metalbear-co/mirrord / skills/skills/mirrord-operator
mirrord-operator
VerifiedHelp users install and configure the mirrord Operator for team/enterprise environments. Use when users ask about operator setup, Helm installation, cloud API key or license configuration, air-gapped/offline licensing, enabling features (queue splitting, DB branching, preview environments, multi-cluster), internal registries, OpenShift/GKE Autopilot, RBAC, or multi-user mirrord deployments. Also use when an AI agent hits a Team or Enterprise feature on a cluster with no license and needs to offer the user a trial (agent-started trial; the trial is Enterprise tier).
| File | Kind | Size |
|---|---|---|
| README.md | markdown | 2.9 KB |
| SKILL.md | skill md | 28.0 KB |
| references/helm-values.md | markdown | 15.4 KB |
| references/troubleshooting.md | markdown | 5.9 KB |
| references/values.yaml | config | 76.7 KB |
references/values.yaml
View on GitHub# Default values for mirrord-operator.
# This is a YAML-formatted file.
# Declare variables to be passed into your templates.
namespace: mirrord
## Whether we should create the namespace or not
## If you set this to false, you must create the namespace manually
createNamespace: true
## Use a list to create one unbound mirrord role in each namespace.
## Use a mapping to create a role and binding for each target and its users or groups.
## Target names must match the operator API name, for example deployment.api.
roleNamespaces:
# roleNamespaces:
# my-namespace:
# deployment.api:
# - type: User
# subject: developer@example.com
# - type: Group
# subject: api-team@example.com
## Limit the operator's own permissions to some namespaces.
##
## By default (`false`), the `mirrord-operator` ClusterRole lets the operator work in all
## namespaces. When `true`, the ClusterRole keeps only the rules for cluster-scoped resources
## (for example nodes, SubjectAccessReviews, MutatingWebhookConfigurations and the mirrord
## `MirrordCluster*` resources). The rules for namespaced resources (for example pods, deployments
## and secrets) go in a `mirrord-operator-namespaced` Role and RoleBinding in each namespace of
## `allowedNamespaces` and in the operator namespace (`namespace`).
##
## The operator then refuses sessions and copy targets in other namespaces. The namespaces must
## exist before you install the chart. These namespaces must also be in the list:
## - the agent namespace, if you set `agent.extraConfig.namespace`;
## - the Argo CD namespace, if you set `operator.applicationPauseAutoSync`;
## - the Flux namespaces, if you set `operator.suspendFluxControllers`.
##
## Not supported with `operator.multiCluster.enabled` or `operator.multiClusterMember`.
namespaced: false
## The namespaces where the operator can work when `namespaced` is `true`.
allowedNamespaces: []
## Additional Kubernetes resources to create with the chart. Entries are rendered as templates,
## so they can reference chart values and release information.
extraObjects: []
role:
# Note there's a cluster role version of it below.
mirrord-operator-user:
# add labels to the role, for i.e aggregate RBAC
labels: {}
clusterRole:
mirrord-operator-user-basic:
# add labels to the role, for i.e aggregate RBAC
labels: {}
mirrord-operator-user:
# add labels to the role, for i.e aggregate RBAC
labels: {}
mirrord-operator-ci:
# add labels to the role, for i.e aggregate RBAC
labels: {}
# Created only with `operator.sessionsManager` enabled.
mirrord-operator-sessions-manager-agent:
# add labels to the role, for i.e aggregate RBAC
labels: {}
operator:
image: ghcr.io/metalbear-co/operator
# Number of replicas of the operator's deployment.
# At all times, only one of the replicas acts as the leader and serves mirrord sessions.
# Additional replicas remain in a standby mode,
# ready to resume the work in case the leader exits (e.g. due to node failure).
replicas: 1
# How long Kubernetes waits for an operator pod to exit after asking it to stop.
# The operator drains its tasks and then releases leadership, needing a little
# over 20 seconds.
terminationGracePeriodSeconds: 25
# Override the pull policy, useful for local development.
# imagePullPolicy: Always
# Set this value to specify a tag to use instead of the `appVersion`.
# imageTag: test
podAnnotations: {}
podLabels: {}
## Custom labels to be added to all Kubernetes resources created by this chart.
## These labels will be applied to higher-level resources like Deployments, Services, etc.
## For pod-specific labels, use operator.podLabels instead.
## Example:
## labels:
## team: platform
labels: {}
## Custom annotations to be added to all Kubernetes resources created by this chart, except the
## CustomResourceDefinitions. Those are left unannotated so that an ordering hint set here cannot
## delay them past the workloads that depend on them.
## For pod-specific annotations, use operator.podAnnotations instead.
## GitOps tools read annotations off the resources themselves, so this is where a sync-wave or a
## similar ordering hint belongs; podAnnotations never leaves the pod template.
## Example:
## annotations:
## argocd.argoproj.io/sync-wave: "1"
annotations: {}
## Optional human-friendly name for this cluster.
## When set, it is reported on the operator's startup telemetry events, so the cluster shows up
## with a recognizable label (instead of only an opaque identifier) in the license server and in
## the mirrord cloud. Leave unset to omit it.
# clusterName: "staging-us-east"
jsonLog: true
# Leave the full HTTP request headers (`request_headers`) and queue message properties
# (`message_properties`) out of the operator's `Message Processing` logs. Set this to `true` when
# headers or properties can contain secrets. The logs still show the trace and correlation
# values (`traceparent`, `tracestate`, `baggage`, `correlation_id`).
hideHeadersAndProperties: false
# Require Kubernetes front-proxy client certificates on operator API routes.
# The operator validates the certificate chain using the front-proxy CA from the cluster's
# extension-apiserver-authentication configmap. Disable this only for environments where
# requests reach the operator without kube-apiserver/front-proxy client certificates, for
# example local development, custom API routing, or debugging through traffic interception.
requireFrontProxyClientCert: true
# Define additional environment variables for the operator. Note that variables which reference
# other variables must be defined _after_ those that they reference, as variables are added to
# the manifest in order.
extraEnv:
# MY_ENV_VAR: "value"
# HOST_IP:
# valueFrom:
# fieldRef:
# fieldPath: status.hostIP
# MY_REFERENCED_VAR: "$(HOST_IP):$(MY_ENV_VAR)"
# Define additional volume mounts for the operator container.
# extraVolumeMounts:
# - name: my-secret
# mountPath: /etc/my-secret
# readOnly: true
# Define additional volumes for the operator deployment.
# extraVolumes:
# - name: my-secret
# secret:
# secretName: my-secret
# Has to be set to `true` in order to use the SQS queue splitting feature.
sqsSplitting: false
# Has to be set to `true` in order to use the Kafka queue splitting feature.
kafkaSplitting: false
# Has to be set to `true`, along with `kafkaSplitting`, in order to use Kafka branching
# (`feature.kafka_branch`).
kafkaBranching: false
# Has to be set to `true` in order to use the RabbitMQ queue splitting feature.
rmqSplitting: false
# Has to be set to `true` in order to use the GCP Pub/Sub queue splitting feature.
gcpPubsubSplitting: false
# Has to be set to `true` in order to use the Azure Service Bus queue splitting feature.
azureServiceBusSplitting: false
# Has to be set to `true` in order to use the Redis Pub/Sub queue splitting feature.
redisPubsubSplitting: false
# Has to be set to `true` in order to use the Temporal queue splitting feature.
temporalSplitting: false
# Has to be set to `true` in order to use the BullMQ queue splitting feature.
bullmqSplitting: false
# Has to be set to `true` in order to use the NATS queue splitting feature.
natsSplitting: false
# Has to be set to `true` in order to use the core NATS (non-JetStream) pub/sub
# queue splitting feature. Best-effort delivery: nothing is persisted.
natsPubsubSplitting: false
# Has to be set to `true` in order to serve the operator-hosted sessions-manager
# (`serverlessagentassignments`, `serverlessclientassignments` and `serverlessdataplanes`),
# which pairs a workload companion's agent with a developer's local mirrord. Grants `proxy` on
# `serverlessclientassignments` and `get` on `serverlessdataplanes` to the mirrord user
# ClusterRoles (see `operator.sessionsManagerUserRbac`), and creates the
# `mirrord-operator-sessions-manager-agent` ClusterRole with
# `proxy` on `serverlessagentassignments` and `get` on `serverlessdataplanes`, to bind to the
# identity the workload companion's agent authenticates to the cluster as.
#
# Registrations are named by the scope they pair in, `<environment>.<service>`, so either side
# can be limited to some scopes with `resourceNames` on its registration rule. Data planes are
# named by per-session allocation ids, so that rule can't be narrowed.
#
# The `mirrord-operator-sessions-manager-agent` ClusterRole covers every scope. To limit an agent
# identity to one, bind it to your own ClusterRole instead:
# - apiGroups: [operator.metalbear.co]
# resources: [serverlessagentassignments]
# resourceNames: [staging.api]
# verbs: [proxy]
# - apiGroups: [operator.metalbear.co]
# resources: [serverlessdataplanes]
# verbs: [get]
#
# The mirrord user ClusterRoles also cover every scope: a developer bound to one can pair with the
# agent of any environment and service and receive that workload's traffic. To limit developers
# to some scopes, grant `proxy` on `serverlessclientassignments` with `resourceNames` (e.g.
# `[staging.api, staging.web]`) and `get` on `serverlessdataplanes` through your own ClusterRole.
# RBAC is additive, so this only narrows access for developers who aren't also bound to a mirrord
# user ClusterRole: set `operator.sessionsManagerUserRbac: false` to leave client access out of
# those roles.
#
# Sessions live in the operator's memory, so an operator rollout or restart ends them: the
# stopping operator closes every assignment stream and data plane, and both sides register again
# with the new one. That drain fits in the default `operator.terminationGracePeriodSeconds`. If
# rollouts cut connections abruptly instead, raise `operator.terminationGracePeriodSeconds` back
# to at least its default, so the operator isn't killed before it finishes.
sessionsManager: false
# With `operator.sessionsManager` on, grants the mirrord user ClusterRoles
# (`mirrord-operator-user`, `-ci` and `-user-basic`) client access to the operator-hosted
# sessions-manager for every scope: `proxy` on `serverlessclientassignments` and `get` on
# `serverlessdataplanes`. Set to `false` to leave it out and grant developers client access
# through your own ClusterRoles, scoped with `resourceNames` (see `operator.sessionsManager`).
# The agent ClusterRole is not affected.
sessionsManagerUserRbac: true
# Temporal queue splitting proxy settings.
temporalProxy:
# Port for the Temporal frontend proxy workers connect to via `TEMPORAL_ADDRESS`.
port: 7233
# Has to be set to `true` in order to use the argocd application auto-sync pause feature.
applicationPauseAutoSync: false
# Has to be set to `true` in order to use the suspend flux controllers feature.
suspendFluxControllers: false
# Has to be set to `true` in order to let the operator modify KEDA ScaledObjects.
manageKedaScaledObjects: false
# Deprecated: use `manageKedaScaledObjects`, which this still enables.
pauseKedaScaleIn: false
# Has to be set to `true` in order to use the MySQL database branching feature.
mysqlBranching: false
# Has to be set to `true` in order to use the MariaDB database branching feature.
mariadbBranching: false
# Has to be set to `true` in order to use the PostgreSQL database branching feature.
pgBranching: false
# Has to be set to `true` in order to use the MongoDB database branching feature.
mongodbBranching: false
# Has to be set to `true` in order to use the DynamoDB database branching feature.
dynamodbBranching: false
# Has to be set to `true` in order to use the MSSQL database branching feature.
mssqlBranching: false
# Has to be set to `true` in order to use the Redis database branching feature.
redisBranching: false
# Has to be set to `true` in order to use the Google Cloud Spanner database branching feature.
spannerBranching: false
# Has to be set to `true` in order to use the ClickHouse database branching feature.
clickhouseBranching: false
# Has to be set to `true` in order to use the CockroachDB database branching feature.
cockroachdbBranching: false
# Has to be set to `true` in order to use the generic (user-supplied image) database
# branching feature. NOTE: enabling this lets users who can create branches run arbitrary
# container images as branch pods (under the namespace default service account). Restrict
# the images with `genericBranchConfig.dbPod.allowedImages` if needed; when that field is
# absent, all images are allowed.
genericBranching: false
# Has to be set to `true` in order to use the S3 bucket branching feature. The operator
# clones buckets in the customer's cloud account with its own AWS credentials (from the
# service account's IAM role annotation); there is no branch pod and no `s3BranchConfig`.
s3Branching: false
# Has to be set to `true` in order to use the turbopuffer namespace branching feature. The
# operator branches namespaces through turbopuffer's API with the API key it reads off the
# target workload; there is no branch pod and no `turbopufferBranchConfig`.
turbopufferBranching: false
# Has to be set to `true` for db branches to run in a namespace of their own per session
# key (`feature.db_branches[].namespace: true` creates `mirrord-<key>`). Grants the
# operator `create`, `list`, `patch` and `delete` on namespaces cluster-wide, and `get`/`patch` on
# ServiceAccounts and ConfigMaps to copy what the branch pod reads into that namespace. The
# operator only deletes namespaces it created (labelled
# `operator.metalbear.co/db-branch-namespace: "true"`). Has no effect with `namespaced: true`.
dbBranchNamespaces: false
# Exactly one {{key}} placeholder, with up to 53 lowercase letters, digits or hyphens
# around it. For example, "t-{{key}}" creates t-alice for key alice. Invalid key
# characters and long names are normalized with a hash suffix. Existing branches keep
# their namespace when this changes; new branches use this pattern.
dbBranchNamespacePattern: "mirrord-{{key}}"
# Settings for the turbopuffer namespace branching feature.
turbopufferOptions:
# Domains a turbopuffer branch may send its API key to, on top of `turbopuffer.com`, which
# is always allowed. Set this only for a dedicated cluster served from another domain: a
# branch names its own endpoint, so anything listed here is a host the operator will hand
# the resolved API key to.
allowedHosts: []
# Grants cluster-wide Secret permissions needed when users pass literal
# database credentials in the mirrord config instead of referencing
# existing K8s Secrets.
dbBranchingLiteralCredentials: true
# Enables the preview environments feature. Set to `false` to leave out the PreviewSession CRD
# and the operator permissions that preview environments need.
previewEnv: true
# Records which Services mirrord sessions talk to, for the cluster topology view in the
# dashboard. Makes the operator watch every Service and EndpointSlice in the cluster, and
# grants it read access to EndpointSlices.
topology: false
# Sharing preview environments and local sessions (`mirrord exec --share`) via a plain link.
# Requires the `mirrord-share-ingress` chart deployed with a matching `shareDomain`. Previews
# also require `previewEnv: true`.
shareIngress:
# Domain that preview and local share hosts are minted under, as `<slug>.<shareDomain>`,
# entered without a leading "*." wildcard. When empty, no share host is minted and neither
# previews nor local sessions are shareable by link.
shareDomain: ""
# Mint share hosts without the random suffix, as `<sanitized key>.<shareDomain>` (the key
# lowercased, other characters turned into "-"), so a link can be built from the session key
# alone - for example by a PR bot from the PR number. One live session per host on the
# cluster: a second session whose key gives the same host fails at creation, naming the
# session that holds it. Sessions that already have a host keep it. The links become
# guessable, so only enable this when the ingress in front of `mirrord-share-ingress`
# authenticates every request.
# Local share hosts always get the random suffix, whatever this is set to.
stableSlugs: false
# Address the operator uses to ask `mirrord-share-ingress` whether it can route local share
# links. Empty uses the `mirrord-share-ingress` Service in the release namespace.
serviceUrl: ""
preview:
# Annotations added to the pods created by preview environments.
annotations: {}
# Labels added to the pods created by preview environments.
labels: {}
# How long to retain "failed" preview sessions before deleting them.
#
# Note that this option controls the time that the session's _internal custom resource_ remains in the cluster, so
# that it's visible to the `mirrord preview` family of commands for inspection. The workload itself is deleted as
# soon as failure is detected.
#
# "Failed" in this case means an operator-side error, such as failing to create the necessary kubernetes resources,
# failing to create DB branches, etc.
#
# Application-side failures such as crashes and evictions are handled automatically by kubernetes through pod
# restarts and the underlying Deployment, this configuration doesn't change that.
cleanupAfterMins: 15
# Bounds on the traffic an idle or waking preview session buffers while its pods boot.
#
# Sessions with idle mode enabled hold incoming traffic instead of failing it while their pods scale up from zero.
# These cap that buffer (per session): when either limit is reached the buffer is flushed and the held requests
# fail with 502 instead. `idleHoldBufferMessages` counts mirrord protocol frames (a whole HTTP request, one body
# chunk of a streamed request, or one data segment of a plain TCP connection — not necessarily one HTTP request);
# `idleHoldBufferBytes` bounds the actual payload memory, since a single frame can carry a large body.
idleHoldBufferMessages: 512
idleHoldBufferBytes: 8388608
# The pull policy to use for preview environment's images.
# imagePullPolicy: "Always"
# Code sync: previews from `mirrord preview sync` for services with `sync` run their image, or
# the target's, on code that the CLI pushes into the preview pod.
# codeSync:
# storage:
# # Where the pushed code lives. `pvc` (the default) keeps it on a claim of the session,
# # so it survives a replaced pod and a session that scales to zero. `emptyDir` needs no
# # StorageClass, but the code is deleted with the pod: the CLI must push again after
# # the pod is replaced, and sessions never scale to zero. An upgrade to another operator
# # image replaces the pods of sync previews, so it fails each `emptyDir` session that no
# # `mirrord preview sync --watch` is attached to, and its code must be pushed again with
# # `mirrord preview sync`. A preview keeps the kind that it started with.
# kind: pvc
# # The StorageClass of the claim, for example `standard-rwo` on GKE. Unset uses the
# # cluster's default class, and a session fails if the cluster has none.
# # storageClassName: standard-rwo
# # The size of the claim, or the size limit of the `emptyDir`. Defaults to 2Gi. A push
# # that changes most of the tree needs room for the old and the new tree at once, so it
# # should be about twice the size of the pushed tree. The kubelet evicts a pod whose
# # `emptyDir` goes past its limit.
# size: 2Gi
# # The fsGroup of a preview pod when neither the pod nor its app container names a
# # `runAsGroup`, `runAsUser` or `fsGroup`, so that a non-root app can write to the code
# # volume. The fsGroup applies to every volume of the pod. Defaults to 65532. `null` sets
# # none, for platforms that assign groups themselves.
# fsGroup: 65532
# # The most bytes, unpacked, that one push of `mirrord preview sync` may write to the code
# # volume. Unset allows as much as the volume holds.
# # maxPushBytes: 1073741824
# # Resources of the `mirrord-sync` sidecar and its `mirrord-sync-install` init container.
# # Unset uses these values.
# sidecarResources:
# requests:
# cpu: 10m
# memory: 32Mi
# limits:
# memory: 256Mi
# dbBranching:
# # Size of the init volume that stages the copy. On emptyDir
# # (`dbPod.storage.kind: emptyDir`, or the automatic fallback on clusters without a
# # default StorageClass) this is the size limit; on the default PVC backing it sizes
# # the claim, unless `initPvcSize` below is also set.
# initPodVolumeLimit: "100Mi"
# # Size of the volume holding the actual branch database. Same scope as above
# # (`databasePvcSize` wins on the PVC path).
# databasePodVolumeLimit: "1G"
# # Requested size of the per-branch data PVC (the default storage backing).
# # Provisioned per branch and deleted with it. Defaults to `databasePodVolumeLimit`
# # if set, else 20Gi.
# databasePvcSize: "20Gi"
# # Requested size of the per-branch init (dump) PVC. The staged dump is roughly the
# # size of the copied data. Defaults to `initPodVolumeLimit` if set, else 20Gi.
# initPvcSize: "20Gi"
# When set to `true`, the operator will use a custom strategy when restarting target workloads.
# This strategy can be used if standard restart procedure is not working for the target workload.
isolatePodsRestart: true
# Enable prometheus metrics endpoint
metrics: false
# When enabled, operator will inject a `mirrord-key` header with the
# value of the current session key into redirected HTTP requests,
# responses to said requests, and the queue-splitting messages routed to
# a session. For queues it uses a native metadata field where the broker
# has one (SQS, Google Pub/Sub, RabbitMQ, Kafka, Azure Service Bus, NATS,
# and Temporal activity tasks), otherwise the JSON payload (BullMQ, Redis
# Pub/Sub); non-JSON payloads are left untouched.
injectSessionKeyHeader: true
# Depth of the single, global `mirrord subscribe` event buffer (shared across all sessions).
# If a subscriber lags by more than this many events, the oldest are dropped and it is sent a
# "lagged" notice. Raise for bursty, slow-consumer cases.
subscribeEventBufferSize: 2048
# Multi-cluster configuration (Envoy)
# Enable this on the PRIMARY cluster only
#
# ## How it works
#
# Clusters are discovered from Secrets with label `operator.metalbear.co/remote-cluster-credentials=true`.
# Each Secret contains credentials for a remote cluster:
# - name: Logical cluster name (e.g., "staging-us-east")
# - server: Kubernetes API server URL
# - caData: CA certificate (base64 encoded)
# - bearerToken: ServiceAccount token
# - isDefault: "true" for the default cluster (stateful operations)
#
# Tokens are auto-refreshed using the TokenRequest API every 45 minutes.
#
# ## Setup
#
# 1. Create a ServiceAccount on each remote cluster (see remoteClusterSetup below)
# 2. Get the token and CA from each remote cluster
# 3. Create a Secret on the primary cluster for each remote cluster
# 4. Enable multiCluster.enabled = true on the primary cluster
#
multiCluster:
# Set to true to enable multi-cluster session orchestration and make this operator
# the primary one.
enabled: false
# CLUSTER NAME: Name of this cluster (optional, defaults to "primary")
# Used for logging and session tracking
# clusterName: "primary"
# DEFAULT CLUSTER: Where stateful operations happen (REQUIRED when multi-cluster enabled)
# - DB branches are created here
# - File operations read/write from here
# - Outgoing traffic originates from here
# Must match the `name` of one of your cluster Secrets
defaultCluster: ""
# Whether the primary cluster is management-only (no workloads)
# When true: Primary only runs Envoy, all sessions go to remote clusters
# When false: Primary is also a workload cluster
managementOnly: true
# These three are commented out on purpose. `remoteSessionTimeoutSeconds` and
# `sessionTtlSeconds` replace `remoteSessionTimeoutSecs` and `sessionTtlSecs`, and both
# spellings are honored. Shipping a value here would make the new name always set, so a values
# file still using the old name would be ignored on upgrade. The defaults below live in the
# ConfigMap template instead.
# Timeout for waiting for remote sessions to become ready (seconds). Defaults to 300.
# remoteSessionTimeoutSeconds: 300
# TTL for multi-cluster sessions (seconds). Defaults to 60.
# Sessions are deleted when no client has been connected for this duration.
# Only applies once a client has connected; until then sessionSetupDeadlineSeconds applies.
# sessionTtlSeconds: 60
# How long a multi-cluster session may wait for its first client connection (seconds).
# Defaults to 180.
# Sessions are created on every cluster before any client connects, and setting them up can
# take a while when queue splitting waits for target pods to roll. Deleting a session also
# deletes its child sessions on every cluster, so this must outlast setup.
# sessionSetupDeadlineSeconds: 180
# Preview environment behavior across the fleet.
preview:
# Where a preview's pods run.
#
# default-cluster (default) - pods on the DEFAULT cluster only
# replicas - a full replica on EVERY workload cluster: own pods,
# own local HTTP steal, own local queue split
#
# Queue-splitting previews apply their splits on every cluster in BOTH modes - so
# matched messages from anywhere keep reaching the preview either way; the modes only
# differ in where pods run.
#
# Use `replicas` when:
# - previews must catch baggage-tagged HTTP arriving through your load balancer,
# which can land on ANY cluster - with default-cluster, matched requests landing
# on other clusters are served by the deployed app instead of the preview
# - previews should survive a cluster outage (the LB fails over onto replicas
# that already exist and are already serving)
# Stay on `default-cluster` when:
# - previews are exercised via their share link or requests you steer yourself -
# that traffic enters one cluster anyway, so extra replicas would sit unused
# - queue splitting is the main feature (its coverage is identical either way)
# - you want the smallest footprint: one pod, no DB tunnel
mode: default-cluster
# Max concurrent branch-DB connections one workload cluster's replicas may hold open
# through the branch tunnel. Only applies with `mode: replicas` AND database branching:
# replicas on non-default clusters reach branch DBs over the tunnel, and each live
# connection holds one stream plus one port-forward on the default cluster's apiserver.
#
# This caps CONNECTIONS, never previews or branches. Size it as
# concurrently running branching previews x their app's DB pool size
# (an idle preview holds none - its pods, and so its connections, are gone). Excess
# opens are rejected and surface as DB connect failures the drivers retry.
#
# The default suits roughly 25-50 concurrently connected replicas per cluster with
# typical pool sizes. Raising it trades more concurrent port-forwards on the default
# cluster's apiserver.
maxTunnelStreams: 256
# Remote cluster configuration
#
# Each cluster requires `authType` to specify how it authenticates:
#
# authType: eks — AWS IAM authentication for EKS clusters
# → Tokens generated locally using the operator's IAM role (IRSA/Pod Identity)
# → Auto-refreshed every 10 min (15-min token lifetime)
# → No secrets to manage — safe to commit to Git
# → Requires `region` field and EKS Access Entry on each remote cluster
# → The cluster key name must match the EKS cluster name
#
# authType: bearerToken — ServiceAccount token
# → Auto-refreshed via TokenRequest API
# → No manual rotation needed
#
# authType: mtls — mTLS client certificate auth
# → NOT auto-refreshed (certs typically valid for months/years)
# → You must manually rotate before expiry
#
# Example with EKS IAM auth (cluster key = EKS cluster name):
# clusters:
# my-eks-cluster: # ← must match the EKS cluster name
# authType: eks
# server: https://ABCDEF.gr7.us-east-1.eks.amazonaws.com
# region: us-east-1
# # caData is optional for EKS — EKS uses Amazon-signed certs trusted by default.
# # caData: LS0tLS1CRUdJTi...
# isDefault: true
#
# EKS IAM setup on each remote cluster:
# 1. Install mirrord operator on the remote cluster with:
# --set operator.multiClusterMember=true \
# --set operator.multiClusterMemberIamGroup=mirrord-operator-envoy
# This creates both the ClusterRole and ClusterRoleBinding for the IAM group.
# 2. Create EKS Access Entry:
# Via AWS Console:
# EKS → select remote cluster → Access tab → Create access entry
# → IAM principal: select the IAM role from sa.roleArn → Type: Standard
# → Groups: mirrord-operator-envoy → Next → skip access policies → Create
# Via AWS CLI:
# aws eks create-access-entry --cluster-name <EKS_CLUSTER> \
# --principal-arn <sa.roleArn value> \
# --type STANDARD --kubernetes-groups mirrord-operator-envoy
#
# Example with AKS Workload Identity auth:
# clusters:
# remote-aks:
# authType: aks
# server: https://remote-aks-xxxxx.hcp.eastus.azmk8s.io:443
# # caData is required for AKS (each cluster has its own self-signed CA).
# # Get it from: az aks get-credentials --name <cluster> -g <rg> --file /dev/stdout | grep certificate-authority-data
# caData: LS0tLS1CRUdJTi...
# # tenantId and clientId are optional -- defaults to env vars injected by webhook
# # tenantId: "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
# # clientId: "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
#
# AKS` Workload Identity setup:
# 1. Create a User-Assigned Managed Identity in Azure
# 2. Create a Federated Identity Credential linking the operator's SA to the identity:
# az identity federated-credential create \
# --name mirrord-operator-federation \
# --identity-name <managed-identity-name> \
# --resource-group <rg> \
# --issuer $(az AKS show -n <primary-cluster> -g <rg> --query oidcIssuerProfile.issuerUrl -o tsv) \
# --subject "system:serviceaccount:mirrord:mirrord-operator" \
# --audiences "api://AzureADTokenExchange"
# 3. Grant the identity access on each remote AKS cluster (Azure RBAC or K8s RBAC)
# 4. Set sa.azureClientId to the Managed Identity's client ID
# 5. Install mirrord on each remote cluster with:
# --set operator.multiClusterMember=true \
# --set operator.multiClusterMemberAzureGroup=mirrord-operator-envoy
#
# Example with bearer token:
# clusters:
# staging-us-east:
# authType: bearerToken
# server: https://staging-us-east.example.com:6443
# caData: LS0tLS1CRUdJTi... # base64 encoded CA cert
# isDefault: true
# bearerToken: eyJhbGci... # initial token (will be auto-refreshed)
#
# Example with mTLS:
# clusters:
# secure-cluster:
# authType: mtls
# server: https://secure.example.com:6443
# caData: LS0tLS1CRUdJTi... # CA cert
# tlsCrt: LS0tLS1CRUdJ... # client cert PEM
# tlsKey: LS0tLS1CRUdJ... # client private key PEM
clusters: {}
# Set to `true` if installing the mirrord-operator on a cluster that is part of a connected
# multi-cluster setup, and this one is not the primary.
multiClusterMember: false
# When using IAM auth for multi-cluster, set this to the Kubernetes group name
# used in the EKS Access Entry. This creates a ClusterRoleBinding that grants
# the IAM group the same permissions as the ServiceAccount-based setup.
# Only used when multiClusterMember is true.
# multiClusterMemberIamGroup: "mirrord-operator-envoy"
# When using AKS Workload Identity for multi-cluster, set this to the Azure AD
# group or object ID used for the ClusterRoleBinding on each remote AKS cluster.
# This creates a ClusterRoleBinding that grants the Azure identity the same
# permissions as the ServiceAccount-based setup.
# Only used when multiClusterMember is true.
# multiClusterMemberAzureGroup: "mirrord-operator-envoy"
# imagePullSecrets:
# - name: value
## stop using container args for agent config, as now you can just use agent.extraConfig
## also imagePullSecrets is re-used from operator.imagePullSecrets
# containerArgs:
# - '--agent-config'
# - '{"image_pull_secrets":[{"name":"docker"}]}'
## You can use this only if using Enterprise license.
disableTelemetries: false
# Sets log level for the logs printed to `stdout` by the operator container.
#
# See reference: https://docs.rs/tracing-subscriber/latest/tracing_subscriber/filter/struct.EnvFilter.html#directives.
logLevel: mirrord=info,operator=info,kube_runtime=warn
## Sets the verbosity of everything the operator exports over OTel: the logs sent to
## `otelLogExportUrl` and the traces sent to `otelTraceExportUrl`. Defaults to `info`.
## This is independent of the logs printed to `stdout` (controlled by `logLevel`).
##
## See reference: https://docs.rs/tracing-subscriber/latest/tracing_subscriber/filter/struct.EnvFilter.html#directives.
# otelLogLevel: mirrord=info
## The URL for OTel logs to be exported to in the cluster. If unset, logs will not be exported.
## This is independent of the logs printed to `stdout` (controlled by `logLevel`).
## This value may reference values in `operator.extraEnv`, as it appears later in the manifest.
# otelLogExportUrl: https://loki:3100/loki/app/v1/push
## The URL for OTel traces to be exported to in the cluster. If unset, traces will not be exported.
## This value may reference values in `operator.extraEnv`, as it appears later in the manifest.
# otelTraceExportUrl: https://tempo:4318/v1/traces
## Controls how long (in seconds) a session is allowed to live for.
# maxSessionTimeSeconds: 3600
## Controls how long (in seconds) a session that a client has already connected to may go
## unused before it is closed. Defaults to 30 seconds.
## Raise this when clients are disconnected by slow or unreliable networks. The cost is that
## agents and patched workloads are held longer after a client really does go away.
# sessionUnusedTtlSeconds: 30
## Controls how long (in seconds) a session may wait for its first client connection before it
## is closed. Defaults to 180 seconds.
## Raise this when setting up a session takes longer than the default, for example when queue
## splitting waits for slow-rolling target pods, or when multi-cluster sessions are created
## well before a client reaches each cluster.
# sessionSetupDeadlineSeconds: 180
## Controls how long (in milliseconds) either side of a connected session - the client and the
## agent - may go without sending a message before the session is closed. The two are timed
## separately, so this is the budget per side, not a combined one. Defaults to 60000 (60 seconds).
## Raise this when developers work over slow or unreliable networks, where a laptop sleeping or
## changing network costs them a running session. The cost is that agents and patched workloads
## are held longer after a client really does go away.
# communicationTimeoutMillis: 60000
## Controls how long (in milliseconds) a session can live without a live mirrord-agent, usually
## because the target has no ready pod to run an agent on. The clock starts when the last agent
## is lost (or at session start) and stops as soon as an agent connects. An agent that is still
## starting on a ready pod when the time runs out gets to finish: the session waits for that
## load before closing. Defaults to 60000 (60 seconds).
## Raise this when a replacement target pod can take longer than that to become ready, for example
## on spot or preemptible nodes with node auto-provisioning, where the replacement pod waits for a
## new node to boot. Rolling updates are not affected: the new pod is ready before the old one goes.
# noPodTargetsSessionTimeoutMillis: 60000
## Default TTL (in milliseconds) for idle Kafka splits.
## For any given topic, starting the first Kafka splitting session requires patching the target workload.
## Similarly, stopping the last Kafka splitting session requires another patch, that reverts the first one.
## If the target workload takes a long time to restart, it may be desirable to keep the Kafka splits alive longer,
## so that the next Kafka splitting session will not have to patch the workload again.
##
## This value can be overridden per topic with the `spec.splitTtl` field in the `MirrordKafkaTopicsConsumer` custom resource.
# idleKafkaSplitTtlMillis: 30000
## You can adjust the format of the created kafka topic names to suit your needs. The default value is:
## `mirrord-tmp-{{RANDOM}}{{FALLBACK}}{{ORIGINAL_TOPIC}}`
## see docs here https://metalbear.co/mirrord/docs/using-mirrord/queue-splitting/#customizing-mirrord-created-kafka-topic-names
## `spec.tmpNameTemplate` of a `MirrordSplitConfig` takes precedence over this format for the workloads it selects.
## That template uses `{{ORIGINAL}}` instead of `{{ORIGINAL_TOPIC}}`, and its `{{FALLBACK}}` is `-main-` rather than `-fallback-`.
# kafkaSplittingTopicFormat: "mirrord-tmp-{{RANDOM}}{{FALLBACK}}{{ORIGINAL_TOPIC}}"
# kafkaBranchConfig:
## example
##
## brokerPod:
## # Any Kafka image configured through `KAFKA_` environment variables, like `apache/kafka`.
## image: "docker.io/apache/kafka:4.4.0"
## imagePullSecrets:
## - name: "kafka-registry-secret"
## imagePullPolicy: "IfNotPresent"
## labels: { "role": "broker" }
## annotations: { "karpenter.sh/do-not-disrupt": "true" }
## # The broker's JVM heap is at most half its memory limit.
## resources:
## requests:
## cpu: "250m"
## memory: "1Gi"
## limits:
## memory: "1Gi"
## # Partitions of every topic created on a branch broker.
## partitions: 1
## # Broker configs of every branch broker. The operator's own listener, KRaft, storage,
## # and partition settings cannot be changed.
## brokerConfig:
## message.max.bytes: "10485760"
kafkaSplittingSidecar:
# Enables a sidecar container responsible for Java Kafka clients.
# This enables Kafka splitting with Kafka Streams consumers.
enabled: false
image: ghcr.io/metalbear-co/operator-kafka-proxy
# Port for the sidecar to listen on.
# The operator communicates with the sidecar through a REST API served on localhost.
port: 33000
# CPU and memory requests for the sidecar.
# Set any of these, or the whole `requests` block, to `null` to leave it out of the pod spec.
requests:
cpu: 100m
memory: 100Mi
# CPU and memory limits for the sidecar.
# Set any of these, or the whole `limits` block, to `null` to leave it out of the pod spec.
limits:
cpu: 200m
memory: 200Mi
# Define additional environment variables for the sidecar, in the same format as
# `operator.extraEnv`.
extraEnv:
# JAVA_TOOL_OPTIONS: "-XX:MaxRAMPercentage=75"
# HOST_IP:
# valueFrom:
# fieldRef:
# fieldPath: status.hostIP
# Linger timeout for legacy SQS splits (in milliseconds).
# Applies only to old MirrordWorkloadQueueRegistry objects: once their last SQS
# session ends, the split lingers, keeping the target patched to read from the
# temporary queue until that queue is empty. This caps how long it lingers.
# With MirrordSplitConfig, set spec.drainTimeout on the config instead.
sqsSplittingLingerTimeout: 60000
## Control whether the operator waits for patched pods to become ready when starting Kafka/SQS splitting sessions.
## This option defaults to true (maintaining existing behavior) but can be set to false to speed up
## session start time in environments where pod readiness waiting is not necessary or causes delays due to cluster conditions.
queueSplittingWaitForReadyTarget: true
## CPU and memory requests and limits for the operator container.
## These defaults should be enough for around 200~ concurrent sessions.
##
## Every quantity is optional. Set one to `null` (`operator.limits.cpu: null`), or a whole block to
## `null` (`operator.limits: null`), and it is left out of the pod spec entirely - for clusters
## that set container resources themselves through a LimitRange or a VPA, or that run the operator
## without a CPU limit so it is never throttled.
requests:
cpu: 100m
memory: 100Mi
limits:
cpu: 500m
memory: 200Mi
## Allow to specify tolerations for operator deployment
# tolerations:
# - key: "key1"
# operator: "Equal"
# value: "value1"
# effect: "NoSchedule"
# affinity:
# podAntiAffinity:
# requiredDuringSchedulingIgnoredDuringExecution:
# - topologyKey: kubernetes.io/hostname
# labelSelector:
# matchLabels:
# k8s-app: mirrord
# nodeSelector:
# kubernetes.io/os: linux
# Port for operator to listen on. If you can't listen on 443 due to privilege issues
# you can change this to 3000/8443 or whatever you want - just make sure that nodes
# can communicate on that port if it doesn't work then.
# The operator binds the kube API on IPv6 (`[::]`) by default, with an IPv4
# (`0.0.0.0`) fallback on the same port — set `OPERATOR_ADDR` /
# `OPERATOR_FALLBACK_ADDR` directly if you need to override either side.
port: 443
copyTarget:
# by default, the operator will create a dummy container using the agent image
# as it is guaranteed to have a sleep binary. If you want the operator
# to use the target's image, set this value to false and make sure it has sleep binary
useAgentImage: true
## The URL of the Jira webhook to enable integration with the mirrord for Jira app.
## If this is set, the operator will attempt to send session duration metrics to the Jira app
## every time a user session ends.
# jiraWebhookUrl: "https://example.atlassian-dev.net/x1/random-hash"
# mysqlBranchConfig:
## example
##
## dbPod:
## image:
## registry: "custom-mysql-registry"
## migrationImages:
## # Registry per schema-migration tool, used by the matching `migrations.flavor`.
## # When unset, each defaults to its upstream image; when set, the registry is used
## # with that tool's pinned tag. The per-branch `migrations` config in mirrord.json
## # can override the full image.
## flyway:
## # Defaults to `flyway/flyway:13`; a registry here is used with the tag `13`.
## registry: "custom-flyway-registry"
## liquibase:
## # Defaults to `liquibase/liquibase:4.33`; a registry here is used with the tag `4.33`.
## registry: "custom-liquibase-registry"
## imagePullSecrets:
## - name: "mysql-registry-secret"
## imagePullPolicy: "IfNotPresent"
## labels: { "role": "database" }
## annotations: { "db.branch/owner-team": "platform" }
## # An explicit volume overrides the default per-branch PVC backing; see `storage`
## # under pgBranchConfig (available on every `<db>BranchConfig`).
## volume:
## name: "mysql-data"
## emptyDir:
## sizeLimit: "1Gi"
## resources:
## requests:
## cpu: "500m"
## memory: "512Mi"
## limits:
## cpu: "1"
## memory: "1Gi"
## # Extra flags for the branch's mysqld, passed as the container's args. The official
## # image's entrypoint hands them to the server, so an image that replaces that entrypoint
## # may not pick them up. Keep the listener on 3306 - the seeding init container, schema
## # migrations, and the operator all connect there.
## # A sourced branch loads `mysql_native_password` through a settings file under
## # `/etc/mysql/conf.d`, which images such as Percona's and Oracle's `community-server`
## # do not read, and a branch with no source gets no file. The `loose-` flag loads the
## # plugin on MySQL 8.4 for those too; 8.0 and 9 only log a warning for it, so it is
## # safe here, where every branch that selects no profile gets it. A profile is a
## # complete `dbPod` and inherits none of these args.
## # To also make the plugin the default for new accounts on 8.4, as
## # `--default-authentication-plugin` did before 8.4 removed it, give a profile only
## # 8.4 branches select both flags in its own `dbServerArgs`: the `loose-` flag, then
## # `--authentication-policy=*:mysql_native_password`. 8.4 refuses a policy naming a
## # plugin that is not loaded, MySQL 8.0 rejects the `*:` form, and MySQL 9 has no
## # such plugin, so a branch on either exits on the policy.
## dbServerArgs:
## - "--loose-mysql-native-password=ON"
## # How branch pods recreate the source database's accounts.
## # `empty` (default): only the user declared in the branch's connection config, with its
## # real password (as a hash) and every privilege; copied objects lose their definers.
## # `full`: source accounts with their grants, and copied objects keep their definers, so
## # the branch enforces the same permissions as the source. Only the declared user can log
## # in; other accounts come along only if it may read the `mysql` schema.
## roles: "empty"
# mariadbBranchConfig:
## example
##
## dbPod:
## image:
## # Registry for the MariaDB branch container.
## # Defaults to `docker.io/library/mariadb` with tag `12`.
## registry: "custom-mariadb-registry"
## migrationImages:
## # Registry per schema-migration tool, used by the matching `migrations.flavor`.
## # When unset, each defaults to its upstream image; when set, the registry is used
## # with that tool's pinned tag. The per-branch `migrations` config in mirrord.json
## # can override the full image.
## flyway:
## # Defaults to `flyway/flyway:13`; a registry here is used with the tag `13`.
## registry: "custom-flyway-registry"
## liquibase:
## # Defaults to `liquibase/liquibase:4.33`; a registry here is used with the tag `4.33`.
## registry: "custom-liquibase-registry"
## imagePullSecrets:
## - name: "mariadb-registry-secret"
## imagePullPolicy: "IfNotPresent"
## labels: { "role": "database" }
## annotations: { "db.branch/owner-team": "platform" }
## # An explicit volume overrides the default per-branch PVC backing; see `storage`
## # under pgBranchConfig (available on every `<db>BranchConfig`).
## volume:
## name: "mariadb-data"
## emptyDir:
## sizeLimit: "1Gi"
## resources:
## requests:
## cpu: "500m"
## memory: "512Mi"
## limits:
## cpu: "1"
## memory: "1Gi"
## # Extra flags for the branch's mariadbd, passed as the container's args. The official
## # image's entrypoint hands them to the server, so an image that replaces that entrypoint
## # may not pick them up. Keep the listener on 3306 - the seeding init container, schema
## # migrations, and the operator all connect there.
## dbServerArgs:
## - "--character-set-server=utf8mb4"
## - "--collation-server=utf8mb4_unicode_ci"
## # How branch pods recreate the source database's accounts.
## # `empty` (default): only the user declared in the branch's connection config, with its
## # real password (as a hash) and every privilege; copied objects lose their definers.
## # `full`: source accounts with their grants, and copied objects keep their definers, so
## # the branch enforces the same permissions as the source. Only the declared user can log
## # in; other accounts come along only if it may read the `mysql` schema.
## roles: "empty"
# pgBranchConfig:
## example
##
## dbPod:
## image:
## # Registry for the PostgreSQL branch container.
## # Defaults to `docker.io/library/postgres`.
## # Tag can be specified by the user when requesting a new branch, defaulting to `17`.
## registry: "custom-postgres-registry"
## initImage:
## # Registry for the PostgreSQL branch initContainer image.
## # Defaults to `docker.io/library/postgres`.
## # Tag can be specified by the user when requesting a new branch, defaulting to `17`.
## registry: "custom-postgres-registry"
## migrationImages:
## # Registry per schema-migration tool, used by the matching `migrations.flavor`.
## # When unset, each defaults to its upstream image; when set, the registry is used
## # with that tool's pinned tag. The per-branch `migrations` config in mirrord.json
## # can override the full image.
## flyway:
## # Defaults to `flyway/flyway:13`; a registry here is used with the tag `13`.
## registry: "custom-flyway-registry"
## liquibase:
## # Defaults to `liquibase/liquibase:4.33`; a registry here is used with the tag `4.33`.
## registry: "custom-liquibase-registry"
## # Environment for container-flavor migration Jobs. `inherit` (default true) hands the
## # Job the target container's env and envFrom, so an app image boots the same way its
## # real pods do; `env`/`envFrom` are admin extras in plain Kubernetes shapes, so
## # `valueFrom`/`secretRef` keep secret values in-cluster. Admin entries override
## # inherited ones by name, but never the operator's redirect of the declared connection
## # vars to the branch; the per-branch `migrations.env` in mirrord.json overrides
## # everything. Available on every `<db>BranchConfig`.
## migrationEnv:
## inherit: true
## env:
## - name: RAILS_ENV
## value: development
## envFrom:
## - configMapRef:
## name: shared-migration-config
## # Glob patterns of images users may run as branch pods via the per-branch `image`
## # field in mirrord.json (which overrides the registry configured above). When this
## # list is absent, all images are allowed. Available on every `<db>BranchConfig`.
## allowedImages:
## - "registry.example.com/postgresql:*"
## - "docker.io/library/postgres:*"
## # Image pull secrets for the PostgreSQL branch pod.
## # The pod will use following images:
## # * .pgBranchConfig.dbPod.image
## # * .pgBranchConfig.dbPod.initImage
## # * .pgBranchConfig.dbPod.migrationImages.flyway
## # * .pgBranchConfig.dbPod.migrationImages.liquibase
## # * .operator.image
## imagePullSecrets:
## - name: "postgres-registry-secret"
## # Image pull policy for all .pgBranchConfig.dbPod images (db, init, and migration).
## imagePullPolicy: "IfNotPresent"
## labels: { "role": "database" }
## annotations: { "db.branch/owner-team": "platform" }
## # Where the branch database pod and the branch's copy and migration Job pods may run,
## # in plain Kubernetes shapes. Needed when every node is tainted, e.g. Karpenter
## # NodePools with a NoSchedule taint: without a matching toleration the branch pod
## # stays Pending and the session times out. Available on every `<db>BranchConfig`.
## tolerations:
## - key: "workload"
## operator: "Equal"
## value: "dev"
## effect: "NoSchedule"
## nodeSelector:
## karpenter.sh/nodepool: "dev-pool"
## affinity:
## nodeAffinity:
## requiredDuringSchedulingIgnoredDuringExecution:
## nodeSelectorTerms:
## - matchExpressions:
## - key: "kubernetes.io/arch"
## operator: "In"
## values: ["amd64"]
## # How the branch pod's data and init (dump) volumes are backed. By default each
## # branch gets its own PersistentVolumeClaims on the cluster's default StorageClass,
## # deleted together with the branch; on clusters without a default StorageClass,
## # branches fall back to node-local emptyDir. Available on every `<db>BranchConfig`.
## storage:
## # `pvc` (default) or `emptyDir` (node-local scratch space, capped by
## # `operator.dbBranching.*PodVolumeLimit`).
## kind: "pvc"
## # StorageClass for the branch PVCs. Unset means the cluster's default.
## storageClassName: "fast-ssd"
## # Requested per-branch PVC sizes. Default to `operator.dbBranching.databasePvcSize`
## # / `initPvcSize` (20Gi each). The dump staged on the init volume is roughly the
## # size of the copied data.
## dataSize: "50Gi"
## initSize: "50Gi"
## # PVC access modes, defaulting to ReadWriteOnce.
## accessModes: ["ReadWriteOnce"]
## # An explicit volume overrides `storage` entirely and is used as given.
## volume:
## name: "pg-data"
## emptyDir:
## sizeLimit: "1Gi"
## resources:
## requests:
## cpu: "500m"
## memory: "512Mi"
## limits:
## cpu: "1"
## memory: "1Gi"
## # How branch pods recreate the source database's roles.
## # `empty` (default): NOLOGIN shell roles, just enough for the schema restore.
## # `full`: real role attributes (LOGIN, ...) and role memberships, and the data copy
## # keeps table ownership and grants, so the branch enforces the same permissions as
## # the source. The user declared in the branch's connection config keeps its real
## # password on the branch (as a salted hash); PostgreSQL never exposes other roles'
## # passwords, so those restore without one.
## roles: "empty"
## # Client authentication on branches whose source uses IAM auth.
## # `trust` (default): the branch accepts any credentials - IAM apps present
## # cloud-minted tokens a plain postgres branch cannot verify, and trust is what
## # lets them connect unchanged.
## # `password`: regular password auth; IAM apps must then read the branch
## # password from an env var mirrord injects (the `password` connection param).
## # Branches without `iam_auth` always use password auth.
## iamBranchAuth: "trust"
## # Extra flags for the branch's postgres server, passed as the container's args. The
## # official image's entrypoint hands them to the server, so an image that replaces that
## # entrypoint may not pick them up. Any file a flag points at (TLS certs below) must
## # exist in the image, and the entrypoint's temporary server that restores the copied
## # data runs with the same flags. Keep the listener on 5432 - the seeding init
## # container, schema migrations, and the operator all connect there.
## dbServerArgs:
## - "-c"
## - "ssl=on"
## - "-c"
## - "ssl_cert_file=/etc/ssl/certs/server.pem"
## - "-c"
## - "ssl_key_file=/etc/ssl/private/server.key"
##
## Named profiles: alternative complete baselines next to the default `dbPod` above,
## selected per branch with `"profile": "<name>"` in mirrord.json. A profile accepts
## the same fields as `dbPod` (registries, migrationImages, allowedImages, pull
## secrets, resources, ...) and is not an overlay - fields left unset get the built-in
## defaults, not the values from the default `dbPod`. Available on every
## `<db>BranchConfig`.
##
## `allowedImages` is the one field a profile inherits when it does not set its own
## list, because an unset list means "any image is allowed" and a profile added for an
## unrelated reason should not quietly widen your image policy. Set
## `allowedImages: ["*"]` in a profile to allow anything on purpose, or `[]` to allow
## no user-supplied images at all.
##
## profiles:
## team-analytics:
## dbPod:
## image:
## registry: "pgvector/pgvector"
## initImage:
## registry: "pgvector/pgvector"
## migrationImages:
## flyway:
## registry: "custom-flyway-registry"
## liquibase:
## registry: "custom-liquibase-registry"
## allowedImages: ["pgvector/*"]
## imagePullSecrets:
## - name: "analytics-registry-secret"
# mongodbBranchConfig:
## example
##
## dbPod:
## image:
## registry: "docker.io/library/mongo"
## imagePullSecrets:
## - name: "mongo-registry-secret"
## imagePullPolicy: "IfNotPresent"
## labels: { "role": "database" }
## annotations: { "db.branch/owner-team": "platform" }
## # An explicit volume overrides the default per-branch PVC backing; see `storage`
## # under pgBranchConfig (available on every `<db>BranchConfig`).
## volume:
## name: "mongo-data"
## emptyDir:
## sizeLimit: "1Gi"
## resources:
## requests:
## cpu: "500m"
## memory: "512Mi"
## limits:
## cpu: "1"
## memory: "1Gi"
# redisBranchConfig:
## example
##
## dbPod:
## image:
## registry: "docker.io/library/redis"
## imagePullSecrets:
## - name: "redis-registry-secret"
## imagePullPolicy: "IfNotPresent"
## labels: { "role": "database" }
## annotations: { "db.branch/owner-team": "platform" }
## # An explicit volume overrides the default per-branch PVC backing; see `storage`
## # under pgBranchConfig (available on every `<db>BranchConfig`).
## volume:
## name: "redis-data"
## emptyDir:
## sizeLimit: "1Gi"
## resources:
## requests:
## cpu: "500m"
## memory: "512Mi"
## limits:
## cpu: "1"
## memory: "1Gi"
##
## For a custom image with TLS baked in, keep the TLS listener on 6379 and
## disable the plaintext one; `tls: true` makes the operator (seeding and the
## connection URL handed to the app) use rediss://.
##
## dbServerArgs:
## - "--tls-port"
## - "6379"
## - "--port"
## - "0"
## - "--tls-cert-file"
## - "/certs/tls.crt"
## - "--tls-key-file"
## - "/certs/tls.key"
## - "--tls-auth-clients"
## - "no"
## tls: true
##
## Named profiles let one cluster hold several complete pod baselines at once
## (e.g. a TLS setup and a plain one). Users select one per branch with
## `"profile": "<name>"` in their mirrord config; without it, the top-level
## dbPod applies. A profile is a complete dbPod (same fields as above), not an
## overlay - fields left unset get the built-in defaults, never the top-level
## values. The exception is `allowedImages`, which a profile inherits unless it
## sets its own list, so adding a profile never quietly widens your image policy;
## use `allowedImages: ["*"]` to allow anything on purpose, or `[]` to allow no
## user-supplied images. Every <db>BranchConfig supports `profiles`, not just Redis.
##
## profiles:
## tls-baseline:
## dbPod:
## allowedImages:
## - "registry.example.com/team/redis-tls:*"
## dbServerArgs:
## - "--tls-port"
## - "6379"
## - "--port"
## - "0"
## - "--tls-cert-file"
## - "/certs/tls.crt"
## - "--tls-key-file"
## - "/certs/tls.key"
## tls: true
##
## Profile settings live under the profile's `dbPod` key. A key placed beside
## `dbPod` (e.g. `profiles.tls-baseline.tls`) is not a valid setting, and a
## branch selecting that profile fails with an error naming the misplaced key.
##
## For `copy.mode: all` against a source Redis that only accepts TLS, provide
## the certificates via a `MirrordPropertyList` named `redis-source-tls` in the
## target's namespace - not via chart values - with the same properties as
## `cockroachdb-source-tls` under cockroachdbBranchConfig below. The object
## name can be overridden with `dbPod.sourceTlsPropertyList: "my-name"`.
##
## `dbPod.sourceSslmode` sets how strictly the copy verifies the source's certificate:
## `verify-full` (chain and hostname, the default) or `verify-ca` (chain only). Use
## `verify-ca` when the source certificate has no subjectAltName covering the host the
## branch dials, where `verify-full` can never pass and `#insecure` in the URL would skip
## the chain check too. It only applies to TLS connections (the property list or a
## `rediss://` URL decide that) and `#insecure` still wins over it:
##
## dbPod:
## sourceSslmode: "verify-ca"
# spannerBranchConfig:
## example
##
## dbPod:
## image:
## registry: "gcr.io/cloud-spanner-emulator/emulator"
## imagePullSecrets:
## - name: "spanner-registry-secret"
## imagePullPolicy: "IfNotPresent"
## labels: { "role": "database" }
## annotations: { "db.branch/owner-team": "platform" }
## # An explicit volume overrides the default per-branch PVC backing; see `storage`
## # under pgBranchConfig (available on every `<db>BranchConfig`).
## volume:
## name: "spanner-data"
## emptyDir:
## sizeLimit: "1Gi"
## resources:
## requests:
## cpu: "500m"
## memory: "512Mi"
## limits:
## cpu: "1"
## memory: "1Gi"
# clickhouseBranchConfig:
## example
##
## dbPod:
## image:
## # Registry for the ClickHouse branch container.
## # Defaults to `docker.io/clickhouse/clickhouse-server`.
## # Tag can be specified by the user when requesting a new branch, defaulting to `26.8`.
## registry: "custom-clickhouse-registry"
## initImage:
## # Registry for the ClickHouse branch initContainer image.
## # Defaults to `docker.io/clickhouse/clickhouse-server`.
## registry: "custom-clickhouse-registry"
## imagePullSecrets:
## - name: "clickhouse-registry-secret"
## imagePullPolicy: "IfNotPresent"
## labels: { "role": "database" }
## annotations: { "db.branch/owner-team": "platform" }
## # An explicit volume overrides the default per-branch PVC backing; see `storage`
## # under pgBranchConfig (available on every `<db>BranchConfig`).
## volume:
## name: "clickhouse-data"
## emptyDir:
## sizeLimit: "1Gi"
## resources:
## requests:
## cpu: "500m"
## memory: "512Mi"
## limits:
## cpu: "1"
## memory: "1Gi"
# cockroachdbBranchConfig:
## example
##
## dbPod:
## image:
## # Registry for the CockroachDB branch container.
## # Defaults to `docker.io/cockroachdb/cockroach`.
## # Tag can be specified by the user when requesting a new branch, defaulting to `latest-v26.2`.
## registry: "custom-cockroachdb-registry"
## initImage:
## # Registry for the CockroachDB branch initContainer image.
## # Defaults to `docker.io/cockroachdb/cockroach`.
## registry: "custom-cockroachdb-registry"
## imagePullSecrets:
## - name: "cockroachdb-registry-secret"
## imagePullPolicy: "IfNotPresent"
## labels: { "role": "database" }
## annotations: { "db.branch/owner-team": "platform" }
## # An explicit volume overrides the default per-branch PVC backing; see `storage`
## # under pgBranchConfig (available on every `<db>BranchConfig`).
## volume:
## name: "cockroachdb-data"
## emptyDir:
## sizeLimit: "1Gi"
## resources:
## requests:
## cpu: "500m"
## memory: "512Mi"
## limits:
## cpu: "1"
## memory: "1Gi"
##
## Sources behind certificate verification (sslmode=verify-ca/verify-full against a
## private CA) or mTLS are supported via a `MirrordPropertyList` named
## `cockroachdb-source-tls` in the target's namespace - not via chart values - so each
## namespace can point at its own cert Secret. The object name can be overridden with
## `dbPod.sourceTlsPropertyList: "my-name"`. tlsClientCert/tlsClientKey must be set
## together and are omitted when the source only needs a trusted CA (tlsCaCert):
##
## apiVersion: mirrord.metalbear.co/v1
## kind: MirrordPropertyList
## metadata:
## name: cockroachdb-source-tls
## namespace: my-app-namespace
## spec:
## properties:
## - name: tlsCaCert
## valueFrom:
## secretKeyRef: { name: my-db-certs, key: ca.crt }
## - name: tlsClientCert
## valueFrom:
## secretKeyRef: { name: my-db-certs, key: client.myservice.crt }
## - name: tlsClientKey
## valueFrom:
## secretKeyRef: { name: my-db-certs, key: client.myservice.key }
##
## `dbPod.sourceSslmode` sets a cluster-wide default sslmode for the copy connection to
## the source. An explicit sslmode in a session's connection URL or params still wins.
## Without it the default is verify-full when the property list provides certs, disable
## otherwise. Note verify-ca/verify-full still need the property list - without a CA the
## copy fails rather than silently downgrading:
##
## dbPod:
## sourceSslmode: "verify-full"
##
## `dbPod.sourceConfigMap` sets the ConfigMap that `configmap` connection param sources
## read when a session's param names none itself: a `configMap` volume of the target
## pod (`volume`) or a ConfigMap name (`name`), plus an optional data `key`. Developers
## then write only the selector per param. Works on the default `dbPod` and on profiles:
##
## profiles:
## app-config:
## dbPod:
## sourceConfigMap:
## volume: app-config
## key: config.yml
# genericBranchConfig:
## example
##
## Generic branches always run a user-supplied image (see the `genericBranching` flag
## note above), so `allowedImages` is the main lever here. `image`/`initImage`
## registries, `dbServerArgs`, `tls`, and default volumes are not consulted for generic
## branches - the user's image, command, and args are used as given.
##
## dbPod:
## allowedImages:
## - "valkey/*"
## - "influxdb:*"
## imagePullSecrets:
## - name: "generic-registry-secret"
## imagePullPolicy: "IfNotPresent"
## labels: { "role": "database" }
## annotations: { "db.branch/owner-team": "platform" }
## resources:
## requests:
## cpu: "250m"
## memory: "256Mi"
## limits:
## cpu: "1"
## memory: "1Gi"
##
## Named profiles work here like for every other database type: keep the default
## locked with `allowedImages: []` and grant each team its own images, pull secrets,
## and resource budget. Users select one per branch with `"profile": "<name>"`.
## A profile that does not set `allowedImages` inherits the default's, so with a
## locked default a profile only ever runs the images it lists itself.
##
## A profile can additionally carry the whole engine setup, so a user's mirrord.json
## shrinks to type/id/profile/connection:
##
## - `dbPod.branch` supplies branch-container defaults (image, port, command, args,
## env, readiness). A value set in the user's mirrord.json wins per field.
## - `dbPod.copy` supplies a one-shot copy Job: an image you author that copies
## schema/data from the source into the branch before it turns Ready. The Job's env
## carries the same `MIRRORD_PARAM_*` vars as the branch container (the source
## connection) plus `MIRRORD_BRANCH_HOST`/`MIRRORD_BRANCH_PORT` (where to write).
## A `copy` in the user's mirrord.json wins over the profile's, and the copy runs at
## most once per branch. The copy image also goes through `allowedImages`. If your
## cluster uses NetworkPolicies, the Job (in the target's namespace) must be able to
## reach both the source database and the branch pod's IP.
## - `dbPod.copyDeadlineSecs` (default 3600) is the longest a copy Job may run. A running
## copy keeps its branch alive past the branch TTL, so this stops a copy that hangs: the
## Job is stopped and the branch fails with a message naming the deadline. It applies to
## the user's `copy` from mirrord.json too, and a profile without it inherits the
## default `dbPod`'s, so it can be set once at the default level.
##
## Both fields are engine-specific, so they are only honored inside a named profile -
## a generic branch that uses no profile fails when the default-level `dbPod` sets
## them (users wanting an empty branch would have no opt-out otherwise).
##
## profiles:
## team-search:
## dbPod:
## allowedImages:
## - "registry.example.com/search/*"
## imagePullSecrets:
## - name: "search-registry-secret"
## resources:
## limits:
## memory: "2Gi"
## opensearch-full:
## dbPod:
## allowedImages:
## - "opensearchproject/*"
## - "registry.example.com/dbtools/*"
## branch:
## image: "opensearchproject/opensearch:2.19.1"
## port: 9200
## env: { "discovery.type": "single-node" }
## copy:
## image: "registry.example.com/dbtools/opensearch-copy:1.0"
## command: ["./copy.sh"]
## args: ["--mode=all"]
agent:
## example
##
## image:
## registry: "your-internal.registry.io/metalbear-co/agent"
## tag: "latest"
##
## or
##
## image: "your-internal.registry.io/metalbear-co/agent:latest"
##
# image:
# registry: "ghcr.io/metalbear-co/mirrord"
# tag: "latest"
# If you want the operator to secure agent connections with TLS, set this value to true.
# This option requires agent version at least 3.97.0.
tls: false
# Configure whether the agent should inject headers by default.
# If `agent.extraConfig.inject_headers` is set, it takes precedence.
# Agent headers mark whether request went thru the agent or not,
# and if they were stolen or handled by original pod
injectHeaders: true
# If you want the agents to accept operator connections on some predefined port,
# you can set it here. If you don't set anything, each agent will be assigned with a random high port.
# port: 9999
## CPU and memory requests and limits for agent pods (not applied to ephemeral agents).
## These defaults match the agent's own built-in defaults; when this block is set it replaces
## them entirely. If `agent.extraConfig.resources` is set, it takes precedence over this block.
##
## Every quantity is optional. Set one to `null` (`agent.resources.limits.cpu: null`), or a whole
## block to `null` (`agent.resources.limits: null`), and it is left out of the agent pod spec
## entirely - for example to run agents without a CPU limit so they are never throttled.
# resources:
# requests:
# cpu: 1m
# memory: 1Mi
# limits:
# cpu: 300m
# memory: 100Mi
# Define additional environment variables for agent pods.
extraEnv: {}
priorityClass:
# Whether to create a PriorityClass for agent pods.
create: true
# PriorityClass name to create/use for agent pods.
# When empty and create=true, defaults to "mirrord-agent-pod".
# When empty and create=false, no priority class is set on the agent.
name: mirrord-agent-pod
# Integer priority value assigned to the class.
value: 1000000000
# Whether this class should be the cluster default.
globalDefault: false
# Pod preemption policy for this class.
preemptionPolicy: PreemptLowerPriority
# Description shown on the class object.
description: "Priority class for mirrord agent pods."
## use this if you want to add settings that aren't covered by the values.yaml
## see possible settings here: https://mirrord.dev/docs/reference/configuration/#root-agent
extraConfig:
# json_log: false
# labels: { "user": "meow" }
# annotations: { "cats.io/inject": "enabled" }
# tolerations:
# - operator: Exists
# privileged: true
# # Max HTTP/2 header size, in bytes, accepted by the agent on intercepted traffic (default 16KB).
# http2_max_header_list_size: 16384
license:
## ID of a Google Secret Manager secret containing the license PEM file.
##
## If provided, the operator will fetch the license file from this secret.
##
## To access the secret, the operator will use Application Default Credentials.
## The easiest way to provide credentials is by allowing the operator's
## Kubernetes ServiceAccount to impersonate a GCP service account.
## This can be done with `.sa.gcpSa` setting.
# pemGsmRef: "projects/PROJECT_ID/secrets/SECRET_NAME/versions/SECRET_VERSION"
## ID of a Google Secret Manager secret containing the license key.
##
## If provided, the operator will fetch the license key from this secret.
##
## Uses the same authentication behavior described for `pemGsmRef` above.
# keyGsmRef: "projects/PROJECT_ID/secrets/SECRET_NAME/versions/SECRET_VERSION"
key: ""
file:
secret: mirrord-operator-license
# data:
# license.pem:
keyRef: ""
pemRef: ""
# For air-gapped systems there is an option to use a license-server for enterprise licenses.
# This must be an accessible endpoint from where the operator is installed
# (can be in cluster or not and via any ingress or node-port as long as communication is possible via http/https)
#
# licenseServer: http://mirrord-operator-license-server.mirrord.svc
# For enterprise licenses the operator can allow usage for more than the max amount of seats.
#
# Set this value to `false` to disable this allowance to go over the max seats count.
allowSeatOverages: true
# Cloud API key — the default credential the operator uses to authenticate to the mirrord cloud
# (RFC 0008). The operator exchanges it (via app.metalbear.com) for a short-lived token and uses
# that token to obtain its license and make its other cloud calls over the API. It is intended to
# replace the standalone license key as the primary credential; existing installs that set only
# `license.*` keep working via license-key authentication.
#
# Not used when `license.licenseServer` is set: such an operator authenticates only against that
# license server, never against the cloud, so the key is ignored and a license source is required.
#
# Provide the key through exactly one of the following:
cloud:
apiKey:
## Direct value (simplest for dev/test; the value lands as plaintext in the pod spec).
key: ""
## Reference to an existing Kubernetes secret holding the key under the `apiKey` data key.
keyRef: ""
## ID of a secret in Google Secret Manager (used by us and by customers on GCP). Accessed
## via Application Default Credentials, like `license.gsmRef` (see `.sa.gcpSa`).
# gsmRef: "projects/PROJECT_ID/secrets/SECRET_NAME/versions/SECRET_VERSION"
gsmRef: ""
## When true, the operator keeps sending only anonymized telemetry to the mirrord cloud even
## when its cloud API key carries identity consent.
anonymizeData: false
service:
name: mirrord-operator
sa:
## Whether the chart should create the operator ServiceAccount.
## Set to `false` to use an existing ServiceAccount named by `sa.name`.
create: true
## Name of the ServiceAccount used by the operator Deployment and RBAC bindings.
name: mirrord-operator
## AWS role ARN to use for IAM role assumption.
# roleArn: arn:aws:iam::111122223333:role/mirrord-operator-role
## GCP service account to impersonate.
# gcpSa: <iam-sa-name>@<iam-sa-project-id>.iam.gserviceaccount.com
## Azure Managed Identity client ID for AKS Workload Identity.
## When set, the operator pod gets the azure.workload.identity/use label.
## If `sa.create` is true, the ServiceAccount also gets the
## azure.workload.identity/client-id annotation. The Workload Identity webhook
## then injects Azure credentials (AZURE_CLIENT_ID, AZURE_TENANT_ID,
## AZURE_FEDERATED_TOKEN_FILE) into the operator pod automatically.
# azureClientId: "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
tls:
secret: mirrord-operator-tls
# Set to `true` to mount the secret named by `tls.secret` without the chart
# creating it. Use this when the TLS secret is provisioned out-of-band (e.g. by
# an ExternalSecret or another controller). When `true`, `tls.data` and
# `tls.certManager` are ignored.
useExistingSecret: false
# if you're using a verified certificate, set this value to false.
apiService:
insecureSkipTLSVerify: true
# if certmanager is disabled and no tls.key and tls.crt is set,
# the operator will generate a self-signed certificate.
certManager:
enabled: false
createIssuer: true
issuer: mirrord-operator-issuer
certificate: mirrord-operator-tls
data:
tls.key: ""
tls.crt: ""
# if you run on OpenShift, set this value to true to get a SCC in the yaml.
openshift: false
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |