Dedicated compute node

By default, a compute node can process both streaming jobs and ad-hoc queries (i.e. serving). Alternatively, you can process streaming and serving by different compute nodes respectively, in order to mitigate resource contention, or to isolate streaming or serving failure. This guides introduces how to set up such dedicated compute nodes to decouple streaming and serving.

1. Specify compute node role

To decouple streaming and serving, you need at least one compute node with the streaming role and one compute node with the serving role.

When launching a compute node, its role can be specified via either the --role command line argument, or RW_COMPUTE_NODE_ROLE environment variable.

You need to restart the node to update the role. A role can be one of:

• both: The default role if not specified. Indicates that the compute node is available for both streaming and serving.
• serving: Indicates that the compute node is only available for serving.
• streaming: Indicates that the compute node is only available for streaming.

2. Specify visibility mode

To decouple streaming and serving, the session variable visibility_mode must be checkpoint, in order to schedule ad-hoc queries to compute nodes with serving role. Otherwise, ad-hoc queries are still served by compute nodes with the streaming role by default (i.e. visibility_mode is all):

• Preferably, you can change the default visibility_mode to checkpoint for all new sessions of a frontend node, via either the --enable-barrier-read=false command line argument or the environment variable RW_ENABLE_BARRIER_READ=false when launching frontend nodes.
• As a session variable, you can modify it during a session. This can be useful when no compute node with the serving role is available. By setting visibility_mode to all temporarily, ad-hoc queries from this session can be served by compute nodes with the streaming role.

About visibility mode

Data visibility mode determines the data freshness perceived in serving. It can be one of:

• all: The default mode. It indicates better data freshness, with the risk of reading data that may not be eventually persisted if a cluster recovery happens later. Under this mode, ad-hoc queries are assigned to compute nodes with either the both or the streaming role.
• checkpoint: It ensures data that you read must be persisted. Under this mode, ad-hoc queries are scheduled to compute nodes with either the both or the serving role.