.
- IndexBuilder gains IndexingOptions.MaxEmbeddingTokens (default 8000)
safety filter — chunks above the cap are warn-logged and dropped
before any batch is sent. New IndexBuildResult.ChunksDropped tracks
how many got skipped.
Goal: notes-md should index 2541/2541 chunks (vs. 2080/2541 last pass)
with zero "Failed to embed batch" 400s.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two-part fix on top of the live Shared.Indexing rollout:
1. Image bump v202604240050corpus -> v202604240108gpu, rebuilt from
FlowerCore.Intranet.Web@feat/shared-indexing-search (HEAD includes
the FilePatterns array-merge fix in IntranetSearchOptions). At
runtime each DocCorpusRoot now sees ONLY the patterns explicitly
set in appsettings.json — notes-md gets ["*.md"], notes-html gets
["*.html"], no accidental cross-bleed.
2. New IntranetSearch__OllamaBaseUrl env var pointing at
http://10.0.56.20:11434 (BLUEJAY-WS GPU, R9700 32GB VRAM). Verified
reachable from the cluster and nomic-embed-text:latest is pulled.
This is the workaround for memory feedback_pi5_nomic_embed_slow:
edge1 Pi 5 takes ~189s per 32-chunk batch, projecting full notes-md
indexing (5665 chunks) at ~9 hours; the GPU should land it in minutes.
Edge1 stays the chat default; this env var only redirects the
indexer's bulk embedding calls.
Image distributed to all three RKE2 nodes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Cluster egress to github.com is fronted by a step-ca TLS proxy that
returns 404 page not found for unmatched routes — git clone of the
public FlowerCore.Notes repo failed inside the pod even with
GIT_SSL_NO_VERIFY=true. Rather than chase the egress NetworkPolicy /
proxy config, bake the docs corpus directly into the image at
/srv/flowercore-notes/docs.
The corpus is just *.md + *.html (369 files, 2.7 MB uncompressed) —
small enough that re-baking on every deploy is fine and avoids any
runtime network dependency.
Manifest changes:
- Image bump: v202604240040search -> v202604240050corpus
- Removed initContainers (clone-notes-corpus is now redundant)
- Removed notes-corpus emptyDir + its volumeMounts
- Vector-store PVC mount stays.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Cluster egress is fronted by a step-ca TLS proxy whose cert doesn't
match github.com. The init container's git clone failed with
"SSL: no alternative certificate subject name matches target hostname
'github.com'". The Notes repo is public — there is no secret to
protect on the wire — so GIT_SSL_NO_VERIFY=true is the right tradeoff
here. Tag at v202604240040search.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phase 3 lane 1 of FlowerCore.Shared.Indexing rollout — wires the new
search consumer in FlowerCore.Intranet.Web to live infrastructure.
Manifest changes:
- Image bump: localhost/fc-intranet-web:latest -> :v202604240040search.
Built from FlowerCore.Intranet.Web@feat/shared-indexing-search and
imported into all three RKE2 nodes (rke2-server, rke2-agent1, rke2-agent2)
via ctr import. Both :latest and :v202604240040search tags are present.
- New PersistentVolumeClaim intranet-vector-store (1Gi, ReadWriteOnce,
Longhorn) mounted at /data for the SQLite vector store
(intranet-vectors.db).
- New emptyDir volume notes-corpus (1Gi sizeLimit) shared between the
init container and main container, mounted at /srv/flowercore-notes
(read-only in the main container).
- New init container clone-notes-corpus (alpine/git) that shallow-clones
https://github.com/astoltz/FlowerCore.Notes.git
(codex/notes-pimanager-live-drift) into /srv/flowercore-notes on every
pod start. Re-clone is cheap (depth=1) and re-runs of git fetch +
reset --hard are idempotent.
- Strategy switched to Recreate for the deployment, since the new RWO
PVC blocks rolling updates — see CLAUDE.md memory "RWO PVC blocks K8s
rolling updates".
- Resource bumps: memory 128Mi -> 256Mi req, 512Mi -> 1Gi limit; CPU
500m -> 1000m limit. The DocsCorpusIndexer + Ollama HTTP calls add
measurable load during the initial index build.
- initialDelaySeconds bumps on both probes (10s -> 30s liveness, 5s ->
10s readiness) to account for startup-time Ollama probing and the
slightly larger image.
The DocsCorpusIndexer waits 15s after host startup before its first
indexing pass, then loops every RescanInterval (default 1h). Its first
run will:
1. Embed all *.md under /srv/flowercore-notes/docs against
nomic-embed-text on edge1 (10.0.57.17:11434).
2. Embed all *.html under /srv/flowercore-notes/docs/dashboards.
3. Persist chunks + embeddings to /data/intranet-vectors.db.
Verify after rollout:
- kubectl -n intranet logs deploy/intranet-web -c clone-notes-corpus
(init container should show the docs/ listing).
- kubectl -n intranet logs deploy/intranet-web -f
(DocsCorpusIndexer should log "Indexing docs root 'notes-md'..." then
"Docs root 'notes-md' indexed: N files, M chunks, M stored").
- curl -sk https://intranet.iamworkin.lan/api/search/indexes
-> ["notes-html","notes-md"]
- curl -sk 'https://intranet.iamworkin.lan/api/search?q=guacamole+single+host&topK=3'
-> hits from docs/infrastructure/guacamole-customization-plan.md
Companion source on FlowerCore.Intranet.Web@feat/shared-indexing-search.
Depends on FlowerCore.Common@feat/shared-indexing.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replaces the 188KB ConfigMap-embedded HTML with a proper Blazor Server
deployment (fc-intranet-web:latest on port 5300). The old nginx deployment,
ConfigMaps (intranet-html, intranet-nginx-conf), and all embedded HTML are
removed. The intranet is now a .NET 10 Blazor app with live health monitoring,
REST API, 49 pages, and the unified Blue Jay theme.
Source: github.com/astoltz/FlowerCore.Intranet.Web
