Batch rollouts¶

A batch rollout pushes the new configuration to a fixed number of agents per batch, then waits for a manual advance.

When to use¶

• Small fleets where percentages do not give meaningful steps.
• Highly regulated environments where every step is an explicit operator decision.
• First runs of a new strategy where you want maximum control.

Configuring¶

In the rollout wizard:

• Strategy: Batch.
• Batch size: integer, default 5.
• Auto-advance: off (default) or on with a dwell timer between batches.
• Order: random (default), by host name, by last seen.

Behaviour¶

1. The rollout's first batch is the first BatchSize agents from the group.
2. Each agent in the batch is pushed in parallel (capacity-limited internally).
3. The rollout reports per-agent state: Pending → Applying → Applied (or Failed).
4. Once every agent in the batch is Applied or Failed, the rollout pauses for operator advance (or auto-advances if configured).
5. The next batch starts.

If health gates fire during a batch, the rollout transitions to Paused immediately rather than waiting for the batch to finish.

Operator decisions per batch¶

After each batch, the operator can:

• Advance — start the next batch.
• Pause — stop here, decide later.
• Rollback — push the previous version to all batched-so-far agents.
• Abort — stop without rolling back. The pushed agents stay on the new config; only later batches are skipped.

Tips¶

• Pick a batch size that gives you a meaningful sample: ~ 1–5 % of the group is a good starting point.
• For fleets larger than ~ 200, switch to Percentage rollouts — batch sizes feel arbitrary at scale.
• Auto-advance with a dwell timer is essentially a stripped-down Canary step-up. For most production cases, prefer the canary strategy.