Skip to main content
Skip to main content

Compute Node

Compute Node

Starting from version 1.2.1, Doris supports the Compute Node.

Starting from this release, BE nodes can be divided into two categories:

  • Mix

    Mix node. The default type of BE node. This type of node can not only participate in calculations, but also be responsible for the storage of user data.

  • Computation

    Compute Node. It is not responsible for data storage, only data calculation.

As a special type of BE node, the computing node has no data storage capability and is only responsible for data calculation. Therefore, the compute node can be regarded as a stateless BE node, and nodes can be easily added and deleted.

Compute nodes are suitable for the following scenarios:

  • Query external data sources

    Compute nodes can be used to query external data sources, such as Hive, Iceberg, JDBC, etc. Doris is not responsible for the storage of external data source data, so you can use compute nodes to easily expand the computing capabilities of external data sources. At the same time, computing nodes can also be configured with cache directories to cache hotspot data from external data sources to further accelerate data reading.

Usage of Compute Node​

Add Compute Node​

Add configuration in BE's be.conf file:

be_node_role=computation

Then start the BE node, which will run as a compute node type.

You can then connect to Doris through the MySQL client and execute:

ALTER SYSTEM ADD BACKEND

Add this BE node. After the addition is successful, you can see that the node type is computation in the NodeRole column of SHOW BACKENDS.

Use Compute Node​

To use compute nodes, the following conditions need to be met:

  • The cluster contains compute nodes.
  • Added configuration in fe.conf: prefer_compute_node_for_external_table = true

At the same time, the following FE configuration will affect the usage strategy of compute node:

  • min_backend_num_for_external_table

    Before Doris 2.0 (inclusive), the default value of this parameter is 3. After version 2.1, the default parameter is -1.

    This parameter indicates: the minimum number of BE nodes expected to participate in external data query. -1 means that the value is equal to the number of compute nodes in the current cluster.

    for example. Assume that there are 3 compute nodes and 5 mix nodes in the cluster.

    If min_backend_num_for_external_table is set to less than or equal to 3. Then the external table's query will only use 3 compute nodes. If the setting is greater than 3, assuming it is 6, in addition to using 3 compute nodes, the external table's query will also select 3 additional mix nodes.

    In summary, this configuration is mainly used for the minimum number of BE nodes that can participate in query process, and will prefer to use compute nodes.

Note:

  1. Only after version 2.1, min_backend_num_for_external_table is supported to be set to -1. In previous versions, this parameter had to be a positive number. And this configuration only takes effect when prefer_compute_node_for_external_table = true.

  2. If prefer_compute_node_for_external_table is false. Then the external table's query will select any BE node.

  3. If there are no compute nodes in the cluster, none of the above configurations will take effect.

  4. If the min_backend_num_for_external_table value is greater than the total number of BE nodes, at most number of BE will be selected.

  5. The above configurations can be modified at runtime without restarting the FE node. And all FE nodes need to be configured.

Best Practices​

Resource isolation and elastic scaling for federated queries​

In federated query scenarios, users can deploy a dedicated group of compute nodes specifically for querying external table data. This allows for the isolation of query loads for external tables (such as large-scale analysis on Hive) from the query loads for internal tables (such as low-latency, fast data analysis).

Moreover, as compute nodes are stateless Backend (BE) nodes, they can be easily scaled up or down. For instance, a cluster of elastic compute nodes can be deployed using Kubernetes, allowing for the utilization of more compute nodes for data lake analysis during peak business periods, and rapid scaling down during off-peak times to reduce costs.

FAQ​

  1. Can compute nodes and mix nodes be interconverted?

    Compute nodes can be converted to mix nodes. However, mix nodes cannot be converted to compute nodes.

    • Converting compute nodes to mix nodes

      1. Stop the BE node.
      2. Remove the be_node_role configuration from be.conf, or set it to be_node_role=mix.
      3. Configure the correct storage_root_path for data storage directory.
      4. Start the BE node.
    • Converting mix nodes to compute nodes

      In principle, this operation is not supported because mix nodes already store data. If conversion is necessary, first perform a safe node decommission, then set up as a compute node in the manner of a new node.

  2. Do compute nodes need to configure a data storage directory?

    Yes. The data storage directory of a compute node will not store user data but will hold some information files of the BE node itself, such as cluster_id, as well as some temporary files generated during running.

    The storage directory for compute nodes requires very little disk space (on the order of MBs) and can be destroyed at any time along with the node without affecting user data.

  3. Can compute nodes and mix nodes configure a file cache directory?

    File cache accelerates subsequent queries for the same data by caching data files from recently accessed remote storage systems (HDFS or object storage).

    Both compute and mix nodes can set up a file cache directory, which needs to be created in advance.

    Additionally, Doris employs strategies like consistent hashing to minimize the probability of cache invalidation when nodes are scaled up or down.

  4. Do compute nodes need to be decommissioned through the DECOMMISSION operation?

    No. Compute nodes can be removed directly using the DROP BACKEND operation.