Hi all. Does anyone know how to serialize an op definition into JSON?

What’s the use case?

We’re building an application on top of Dagster and would like a way to get a list of available op definitions without having to pre-load dagster/dagit with dummy jobs. Currently we create a dummy job with all ops and then query the GraphQL api for the available ops, but it would be more convenient if we could just serialize the ops directly and send them to our client.

• If all OpDefinition objects are available within the `repo`/`definitions` object, via various jobs, you can just use the existing functionality to iterate over them and retrieve their information:

defs = Definitions(jobs=...)

available_ops = set()
for job in defs.get_all_job_defs():
    available_ops.extend(node_def.name for node_def in job.graph.node_defs)

That would still require pre-populating the repo with jobs. We are building an app that will allow you to assemble jobs from the ops using a DAG editor. The dummy jobs works ok for now but it would be good to know if there was an option to just call

op.to_dict()

or something similar. Does anything like that exist?

Hmmm.. We do have internal snapshotting functionality to do this but I wouldn’t recommend using it as it can change at any point / isn’t considered a public API. How about the following things: • A function

get_op_metadata

which, given an

OpDefinition

, outputs the dictionary of information you seek (can help you retrieve whatever that information might be hopefully) • Have a module where you export all ops to. A la in some

__init__.py

file:

from somewhere import some_op
from somewhere_else import other_op

then you can iterate over everything exported from that module, if it’s an

OpDefinition

, run

get_op_metadata

Yes that's exactly what I have in mind.

It does exist but is used only internally and we have no plans to export it. That is to say, you’re better off writing your own

Got it. Can you point me to the internal metho? I'll use it as a reference

https://github.com/dagster-io/dagster/blob/73eede0bc1667508445a70d559f0d80ba2fafd94/python_modules/dagster/dagster/_core/snap/solid.py#L386 - we use some serializable NamedTuples instead of dicts, and things are called

solid

rather than

op

for backcompat purposes, but you could imagine building something similar with just dictionaries

Perfect. Thanks, Chris