DEADLINE_EXCEEDED on checkCollectionLimit() - gRPC timeout on collection creation (Free tier, macOS)

[gvorwaller]

Bug Description

Consistent DEADLINE_EXCEEDED error when attempting to index a codebase. The error occurs during the pre-flight checkCollectionLimit() call before any actual indexing begins. A different codebase on the same cluster was successfully indexed 4 days prior.

Environment

• OS: macOS Darwin 25.3.0 (Apple Silicon)
• Node.js: v24.3.0
• Package: @zilliz/claude-context-mcp@0.1.6
• Zilliz Cloud: Free tier, gcp-us-west1, Milvus 2.6.x compatible
• Cluster: in03-e63f077fc4f1b69

Cluster State

• Collections: 1 of 5 allowed (hybrid_code_chunks_9429127a, status: Loaded)
• Storage: 0.01 GB of 5 GB
• Read vCUs: ~12 (minimal load)
• No resource limits hit

Error

Error validating collection creation: 4 DEADLINE_EXCEEDED: Deadline exceeded after 15.001s,remote_addr=34.111.198.99:443

Root Cause Trace

The error originates in milvus-vectordb.js:607 (checkCollectionLimit()), which creates a dummy collection as a pre-flight limit check:

async checkCollectionLimit() {
    const collectionName = `dummy_collection_${Date.now()}`;
    // creates a trivial 2-field collection, then drops it
    await this.client.createCollection(createCollectionParams);  // <-- times out here
}

This is called from handlers.js:249 before any real indexing starts:

const canCreateCollection = await this.context.getVectorDatabase().checkCollectionLimit();

The MilvusClient at milvus-vectordb.js:20 is created with no timeout parameter, so the 15s gRPC deadline is the SDK default.

Reproduction Steps

1. Have an existing indexed codebase (collection loaded and working)
2. Attempt to index a second, different codebase: index_codebase({ path: "/path/to/project" })
3. Consistently fails with DEADLINE_EXCEEDED on the checkCollectionLimit() dummy collection creation
4. Tried with both ast and langchain splitters, with and without force: true, with and without ignorePatterns — all produce the same error
5. Indexing a subdirectory also fails identically

Key Observation

• Another codebase (~/gaylonphotos, 297 files, 1241 chunks) was successfully indexed on this same cluster on Mar 22, 2026
• The failing codebase (~/BTC-dashboard) was previously indexed (196k chunks, Feb 20) but was cleared during a force-reindex attempt
• Since then, every attempt to create any new collection times out
• get_indexing_status confirms the codebase is not indexed
• clear_index confirms nothing to clear

Related

• Issue MCP Fails when started to index the codebase #154 — same DEADLINE_EXCEEDED error, also on Node v24.x
• Issue Error validating collection creation during indexing #126 — similar collection creation failure (different error format)

Suggested Improvements

1. Configurable gRPC timeout — allow setting a timeout > 15s via env var (e.g., MILVUS_TIMEOUT_MS)
2. Skip dummy collection check — checkCollectionLimit() could use listCollections() + count instead of create/drop, avoiding the gRPC-heavy collection creation path
3. REST API fallback — MilvusRestfulVectorDatabase exists in claude-context-core but isn't wired up as a fallback when gRPC fails