Bindplane at scale: fleets, gateways and rollouts

1. Organizing the fleet #

Labels, fleets and gateways.

Edge and gateway #

• Labels (env, team, region) are set at installation (OPAMP_LABELS) or in the interface.
• An edge fleet groups collectors that share a configuration: you manage a group, not machines.
• A gateway is a named group of gateway collectors, to be placed behind an OTLP load balancer.

Good practice: With gateways, backend API tokens are only present on a few machines, and outbound traffic to the Internet goes through a single point.

2. Rollouts #

Deploying without breaking everything.

Standard #

A standard rollout (all editions) deploys in growing batches: 3 collectors, then 15, 75, and 100 per batch after that, one batch every few seconds. If a collector reports an error with the new version, the rollout pauses automatically.

bindplane rollout start web-logs --initial 2 --multiplier 3 --max 50 --max-errors 0
bindplane rollout status web-logs

Progressive (Enterprise) #

A progressive rollout is split into stages defined by labels (env=dev, then env=preprod, then env=prod). Each stage waits for a click on Continue Rollout. A final stage always picks up the collectors that match no other stage.

When an error occurs #

1. Click the rollout's error text to list the failing collectors, then open a collector to read the message.
2. Fix the configuration and start a new rollout, or go back to a previous version with Roll forward to this version.

Common mistake: A configuration error (port already in use, file not found, invalid token) often only shows up when the collector starts: this is precisely what the first batch of 3 collectors should reveal. Do not start a rollout with --initial equal to the whole fleet.

Permissions #