TaskManager

View Licence Agreement

sosw - Serverless Orchestrator of Serverless Workers

The MIT License (MIT)
Copyright (C) 2024  sosw core contributors <info@sosw.app>

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

class sosw.managers.task.TaskManager(custom_config=None, **kwargs)

TaskManager is the core class used by most sosw Lambdas. It handles all the operations with tasks thus the configuration of this Manager is essential during your sosw implementation.

The default version of TaskManager works with DynamoDB tables to store and analyze the state of Tasks. This could be upgraded in future versions to work with other persistent storage or DBs.

The very important concept to understand about Task workflow is greenfield. Read more.

construct_payload_for_task(**kwargs) → str

Combines remaining kwargs to a singular JSON payload.

create_task(labourer: Labourer, strict: bool = True, **kwargs)

Schedule a new task.

Parameters:

• labourer – Labourer object of Lambda to execute the task.

• strict (bool) – By default (True) prohibits specifying in the task (kwargs) the fields that are supposed to be autogenerated. Only if they match with autogen - then pass. You can override this and pass custom task properties setting strict = False

get_completed_tasks_for_labourer(labourer: Labourer) → List[Dict]

Return a list of tasks of the Labourer marked as completed. Scavenger is supposed to archive them all so no special filtering is required here.

In order to be able to use the already existing index_greenfield, we sort tasks only in invoked stages (greenfield > now()). This number is supposed to be small, so filtering by an un-indexed field will be fast.

get_count_of_running_tasks_for_labourer(labourer: Labourer) → int

Returns a number of tasks we assume to be still running. Theoretically they can be dead with Exception, but not yet expired.

get_db_field_name(key: str) → str

Could be useful if you overwrite field names with your own ones (e.g. for tests).

get_expired_tasks_for_labourer(labourer: Labourer) → List[Dict]

Return a list of tasks of Labourer previously invoked, and expired without being closed.

get_invoked_tasks_for_labourer(labourer: Labourer, completed: bool | None = None) → List[Dict]

Return a list of tasks of current Labourer invoked during the current run of the Orchestrator.

If completed is provided: * True - filter completed ones * False - filter NOT completed ones * None (default) - do not care about completed status.

get_labourers() → List[Labourer]

Return configured Labourers. Config of the TaskManager expects ‘labourers’ as a dict ‘name_of_lambda’: {‘some_setting’: ‘value1’}

get_length_of_queue_for_labourer(labourer: Labourer) → int

Approximate count of tasks still in queue for labourer. Tasks with greenfield <= now()

Parameters:

labourer –

Returns:

get_newest_greenfield_for_labourer(labourer: Labourer) → int

Return value of the newest greenfield in queue. This means the end of the queue or latest added.

get_next_for_labourer(labourer: Labourer, cnt: int = 1, only_ids: bool = False) → List[str | Dict]

Fetch the next task(s) from the queue for the Labourer.

Parameters:

• labourer – Labourer to get next tasks for.

• cnt – Optional number of Tasks to fetch.

• only_ids – If explicitly set True, then returns only the IDs of tasks. This could save some transport if you are sending big batches of tasks between Lambdas.

get_oldest_greenfield_for_labourer(labourer: Labourer, reverse: bool = False) → int

Return value of oldest greenfield in queue. This means the beginning of the queue if you need FIFO behaviour.

get_running_tasks_for_labourer(labourer: Labourer, count: bool = False) → List[Dict] | int

Return a list of tasks of Labourer previously invoked, but not yet closed or expired. We assume they are still running.

If count is specified as True will return just the number of tasks, not the items themselves. Much cheaper.

get_task_by_id(task_id: str) → Dict

Fetches the full data of the Task.

invoke_task(labourer: Labourer, task_id: str | None = None, task: Dict | None = None)

Invoke the Lambda Function execution for task. Providing the ID is more expensive, but safer from “task injection” attacks method that prefetches Task from the table before trying to invoke.

Skips already running tasks with no exception, thus concurrent Orchestrators (or whoever else) should not duplicate invocations.

is_valid_task(task: Dict) → bool

Simple validation for required fields.

mark_task_invoked(labourer: Labourer, task: Dict, check_running: bool | None = True)

Update the greenfield with the latest invocation timestamp + invocation_delta

By default updates with a conditional expression that fails in case the current greenfield is already in invoked state. If this check fails the function raises RuntimeError that should be handled by the Orchestrator. This is very important to help duplicate invocations of the Worker by simultaneously running Orchestrators.

Parameters:

• labourer – Labourer for the task

• task – Task dictionary

• check_running – If True (default) updates with conditional expression.

:raises RuntimeError

move_task_to_retry_table(task: Dict, wanted_delay: int)

Put the task to a Dynamo table sosw_retry_tasks, with the wanted delay: labourer.max_runtime * attempts. Delete it from sosw_tasks table.

register_labourers() → List[Labourer]

Sets timestamps, health status and other custom attributes on Labourer objects passed for registration.

We also send a pointer to the TaskManager (aka self) to Ecology Manager. The latter will have to make some queries, and we don’t want him to initialise another TaskManager for himself.

retry_task(task: Dict, labourer_id: str, greenfield: int)

Move task to tasks table, in beginning of the queue (with greenfield of a task that will be invoked next). This method is called by Scavenger.