Scale-out Cluster

Through a series of operations, add database nodes in the original cluster.

Purpose

PTK provides users with the function of cluster scale-out to meet their needs for business, cost, resource allocation, risk and other factors.

Scale-out Logic

When PTK scales-out, it packages static files such as application directories and tool directories on the primary node, copies them to the target machine and then unpacks them into the corresponding directories. Then the kernel tool is used to initialize a new data directory, and finally the cluster configuration is refreshed according to the topology of the new cluster as a whole.

Note that the scale-out is done in a node-by-node mode. When the scale-out fails at a node, PTK will stop the scale-out immediately and refresh the configuration according to the completed cluster.

Scale-out Process Demonstration

Create The Configuration File Needed for Scaling-out

Use the following command to create a scale-out configuration file

ptk template scale-out > scale-out.yaml

At this point a scale-out configuration file scale-out.yaml has been created, modify this configuration file as needed.

The contents of the scale-out template are described below.

# New list of database servers added, with the same fields supported as in the installation
db_servers:
  - host: "replace host ip here"
    # Roles only support "standby" (default) or "cascade_standby"
    role: standby
    upstream_host: ""
    ssh_option:
      port: 22
      user: root
      password: "encrypted ssh password by ptk"

# List of CM servers
# If the cluster before scale-out has CM installed, you need to specify the list of CM servers for scale-out.
# In general, the list of CM servers is the same as the list of database servers.
# But if you scale-out only the database or only the CM on the new server, the list of machines in the two lists can be different.
cm_servers:
  - host: "replace host ip here"

Perform Scale-out Operations On Clusters

Scale-out command:

ptk cluster -n <CLUSTER_NAME> scale-out -c CONFIG.yaml [--build-from BUILD_FROM_HOST] [--skip-create-user] [--skip-check-os] [--skip-check-distro] [--default-guc] [--skip-rollback] [--skip-gen-ptkc] [--cpu CPU_MODEL] [--not-limit-cm-nodes]

Options:

option name	option type	description
-c	String	Specify the configuration file to use when scaling-out
--build-from	String	Specify the data source node from which the cascade_standby node will be built when scaling-out; if not specified, build from the primary node by default.
--skip-create-user	Bool	Skip user creation for scale-out nodes
--skip-check-os	Bool	Skip OS-related checks for scale-out nodes
--skip-check-distro	Bool	Skip distro checks for scale-out nodes
--default-guc	Bool	Use the database default parameter configuration for scale-out nodes
--skip-rollback	Bool	Skip rollback if errors are reported during scale-out
--skip-gen-ptkc	Bool	Skip transmitting ptkc for scale-out nodes
--cpu	String	Specify the CPU Model of the scale-out node.
--not-limite-cm-nodes	Bool	No limitation on the number of CM nodes, the default number of CM nodes must be odd if the number of CM nodes is greater than 3

Scale-out Support List

original cluster	scale-out condition	Whether to support	solution
db1,db2	db3	Yes
db1+cm1,db2+cm2,db3+cm3	db4+cm4	Yes
db1+cm1,db2+cm2,db3+cm3	db4	Yes
db1+cm1,db2+cm2,db3+cm3	cm4	Yes
db1+cm1,db2+cm2,db3	cm3	No	Scale-in db3, scale-out db3+cm3
db1+cm1,db2+cm2,cm3	db3	No	Scale-in cm3, scale-out db3+cm3

QA

Can a cluster be scaled-in?

Yes. Refer to Scale-in Cluster

What is the maximum number of nodes for scale-out?

Up to 9 nodes.

Can I scale-out a primary node?

No.

Can I scale-out a cluster with CM? If yes, is there a restriction?

Yes. See scale-out support list for more details.

What does `--skip-rollback` do? When should it be used?

Usage: PTK performs scale-out on a node-by-node basis. If a node fails when performing a scale-out operation, it rolls back the operations that have already been performed on that node. The role of --skip-rollback is to not roll back operations that have already been performed on a node if the scale-out fails.

When to use: Users can use this option if they want to see why the scale-out failed on the target node.

What does `--skip-check-distro` do? When should it be used?

Usage: When scaling-out, PTK will check if the operating system of the node to be scaled-out is the same as the primary node by default. If it is not, the check fails. The role of --skip-check-distro is to skip this check. When to use:

Misjudgment caused by incomplete judgment of the PTK on an homogeneous system.
In a heterogeneous system with consistent MogDB packages, this option can be used to skip the check, but PTK does not guarantee the availability and correctness of the database nodes after a successful scale-out.

What does `--not-limite-cm-nodes` do? When should it be used?

Usage: When PTK scales-out, if the number of CM nodes is greater than 3, it will force the number of CM nodes to be an odd number, in order to prevent the primary selection problem caused by the CM cluster brain-split. Use this option to avoid the mandatory check for an odd number of CM nodes.

When to use: It is not recommended to use this option, forcing the use of this option is likely to encounter problem with primary selection caused by brain-split of CM clusters, which can then cause CM to fail.

If CM is currently deployed in the cluster and VIPs are configured, will scale-out affect VIPs?

No. PTK refreshes CM-VIP after scale-out for clusters that have VIPs configured to ensure that the basic CM-VIP information is correct.