I have a resource that acts as a context manager it has an e

Question

I have a resource that acts as a context manager it has an enter and exit method If the op fails will the exit method still be called

Roel Hogervorst · Answer

alright this was dumb I refrased my question and the search engine gave me an answer from stack overflow <https stackoverflow com questions 62086216 is exit method called when theres an exception inside context manager> yes it is still called

Roel Hogervorst · Answer

exit returns 3 things if the internal work was succesfull these things are None so you can add a clause in the exit to check for None For instance if you want to commit database transactions only on success

Roel Hogervorst · Answer

this does not actually work because dagster wraps everything up to continue executing

Roel Hogervorst · Answer

Is there a way to signal to a resource that an OP was successful Usecase I want to commit a database transaction on resource teardown but ONLY if the op was successful I have looked into dagster run object that is available through context but the failure event happens after teardown

nickvazz · Answer

I think <https docs dagster io apidocs hooks dagster success hook| success hook> might work

Roel Hogervorst · Answer

hi Nick yes normally that would be a great solution But can I access the same things inside the resource

Roel Hogervorst · Answer

looking at the source code the hook looks at events in the context

Roel Hogervorst · Answer

could I access the events from the resource

nickvazz · Answer

Oh that is an interesting question I did not see the rest of the thread and have not tried using the ` success hook` with a resource before

claire · Answer

Hi Roel Not sure if this is what you are asking but since a hook executes in the same process as the step it is attached to you ll be able to access write to a resource within an op and then access the same instance of the resource in the hook that executes after the op

Roel Hogervorst · Answer

Hi < claire> thank you My team and I have come to the conclusion that although it would be technically possible to create a database commit at success of an op the way to do this is way too complex It is much simpler and debuggable to do it inside the op and without the magic magic of hooks and resources combined If it feels like i m fighting the dagster system than this is a strong signal that we are trying to solve the problem at the wrong level thinking face