mml.core.scripts.notifier

Module to send notifications of MML execution via various media to the user.

class BaseNotifier

Bases: ABC

__init__(on_start: bool = False, on_end: bool = False, on_failure: bool = True)

The notifier base class provides utilities to send messages for mml events.

Parameters:

• on_start – emit message on mml start

• on_end – emit message on (successful) mml end

• on_failure – emit message on mml failure (except keyboard interrupt / user caused SIGTERM signal)

abstract emit_message(text: str) → None

Abstract method that must implement the message delivery mechanism for inherited classes.

Parameters:

text – the text to be sent

Returns:

get_message_header() → str

Helper to return uniform message header for all kinds of messages.

Returns:

A header string to be used by notifiers.

static is_master() → bool

Checks if the current process is the master process - used e.g. in Multi-GPU settings to prevent sending multiple messages from each node / process.

Returns:

notify_on_end(return_value: float | None) → None

The notification wrapper function to be called from outside in case MML end.

Parameters:

return_value (Optional[float]) – the return value of the scheduler

Returns:

notify_on_failure(error: BaseException) → None

The notification wrapper function to be called from outside in case MML fails.

Parameters:

error (Exception) – an error that caused MML to fail

Returns:

notify_on_start() → None

The notification wrapper function to be called from outside in case MML starts.

Returns:

static search_overrides() → str

For enriching messages the used overrides to call MML are tried to be read from the hydra config. If reading is not successful the returned string is a message describing the failed attempt to read

Returns:

if successful the string that can be used to reproduce the MML call, otherwise a failure description

static search_run_path() → str

For enriching messages the run path of the current experiment is tried to be read from various sources.

Returns:

if successful the experiment path is extracted, otherwise a fallback notification

class DummyNotifier

Bases: BaseNotifier

emit_message(text: str) → None

The dummy notifier does not emit any message. It may be used in testing or to avoid being messaged.

class EMailNotifier

Bases: BaseNotifier

emit_message(text: str) → None

A function to send a simple email notification.

All email settings are read from the mml.env environment variables (thus they have to be set beforehand): MML_SMTP_SERVER MML_SMTP_SERVER_PORT MML_SENDER_EMAIL MML_RECEIVER_EMAIL MML_MAIL_PASSWORD=NO_PASSWORD (set to NO_PASSWORD if there is no password necessary, otherwise set password)

Parameters:

text – message to be sent via email

Returns:

class SlackNotifier

Bases: BaseNotifier

emit_message(text: str) → None

A function to send a simple slack notification. All slack settings are read from the mml.env environment variables (thus they have to be set beforehand).

A slack app has to be created via the web interface and added to the workspace. Here are some example settings:

display_information:
  name: MML Monitoring
  description: Alerts from the mml monitoring system.
  background_color: "#8B0000"
features:
  bot_user:
    display_name: MML Monitoring Alert
    always_online: true
oauth_config:
  scopes:
    bot:
      - incoming-webhook
settings:
  org_deploy_enabled: false
  socket_mode_enabled: false
  token_rotation_enabled: false

Generate the webhook and store it inside the mml.env under “MML_SLACK_WEBHOOK_URL” to enable the notifier.

Parameters:

text – message to be sent via slack

Returns: