Scaling and cache

Choose the replica count and size the directory snapshot cache.

Before you start

• Replicas above one need a shared State: config.ports.adapter: dynamodb or the k8s-aws preset. The chart refuses a controller with replicaCount above 1 on any other adapter.

• A single replica may run on memory, or legacy with no valkey.address. It starts cold every time and logs keeping snapshots in memory.

• Every replica serves every workspace and opens a console-connected one from its stored credential on first use.

Steps

1. Set replicaCount (default 2). For more than one replica follow high availability. Each pod logs keeping snapshots and the refresh lease in the state ports.

2. Estimate the snapshot size as the directories' size. The Blob port holds one gzip object google/<workspace> per workspace, of the order of the number of accounts and groups.

Verify

Read the object sizes in the bucket. After raising the count, watch sluis.leases.contended and SluisLeaseLost (telemetry).

Roll back

Set replicaCount back.

Decided in

Why more than one replica is safe.