Hello guys. I am trying to interact with command line through

subprocess

to select which

bentoctl operator

to use. Unfortunately I could not find one way to interact with this menu when I call

bentoctl init

. Do you have any experience on how can I do this? If I only have one operator installed this menu never appears. But I want to do multiple calls in parallel, so I first install all the operators, and after I call

bentoctl init

. If I uninstall one operator in any moment, I will break a parallel process. This is why I am forced to understand how to interact with this menu.

The way I am doing this is using

subprocess.Popen

that works completely fine if I should not interact with this menu.

I'm not sure if I understand what u are trying to do but if you want to use 2 different operators you can use 2 different deployment config files right?

You will have to create different folders for each of the generated files

@Matheus You could also generate the yaml file without using

bentoctl init

. It is probably an easier way to automate this process

@Bo I just solved the problem in the way you suggested. I created the

deployment_config.yaml

using

pyYaml

and after used

bentoctl generate

to create

bentoctl.tfvars

and

<http://main.tf|main.tf>

. It worked!

@jjmachan that is exactly what I am doing. I am creating one folder that register every process creating sub-folders with id of each run. In each sub-folder I have the terraform files. You can see it below

And now it comes my second issue.

bentoctl build

updates bentoctl.tfvar with those 3 informations: • image_tag = <image_tag> • image_repository = <image_repository> • image_version = <image_version> Unfortunately

bentoctl build

does not have an option to specify the output location of the

bentoctl.tfvars

file, so it is being saved in my root path (what completely breaks the idea of parallelism that I should have inn my project). In other words, I am only able to specify the input --deployment-config-file but I am not able to specify where I want that the

bentoctl.tfvars

are generated.

For the moment I am only appending those lines to the

bentoctl.tfvars

file as a turnaround. But I believe that enable users specify the output path is something useful for

bentoctl

Guys do you have any solution for periodic calls of a function using

bentoml

? I am trying to use the lib

AsyncIOScheduler

but it does not work if I serve my app with

bentoml

. It works if I serve my app with

fastapi

though

I basically applied with this guy said https://github.com/tiangolo/fastapi/issues/520#issuecomment-716969948

But it not works when I serve my app with

bentoml

When you say periodic calls to a function, do you mean from an external source?

No, I mean that the bentoml app calls a route periodically

I opened this bug on github. Can try to replicate it and verify what I mean?

I think you can use FastAPI to achieve scheduled task. In FastAPI, you can add

startup

and

shutdown

event to an app. You can schedule the task against a

BackgroundScheduler

in the startup and shutdown the scheduler in the

shutdown

event. https://fastapi.tiangolo.com/advanced/events/

Though running scheduled task in a service is not well supported use case.

Hello @Sean

The code i shared is perfecly running with fastapi, The problem is that it is not running with bentoml, and that is the problem to me because I have a bento app that mounts a fastapi app

Hello guys

Is there a way I can specify the path of a file in the deployed machine differently than the calling machine in the

bentofile.yaml

?

Something like this

The name in the left should be the file location in the calling machine and the second should be the path in the deployed machine

Hi @Matheus, this is not possible. The problem is more fundamental with Python. If a module needs to import

src.job.2222.file.py

, how would you write the

import

statement so that the module can be imported correctly in both the calling and deployed machine?

In details my problem is that I am deploying multiple models in pararell with

bentoctl

in GCP. For this each model has a folder with a job_id in my source machine, where terraform files are created. But in the

deployment

machine there is not a way to know the

job_id

number. My solutions was to read the content of the FIRST

job_id

folder even thought we dont know its number. But the problem it that it can not be done in the

__init__

method because it is called in the source machine, and the deployment does not work. I need to read the content of this folder only in my

predict

route, but if have multiple routes i need to repeat the same code in all of them. I also realised that even with I could change the path I would not be able to read it in the

init

method of my router 😅, so I will keep my solution for now.

This message was deleted.

deployment_config.yaml have the configurations below