Skip to main content

Indexing stuck / stale results

Covers "my edits don't show up in search results" and watcher issues.

Symptom: recent edits don't appear in search results​

Likely causes (ranked)​

  1. You're not running the daemon. The CLI (shaktiman search) reads a static index. If nothing's re-indexing after your edits, nothing can show up. Check with ls /tmp/shaktiman-*.sock or look for shaktimand in ps.
  2. The watcher is disabled in config. watcher.enabled = false in .shaktiman/shaktiman.toml (not the default) turns off auto-reindex.
  3. The file extension isn't recognised. Shaktiman indexes only the languages wired in internal/daemon/scan.go; arbitrary .txt / .md / config files aren't indexed. See Supported Languages.
  4. The file matches an ignore rule — .gitignore, .shaktimanignore, or a pattern in [test].patterns if you searched with scope:"impl" (default).

Diagnostic​

# Is the daemon watching?
tail -f /path/to/project/.shaktiman/shaktimand.log | jq 'select(.component == "watcher")'

# Save a file and watch for the event
touch /path/to/project/src/<your-file>
# Expect to see: {"msg": "file change detected", ...} within ~200ms

# Check the scope you searched in
shaktiman search "your-query" --root /path/to/project --scope all
#                                                      ^^^^^^^^^^^^ no filter

Fix​

  • Start the daemon (open Claude Code, or run shaktimand /path/to/project in another terminal).
  • If the file extension isn't wired: see Supported Languages → Adding a language.
  • If scope is hiding it, re-query with scope: "all".
  • If .shaktimanignore matches when it shouldn't, edit or remove the pattern.

Symptom: enrichment_status pending count keeps growing​

Chunks are being added faster than the embedding worker drains them.

Likely causes​

  1. A cold index is in progress — this is normal and will catch up.
  2. Ollama is responding slowly (weak GPU, CPU-only mode, or a very large batch).
  3. embedding.batch_size is too small for your hardware, so each HTTP round-trip dominates.

Diagnostic​

# Watch pending count over time
watch -n 5 'shaktiman enrichment-status --root .'

# How fast is each batch?
grep "embed_batch" /path/to/project/.shaktiman/shaktimand.log | tail -n 20

Fix​

  • Raise [embedding].batch_size if your Ollama can handle it.
  • Move Ollama to a faster machine if CPU-bound.
  • Accept that it takes a while on a huge repo; keyword search works in the meantime.

Symptom: the watcher is dropping events​

The log shows "dropped file change event" warnings.

Likely cause​

A burst of file changes exceeded the watcher's buffered channel (capacity 100). Common causes: npm install, git reset --hard, extracting a big tarball into the project, or a build tool writing generated files in-tree.

Fix​

  • Add noise to .shaktimanignore — node_modules, build/, dist/, generated directories. This prevents them from entering the watcher in the first place.
  • Ignore transient noise: the watcher intentionally drops silently under load rather than blocking indexing; a periodic full scan would pick up genuine missed changes, but that scan isn't implemented today. For now, if you suspect a file was missed, touch it to force a re-index.

See also​