Skip to main content

Kubernetes topology reference

MintPress is deployed to Kubernetes as a set of workloads in a single namespace, together with a small number of objects that live outside it. This page lists those objects and explains what each one does, so that you can tell at a glance whether what is running in your cluster is what you expect.

Not every object listed here is present in every installation. Several are optional, and are noted as such along with the values.yaml setting that controls them. Others are created only while MintPress is doing work and disappear again afterwards — see pods created while MintPress runs.

To see what your own installation is running:

kubectl -n ${KUBERNETES_NAMESPACE} get deployment,statefulset,daemonset,cronjob,job
kubectl -n ${KUBERNETES_NAMESPACE} get pods

Deployments​

NamePurposePresent
opschain-apiServes the OpsChain API, which the GUI and the CLI both use.Always
opschain-api-workerRuns MintPress's background jobs — starting changes, refreshing actions, requesting image builds, sending notifications. Despite the name, it does not run your steps — those run in separate pods.Always
opschain-build-serviceBuilds the runner and agent container images. Builds run inside this pod rather than in a pod of their own.Always
opschain-log-aggregatorForwards copies of stored log output to any additional outputs you have configured.Always
opschain-log-ingestStores the log output sent by the pods running your steps and agents, in place of the API. See log ingestion.Optional — logIngest.enabled. Only enable it if LimePoint support ask you to
mintpress-ingressThe cluster's entry point for the API, the GUI and the image registry. Runs the ingress controller and its proxy.Always
opschain-ldapThe bundled LDAP directory used to authenticate users.Optional — ldap.enabled. Disable it when using your own LDAP directory
opschain-mintmodel-apiRenders MintModels for templated assets and computes MintModel diffs. See MintModel API.Always
mintpress-reloaderRestarts a workload when a secret named in its secret.reloader.opschain.io/reload annotation changes. The secret vault uses it to restart when its addresses or the database password change.Always
opschain-db-pooler-rwA connection pooler in front of the database primary. See database connection pooling.Feature preview — db.cnpg.pooler.enabled
opschain-db-recoveryAn idle pod with the backup volume mounted read only, used to restore a backup. See database recovery.With backups enabled (the default), on the primary cluster only. Scaled to 0 unless db.backup.recovery.enabled is set
opschain-debug-toolboxAn idle pod containing diagnostic tooling. See debug toolbox.Always. Scaled to 0 unless debugToolbox.enabled is set

Stateful sets​

NamePurposePresent
opschain-image-registryThe internal container image registry that holds your built runner and agent images.Always. Its pod is opschain-image-registry-0
opschain-secret-vaultThe bundled secret vault, which stores your MintPress secrets. Its data is kept in the database rather than on a volume of its own.Always
Secret vault replicas

The secret vault runs a single replica, opschain-secret-vault-0. While it is being replaced during an upgrade, vault requests are briefly unavailable, and MintPress retries them. In a high availability setup each cluster runs its own secret vault: one holds leadership at a time and the others stand by.

Daemon sets​

NamePurposePresent
fuse-device-plugin-daemonsetGives the build service access to FUSE on older kernels that cannot run rootless builds with overlayfs.Optional — fuseDevicePlugin.enabled

The database​

The database is managed by the CloudNativePG operator rather than by a stateful set in the MintPress Helm chart, so it does not appear in a kubectl get statefulset listing. It is created from a Cluster resource named opschain-db, which produces one pod per instance:

kubectl -n ${KUBERNETES_NAMESPACE} get cluster
kubectl -n ${KUBERNETES_NAMESPACE} get pods -l cnpg.io/cluster=opschain-db
NamePurpose
opschain-dbThe Cluster resource describing the database.
opschain-db-1, opschain-db-2, …One pod per database instance. A single instance installation has only opschain-db-1.

The operator itself runs outside the MintPress namespace — see objects outside the MintPress namespace.

Cron jobs and jobs​

NameKindPurposePresent
opschain-db-backupCron jobTakes a scheduled database backup. See automated backups.With backups enabled (the default), on the primary cluster only. Suspended when db.backup.periodic.enabled is false or the database is stopped
opschain-registry-reconcileCron jobRemoves image tags from the internal registry that MintPress no longer needs. See container image cleanup.Always. Suspended when registryReconcile.enabled is false or MintPress is stopped
opschain-cnpg-webhook-ready-checkJobRuns at the start of every install and upgrade, and waits for the CloudNativePG operator to be reachable before MintPress's database resources are created or updated.Always
opschain-mintmodel-api-isolation-checkJobRuns at the start of every install and upgrade, and stops it if the nodes cannot isolate MintModel renders. See the isolation check.Optional — skipped when mintModelApi.isolation is never
opschain-runner-image-jobJobCopies the runner image into the internal image registry during an upgrade or rollback.Unless MintPress is stopped
opschain-db-pre-deploy-backupJobTakes a backup before an upgrade begins. A failure stops the upgrade.Optional — db.backup.preDeploy.enabled
opschain-runner-image-install-jobJobCopies the runner image into the internal image registry during an install.Install only
opschain-api-worker-drainJobRuns at the start of an upgrade that needs it, and stops the API workers before the database is migrated.Unless rolloutGuard.enabled is false
opschain-transportable-keyJobRuns at the start of every install and upgrade, and makes sure the MintPress transportable key secret exists.Always
patch-service-accountJobRuns at the start of an install, and adds MintPress's image pull secret to the namespace's default service account.Install only
opschain-image-warm-jobJobRuns after every install, upgrade and rollback, and pulls the images MintPress needs so the first change does not wait for them.Unless imageWarmJob.enabled is false
opschain-recovery-resetJobRuns before MintPress starts after a backup has been restored in a high availability setup. See resetting sites and work after a recovery.Only while db.backup.recovery.resetAfterRecovery is set
opschain-db-backup-initJobPrepares the backup volume the first time backups are enabled, then removes itself.With backups enabled, until the volume is ready
Completed pods you will see

opschain-cnpg-webhook-ready-check leaves its pod in the namespace in a Completed state after it finishes, so that its outcome can be inspected if an upgrade fails. The next install or upgrade replaces it. opschain-runner-image-install-job also leaves a completed pod, which remains until the first upgrade. The cron jobs above also leave a small number of completed pods behind, according to their history limits. Completed pods of either kind are expected and do not need to be cleaned up.

Pods created while MintPress runs​

This is where your changes actually run. The deployments listed above start and monitor this work rather than performing it themselves, so an installation can be busy without any of them looking any different.

MintPress creates these pods directly while it works. They are not created by a deployment, stateful set or job, so they never appear in a kubectl get deployment or kubectl get job listing — only in kubectl get pods, and only while they are running. A listing taken while an installation is idle therefore shows none of them.

NameCreated for
change-<change id>Each running change. This pod orchestrates the change and, unless pod_per_change_step is true, runs its steps too.
opschain-step-<step id>Only when pod_per_change_step is true: each step that runs an action in the step runner. A dry run adds a -dry suffix.
opschain-tva-<background task id>Each refresh of an asset's available actions.
executor-<digest>-0A pod shared by changes or action refreshes when the executor pool is enabled. Work that uses one has no change- or opschain-tva- pod of its own.
opschain-agent-<node id>Each running agent. Unlike the others, this pod is long lived.

Building a runner or agent image does not create a pod — image builds run inside the opschain-build-service pod.

These pods can be seen in the GUI as well as with kubectl, including after they have finished. See pods.

Services​

NamePurposePresent
opschain-apiThe API.Always
opschain-build-serviceThe image build service.Always
opschain-image-registryThe internal image registry.Always
opschain-log-aggregatorThe log aggregator.Always
opschain-log-ingestThe dedicated log ingest deployment.With opschain-log-ingest
mintpress-ingress-proxyThe external entry point for all MintPress traffic.Always
mintpress-ingress-validation-webhookUsed by Kubernetes to validate ingress configuration.Always
opschain-db-rwThe database primary.Always
opschain-db-rw-headlessA headless service that resolves straight to the database primary.Optional — db.cnpg.primaryHeadlessService.enabled
opschain-ldapThe bundled LDAP directory.With opschain-ldap
opschain-mintmodel-apiThe MintModel API.Always
opschain-db-pooler-rwThe database connection pooler.With opschain-db-pooler-rw
opschain-secret-vaultThe secret vault.Always
opschain-secret-vault-internalAddresses the vault replicas individually.Always
opschain-secret-vault-activeAddresses whichever vault replica currently holds leadership.Always
opschain-secret-vault-standbyAddresses the vault replicas that do not hold leadership.Always
opschain-db-externalDirect access to the database from outside the cluster.Optional — db.cnpg.externalService.enabled
opschain-secret-vault-externalDirect access to the secret vault from outside the cluster.Optional — secretVault.externalService.enabled

Storage​

Each of these is a persistent volume claim. Deleting one loses the data it holds, so treat them as you would any other persistent data — see persistent data.

NameHoldsPresent
opschain-db-1, opschain-db-2, …The database's data — one volume per database instance.Always
opschain-project-git-repos-claimThe Git repositories MintPress has cloned for your projects.Always
opschain-step-data-claimData shared between the pods running a change's steps.Always
opschain-build-service-cache-claimThe image build cache.Always
opschain-log-aggregator-volumeLog output buffered before it reaches your additional outputs.Always
data-vol-opschain-image-registry-0The images in the internal registry.Always
opschain-ldap-claimThe bundled LDAP directory's data.With opschain-ldap
opschain-db-backupDatabase backups.With backups enabled (the default), on the primary cluster only

The secret vault has no volume of its own — it stores its data in the MintPress database.

Configuration and secrets​

opschain-config is the config map holding the settings your installation was deployed with, and nearly every MintPress workload reads it. The encryption keys and passwords from your values.yaml are in the opschain-secret-env secret. The database password is in the opschain-db-credentials secret. The other config maps hold configuration for a single component: opschain-build-service-config, opschain-log-aggregator-additional-output-config, opschain-secret-vault-config, opschain-ssh-known-hosts (the Git host keys MintPress trusts) and opschain-trust-store (the certificate authorities you have uploaded).

Secrets are created both by the installation and by the components that manage certificates, so the set present in your namespace varies. The ones that hold data you supplied, and that are preserved across an uninstall, are described in persistent data.

Objects outside the MintPress namespace​

ComponentNamespaceInstalled by
CloudNativePG operatorcnpg-systemApplied with kubectl, separately from the MintPress Helm chart. See install the CNPG operator.
cert-managercert-managerInstalled with Helm, separately from the MintPress Helm chart, when MintPress manages its own certificates.

Because the CloudNativePG operator is installed separately, a helm upgrade does not update it. A MintPress release that changes the operator says so under Before upgrading in the release notes.

Cluster-scoped objects​

A small number of objects are not namespaced. These matter mainly when uninstalling, and when deciding how many MintPress installations a cluster can host.

  • opschain-api-cluster-role, opschain-worker-cluster-role, mintpress-ingress and their bindings, plus opschain-secret-vault-server-binding — the cluster-wide access the API, the API workers, the ingress and the secret vault need.
  • The priority classes opschain-system, opschain-runner and opschain-backup, unless priorityClasses.enabled is false.
  • An ingress class and a validating webhook configuration for the ingress.
  • Custom resource definitions for the ingress, and for the CloudNativePG operator.
One MintPress installation per cluster

These objects are named without reference to the Helm release, so a second MintPress release in the same cluster fails to install because those names are already taken. Install at most one MintPress release per cluster.