Golem is a global, open-source, decentralized supercomputer that anyone can access. It connects individual machines - be that laptops, home PCs or even data centers - to form a vast network, the purpose of which is to provide a way to distribute computations to its provider nodes and allow requestors to utilize its unique potential - which can lie in its combined computing power, the geographical distribution or its censorship resistance.
Golem's requestor-side configuration consists of two separate components:
- the
yagnadaemon - your node in the new Golem network, responsible for communication with the other nodes, running the market and providing easy access to the payment mechanisms. - the requestor agent - the part that the developer of the specific Golem application is responsible for.
The daemon and the requestor agent communicate using three REST APIs which
yapapi - Golem's Python high-level API - aims to abstract to large extent to make
application development on Golem as easy as possible.
Assuming you have your Golem node up and running (you can find instructions on how to do that in the yagna repository and in our handbook), what you need to do is:
- prepare your payload - this needs to be a Docker image containing your application
that will be executed on the provider's end. This image needs to have its volumes
mapped in a way that will allow the supervisor module to exchange data (write and
read files) with it. This image needs to be packed and uploaded into Golem's image repository
using our dedicated tool -
gvmkit-build. - create your requestor agent - this is where
yapapicomes in. Utilizing our high-level API, the creation of a requestor agent should be straighforward and require minimal effort. You can use examples contained in this repository (blender and hashcat) as references.
There are a few components that are crucial for any requestor agent app:
The heart of the high-level API is the requestor's task executor (yapapi.Executor).
You tell it, among others, which package (VM image) will be used to run your task,
how much you'd like to pay and how many providers you'd like to involve in the execution.
Finally, you feed it the worker script and a list of Task objects to execute on providers.
The worker will most likely be the very core of your requestor app. You need to define
this function in your agent code and then you pass it to the Executor.
It receives a WorkContext (yapapi.WorkContext) object that serves
as an interface between your script and the execution unit within the provider.
Using the work context, you define the steps that the provider needs to execute in order
to complete the job you're giving them - e.g. transferring files to and from the provider
or running commands within the execution unit on the provider's end.
Depending on the number of workers, and thus, the maximum number of providers that your
Executor utilizes in parallel, a single worker may tackle several tasks
(units of your work) and you can differentiate the steps that need to happen once
per worker run, which usually means once per provider node - but that depends on the
exact implementation of your worker function - from those that happen for each
individual unit of work. An example of the former would be an upload of a source
file that's common to each fragment; and of the latter - a step that triggers the
processing of the file using a set of parameters specified in the Task data.
The Task (yapapi.Task) object describes a unit of work that your application needs
to carry out.
The Executor will feed an instance of your worker - bound to a single provider node -
with Task objects. The worker will be responsible for completing those tasks. Typically,
it will turn each task into a sequence of steps to be executed in a single run
of the execution script on a provider's machine, in order to compute the task's result.
An example Golem application, using a Docker image containing the Blender renderer:
import asyncio
from yapapi import Executor, Task, WorkContext
from yapapi.log import enable_default_logger, log_summary, log_event_repr
from yapapi.package import vm
from datetime import timedelta
async def main(subnet_tag: str):
package = await vm.repo(
image_hash="9a3b5d67b0b27746283cb5f287c13eab1beaa12d92a9f536b747c7ae",
min_mem_gib=0.5,
min_storage_gib=2.0,
)
async def worker(ctx: WorkContext, tasks):
ctx.send_file("./scene.blend", "/golem/resource/scene.blend")
async for task in tasks:
frame = task.data
crops = [{"outfilebasename": "out", "borders_x": [0.0, 1.0], "borders_y": [0.0, 1.0]}]
ctx.send_json(
"/golem/work/params.json",
{
"scene_file": "/golem/resource/scene.blend",
"resolution": (400, 300),
"use_compositing": False,
"crops": crops,
"samples": 100,
"frames": [frame],
"output_format": "PNG",
"RESOURCES_DIR": "/golem/resources",
"WORK_DIR": "/golem/work",
"OUTPUT_DIR": "/golem/output",
},
)
ctx.run("/golem/entrypoints/run-blender.sh")
output_file = f"output_{frame}.png"
ctx.download_file(f"/golem/output/out{frame:04d}.png", output_file)
yield ctx.commit()
task.accept_result(result=output_file)
ctx.log("no more frames to render")
# iterator over the frame indices that we want to render
frames: range = range(0, 60, 10)
init_overhead: timedelta = timedelta(minutes=3)
# By passing `event_consumer=log_summary()` we enable summary logging.
# See the documentation of the `yapapi.log` module on how to set
# the level of detail and format of the logged information.
async with Executor(
package=package,
max_workers=3,
budget=10.0,
timeout=init_overhead + timedelta(minutes=len(frames) * 2),
subnet_tag=subnet_tag,
event_consumer=log_summary(),
) as executor:
async for task in executor.submit(worker, [Task(data=frame) for frame in frames]):
print(f"Task computed: {task}, result: {task.result}")
enable_default_logger()
loop = asyncio.get_event_loop()
task = loop.create_task(main(subnet_tag="community.3"))
try:
asyncio.get_event_loop().run_until_complete(task)
except (Exception, KeyboardInterrupt) as e:
print(e)
task.cancel()
asyncio.get_event_loop().run_until_complete(task)