openclaw - 💡(How to fix) Fix memory index --force crashes with EMFILE: too many open files, watch across builtin, Ollama, and QMD backends on macOS [1 participants]

Summary

openclaw memory index --force crashes with:

Error: EMFILE: too many open files, watch
    at FSWatcher._handle.onchange (node:internal/fs/watchers:267:21)

This reproduces across multiple memory provider/backend combinations and still happens after rebuilding the memory DB from scratch.

Environment

• OpenClaw: 2026.3.31 (213a704)
• OS: macOS
• Node: v24.14.1
• Install method: global npm/Homebrew-managed Node install
• Current memory provider tested:
  • builtin + provider: local
  • builtin + provider: ollama
  • memory.backend = "qmd"

What I tested

1. Original local embeddings path

Configured:

{
  "agents": {
    "defaults": {
      "memorySearch": {
        "provider": "local"
      }
    }
  }
}

Observed:

• openclaw memory status --json worked
• openclaw memory index --force crashed with EMFILE: too many open files, watch

2. Clean-room repro

Created an isolated temp profile and workspace with only two tiny markdown files in workspace/memory/.

Observed:

• memory status --json saw scan.totalFiles = 2
• memory index --force still crashed with the same EMFILE error

3. Rebuilt memory DB from scratch

Backed up and deleted:

• ~/.openclaw/memory/main.sqlite
• ~/.openclaw/memory/main.sqlite.tmp-*

Observed after rebuild:

• fresh status showed files: 0, chunks: 0, dirty: true, scan.totalFiles: 2
• memory index --force still crashed with the same EMFILE

4. Switched to Ollama

Installed Ollama and nomic-embed-text, configured:

{
  "agents": {
    "defaults": {
      "memorySearch": {
        "provider": "ollama"
      }
    }
  }
}

Also set:

OLLAMA_API_KEY=ollama-local

Observed:

• openclaw memory status --json showed:
  • provider: "ollama"
  • model: "nomic-embed-text"
• openclaw memory index --force still crashed with the same EMFILE

5. Switched to QMD backend

Configured:

{
  "memory": {
    "backend": "qmd",
    "qmd": {
      "command": "/opt/homebrew/bin/qmd",
      "includeDefaultMemory": true
    }
  }
}

Observed:

• openclaw memory status --deep showed:
  • Provider: qmd
  • Store: ~/.openclaw/agents/main/qmd/xdg-cache/qmd/index.sqlite
  • Indexed: 0/2 files
  • Embeddings error: QMD index has 0 vectors; semantic search is unavailable until embeddings finish
• openclaw memory index --force still crashed with the same EMFILE
• before the crash, OpenClaw also logged a QMD collection conflict warning:
  • qmd collection add skipped for memory-alt-main ... A collection already exists for this path and pattern

Important observations

• This does not appear tied to:
  • node-llama-cpp
  • Ollama
  • the builtin SQLite memory DB
  • one corrupt memory markdown file
• The same failure happens with:
  • builtin memory + local provider
  • builtin memory + ollama provider
  • QMD backend
• The same failure also reproduces in a clean temporary OpenClaw state directory with only two small memory files

Current memory files

Only two files under workspace/memory/, both normal markdown and readable.

Related status outputs

Example after DB rebuild:

files: 0
chunks: 0
dirty: true
scan.totalFiles: 2
issues: []

Example with Ollama:

provider: "ollama"
model: "nomic-embed-text"
scan.totalFiles: 2

Example with QMD:

Provider: qmd
Indexed: 0/2 files
Embeddings error: QMD index has 0 vectors

Expected behavior

openclaw memory index --force should rebuild the memory index successfully.

Actual behavior

It crashes consistently with:

Error: EMFILE: too many open files, watch
    at FSWatcher._handle.onchange (node:internal/fs/watchers:267:21)

Notes

This looks like a problem in OpenClaw’s memory indexing/orchestration/watch layer rather than in a specific provider/backend.