OpenClaw Service Migration: From pnode200 to openclaw-node01

3 minute read

What started as a routine cron job review turned into a full infrastructure migration session. Here’s how we moved core OpenClaw services from pnode200 to openclaw-node01 in a single evening.

The Trigger

We discovered that QMD memory sync was still running on pnode200 - the old OpenClaw host. The sync was pushing outdated workspace content to MS-S1-MAX. That kicked off a review of what else was running on pnode200 that should have been on openclaw-node01.

Cron Jobs

What Was Still on pnode200

Job	Schedule	Why It Matters
Whisper-watch	Every 5 min	Audio transcription queue
Memory distill	Sunday 22:00	Weekly knowledge consolidation
Workspace → Obsidian	Daily 06:00	Note synchronization

All still running on old hardware, syncing content that was increasingly stale.

What Was Missing

OpenClaw cron jobs hadn’t been fully set up on the new host:

Morning Briefing at 08:00 Mon-Fri
QMD Index Maintenance at 06:30 daily

We added all of these to openclaw-node01’s crontab.

The SSH Key Detail

The QMD sync needed SSH access from openclaw-node01 to MS-S1-MAX. One ssh-copy-id later and the memory sync was running on the right machine.

Container Migration

markitdown-mcp

Microsoft’s MarkItDown MCP server - converts PDFs, DOCX, PPTX to Markdown. Built from source:

git clone https://github.com/microsoft/markitdown.git
cd markitdown/packages/markitdown-mcp
docker build -t markitdown-mcp:0.1.0 .

This is a document ingestion tool for agents. PDF extraction works well. YouTube transcripts are hit-or-miss due to anti-bot measures.

yt-dlp

Not a persistent service - this is CLI-only. The container exits immediately. Running docker compose up -d causes restart loops.

Instead, use the wrapper script:

/opt/stacks/yt-dlp/yt-dlp-wrap.sh "https://youtube.com/watch?v=VIDEO_ID"

Downloads to ~/workspace/downloads/.

mkdocs-material

This one mattered for Tailscale integration.

The old instance on pnode200 was hosting the docs site at docs.<tailscale>.net. We needed to:

Stop the container on pnode200
Sync the docs content via rsync
Start the new container on openclaw-node01 with Docktail labels
Tag openclaw-node01 in Tailscale with tag:server

Docktail reads the container labels and creates Tailscale Funnel services automatically. Once the node had the right tag, docs.<tailscale>.net pointed to the new host.

What We Skipped

openclaw-dashboard - Phase 1 was rough, not ready
apprise-notification-server - Future use only
rackpeek - Not fully functional

discrawl: Discord Archive Search

This was the evening’s bonus. discrawl mirrors Discord into SQLite for local search.

brew tap steipete/tap
brew install steipete/tap/discrawl
discrawl init --from-openclaw ~/.openclaw/openclaw.json
discrawl sync --full

It pulled 113,842 messages from the #openclaw guild. Now we can search locally:

discrawl search "migration" --channel openclaw

The --from-openclaw flag reuses OpenClaw’s Discord bot token, so no separate configuration needed.

We set it up as a systemd service to tail live events:

sudo systemctl enable --now discrawl-tail

Now every message gets archived in real-time.

Go and Homebrew

discrawl required Go 1.26+. Installed via Homebrew:

brew install go
# go version go1.26.1 linux/amd64

Had to add --break-system-packages for AgentMail’s Python package on Ubuntu 24.04. The PEP 668 restrictions caught us there.

What Else

Created .gitignore for the workspace - .env was not being ignored (!)
Documented MCP code mode pattern for reducing context bloat
Fixed AgentMail Python package installation
Added context-hub skill for API documentation fetching

The Numbers

Metric	Count
Containers deployed	5
Cron jobs migrated/added	8
Tailscale services exposed	1
Messages in discrawl	113,842
Session duration	~2 hours

Lessons Learned

Docktail needs node tags - Tailscale Services require tag:server on the host before they’ll create
yt-dlp is CLI-only - Don’t run as a service
Ubuntu 24.04 + pip - The --break-system-packages flag is necessary
MCP context bloat is real - We documented a pattern for future reference

What’s Next

Wire discrawl into the agent research workflow
Migrate remaining containers (dashboard, apprise, rackpeek) when ready
Monitor discrawl-tail for stability over longer runs

Session duration: approximately 2 hours

Derek Leeds