With a pool claim you state which nodes a purpose needs – for a scanner, a batch run or a single tenant. Clusterward creates a Kapsule node pool of its own for it, keeps it matching and releases it again. Ephemeral claims start from zero nodes and release themselves when idle, so you pay only while they are used.
A pool claim is a request for node capacity of its own on a cluster that Clusterward created. You state who needs it and what for, the node type and the maximum number of nodes. Clusterward turns it into a real Kapsule node pool, tags it with its purpose and keeps it in line with the claim: create, adjust the bounds, release. With a label and a taint the pool belongs to that purpose alone.
Cluster page → “Pool claims” → “New claim”: consumer, purpose, node type, size, lifecycle and – for isolation – label and taint.
Clusterward creates the Kapsule pool with autoscaling bounds and tags it with its purpose. The status changes from “pending” to “active”.
Pods with a matching nodeSelector and toleration run on the pool; new nodes get the label and taint within a minute.
Without a heartbeat an ephemeral claim ends after its idle time; “Release…” ends any claim. Clusterward drains and deletes the pool.
Consumer and purpose are free text – the same mechanism serves any workload.
Security scans run on nodes of their own instead of next to your applications – and the pool disappears once the scan is over.
Nightly runs or larger calculations get nodes only when pods are waiting for them. Afterwards the pool scales to zero.
A persistent claim with a label and a taint gives a customer nodes of their own – the isolation some contracts require.
An ephemeral claim starts at zero nodes. The cluster autoscaler adds nodes only for waiting pods, and Clusterward releases the pool when it is idle.
Clusterward sets the label and the taint on every node of the pool and repeats this on every pass, also for nodes the autoscaler adds later. Foreign taints are kept.
A program creates claims through the REST API, keeps them alive with a heartbeat and releases them – with an API token whose role allows it.
The “Pool claims” card shows every claim with size, lifecycle, status and pool. Creating and releasing are recorded in the audit log.
“Release…” asks for the claim’s name as confirmation and never touches another pool or the cluster’s default capacity.
You pick the Scaleway instance type that fits the purpose – from a small node for a scanner to a large machine for compute jobs.
The lifecycle decides whether the pool disappears when idle or is always ready. An ephemeral claim lives as long as its consumer sends a heartbeat – “Touch (keep alive)” in the cockpit, the API’s touch call for a program. Through the API, a touch also revives a released claim.
| Lifecycle | Size | When it fits |
|---|---|---|
| Ephemeral | 0 to maximum, autoscaler on demand | Scans, batch runs, tests – anything with a start and an end |
| Persistent | Minimum to maximum, never below the minimum | Capacity that must always be ready, e.g. a pool for one tenant |
| Idle time | in minutes, ephemeral only | How long the heartbeat may be missing before the pool is released |
Pool claims work with Kapsule clusters that Clusterward created. Pods get onto the pool when they target it with a nodeSelector and tolerate the taint; all other pods stay away.
| Topic | How it is |
|---|---|
| Cluster | only Kapsule clusters created by Clusterward |
| Size | up to 50 nodes per pool |
| Isolation | up to 20 labels and 10 taints; NoSchedule, PreferNoSchedule or NoExecute |
| Release | Pods move to other pools if they fit; otherwise they wait |
| Cost | no extra charge at Clusterward; Scaleway bills the running nodes |
| Rights | Create and “Touch” with “Operate”, “Release…” with “Manage” in the provisioning area |
Pool claims complement the node pools you manage under Cluster updates by hand: there you control size and Kubernetes version, a claim takes care of capacity that comes and goes. How a cluster that runs claims is created is shown under Cluster provisioning.
For infrastructure of its own per customer, pool claims go with the tenant pipelines and the page for SaaS providers.
Kubernetes upgrade, node pools and add-ons from the cockpit.
Go to Cluster updatesKapsule clusters in minutes, secure by default.
Go to Cluster provisioningCustomer onboarding with a database, storage and domain of their own per tenant.
Go to SaaS providersYour question isn’t here? Write to us – we usually reply the same working day.
Ask a questionWe create an ephemeral claim for a scanner, show the label and taint on the nodes and let the pool disappear again when idle.
