Skip to content

Latest commit

 

History

History
703 lines (551 loc) · 31.1 KB

cluster_manifest.md

File metadata and controls

703 lines (551 loc) · 31.1 KB
Raw

Cluster manifest reference

Individual Postgres clusters are described by the Kubernetes cluster manifest that has the structure defined by the postgresql CRD (custom resource definition). The following section describes the structure of the manifest and the purpose of individual keys. You can take a look at the examples of the minimal and the complete cluster manifests.

When Kubernetes resources, such as memory, CPU or volumes, are configured, their amount is usually described as a string together with the units of measurements. Please, refer to the Kubernetes documentation for the possible values of those.

❗ If both operator configmap/CRD and a Postgres cluster manifest define the same parameter, the value from the Postgres cluster manifest is applied.

Manifest structure

A Postgres manifest is a YAML document. On the top level both individual parameters and parameter groups can be defined. Parameter names are written in camelCase.

Cluster metadata

Those parameters are grouped under the metadata top-level key.

  • name the name of the cluster. Must start with the teamId followed by a dash. Changing it after the cluster creation is not supported. Required field.

  • namespace the namespace where the operator creates Kubernetes objects (i.e. pods, services, secrets) for the cluster. Changing it after the cluster creation results in deploying or updating a completely separate cluster in the target namespace. Optional (if present, should match the namespace where the manifest is applied).

  • labels if labels are matching one of the inherited_labels configured in the operator parameters, they will automatically be added to all the objects (StatefulSet, Service, Endpoints, etc.) that are created by the operator. Labels that are set here but not listed as inherited_labels in the operator parameters are ignored.

Top-level parameters

These parameters are grouped directly under the spec key in the manifest.

  • teamId name of the team the cluster belongs to. Required field.

  • numberOfInstances total number of instances for a given cluster. The operator parameters max_instances and min_instances may also adjust this number. Required field.

  • dockerImage custom Docker image that overrides the docker_image operator parameter. It should be a Spilo image. Optional.

  • schedulerName specifies the scheduling profile for database pods. If no value is provided K8s' default-scheduler will be used. Optional.

  • spiloRunAsUser sets the user ID which should be used in the container to run the process. This must be set to run the container without root. By default the container runs with root. This option only works for Spilo versions >= 1.6-p3.

  • spiloRunAsGroup sets the group ID which should be used in the container to run the process. This must be set to run the container without root. By default the container runs with root. This option only works for Spilo versions >= 1.6-p3.

  • spiloFSGroup the Persistent Volumes for the Spilo pods in the StatefulSet will be owned and writable by the group ID specified. This will override the spilo_fsgroup operator parameter. This is required to run Spilo as a non-root process, but requires a custom Spilo image. Note the FSGroup of a Pod cannot be changed without recreating a new Pod. Optional.

  • enableMasterLoadBalancer boolean flag to override the operator defaults (set by the enable_master_load_balancer parameter) to define whether to enable the load balancer pointing to the Postgres primary. Optional.

  • enableMasterPoolerLoadBalancer boolean flag to override the operator defaults (set by the enable_master_pooler_load_balancer parameter) to define whether to enable the load balancer for master pooler pods pointing to the Postgres primary. Optional.

  • enableReplicaLoadBalancer boolean flag to override the operator defaults (set by the enable_replica_load_balancer parameter) to define whether to enable the load balancer pointing to the Postgres standby instances. Optional.

  • enableReplicaPoolerLoadBalancer boolean flag to override the operator defaults (set by the enable_replica_pooler_load_balancer parameter) to define whether to enable the load balancer for replica pooler pods pointing to the Postgres standby instances. Optional.

  • allowedSourceRanges when one or more load balancers are enabled for the cluster, this parameter defines the comma-separated range of IP networks (in CIDR-notation). The corresponding load balancer is accessible only to the networks defined by this parameter. Optional, when empty the load balancer service becomes inaccessible from outside of the Kubernetes cluster.

  • maintenanceWindows a list which defines specific time frames when certain maintenance operations are allowed. So far, it is only implemented for automatic major version upgrades. Accepted formats are "01:00-06:00" for daily maintenance windows or "Sat:00:00-04:00" for specific days, with all times in UTC.

  • users a map of usernames to user flags for the users that should be created in the cluster by the operator. User flags are a list, allowed elements are SUPERUSER, REPLICATION, INHERIT, LOGIN, NOLOGIN, CREATEROLE, CREATEDB, BYPASSRLS. A login user is created by default unless NOLOGIN is specified, in which case the operator creates a role. One can specify empty flags by providing a JSON empty array '[]'. If the config option enable_cross_namespace_secret is enabled you can specify the namespace in the user name in the form {namespace}.{username} and the operator will create the K8s secret in that namespace. The part after the first . is considered to be the user name. Optional.

  • usersWithSecretRotation list of users to enable credential rotation in K8s secrets. The rotation interval can only be configured globally. On each rotation a new user will be added in the database replacing the username value in the secret of the listed user. Although, rotation users inherit all rights from the original role, keep in mind that ownership is not transferred. See more details in the administrator docs.

  • usersWithInPlaceSecretRotation list of users to enable in-place password rotation in K8s secrets. The rotation interval can only be configured globally. On each rotation the password value will be replaced in the secrets which the operator reflects in the database, too. List only users here that rarely connect to the database, like a flyway user running a migration on Pod start. See more details in the administrator docs.

  • usersIgnoringSecretRotation if you have secret rotation enabled globally you can define a list of of users that should opt out from it, for example if you store credentials outside of K8s, too, and corresponding deployments cannot dynamically reference secrets. Note, you can also opt out from the rotation by removing users from the manifest's users section. The operator will not drop them from the database. Optional.

  • databases a map of database names to database owners for the databases that should be created by the operator. The owner users should already exist on the cluster (i.e. mentioned in the user parameter). Optional.

  • tolerations a list of tolerations that apply to the cluster pods. Each element of that list is a dictionary with the following fields: key, operator, value, effect and tolerationSeconds. Each field is optional. See Kubernetes examples for details on tolerations and possible values of those keys. When set, this value overrides the pod_toleration setting from the operator. Optional.

  • podPriorityClassName a name of the priority class that should be assigned to the cluster pods. When not specified, the value is taken from the pod_priority_class_name operator parameter, if not set then the default priority class is taken. The priority class itself must be defined in advance. Optional.

  • podAnnotations A map of key value pairs that gets attached as annotations to each pod created for the database.

  • serviceAnnotations A map of key value pairs that gets attached as annotations to the services created for the database cluster. Check the administrator docs for more information regarding default values and overwrite rules.

  • masterServiceAnnotations A map of key value pairs that gets attached as annotations to the master service created for the database cluster. Check the administrator docs for more information regarding default values and overwrite rules. This field overrides serviceAnnotations with the same key for the master service if not empty.

  • replicaServiceAnnotations A map of key value pairs that gets attached as annotations to the replica service created for the database cluster. Check the administrator docs for more information regarding default values and overwrite rules. This field overrides serviceAnnotations with the same key for the replica service if not empty.

  • enableShmVolume Start a database pod without limitations on shm memory. By default Docker limit /dev/shm to 64M (see e.g. the docker issue, which could be not enough if PostgreSQL uses parallel workers heavily. If this option is present and value is true, to the target database pod will be mounted a new tmpfs volume to remove this limitation. If it's not present, the decision about mounting a volume will be made based on operator configuration (enable_shm_volume, which is true by default). It it's present and value is false, then no volume will be mounted no matter how operator was configured (so you can override the operator configuration). Optional.

  • enableConnectionPooler Tells the operator to create a connection pooler with a database for the master service. If this field is true, a connection pooler deployment will be created even if connectionPooler section is empty. Optional, not set by default.

  • enableReplicaConnectionPooler Tells the operator to create a connection pooler with a database for the replica service. If this field is true, a connection pooler deployment for replica will be created even if connectionPooler section is empty. Optional, not set by default.

  • enableLogicalBackup Determines if the logical backup of this cluster should be taken and uploaded to S3. Default: false. Optional.

  • logicalBackupRetention You can set a retention time for the logical backup cron job to remove old backup files after a new backup has been uploaded. Example values are "3 days", "2 weeks", or "1 month". It takes precedence over the global logical_backup_s3_retention_time configuration. Currently only supported for AWS. Optional.

  • logicalBackupSchedule Schedule for the logical backup K8s cron job. Please take the reference schedule format into account. It takes precedence over the global logical_backup_schedule configuration. Optional.

  • additionalVolumes List of additional volumes to mount in each container of the statefulset pod. Each item must contain a name, mountPath, and volumeSource which is a kubernetes volumeSource. It allows you to mount existing PersistentVolumeClaims, ConfigMaps and Secrets inside the StatefulSet. Also an emptyDir volume can be shared between initContainer and statefulSet. Additionaly, you can provide a SubPath for volume mount (a file in a configMap source volume, for example). Set isSubPathExpr to true if you want to include API environment variables. You can also specify in which container the additional Volumes will be mounted with the targetContainers array option. If targetContainers is empty, additional volumes will be mounted only in the postgres container. If you set the all special item, it will be mounted in all containers (postgres + sidecars). Else you can set the list of target containers in which the additional volumes will be mounted (eg : postgres, telegraf)

Prepared Databases

The operator can create databases with default owner, reader and writer roles without the need to specifiy them under users or databases sections. Those parameters are grouped under the preparedDatabases top-level key. For more information, see user docs.

  • defaultUsers The operator will always create default NOLOGIN roles for defined prepared databases, but if defaultUsers is set to true three additional LOGIN roles with _user suffix will get created. Default is false.

  • extensions map of extensions with target database schema that the operator will install in the database. Optional.

  • schemas map of schemas that the operator will create. Optional - if no schema is listed, the operator will create a schema called data. Under each schema key, it can be defined if defaultRoles (NOLOGIN) and defaultUsers (LOGIN) roles shall be created that have schema-exclusive privileges. By default, defaultRoles is true and defaultUsers is false.

  • secretNamespace for each default LOGIN role the operator will create a secret. You can specify the namespace in which these secrets will get created, if enable_cross_namespace_secret is set to true in the config. Otherwise, the cluster namespace is used.

Postgres parameters

Those parameters are grouped under the postgresql top-level key, which is required in the manifest.

  • version the Postgres major version of the cluster. Looks at the Spilo project for the list of supported versions. Changing the cluster version once the cluster has been bootstrapped is not supported. Required field.

  • parameters a dictionary of Postgres parameter names and values to apply to the resulting cluster. Optional (Spilo automatically sets reasonable defaults for parameters like work_mem or max_connections).

Patroni parameters

Those parameters are grouped under the patroni top-level key. See the Patroni documentation for the explanation of ttl and loop_wait parameters.

  • initdb a map of key-value pairs describing initdb parameters. For data-checksums, debug, no-locale, noclean, nosync and sync-only parameters use true as the value if you want to set them. Changes to this option do not affect the already initialized clusters. Optional.

  • pg_hba list of custom pg_hba lines to replace default ones. Note that the default ones include

    hostssl all +pamrole all pam

    where pamrole is the name of the role for the pam authentication; any custom pg_hba should include the pam line to avoid breaking pam authentication. Optional.

  • ttl Patroni ttl parameter value, optional. The default is set by the Spilo Docker image. Optional.

  • loop_wait Patroni loop_wait parameter value, optional. The default is set by the Spilo Docker image. Optional.

  • retry_timeout Patroni retry_timeout parameter value, optional. The default is set by the Spilo Docker image. Optional.

  • maximum_lag_on_failover Patroni maximum_lag_on_failover parameter value, optional. The default is set by the Spilo Docker image. Optional.

  • slots permanent replication slots that Patroni preserves after failover by re-creating them on the new primary immediately after doing a promote. Slots could be reconfigured with the help of patronictl edit-config. It is the responsibility of a user to avoid clashes in names between replication slots automatically created by Patroni for cluster members and permanent replication slots. Optional.

  • synchronous_mode Patroni synchronous_mode parameter value. The default is set to false. Optional.

  • synchronous_mode_strict Patroni synchronous_mode_strict parameter value. Can be used in addition to synchronous_mode. The default is set to false. Optional.

  • synchronous_node_count Patroni synchronous_node_count parameter value. Note, this option is only available for Spilo images with Patroni 2.0+. The default is set to 1. Optional.

  • failsafe_mode Patroni failsafe_mode parameter value. If enabled, Patroni will cope with DCS outages by avoiding leader demotion. See the Patroni documentation here for more details. This feature is included since Patroni 3.0.0. Hence, check the container image in use if this feature is included in the used Patroni version. The default is set to false. Optional.

Postgres container resources

Those parameters define CPU and memory requests and limits for the Postgres container. They are grouped under the resources top-level key with subgroups requests and limits.

Requests

CPU and memory requests for the Postgres container.

  • cpu CPU requests for the Postgres container. Optional, overrides the default_cpu_requests operator configuration parameter.

  • memory memory requests for the Postgres container. Optional, overrides the default_memory_request operator configuration parameter.

  • hugepages-2Mi hugepages-2Mi requests for the sidecar container. Optional, defaults to not set.

  • hugepages-1Gi 1Gi hugepages requests for the sidecar container. Optional, defaults to not set.

Limits

CPU and memory limits for the Postgres container.

  • cpu CPU limits for the Postgres container. Optional, overrides the default_cpu_limits operator configuration parameter.

  • memory memory limits for the Postgres container. Optional, overrides the default_memory_limits operator configuration parameter.

  • hugepages-2Mi hugepages-2Mi requests for the sidecar container. Optional, defaults to not set.

  • hugepages-1Gi 1Gi hugepages requests for the sidecar container. Optional, defaults to not set.

Parameters defining how to clone the cluster from another one

Those parameters are applied when the cluster should be a clone of another one that is either already running or has a basebackup on S3. They are grouped under the clone top-level key and do not affect the already running cluster.

  • cluster name of the cluster to clone from. Translated to either the service name or the key inside the S3 bucket containing base backups. Required when the clone section is present.

  • uid Kubernetes UID of the cluster to clone from. Since cluster name is not a unique identifier of the cluster (as identically named clusters may exist in different namespaces) , the operator uses UID in the S3 bucket name in order to guarantee uniqueness. Has no effect when cloning from the running clusters. Optional.

  • timestamp the timestamp up to which the recovery should proceed. The operator always configures non-inclusive recovery target, stopping right before the given timestamp. When this parameter is set the operator will not consider cloning from the live cluster, even if it is running, and instead goes to S3. Optional.

  • s3_wal_path the url to S3 bucket containing the WAL archive of the cluster to be cloned. Optional.

  • s3_endpoint the url of the S3-compatible service should be set when cloning from non AWS S3. Optional.

  • s3_access_key_id the access key id, used for authentication on S3 service. Optional.

  • s3_secret_access_key the secret access key, used for authentication on S3 service. Optional.

  • s3_force_path_style to enable path-style addressing(i.e., http://s3.amazonaws.com/BUCKET/KEY) when connecting to an S3-compatible service that lack of support for sub-domain style bucket URLs (i.e., http://BUCKET.s3.amazonaws.com/KEY). Optional.

Standby cluster

On startup, an existing standby top-level key creates a standby Postgres cluster streaming from a remote location - either from a S3 or GCS WAL archive or a remote primary. Only one of options is allowed and required if the standby key is present.

  • s3_wal_path the url to S3 bucket containing the WAL archive of the remote primary.

  • gs_wal_path the url to GS bucket containing the WAL archive of the remote primary.

  • standby_host hostname or IP address of the primary to stream from.

  • standby_port TCP port on which the primary is listening for connections. Patroni will use "5432" if not set.

Volume properties

Those parameters are grouped under the volume top-level key and define the properties of the persistent storage that stores Postgres data.

  • size the size of the target volume. Usual Kubernetes size modifiers, i.e. Gi or Mi, apply. Required.

  • storageClass the name of the Kubernetes storage class to draw the persistent volume from. See Kubernetes documentation for the details on storage classes. Optional.

  • subPath Subpath to use when mounting volume into Spilo container. Optional.

  • isSubPathExpr Set it to true if the specified subPath is an expression. Optional.

  • iops When running the operator on AWS the latest generation of EBS volumes (gp3) allows for configuring the number of IOPS. Maximum is 16000. Optional.

  • throughput When running the operator on AWS the latest generation of EBS volumes (gp3) allows for configuring the throughput in MB/s. Maximum is 1000. Optional.

  • selector A label query over PVs to consider for binding. See the Kubernetes documentation for details on using matchLabels and matchExpressions. Optional

Sidecar definitions

Those parameters are defined under the sidecars key. They consist of a list of dictionaries, each defining one sidecar (an extra container running along the main Postgres container on the same pod). The following keys can be defined in the sidecar dictionary:

Requests

CPU and memory requests for the sidecar container.

  • cpu CPU requests for the sidecar container. Optional, overrides the default_cpu_requests operator configuration parameter. Optional.

  • memory memory requests for the sidecar container. Optional, overrides the default_memory_request operator configuration parameter. Optional.

  • hugepages-2Mi hugepages-2Mi requests for the sidecar container. Optional, defaults to not set.

  • hugepages-1Gi 1Gi hugepages requests for the sidecar container. Optional, defaults to not set.

Limits

CPU and memory limits for the sidecar container.

  • cpu CPU limits for the sidecar container. Optional, overrides the default_cpu_limits operator configuration parameter. Optional.

  • memory memory limits for the sidecar container. Optional, overrides the default_memory_limits operator configuration parameter. Optional.

  • hugepages-2Mi hugepages-2Mi requests for the sidecar container. Optional, defaults to not set.

  • hugepages-1Gi 1Gi hugepages requests for the sidecar container. Optional, defaults to not set.

Connection pooler

Parameters are grouped under the connectionPooler top-level key and specify configuration for connection pooler. If this section is not empty, a connection pooler will be created for master service only even if enableConnectionPooler is not present. But if this section is present then it defines the configuration for both master and replica pooler services (if enableReplicaConnectionPooler is enabled).

  • numberOfInstances How many instances of connection pooler to create.

  • schema Database schema to create for credentials lookup function.

  • user User to create for connection pooler to be able to connect to a database. You can also choose a role from the users section or a system user role.

  • dockerImage Which docker image to use for connection pooler deployment.

  • maxDBConnections How many connections the pooler can max hold. This value is divided among the pooler pods.

  • mode In which mode to run connection pooler, transaction or session.

  • resources Resource configuration for connection pooler deployment.

Custom TLS certificates

Those parameters are grouped under the tls top-level key. Note, you have to define spiloFSGroup in the Postgres cluster manifest or spilo_fsgroup in the global configuration before adding the tls section'.

  • secretName By setting the secretName value, the cluster will switch to load the given Kubernetes Secret into the container as a volume and uses that as the certificate instead. It is up to the user to create and manage the Kubernetes Secret either by hand or using a tool like the CertManager operator.

  • certificateFile Filename of the certificate. Defaults to "tls.crt".

  • privateKeyFile Filename of the private key. Defaults to "tls.key".

  • caFile Optional filename to the CA certificate (e.g. "ca.crt"). Useful when the client connects with sslmode=verify-ca or sslmode=verify-full. Default is empty.

  • caSecretName By setting the caSecretName value, the ca certificate file defined by the caFile will be fetched from this secret instead of secretName above. This secret has to hold a file with that name in its root.

    Optionally one can provide full path for any of them. By default it is relative to the "/tls/", which is mount path of the tls secret. If caSecretName is defined, the ca.crt path is relative to "/tlsca/", otherwise to the same "/tls/".

Change data capture streams

This sections enables change data capture (CDC) streams via Postgres' logical decoding feature and pgoutput plugin. While the Postgres operator takes responsibility for providing the setup to publish change events, it relies on external tools to consume them. At Zalando, we are using a workflow based on Debezium Connector which can feed streams into Zalando’s distributed event broker Nakadi among others.

The Postgres Operator creates custom resources for Zalando's internal CDC operator which will be used to set up the consumer part. Each stream object can have the following properties:

  • applicationId The application name to which the database and CDC belongs to. For each set of streams with a distinct applicationId a separate stream resource as well as a separate logical replication slot will be created. This means there can be different streams in the same database and streams with the same applicationId are bundled in one stream resource. The stream resource will be called like the Postgres cluster plus "-" suffix. Required.

  • database Name of the database from where events will be published via Postgres' logical decoding feature. The operator will take care of updating the database configuration (setting wal_level: logical, creating logical replication slots, using output plugin pgoutput and creating a dedicated replication user). Required.

  • tables Defines a map of table names and their properties (eventType, idColumn and payloadColumn). Required. The CDC operator is following the outbox pattern. The application is responsible for putting events into a (JSON/B or VARCHAR) payload column of the outbox table in the structure of the specified target event type. The operator will create a PUBLICATION in Postgres for all tables specified for one database and applicationId. The CDC operator will consume from it shortly after transactions are committed to the outbox table. The idColumn will be used in telemetry for the CDC operator. The names for idColumn and payloadColumn can be configured. Defaults are id and payload. The target eventType has to be defined. One can also specify a recoveryEventType that will be used for a dead letter queue. By enabling ignoreRecovery, you can choose to ignore failing events.

  • filter Streamed events can be filtered by a jsonpath expression for each table. Optional.

  • enableRecovery Flag to enable a dead letter queue recovery for all streams tables. Alternatively, recovery can also be enable for single outbox tables by only specifying a recoveryEventType and no enableRecovery flag. When set to false or missing, events will be retried until consuming succeeded. You can use a filter expression to get rid of poison pills. Optional.

  • batchSize Defines the size of batches in which events are consumed. Optional. Defaults to 1.

  • cpu CPU requests to be set as an annotation on the stream resource. Optional.

  • memory memory requests to be set as an annotation on the stream resource. Optional.