Writer scaling properties#
Writer scaling allows Trino to dynamically scale out the number of writer tasks rather than allocating a fixed number of tasks. Additional tasks are added when the average amount of physical data per writer is above a minimum threshold, but only if the query is bottlenecked on writing.
Writer scaling is useful with connectors like Hive that produce one or more files per writer – reducing the number of writers results in a larger average file size. However, writer scaling can have a small impact on query wall time due to the decreased writer parallelism while the writer count ramps up to match the needs of the query.
Enable writer scaling by dynamically increasing the number of writer tasks on
the cluster. This can be specified on a per-query basis using the
Enable scaling the number of concurrent writers within a task. The maximum writer
count per task for scaling is
writers are added only when the average amount of physical data written per writer
is above the minimum threshold of
writer-min-size and query is bottlenecked on
writing. This can be specified on a per-query basis using the
Restrictions: Must be a power of two
Maximum number of concurrent writers per task upto which the task can be scaled when
task.scale-writers.enabled is set. Increasing this value may improve the
performance of writes when the query is bottlenecked on writing. Setting this too high
may cause the cluster to become overloaded due to excessive resource utilization.
Type: data size
The minimum amount of data that must be written by a writer task before
another writer is eligible to be added. Each writer task may have multiple
writers, controlled by
task.writer-count, thus this value is effectively
divided by the number of writers per task. This can be specified on a
per-query basis using the
writer_min_size session property.