Prevent silent O(n) search fallback when sqlite-vec is unavailable #17

New Issue

Francy51 · 2026-03-15T01:25:57Z

Francy51 commented

2026-03-15 01:25:57 +00:00

When sqlite-vec is unavailable, search silently degrades to a JavaScript full scan over stored JSON embeddings.

Why this is a problem:

Query cost grows linearly with corpus size.
Each search parses stored embedding JSON and computes cosine distance in-process.
This creates a severe performance cliff that is easy to miss in development and difficult to operate safely at scale.

Observed in:

lib/server/db/index.ts
- logs a warning and falls back to table-backed vector storage
lib/server/search.ts
- vectorSearch() loads candidate rows, parses JSON embeddings, computes cosine distance, sorts, and slices in JS when the extension is unavailable

Suggested direction:

Treat missing vector capability as a degraded mode that is explicit, gated, or disabled.
Avoid silently preserving the same feature contract with radically different performance.
Consider:
- failing startup in environments that require semantic search
- feature-flagging search when vector support is missing
- providing a clearly limited lexical-only fallback instead of faux vector search

Acceptance criteria:

Search does not silently drop to O(n) vector scans.
Missing vector support is visible as an explicit degraded capability.
Performance expectations are preserved or the feature is clearly disabled.

When `sqlite-vec` is unavailable, search silently degrades to a JavaScript full scan over stored JSON embeddings. Why this is a problem: - Query cost grows linearly with corpus size. - Each search parses stored embedding JSON and computes cosine distance in-process. - This creates a severe performance cliff that is easy to miss in development and difficult to operate safely at scale. Observed in: - `lib/server/db/index.ts` - logs a warning and falls back to table-backed vector storage - `lib/server/search.ts` - `vectorSearch()` loads candidate rows, parses JSON embeddings, computes cosine distance, sorts, and slices in JS when the extension is unavailable Suggested direction: - Treat missing vector capability as a degraded mode that is explicit, gated, or disabled. - Avoid silently preserving the same feature contract with radically different performance. - Consider: - failing startup in environments that require semantic search - feature-flagging search when vector support is missing - providing a clearly limited lexical-only fallback instead of faux vector search Acceptance criteria: - Search does not silently drop to O(n) vector scans. - Missing vector support is visible as an explicit degraded capability. - Performance expectations are preserved or the feature is clearly disabled.

Francy51 added the P1 label 2026-03-15 01:25:57 +00:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: Francy51/Neon-Desk#17