Dask Executor¶
flowdapt.compute.executor.dask.DaskExecutor
¶
Bases: Executor
An Executor built on top of Dask. Supports both local and distributed execution, and can be configured to use GPUs. Often best for larger than memory workloads.
To use this Executor, set the services
> compute
> executor
config
target, for example:
services:
compute:
executor:
target: flowdapt.compute.executor.dask.DaskExecutor
Parameters:
Name | Type | Description | Default |
---|---|---|---|
cluster_address |
str | None
|
The address of an existing Dask cluster to connect to. |
None
|
scheduler_host |
str
|
The host to start the scheduler on. |
'127.0.0.1'
|
scheduler_port |
int
|
The port to start the scheduler on. |
6684
|
scheduler_scheme |
str
|
The protocol to use for the scheduler. |
'tcp'
|
dashboard_port |
int
|
The port to start the dashboard on. |
9968
|
cpus |
int | Literal['auto']
|
The number of CPUs to use. If set to "auto", will use the number of CPUs on the machine. |
'auto'
|
gpus |
int | str
|
The number of GPUs to use, or a comma delimited list of device ID's. If set to "auto", will use the number of GPUs on the machine. Defaults to "disabled". |
'disabled'
|
threads |
int | Literal['auto']
|
The number of threads to use. Defaults to all available. |
'auto'
|
memory |
str
|
The amount of memory to use. Defaults to all available. |
'auto'
|
adaptive |
bool
|
Whether to use adaptive scaling. Defaults to False. |
False
|
resources |
dict[str, float]
|
A dictionary of custom resource labels to use for the cluster. These labels are logical and not physical, and are used to determine if a stage can be run on a worker. Base resources such as CPUs and memory are automatically added. Defaults to an empty dictionary. |
{}
|
pip |
list[str]
|
A list of pip packages to install on the workers. Defaults to an empty list. |
[]
|
env_vars |
dict[str, str]
|
A dictionary of environment variables to set on the workers. Defaults to an empty dictionary. |
{}
|
upload_plugins |
bool
|
Whether to upload plugins to the workers. This is ignored if running an in-process cluster. Defaults to False. |
False
|