How to start a private clique

A private clique is a clique that can be accessed only by certain users or units.

Start conditions

To start a private clique, you need a dedicated compute pool with CPU guarantees (strong_guarantee_resources). If there is not any, you need to:

Ask colleagues whether it is possible to find a compute pool with available resources.
Order new resources for the future.
Start a clique without guarantees (not recommended).

How to start

Install the CHYT CLI by installing the ytsaurus-client package.
Start the clique. Below is an example of a clique of five instances with default settings (16 cores per instance) in the chyt pool called chyt_example_clique.

CPU quota request

To create a clique with N instances on the <cluster_name> cluster, you need to:

<cluster_name> <pool_tree_name> <running_operations> : 1 (a clique is a single YTsaurus operation that is always running).
<cluster_name> <pool_tree_name> <total_operations>: 1 (must be greater than or equal to the limit on the number of running operations).
<cluster_name> <pool_tree_name> <CPU strong guarantee>: N * 16 (there are 16 CPU cores per instance by default).

To avoid mentioning the YTsaurus cluster name in each command (in the --proxy argument), as the first step, set the default value by the environment variable:

export YT_PROXY=<cluster_name>

Next, create a clique, using the command from the CHYT Controller CLI:

yt clickhouse ctl create chyt_example_clique

Once you've created the clique, configure it with the relevant options: Every clique has a mandatory pool option with the name of the compute pool where the operation on instances will be started for the clique. Because the operations for the clique are started by the robot-strawberry-controller system robot, first you need to grant the Use permission for the specified pool to this robot.

When setting the pool option, make sure that you have the Use right for the specified pool (otherwise, the command will fail):

yt clickhouse ctl set-option pool chyt_example_pool --alias chyt_example_clique

To set the instance count, use instance_count:

yt clickhouse ctl set-option instance_count 2 --alias chyt_example_clique

The clique is now configured, and all its settings are saved in Cypress. The only thing left to do is to start the configured clique:

yt clickhouse ctl start chyt_example_clique

To view the clique status, use the status command. Once the clique's operation is started, the value of the status field should change to Ok, and operation_state should change to running:

$ yt clickhouse ctl status chyt_example_clique
{
    "status" = "Waiting for restart: oplet does not have running yt operation";
}
# a few moments later
$ yt clickhouse ctl status chyt_example_clique
{
    "status" = "Ok";
    "operation_state" = "running";
    "operation_url" = "https://domain.com/<cluster_name>/operations/48bdec5d-ed641014-3fe03e8-4289d62e";
}

After starting the operation, you need to wait for the clique instances to start up and start accepting incoming queries. To make sure that the clique works, make a test query to the table //sys/clickhouse/sample_table. This table is available on all clusters with CHYT:

$ yt clickhouse execute --proxy <cluster_name> --alias chyt_example_clique 'select avg(a) from `//sys/clickhouse/sample_table`'
224.30769230769232

If the clique is unavailable for more than 10 minutes, try to identify the problem by following the operation_url from the status command in the operation's web interface. If you don't manage to work it out on your own, write to the YTsaurus chat.