Skip to content

Agent Helm values

The values of the seylo-agent chart, oci://ghcr.io/seylo-inc/charts/seylo-agent, as of 0.2.22. To see them for the version you install:

Terminal window
helm show values oci://ghcr.io/seylo-inc/charts/seylo-agent --version 0.2.22

The install page in Seylo fills in the required values for you. To change a value on an existing install, upgrade with --reset-then-reuse-values and --set the value, passing the image tag and digest too; see upgrade the agent.

Value Default What it does
serverUrl "" Required. Where the agent connects, such as wss://agent.seylo.in/agent/connect. http(s):// is upgraded to ws(s)://, and /agent/connect is added if missing
clusterId "" Required. The cluster’s ID, as shown on Admin → Clusters
enrollmentToken "" The one-time enrollment token. Stored in a Secret the chart creates. After the agent enrolls it keeps its own credential, and this value is no longer used
existingEnrollmentSecret.name, .key "", enrollmentToken Read the enrollment token from a Secret you created instead, for example one an external secret manager delivers
heartbeatSeconds 30 How often the agent reports. The server’s own cadence, sent when the agent connects, wins
connection.pingSeconds 15 How often the agent pings its connection (0.2.17 and later). 0 turns the ping off, for a network path that drops WebSocket control frames
connection.pongTimeoutSeconds 10 How long the agent waits for an answer before it reconnects. Keep it below pingSeconds
logLevel info The agent’s own log level
Value Default What it does
image.repository ghcr.io/seylo-inc/seylo-agent Change it only to install from your own mirror
image.tag "" Empty means the chart’s version. Set it with every upgrade that sets a digest: reused values keep an earlier tag, and the image is then named for the old version. See upgrade the agent
image.digest "" in the source; the chart published to the registry carries the digest of the image it was released with sha256: and 64 hex digits. With a digest, Kubernetes pulls exactly those bytes and ignores the tag. See pinning a digest
image.pullPolicy IfNotPresent
imagePullSecrets [] Pull Secrets for a private mirror, created in the agent’s namespace
Value Default What it does
rbac.create true Create the agent’s Roles and bindings
rbac.scope cluster cluster: one ClusterRole. namespaced: a Role in each namespace of watchNamespaces, and nothing outside them. See cluster-wide vs namespace-scoped
watchNamespaces [] The namespaces the agent manages. Optional with cluster scope (empty means all it can see); required with namespaced
rbac.posture "" read-only, deploy or admin, instead of setting the write tiers below one by one. Empty leaves the tiers below in charge, so an existing install keeps exactly the Role it has. No posture grants the RBAC tier. See choose an RBAC posture
rbac.genericRead.enabled false Broad read, for browsing any kind on Resources
rbac.genericRead.mode wildcard wildcard reads every kind, Secrets included at the API level; enumerated reads the core group without Secrets plus the groups you list. See browsing arbitrary resource kinds
rbac.genericRead.extraApiGroups [] With enumerated, the custom API groups that may be browsed

What the agent is able to change. Seylo’s own rules decide what a person may ask for; these decide what Kubernetes lets the agent do at all.

Value Default What it does
rbac.write.tier1.enabled true Create, update and delete the kinds an application team edits: workloads, Services, ConfigMaps, Secrets, Jobs, Ingresses, autoscalers and similar; create and patch (never delete) PersistentVolumeClaims and SecretProviderClasses; read ReplicaSets. See Secrets in the default grant
rbac.write.tier1.customResources [] The custom resources the agent may write, as an explicit list of {apiGroup, resources}. See custom resources
rbac.write.pods.enabled true Delete and evict a pod, and read pod logs
rbac.write.nodes.enabled false Cordon, uncordon and drain nodes
rbac.write.destructive.enabled false The destructive tier. Delete PersistentVolumeClaims, PersistentVolumes, StorageClasses, Namespaces, CustomResourceDefinitions and VolumeSnapshots. Needed for Nightshift to recreate a claim whose StorageClass does not exist; without it, that fix is refused and the claim is left as it was. The admin posture includes it
rbac.write.autoscaler.enabled false Create and update, never delete, your node autoscaler’s objects: Karpenter NodePools, AKS node auto-provisioning AKSNodeClasses, EKS EC2NodeClasses and GKE ComputeClasses. Needed for managed savings to run spot or better machine sizes. Cluster-wide installs only. No posture turns it on, admin included. Agent 0.2.21 or later
rbac.write.rbac.enabled false Write Roles, RoleBindings, ClusterRoles and ClusterRoleBindings, including ones that widen the agent’s own permissions. Read the warning first
rbac.write.rbac.allowEscalate true With the RBAC tier on, also grant escalate and bind. false grants RBAC write without them, so most useful changes are refused by the API server

Whatever you set, the chart never grants pods/attach, and never pods/exec except through sre.exec.enabled below.

Value Default What it does
credentialStorage.mode secret secret: the agent keeps its credential in its own Secret, and its RBAC for it is pinned to that one name. file: on the volume below, and the agent gets no Secret access for it
credentialStorage.secretName seylo-agent-credential The Secret, in secret mode
credentialStorage.file.path /var/lib/agent/credential.json The file, in file mode
credentialStorage.file.volume a PersistentVolumeClaim named seylo-agent Any volume. A claim survives rescheduling; an emptyDir does not, and the agent would need a new enrollment token after a restart
Value Default What it does
proxy.url "" A forward proxy the agent tunnels its connection through, such as http://proxy.corp.local:3128. Wins over HTTPS_PROXY in the pod
proxy.noProxy "" Hosts that bypass the proxy. A match always wins
proxy.existingSecret.name, .key "", proxyUrl Read the whole proxy URL, with its password, from a Secret instead of the values
ca.existingSecret, ca.key "", ca.crt A Secret holding a CA bundle to trust, for a private CA or a TLS-inspecting proxy
ca.pem "" The bundle inline; the chart creates the Secret. Use --set-file ca.pem=corp-ca.crt
tls.caSecret.name, .key "", ca.crt The older spelling of ca.existingSecret. ca wins when both are set
tls.insecureSkipVerify false Diagnostics only. Turns off certificate checks and logs a warning. Not the fix for an untrusted CA: set ca

See restricted networks.

Value Default What it does
observability.enabled true Look for the seylo-observability chart’s stores. With nothing installed, Seylo says “not installed”
observability.namespace seylo-observability Where the stores are
observability.metricsUrl, .logsUrl, .tracesUrl "" Point at a VictoriaMetrics, VictoriaLogs or VictoriaTraces you already run. The agent dials only these addresses
observability.opencostUrl "" OpenCost’s API (0.2.11 and later). Empty means the chart’s own OpenCost Service
observability.logsRehydratedUrl "" The temporary store archived log days are loaded into (0.2.16 and later). Empty means the chart’s own
datadog.enabled, .site, .clusterName, .existingSecret false, "", "", "" The optional Datadog connector (0.2.10 and later). The keys stay in a Secret in your cluster
connections.secretName seylo-connections The Secret holding keys for Grafana, Prometheus and Alertmanager connections made through the agent (0.2.15 and later)
sre.exec.enabled false Allow Nightshift’s fixed list of read-only commands in containers (0.2.10 and later). A Seylo admin must also allow it for the cluster
Value Default What it does
resources requests 25m CPU, 64Mi; limits 250m CPU, 256Mi The agent’s own resources
serviceAccount.create, .name, .annotations true, "", {} The agent’s ServiceAccount
podAnnotations, podLabels {} Added to the agent’s pod
nodeSelector, tolerations, affinity {}, [], {} Scheduling. The image runs on linux/amd64 and linux/arm64, so no architecture pin is needed
priorityClassName "" A PriorityClass for the agent’s pod
nameOverride seylo-agent Do not change. It is part of the Deployment’s selector, which cannot change on upgrade
fullnameOverride ""
mockK8s false For Seylo’s own tests. Never set it in a real cluster