Node management interfaces v5

You can add and remove nodes dynamically using the SQL interfaces.

bdr.create_node

This function creates a node.

Synopsis

bdr.create_node(node_name text, local_dsn text)

Parameters

  • node_name Name of the new node. Only one node is allowed per database. Valid node names consist of lowercase letters, numbers, hyphens, and underscores.
  • local_dsn Connection string to the node.

Notes

This function creates a record for the local node with the associated public connection string. There can be only one local record, so once it's created, the function reports an error if run again.

This function is a transactional function. You can roll it back and the changes made by it are visible to the current transaction.

The function holds lock on the newly created bdr node until the end of the transaction.

bdr.drop_node

Drops a node.

Warning

This function isn't intended for regular use. Execute it only if instructed by Technical Support.

This function removes the metadata for a given node from the local database. The node can be either:

  • The local node, in which case all the node metadata is removed, including information about remote nodes.
  • A remote node, in which case only metadata for that specific node is removed.

Synopsis

bdr.drop_node(node_name text, cascade boolean DEFAULT false, force boolean DEFAULT false)

Parameters

  • node_name Name of an existing node.
  • cascade Deprecated, will be removed in the future.
  • force Circumvents all sanity checks and forces the removal of all metadata for the given PGD node despite a possible danger of causing inconsistencies. Only Technical Support uses a forced node drop in case of emergencies related to parting.

Notes

Before you run this, part the node using bdr.part_node().

This function removes metadata for a given node from the local database. The node can be the local node, in which case all the node metadata are removed, including information about remote nodes. Or it can be the remote node, in which case only metadata for that specific node is removed.

Note

PGD4 can have a maximum of 1024 node records (both ACTIVE and PARTED) at one time because each node has a unique sequence number assigned to it, for use by snowflakeid and timeshard sequences. PARTED nodes aren't automatically cleaned up. If this becomes a problem, you can use this function to remove those records.

bdr.create_node_group

This function creates a PGD group with the local node as the only member of the group.

Synopsis

bdr.create_node_group(node_group_name text,
                      parent_group_name text DEFAULT NULL,
                      join_node_group boolean DEFAULT true,
                      node_group_type text DEFAULT NULL)

Parameters

  • node_group_name Name of the new PGD group. As with the node name, valid group names must consist of only lowercase letters, numbers, and underscores.
  • parent_group_name The name of the parent group for the subgroup.
  • join_node_group This parameter helps a node to decide whether to join the group being created by it. The default value is true. This is used when a node is creating a shard group that it doesn't want to join. This can be false only if you specify parent_group_name.
  • node_group_type The valid values are NULL, subscriber-only, datanode, read coordinator, and write coordinator. subscriber-only type is used to create a group of nodes that receive changes only from the fully joined nodes in the cluster, but they never send replication changes to other nodes. See Subscriber-only groups for more details. Datanode implies that the group represents a shard, whereas the other values imply that the group represents respective coordinators. Except subscriber-only, the other values are reserved for future use. NULL implies a normal general-purpose node group is created.

Notes

This function passes a request to the local consensus worker that's running for the local node.

The function isn't transactional. The creation of the group is a background process, so once the function finishes, you can't roll back the changes. Also, the changes might not be immediately visible to the current transaction. You can call bdr.wait_for_join_completion to wait until they are.

The group creation doesn't hold any locks.

bdr.alter_node_group_config

This function changes the configuration parameters of an existing PGD group. Options with NULL value (default for all of them) aren't modified.

Synopsis

bdr.alter_node_group_config(node_group_name text,
                            insert_to_update boolean DEFAULT NULL,
                            update_to_insert boolean DEFAULT NULL,
                            ignore_redundant_updates boolean DEFAULT NULL,
                            check_full_tuple boolean DEFAULT NULL,
                            apply_delay interval DEFAULT NULL,
                            check_constraints boolean DEFAULT NULL,
                            num_writers int DEFAULT NULL,
							enable_wal_decoder boolean DEFAULT NULL,
							streaming_mode text DEFAULT NULL,
                            default_commit_scope text DEFAULT NULL)

Parameters

  • node_group_name Name of an existing PGD group. The local node must be part of the group.

  • insert_to_update Reserved for backward compatibility.

  • update_to_insert Reserved for backward compatibility. This option is deprecated and will be disabled or removed in future versions of PGD. Use bdr.alter_node_set_conflict_resolver instead.

  • ignore_redundant_updates Reserved for backward compatibility.

  • check_full_tuple Reserved for backward compatibility.

  • apply_delay Reserved for backward compatibility.

  • check_constraints Whether the apply process checks the constraints when writing replicated data. This option is deprecated and will be disabled or removed in future versions of PGD.

  • num_writers Number of parallel writers for subscription backing this node group. -1 means the default (as specified by the GUC bdr.writers_per_subscription) is used. Valid values are either -1 or a positive integer.

  • enable_wal_decoder Enables/disables the decoding worker process. You can't enable the decoding worker process if streaming_mode is already enabled.

  • streaming_mode Enables/disables streaming of large transactions. When set to off, streaming is disabled. When set to any other value, large transactions are decoded while they're still in progress, and the changes are sent to the downstream. If the value is set to file, then the incoming changes of streaming transactions are stored in a file and applied only after the transaction is committed on upstream. If the value is set to writer, then the incoming changes are directly sent to one of the writers, if available. If parallel apply is disabled or no writer is free to handle streaming transaction, then the changes are written to a file and applied after the transaction is committed. If the value is set to auto, PGD tries to intelligently pick between file and writer, depending on the transaction property and available resources. You can't enable streaming_mode if the WAL decoder is already enabled.

    For more details, see Transaction streaming.

  • default_commit_scope The commit scope to use by default, initially the local commit scope. This applies only to the top-level node group. You can use individual rules for different origin groups of the same commit scope. See Origin groups for more details.

Notes

This function passes a request to the group consensus mechanism to change the defaults. The changes made are replicated globally using the consensus mechanism.

The function isn't transactional. The request is processed in the background so you can't roll back the function call. Also, the changes might not be immediately visible to the current transaction.

This function doesn't hold any locks.

Warning

When you use this function to change the apply_delay value, the change doesn't apply to nodes that are already members of the group. This restriction has little consequence on production use because this value normally isn't used outside of testing.

bdr.join_node_group

This function joins the local node to an already existing PGD group.

Synopsis

bdr.join_node_group (
    join_target_dsn text,
    node_group_name text DEFAULT NULL,
    pause_in_standby boolean DEFAULT false,
    wait_for_completion boolean DEFAULT true,
    synchronize_structure text DEFAULT 'all'
)

Parameters

  • join_target_dsn Specifies the connection string to an existing (source) node in the PGD group you want to add the local node to.
  • node_group_name Optional name of the PGD group. Defaults to NULL, which tries to detect the group name from information present on the source node.
  • pause_in_standby Optionally tells the join process to join only as a logical standby node, which can be later promoted to a full member.
  • wait_for_completion Wait for the join process to complete before returning. Defaults to true.
  • synchronize_structure Set the kind of structure (schema) synchronization to do during the join. Valid options are all, which synchronizes the complete database structure, and none, which doesn't synchronize any structure. However, it still synchronizes data.

If wait_for_completion is specified as false, this is an asynchronous call that returns as soon as the joining procedure starts. You can see progress of the join in logs and the bdr.event_summary information view or by calling the bdr.wait_for_join_completion() function after bdr.join_node_group() returns.

Notes

This function passes a request to the group consensus mechanism by way of the node that the join_target_dsn connection string points to. The changes made are replicated globally by the consensus mechanism.

The function isn't transactional. The joining process happens in the background and you can't roll it back. The changes are visible only to the local transaction if wait_for_completion was set to true or by calling bdr.wait_for_join_completion later.

Node can be part of only a single group, so you can call this function only once on each node.

Node join doesn't hold any locks in the PGD group.

bdr.switch_node_group

This function switches the local node from its current subgroup to another subgroup within the same existing PGD node group.

Synopsis

bdr.switch_node_group (
    node_group_name text,
    wait_for_completion boolean DEFAULT true
)

Parameters

  • node_group_name Name of the PGD group or subgroup.
  • wait_for_completion Wait for the switch process to complete before returning. Defaults to true.

If wait_for_completion is specified as false, this is an asynchronous call that returns as soon as the switching procedure starts. You can see progress of the switch in logs and the bdr.event_summary information view or by calling the bdr.wait_for_join_completion() function after bdr.switch_node_group() returns.

Notes

This function passes a request to the group consensus mechanism. The changes made are replicated globally by the consensus mechanism.

The function isn't transactional. The switching process happens in the background and you can't roll it back. The changes are visible only to the local transaction if wait_for_completion was set to true or by calling bdr.wait_for_join_completion later.

The local node changes membership from its current subgroup to another subgroup within the same PGD node group without needing to part the cluster. The node's kind must match that of existing nodes within the target subgroup.

Node switching doesn't hold any locks in the PGD group.

Restrictions: Currently, the function only allows switching between a subgroup and its PGD node group. To effect a move between subgroups it is necessary to make two separate calls: 1) switch from subgroup to node group and, 2) switch from node group to other subgroup.

bdr.promote_node

This function promotes a local logical standby node to a full member of the PGD group.

Synopsis

bdr.promote_node(wait_for_completion boolean DEFAULT true)

Notes

This function passes a request to the group consensus mechanism to change the defaults. The changes made are replicated globally by the consensus mechanism.

The function isn't transactional. The promotion process happens in the background, and you can't roll it back. The changes are visible only to the local transaction if wait_for_completion was set to true or by calling bdr.wait_for_join_completion later.

The promotion process holds lock against other promotions. This lock doesn't block other bdr.promote_node calls but prevents the background process of promotion from moving forward on more than one node at a time.

bdr.wait_for_join_completion

This function waits for the join procedure of a local node to finish.

Synopsis

bdr.wait_for_join_completion(verbose_progress boolean DEFAULT false)

Parameters

  • verbose_progress Optionally prints information about individual steps taken during the join procedure.

Notes

This function waits until the checks state of the local node reaches the target state, which was set by bdr.create_node_group, bdr.join_node_group, or bdr.promote_node.

bdr.part_node

Removes (parts) the node from the PGD group but doesn't remove data from the node.

You can call the function from any active node in the PGD group, including the node that you're removing. However, once the node is parted, it can't part other nodes in the cluster.

Note

If you're parting the local node, you must set wait_for_completion to false. Otherwise, it reports an error.

Warning

This action is permanent. If you want to temporarily halt replication to a node, see bdr.alter_subscription_disable().

Synopsis

bdr.part_node (
    node_name text,
	wait_for_completion boolean DEFAULT true,
	force boolean DEFAULT false
)

Parameters

  • node_name Name of an existing node to part.
  • wait_for_completion If true, the function doesn't return until the node is fully parted from the cluster. Otherwise the function starts the parting procedure and returns immediately without waiting. Always set to false when executing on the local node or when using force.
  • force Forces removal of the node on the local node. This sets the node state locally if consensus can't be reached or if the node parting process is stuck.
Warning

Using force = true can leave the PGD group in an inconsistent state. Use it only to recover from failures in which you can't remove the node any other way.

Notes

This function passes a request to the group consensus mechanism to part the given node. The changes made are replicated globally by the consensus mechanism. The parting process happens in the background, and you can't roll it back. The changes made by the parting process are visible only to the local transaction if wait_for_completion was set to true.

With force set to true, on consensus failure, this function sets the state of the given node only on the local node. In such a case, the function is transactional (because the function changes the node state) and you can roll it back. If the function is called on a node that is already in process of parting with force set to true, it also marks the given node as parted locally and exits. This is useful only when the consensus can't be reached on the cluster (that is, the majority of the nodes are down) or if the parting process is stuck. But it's important to take into account that when the parting node that was receiving writes, the parting process can take a long time without being stuck. The other nodes need to resynchronize any missing data from the given node. The force parting completely skips this resynchronization and can leave the other nodes in an inconsistent state.

The parting process doesn't hold any locks.

bdr.alter_node_interface

This function changes the connection string (DSN) of a specified node.

Synopsis

bdr.alter_node_interface(node_name text, interface_dsn text)

Parameters

  • node_name Name of an existing node to alter.
  • interface_dsn New connection string for a node.

Notes

Run this function and make the changes only on the local node. This means that you normally execute it on every node in the PGD group, including the node that is being changed.

This function is transactional. You can roll it back, and the changes are visible to the current transaction.

The function holds lock on the local node.

bdr.alter_subscription_enable

This function enables either the specified subscription or all the subscriptions of the local PGD node. This is also known as resume subscription. No error is thrown if the subscription is already enabled. Returns the number of subscriptions affected by this operation.

Synopsis

bdr.alter_subscription_enable(
    subscription_name name DEFAULT NULL,
    immediate boolean DEFAULT false
)

Parameters

  • subscription_name Name of the subscription to enable. If NULL (the default), all subscriptions on the local node are enabled.
  • immediate This currently has no effect.

Notes

This function isn't replicated and affects only local node subscriptions (either a specific node or all nodes).

This function is transactional. You can roll it back, and the current transaction can see any catalog changes. The subscription workers are started by a background process after the transaction has committed.

bdr.alter_subscription_disable

This function disables either the specified subscription or all the subscriptions of the local PGD node. Optionally, it can also immediately stop all the workers associated with the disabled subscriptions. This is also known as pause subscription. No error is thrown if the subscription is already disabled. Returns the number of subscriptions affected by this operation.

Synopsis

bdr.alter_subscription_disable(
    subscription_name name DEFAULT NULL,
    immediate boolean DEFAULT false,
    fast boolean DEFAULT true
)

Parameters

  • subscription_name Name of the subscription to disable. If NULL (the default), all subscriptions on the local node are disabled.
  • immediate Used to force the action immediately, stopping all the workers associated with the disabled subscription. When this option is true, you can't run this function inside of the transaction block.
  • fast This argument influences the behavior of immediate. If set to true (the default) it stops all the workers associated with the disabled subscription without waiting for them to finish current work.

Notes

This function isn't replicated and affects only local node subscriptions (either a specific subscription or all subscriptions).

This function is transactional. You can roll it back, and the current transaction can see any catalog changes. However, the timing of the subscription worker stopping depends on the value of immediate. If set to true, the workers receive the stop without waiting for the COMMIT. If the fast argument is set to true, the interruption of the workers doesn't wait for current work to finish.

bdr.alter_node_group_option

Modify the PGD node group routing configuration

Synopsis

bdr.alter_node_group_option(node_group_name text,
    config_key text,
    config_value text);

Parameters

  • node_group_name Name of the group to be changed.
  • config_key Key of the option in the node group to be changed.
  • config_value New value to be set for the given key.

bdr.alter_node_option

Modify the PGD node routing configuration

Synopsis

bdr.alter_node_option(node_name text,
    config_key text,
    config_value text);

Parameters

  • node_name Name of the node to be changed.
  • config_key Key of the option in the node to be changed.
  • config_value New value to be set for the given key.