Skip to content

Routers

Trino Gateway offers two entry-level router options, providing users with a straightforward and easy-to-use starting point for their routing needs. The routers make the decision based on the clusters load reported in ClusterStats.

StochasticRoutingManager

This primary routing mechanism employs a straightforward, decentralized approach, dispatching incoming queries in a random manner without utilizing advanced optimization techniques.

QueryCountBasedRouterProvider

This routing mechanism utilizes near real-time, user-level cluster load statistics. It uses running queries count and query queue lengths to determine the most suitable cluster for each individual user. With is information it directs queries to the least loaded cluster for that user, optimizing the likelihood of successful execution.

Adding a routing mechanism

To enhance Trino Gateway's capabilities, you can create and contribute new and advanced router modules with intelligent routing features. To integrate a new router, you need to create a provider module that can be configured via the configuration file. This allows for seamless addition of new routers without disrupting existing functionality.

Add router Provider module

Use the following steps to incorporate a new routing mechanism with advanced capabilities:

  • Derive a class from RouterBaseModule
  • The module must instantiate the router and hold a reference to it.
  • Add the module name to the modules section of the configuration file to load the provider module and make the new router available.
  • For example, QueryCountBasedRouterProvider and refer to the config file in the following sections.

Add router class

Use the following steps to create a new router:

  • Derive a class from StochasticRoutingManager to create the router that does the actual work.
  • Override the methods provideAdhocBackend and provideBackendForRoutingGroup and implement the new smarter logic
  • The router listens to the list of ClusterStats via theupdateBackEndStats method.
  • This method is called on regular intervals defined in the config parameter monitor=>taskDelay.
  • Each element in the list corresponds to each backend cluster.
  • Only the stats from the healthy cluster are reported, unhealthy clusters are not included in the list. If you have three cluster backends and one is unhealthy, then the parameter List<ClusterStats> stats has only two elements.
  • To get the cluster stats set the parameter clusterStatsConfiguration=>monitorType to UI_API or JDBC which in turn needs the setup of backendState section in the config file.

Configuration file reference

backendState:
  username: <usernme>
  password: <password>
  ssl: <false/true>
  xForwardedProtoHeader: <false/true>

clusterStatsConfiguration:
  monitorType: UI_API

modules:
  - io.trino.gateway.ha.module.QueryCountBasedRouterProvider