Scale-In/Out

What Is Scale-In/Out?

Scale-in/out indicates that you can add database nodes to or delete database nodes from the existing database cluster to meet different service requirements.

Terms:

Scale-out: Adds database nodes to the existing database cluster.
Scale-in: Deletes database nodes from the existing database cluster.

Prerequisites

Assume that you have a MogDB database cluster which can include a node or multiple nodes, and the cluster name is CLUSTER_NAME.

During scale-out, make sure that the database version and configuration is consistent with the system environment, avoiding the incompatibility problem.

For details about how to install a database cluster, see Database Installation.

Scale-Out

Scale-Out Parameters

# ptk cluster scale-out -h
Perform scale-out for a MogDB cluster.

Usage:
  ptk cluster scale-out [flags]

Examples:
ptk cluster -n <cluster_name> scale-out -c scale-out.yaml

Flags:
  -y, --assumeyes           Replies yes to all questions automatically. 
  -c, --config string       Specifies the scale-in and scale-out configuration file path.
      --cpu string          Specifies the CPU model. 
                            'ptk candidate cpu' can be used for querying the supported CPU models.
      --default-guc         Specifies that GUC parameters are not automatically optimized. The default database parameters are used.                   
  -h, --help                help for scale-out
      --skip-check-distro   Skips the system release check. 
      --skip-check-os       Skips the system check. 
      --skip-create-user    Skips user creation.
      --skip-rollback       Skips rollback for installation failures.

Scale-Out Principle

During scale-out, PTK randomly chooses a node in a cluster, packages static files such as the application and tool directories of the node, copies the package to the target machine, and then decompress it to a directory.

PTK then uses a kernel tool to initiate a new data directory, and refreshes the cluster configuration file based on the topology of the new cluster.

Note that nodes are scaled out one by one. When a node fails scale-out, scale-out is stopped, and the cluster configuration is refreshed for only cluster nodes that finish scale-out.

Create a Configuration File for Scale-Out

The ptk template subcommand includes scale-out, and it can generate a basic scale-out configuration file template.

ptk template scale-out > scale-out.yaml

In this case, a scale-out configuration file named scale-out.yaml is created. You need to modify it as needed.

The following introduces the scale-out configuration file template.

# The newly added database server list is the same as that during installation. 
db_servers:
  - host: "replace host ip here"
    # The role can be "standby" (default) or "cascade_standby".
    role: standby
    ssh_option:
      port: 22
      user: root
      password: "encrypted ssh password by ptk"

# CM component server list
# If a CM component is installed in a cluster before scale-out,, the CM server list needs to be specified.
# Typically, the CM server list is the same as that of the database. 
# If only database or CM component on the new server is scaled out, the two server lists can be inconsistent. 
cm_servers:
  - host: "replace host ip here"

Scale-Out Support Description

Original Cluster	Scale-Out Condition	Supported or Not	Solution
db1, db2	db3	Supported
db1, db2	cm1, cm2	Not Supported	Uninstall the cluster and specify the `--install-cm` parameter to install the cluster again.
db1+cm1, db2+cm2, db3+cm3	db4+cm4	Supported
db1+cm1, db2+cm2, db3+cm3	db4	Supported
db1+cm1, db2+cm2, db3+cm3	cm4	Supported
db1+cm1, db2+cm2, db3	cm3	Not Supported	Perform scale-in for db3 and perform scale-out for db3+cm3.
db1+cm1, db2+cm2, cm3	db3	Not Supported	Perform scale-in for cm3 and perform scale-out for db3+cm3.

Scale-Out Command

ptk cluster -n CLUSTER_NAME scale-out -c scale-out.yaml

Scale-Out Demonstration

Scale-In

View the Help Information

# ptk cluster scale-in -h

Perform scale-in for a MogDB cluster

Usage:
  ptk cluster scale-in [flags]

Examples:
ptk cluster -n <cluster_name> scale-in -H <IP1> -H <IP2>

Flags:
  -y, --assumeyes          Replies yes to all questions automatically.
      --force              If the target host cannot be connected using SSH, forcibly delete the target instance from the configuration of the upgraded cluster node. 
  -h, --help               help for scale-in
  -H, --host stringArray   Specifies the target IP address to be deleted. 
      --skip-clear-db      Skips deletion of the directory of the target database. 
      --skip-clear-user    Skips deletion of the system user of the target database.

Scale-In Command

ptk cluster -n CLUSTER_NAME scale-in -H IP1 -H IP2

The above command can remove database nodes whose IP addresses are IP1 and IP2 from a cluster.

Scale-in description

Scale-in for one primary server, multiple standby servers, and multiple cascaded servers

Assume that the cluster topology is as follows:

         standby1 —— cascade_standby1
       /
primary 
       \
         standby2 —— cascade_standby2

After deleting standby1, the topology is changed to the following one:

         standby1   cascade_standby1
                  /
primary          /
       \        /
         standby2 —— cascade_standby2

At this time, standby1 is not in the cluster but is independent of the cluster.

That is to say, if a standby node is to be deleted, there is a cascaded standby node under the standby node, and there are multiple standby nodes, PTK will randomly choose a standby node to connect the cascaded standby node.

Scale-in for one primary server, one standby servers, and one cascaded servers

Assume that the cluster topology is as follows:

primary —— standby1 —— cascade_standby1

After deleting standby1, the topology is changed to the following one:

primary    standby1    cascade_standby1

At this time, three nodes are independent of each other.

Scale-In Support Description

Note: Scale-In supports only deletion of a node based on the node level of the IP address.

Original Cluster	Scale-In Condition	Supported or Not	Solution
db1, db2, db3	db3	Supported
db1+cm1, db2+cm2, db3+cm3	db3+cm3 (same IP address)	Supported
db1+cm1, db2+cm2, db3+cm3	db3	Not Supported	Perform scale-in for db3+cm3 and perform scale-out for cm3.
db1+cm1, db2+cm2, db3+cm3	cm3	Not Supported	Perform scale-in for db3+cm3 and perform scale-out for db3.
db1+cm1, db2+cm2, db3	db3	Supported
db1+cm1, db2+cm2, cm3	cm3	Supported