Callbacks

The Callback interface enables custom code to be triggered by various Hydra events.

To use the callback API, one should import Hydra's Callback class:

from hydra.experimental.callback import Callback

Users can then create subclasses of this Callback class, overriding one or more of the methods defined by Callback. For the methods of a subclass to be called at the appropriate time, the subclass must be registered with Hydra in the hydra.callbacks config (see examples below).

The full API exposed by the hydra.experimental.callback.Callback class is listed below:

Events supported (Click to expand)

class Callback:
    def on_run_start(self, config: DictConfig, **kwargs: Any) -> None:
        """
        Called in RUN mode before job/application code starts. `config` is composed with overrides.
        Some `hydra.runtime` configs are not populated yet.
        See hydra.core.utils.run_job for more info.
        """
        ...

    def on_run_end(self, config: DictConfig, **kwargs: Any) -> None:
        """
        Called in RUN mode after job/application code returns.
        """
        ...

    def on_multirun_start(self, config: DictConfig, **kwargs: Any) -> None:
        """
        Called in MULTIRUN mode before any job starts.
        When using a launcher, this will be executed on local machine before any Sweeper/Launcher is initialized.
        """
        ...

    def on_multirun_end(self, config: DictConfig, **kwargs: Any) -> None:
        """
        Called in MULTIRUN mode after all jobs returns.
        When using a launcher, this will be executed on local machine.
        """
        ...

    def on_job_start(self, config: DictConfig, **kwargs: Any) -> None:
        """
        Called in both RUN and MULTIRUN modes, once for each Hydra job (before running application code).
        This is called from within `hydra.core.utils.run_job`. In the case of remote launching, this will be executed
        on the remote server along with your application code.
        """
        ...

    def on_job_end(
        self, config: DictConfig, job_return: JobReturn, **kwargs: Any
    ) -> None:
        """
        Called in both RUN and MULTIRUN modes, once for each Hydra job (after running
        application code).
        This is called from within `hydra.core.utils.run_job`. In the case of remote launching, this will be executed
        on the remote server after your application code.

        `job_return` contains info that could be useful for logging or post-processing.
        See hydra.core.utils.JobReturn for more.
        """
        ...

Configure Callback

Say we have MyCallback so after every job ends we can upload a certain file to a S3 bucket. For simplicity we include this Callback class within the application, in real life you should have the Callback in a separate file. Running the application, we can see our custom method on_job_end was called.

my_app.py

class MyCallback(Callback):
   def __init__(self, bucket: str, file_path: str) -> None:
        self.bucket = bucket
        self.file_path = file_path

   def on_job_end(self, config: DictConfig, **kwargs: Any) -> None:
        print(f"Job ended,uploading...")
        # uploading...

@hydra.main(config_path="conf", config_name="config")
def my_app(cfg: DictConfig) -> None:
    print(OmegaConf.to_yaml(cfg))


if __name__ == "__main__":
    my_app()

output

$ python  my_app.py
foo: bar

Job ended,uploading...

Now let's take a look at the configurations.

$ tree conf

conf
├── config.yaml
└── hydra
    └── callbacks
        └── my_callback.yaml

conf/config.yaml

defaults:
 - /hydra/callbacks:
    - my_callback

foo: bar

conf/hydra/callbacks/my_callback.yaml

# @package _global_
hydra:
  callbacks:
    my_callback:
      _target_: my_app.MyCallback
      bucket: my_s3_bucket
      file_path: ./test.pt

Callback ordering

The on_run_start or on_multirun_start method will get called first, followed by on_job_start (called once for each job). After each job on_job_end is called, and finally either on_run_end or on_multirun_end is called one time before the application exits.

In the hydra.callbacks section of your config, you can use a list to register multiple callbacks. They will be called in the final composed order for start events and in reversed order for end events. So, for example, suppose we have the following composed config:

python my_app.py --cfg hydra -p hydra.callbacks

# @package hydra.callbacks
my_callback1:
  _target_: my_app.MyCallback1
  param1: val1
my_callback2:
  _target_: my_app.MyCallback2
  param2: val2

Before each job starts, MyCallback1.on_job_start will get called first, followed by MyCallback2.on_job_start. After each job ends, MyCallback2.on_job_end will get called first, followed by MyCallback1.on_job_end.