`labtasker task`¶

Abstract

This page details the task related operations.

Submit tasks¶

Specifying task args via `--args` option or `[ARGS]` argument¶

You can submit the task args that you want to run.

Specify args via --args optionSpecify args via [ARGS] argument

labtasker task submit --args '{"foo.bar": 0.1}'

Tip

Specify task args via [ARGS] argument is convenient if you have a job script that takes CLI arguments directly.

Normally, you would run it like this (suppose your job script takes --foo.bar as a CLI argument):

python train.py --foo.bar 0.1

You can directly copy and paste the --foo.bar argument as a positional argument to labtasker task submit.

labtasker task submit -- --foo.bar 0.1

arg cast to python primitive

By default, labtasker task submit will cast the args to python primitive types.

Therefore, if you specify -- --foo 0.1, it will be cast to float type rather than remain as a string.

The same principle applies to labtasker task update.

Dict nesting

Labtasker parses the dot-separated top-level keys as keys to nested dicts.

For example:

--foo.bar 0.1 would give you task args as {"foo": {"bar": 0.1}}.

If you are accessing task args via labtasker.task_info().args, remember to access like labtasker.task_info().args["foo"]["bar"] rather than labtasker.task_info().args["foo.bar"].

The same principle applies to args, metadata, summary for queues, tasks and workers.

Metadata¶

Metadata is handy if you want to filter tasks according to certain conditions.

For example, you may implement a custom tag system to manage your tasks through metadata.

Example

# submit a task 1
labtasker task submit --metadata '{"tags": ["test", "experimental"]}' -- --foo.bar 0.1
# submit a task 2
labtasker task submit --metadata '{"tags": ["tag-baz"]}' -- --foo.baz 0.2

# loop up tasks that contains 'experimental' or 'tag-baz' (1)
labtasker task ls --extra-filter '{"metadata.tags": {"$in": ["experimental", "tag-baz"]}}' --quiet --no-pager

# output:
# dd4e3ce3-25a8-4176-bf4b-dba82187679d
# 155f3872-1a07-45c9-85fb-51fce5cc29b5

See more about labtasker task ls in List tasks.

List (query) tasks¶

By default, labtasker task ls displays all tasks in the queue. Output is shown through a pager like less by default. Add --no-pager to display directly in the terminal.

You can filter tasks using:

--task-id or --task-name for basic filtering
--extra-filter / -f for advanced queries

Using Filters¶

Choose between two filter syntaxes:

Python Native Syntax: Intuitive to use but less powerful.
```
# Find tasks where args.foo.bar > 0.1
labtasker task ls -f 'args.foo.bar > 0.1' --quiet --no-pager
```
Note: Does not support not in, not expr, or != due to null value ambiguities

MongoDB Syntax: More powerful but requires MongoDB knowledge.

# Find tasks where args.foo.bar > 0.1
labtasker task ls -f '{"args.foo.bar": {"$gt": 0.1}}' --quiet --no-pager

You can see the transpiled query using --verbose option.

See detail in How to Use Filter.

Modify (update) tasks¶

By default, labtasker task update will open terminal editor (such as vim) to allow you edit the task info.

The read-only fields are marked with a comment # Read-only. DO NOT modify!.

However, you may specify the --update option or the [UPDATES] argument to specify the fields to update instead of opening the editor.

Specify args via --update/-u optionSpecify args via [UPDATES] argument

# Example of updating task
labtasker task update --id 9ca765ce-94fe-4e2f-b88b-954b3412e607 -u 'args.arg1=0.3' -u 'args.arg2={"arg21": 0.4}'

# Example of updating task
labtasker task update --id 9ca765ce-94fe-4e2f-b88b-954b3412e607 -- args.arg1=0.3 args.arg2='{"arg21": 0.4}'

If you wish to use this command in a bash script, use --quiet option to disable unnecessary output and confirmations.

If you want to filter out which tasks to update, use --extra-filter / -f option.

See detail in How to Use Filter.

Delete tasks¶

labtasker task delete --help

To batch delete tasks, you can use Unix pipes.

Example:

# -f: filter out tasks; -q: quiet so that only task_ids are printed; -y: skip confirmation
labtasker task ls -f 'created_at > date("10 minutes ago")' -q | labtasker task delete -y

labtasker task¶