Migrating to the Agent Orchestrator

This guide walks an existing Bigeye agent host through registering with the Agent Orchestrator so Bigeye can run that host's batch agent jobs — Lineage Plus, DataHealth, and External Monitors — on demand, instead of you triggering them with per-agent commands or cron.

🚧

Current availability

Orchestration is available today for batch agents only: Lineage Plus, DataHealth, and External Monitors. This is the completed first phase of the orchestrator.

Lifecycle management of the persistent agents (Data Source, SDS, Cross-Source) through the orchestrator is in progress and not yet available. Continue to run and manage those agents with their existing per-agent commands — registering a host does not put them under orchestration.

Migration is non-destructive: your bigeye_agent.yml, connectors, and mTLS keys are reused. It is also opt-in — a host keeps working exactly as it does today until you run bigeye-agent register.

Who should migrate

You runMigrate now?Notes
Batch agents (Lineage Plus, DataHealth, External Monitors)YesRegister so Bigeye can dispatch their jobs to this host.
Only persistent agents (Data Source, SDS, Cross-Source)Not yetThe orchestrator doesn't manage these yet. Keep using their per-agent commands and revisit once persistent-agent orchestration ships.
Kubernetes hostNot supportedThe orchestrator runs as a Docker Compose stack; Kubernetes hosts can't register.

Your bigeye_agent.yml, connectors, and agent configuration all carry over. Persistent agents you've already installed keep running and stay under your control — the orchestrator does not start, stop, or upgrade them.

Prerequisites

Complete the orchestrator Prerequisites before migrating — most importantly:

  • Docker Compose V2 (docker compose version reports v2.x.x).
  • Docker usable without sudo — the user running bigeye-agent must be in the docker group (or use rootless Docker). The orchestrator never runs Docker with sudo; if your host currently relies on sudo docker, fix this first or the orchestrator cannot start or manage containers.
  • A Bigeye Personal API Key for an Admin user (register rejects non-admin keys).

Migration-specific preparation:

  • Back up your configuration. Copy bigeye_agent.yml and the entire agent_config/ directory (it holds your mTLS keys) somewhere safe before you start.
  • Pick a host id — a stable, unique, human-readable name for this host (alphanumeric + hyphens, ≤256 chars), e.g. prod-agent-01. It identifies the host to Bigeye and cannot collide with another registered host on the account.

Migration steps

Run all commands from the directory that contains your bigeye_agent.yml.

1. Verify Docker access

docker compose version   # must be v2.x.x
docker ps                # must succeed WITHOUT sudo

If docker ps only works with sudo, add your user to the docker group and start a new login session before continuing:

sudo usermod -aG docker $USER
# log out and back in (or run `newgrp docker`), then re-check `docker ps`

2. Register the host

./bigeye-agent register --host-id prod-agent-01

On a host that already has a bigeye_agent.yml, register:

  1. Reuses the existing mTLS keys in agent_config/ (it does not regenerate them).
  2. Registers the host and its installed agent set with Bigeye so batch-agent jobs can be dispatched here.
  3. Enables Docker Compose mode (docker_compose_enabled: true) in bigeye_agent.yml.
  4. Regenerates docker-compose.bigeye.yml with the orchestrator and docker-socket-proxy services (on an isolated orchestrator-internal network) and starts the orchestrator stack.
⚠️

If this host also runs persistent agents

The orchestrator stack runs under Docker Compose, but the orchestrator does not manage persistent agents (Data Source, SDS, Cross-Source) yet — keep starting, stopping, and upgrading them with their per-agent commands. While persistent-agent orchestration is still in development, hosts dedicated to batch agents are the simplest to register.

3. Verify the migration

Confirm the orchestrator is running:

./bigeye-agent compose status

You should see bigeye-agent-orchestrator and docker-socket-proxy running (plus any long-running agent services already configured on the host).

Check that the orchestrator connected to Bigeye:

./bigeye-agent compose logs -s bigeye-agent-orchestrator

Look for the Temporal worker starting and a successful heartbeat to Bigeye. The host should also appear in your account's orchestrator host list in Bigeye.


After migrating, the host appears in the Hosts tab under Settings → Agents. Representative mockup; the shipped UI may differ.

After migration

  • Batch jobs run on demand. Lineage Plus, DataHealth, and External Monitors jobs are triggered by Bigeye and executed on this host — you can retire the local cron entries or manual lineage run / datahealth run invocations that drove them. See Creating & Managing Agent Jobs.
  • Persistent agents are unchanged. Data Source, SDS, and Cross-Source are not orchestrated yet — keep starting, stopping, and upgrading them with their per-agent commands. Orchestrated lifecycle management for these is planned for a later phase.
  • Observe the orchestrator with compose status and compose logs -s bigeye-agent-orchestrator.
  • Keep the host in sync. bigeye-agent install and bigeye-agent sync continue to work; on a registered host they also regenerate the compose files and push the updated agent set to Bigeye automatically.
  • Upgrade the orchestrator with bigeye-agent compose upgrade.

See the Registering a Host page for the full command reference and the security boundary details.

Migrating additional hosts

mTLS keys are shared across all orchestrator hosts on a Bigeye account. To bring a second host under orchestration:

  1. Copy the entire agent_config/ directory from an already-registered host to the new host.
  2. Run bigeye-agent register --host-id <unique-name> on the new host.

Registering on a host that has no agent_config/ while other orchestrator hosts already exist fails with an error directing you to copy the directory first.

Rolling back

There is no CLI command to undo a migration; roll back manually:

  1. Stop the orchestrator stack:

    ./bigeye-agent compose down
  2. Restore the bigeye_agent.yml you backed up (which has docker_compose_enabled unset and no orchestrator section), or set docker_compose_enabled: false and remove the orchestrator: block.

  3. Restart agents using your previous method (per-agent start commands, or compose up if you were already on Docker Compose mode).

The host record remains registered with Bigeye until it's removed server-side; contact Bigeye support if you need it deregistered.

Troubleshooting

  • "Docker Compose mode is not enabled" when running compose commands — re-run bigeye-agent register; it sets and persists docker_compose_enabled: true.
  • register aborts on the Docker Compose check — install Docker Compose V2 (docker compose, not the legacy docker-compose) and re-run.
  • Orchestrator container won't start or can't manage agents — confirm docker ps works without sudo for the user running the CLI.
  • "Host is already registered" — choose a different --host-id, or reuse the same id on the host that already owns it.

See the Troubleshooting page for more.