Skip to main content

User code executors for Backend.AI kernels

Project description

A common base runner for various programming languages.

It manages an internal task queue so that multiple command/code execution requests are processed in the FIFO order, without garbling the console output.

How to write a new computation kernel

Inherit ai.backend.kernel.BaseRunner and implement the following methods:

  • async def init_with_loop(self)

    • Called after the asyncio event loop becomes available.

    • Mostly just pass.

    • If your kernel supports interactive user input, then put set self.user_input_queue as an asyncio.Queue object. It’s your job to utilize the queue object for waiting for the user input. (See handle_input() method in ai/backend/kernel/python/inproc.py for reference) If it’s not set, then any attempts for getting interactive user input will simply return "<user-input is unsupported>".

  • async def build_heuristic(self)

    • (Batch mode) Write a heuristic code to find some build script or run a good-enough build command for your language/runtime.

    • (Blocking) You don’t have to worry about overlapped execution since the base runner will take care of it.

  • async def execute_heuristic(self)

    • (Batch mode) Write a heuristic code to find the main program.

    • (Blocking) You don’t have to worry about overlapped execution since the base runner will take care of it.

  • async def query(self, code_text)

    • (Query mode) Directly run the given code snippet. Depending on the language/runtime, you may need to create a temporary file and execute an external program.

    • (Blocking) You don’t have to worry about overlapped execution since the base runner will take care of it.

  • async def complete(self, data)

    • (Query mode) Take a dict data that includes the current line of code where the user is typing and return a list of strings that can auto-complete it.

    • (Non-blocking) You should implement this method to run asynchronously with ongoing code execution.

  • async def interrupt(self)

    • (Query mode) Send an interruption signal to the running program. The implementation is up to you. The Python runner currently spawns a thread for in-process query-mode execution and use a ctypes hack to throw KeyboardInterrupt exception into it.

    • (Non-blocking) You should implement this method to run asynchronously with ongoing code execution.

NOTE: Existing codes are good referecnes!

How to use in your Backend.AI computation kernels

Install this package using pip via a RUN instruction in Dockerfile. Then, set the CMD instruction like below:

CMD ["/home/sorna/jail", "-policy", "/home/sorna/policy.yml", \
     "/usr/local/bin/python", "-m", "ai.backend.kernel", "<language>"]

where <language> should be one of the supported language names defined in lang_map variable in ai/backend/kernel/__main__.py file.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

backend.ai-kernel-runner-1.0.8.tar.gz (19.2 kB view details)

Uploaded Source

Built Distribution

backend.ai_kernel_runner-1.0.8-py3-none-any.whl (36.0 kB view details)

Uploaded Python 3

File details

Details for the file backend.ai-kernel-runner-1.0.8.tar.gz.

File metadata

File hashes

Hashes for backend.ai-kernel-runner-1.0.8.tar.gz
Algorithm Hash digest
SHA256 6ef9e918fb671e720a64808537a2919f04b125efc1421b07160d086136ec881f
MD5 a8785324a7353c05c008a30363ab42a3
BLAKE2b-256 283af728280149a190a802cb6c6808faf881f74fa44522646882e9ff64dfca25

See more details on using hashes here.

Provenance

File details

Details for the file backend.ai_kernel_runner-1.0.8-py3-none-any.whl.

File metadata

File hashes

Hashes for backend.ai_kernel_runner-1.0.8-py3-none-any.whl
Algorithm Hash digest
SHA256 f090d38ae0a8e00221edf37279aa21ee36116fcf11b7bc6613f19f1ae08e220b
MD5 19c162dda1e8b14703373c2c53d60ac1
BLAKE2b-256 a0958d85275c88d32e857a444f585f03f4255ef1fbe9d58e3a5af960e3a3aebf

See more details on using hashes here.

Provenance

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page