bug: query with empty filter returns empty results on local Milvus (syncIndexedCodebasesFromCloud)

[pablogravielseo]

Problem

syncIndexedCodebasesFromCloud() in handlers.ts calls vectorDb.query() with an empty filter string to extract codebasePath from collection metadata:

const results = await vectorDb.query(
    collectionName,
    '', // Empty filter to get all results
    ['metadata'],
    1
);

This query returns empty results on local Milvus standalone, even when the collection has data. This was also reported by @juanmlarios in #246.

Root Cause

In milvus-vectordb.ts:435-437, the empty string is passed as filter:

const queryParams: any = {
    collection_name: collectionName,
    filter: filter,  // '' (empty string)
    output_fields: outputFields,
};

The Milvus Node SDK (@zilliz/milvus2-sdk-node) then processes this with:

data.expr = data.filter || data.expr || primaryKeyInIdsExpression;

Since '' is falsy in JavaScript, '' || data.expr falls through to data.expr (which is undefined), then to primaryKeyInIdsExpression (also undefined). The final expression sent to the Milvus gRPC server is effectively empty/undefined, which may return no results depending on the Milvus server version.

Impact

This bug prevents syncIndexedCodebasesFromCloud() from discovering codebases in Milvus, which causes:

• cloudCodebases set remains empty
• The sync removes all local codebases from the snapshot
• search_code returns "not indexed" error

This is the underlying cause of #145 and the blocker that stalled #246.

Suggested Fix

Replace the empty filter with an expression that is always truthy and matches all documents. For example:

// Option A: use a non-empty expression that matches everything
const results = await vectorDb.query(
    collectionName,
    'id != ""',
    ['metadata'],
    1
);

Or fix at the milvus-vectordb.ts level to handle empty filter correctly:

// Option B: in milvus-vectordb.ts query(), use 'expr' key instead of 'filter'
const queryParams: any = {
    collection_name: collectionName,
    expr: filter || undefined,  // Let SDK handle the default
    output_fields: outputFields,
};

Related

• Indexing succeds but then search_code right after that says code base is not indexed #145 — search fails with "not indexed" after indexing
• fix: sync cloud codebases to local snapshot after MCP restart #246 — PR stalled because query returns empty
• fix: restore indexed codebases from Milvus to local snapshot after restart #278 — PR with defensive fixes that work around this issue

[zc277584121]

Thanks for reporting! This was fixed in v0.1.4 (PR #280) and further improvements were added in v0.1.5.

The empty filter query issue has been resolved, and the cloud sync deletion logic now uses collection description-based extraction with proper safety guards.

Please update to v0.1.5 and test. If no further issues are reported, this will be closed after a period of time.

[pablogravielseo]

Confirmed fixed in v0.1.4 (PR #280) and v0.1.5.

The fix omits the filter field entirely instead of sending an empty string, avoiding the JS falsy '' issue. Additionally, the sync now uses getCollectionDescription() for path extraction, making the query() call a fallback for older collections only.

Thanks for the fix @zc277584121!