bedrock

Resources and priorities

Every platform container has requests and limits. Requests are what the scheduler reserves; limits are the most the container can use before it is throttled, or killed for memory. Replica counts are left out, because they follow the number of nodes.

#Priority classes

When a node runs out of room, Kubernetes evicts lower priority pods first. bedrock defines four classes in tenant-workload/platform/priorityclasses.yaml, and the restrict-priority-classes policy stops anything outside their namespaces from claiming the higher ones.

ClassValueReserved for
bedrock-storage90000000Garage
bedrock-data80000000Database clusters, their poolers, Falco and the failover controller
bedrock-workload50000000Event buses
bedrock-default10000000Everything else. This is the global default.

Kubernetes components and Kyverno’s admission controller run at the built-in system classes, above all of these.

#Storage and data

WorkloadContainerRequestsLimitsPriority
Garagegarage100m, 256Mi1 CPU, 1Gibedrock-storage
Database clusterspostgres100m, 256Mi1 CPU, 1Gibedrock-data
Falcofalco100m, 256Mi1 CPU, 1Gibedrock-data
PgBouncer poolerspgbouncer50m, 64Mi500m, 256Mibedrock-data

#Failover controller

All run at bedrock-data.

WorkloadContainerRequestsLimits
Observers, a and bobserver20m, 32Mi200m, 128Mi
Decidersdecider20m, 48Mi300m, 192Mi
Verifierverifier20m, 48Mi300m, 192Mi
Checkerchecker10m, 24Mi100m, 96Mi

#Secrets

WorkloadContainerRequestsLimitsPriority
OpenBaoopenbao100m, 256Mi1 CPU, 512Mibedrock-default
OpenBaodb-ca-bundle5m, 32Mi100m, 128Mibedrock-default
CSI driversecrets-store50m, 100Mi200m, 200Mibedrock-default
CSI drivernode-driver-registrar, liveness-probe10m, 20Mi100m, 100Mibedrock-default
OpenBao CSI providerprovider-openbao50m, 100Mi200m, 200Mibedrock-default

#Monitoring

WorkloadContainerRequestsLimits
Prometheusprometheus100m, 400Mi1 CPU, 1Gi
Alertmanageralertmanager25m, 64Mi250m, 128Mi
Lokiloki100m, 128Mi500m, 512Mi
Tempotempo250m, 512Mi1 CPU, 2Gi
Grafanagrafana250m, 256Mi1 CPU, 1Gi
Grafanasc-dashboards10m, 64Mi250m, 256Mi
kube-state-metricskube-state-metrics50m, 64Mi500m, 256Mi
Blackbox exporterblackbox-exporter25m, 32Mi200m, 128Mi
OpenCostopencost10m, 64Mi500m, 1Gi
OpenCostopencost-ui10m, 32Mi100m, 128Mi
Alloyalloy50m, 128Mi500m, 256Mi
node-exporternode-exporter50m, 32Mi200m, 128Mi

#Platform

WorkloadContainerRequestsLimits
Reloaderreloader10m, 64Mi200m, 128Mi

#Event buses

These run only when turned on. All run at bedrock-workload.

WorkloadContainerRequestsLimits
NATSnats500m, 1Gi2 CPU, 2Gi
Valkeyvalkey250m, 512Mi1 CPU, 1Gi
Valkeysentinel50m, 64Mi200m, 128Mi
Kafkabroker500m, 2Gi2 CPU, 3Gi

#Changing a value

Edit the resources block in the component’s file and apply it. In application and node-agents, the require-resource-limits policy rejects any container without CPU and memory limits, so a limit can be changed but never removed.