Fleet Lifecycle
Rolling drain, cordon and upgrade across the GPU fleet without interrupting running jobs.
What it is
When we patch kernels, swap drivers or roll a new node image, Fleet Lifecycle drains workloads off nodes in a controlled order: cordon → evict pods with grace → reschedule → reboot → uncordon. You see per-node events in the audit log and can register a webhook to react (checkpoint, fail-over).
When to use it
- Subscribe to maintenance windows for your training jobs.
- Force a manual drain to test failure modes.
- Pin a workload to a node group that opts out of automatic upgrades.
Quickstart
asc fleet maintenance list
asc fleet drain node_abc --grace 60sclient.fleet.maintenance.list(region='apac')
client.fleet.drain(node_id='node_abc', grace_seconds=60)await client.fleet.maintenance.list({ region: 'apac' });
await client.fleet.drain('node_abc', { graceSeconds: 60 });Limits & quotas
| Limit | Default | Burst | Notes |
|---|---|---|---|
| Grace period | 1 s – 600 s | — | Per drain |
Pricing
See pricing. Pay-as-you-go, billed monthly via Stripe.
API surface
GET /v1/atlas/fleet/maintenancePOST /v1/atlas/fleet/drain
Required scope(s): fleet:read, fleet:write. See Scopes.
Security
All access is authenticated and scoped. See Auth & scopes and Network controls.