Open bachlean opened 1 year ago
route to CXP team
@bachlean Thank you for reaching out, we are looking into it.
Thanks for the feedback! We are routing this to the appropriate team for follow-up. cc @azureml-github.
Author: | bachlean |
---|---|
Assignees: | RakeshMohanMSFT |
Labels: | `Service Attention`, `Machine Learning`, `customer-reported`, `Auto-Assign` |
Milestone: | - |
Hello @bachlean, thanks for your feedback. Sorry for the confusion, we had chosen to use create
only for new endpoints/deployments, and we're guiding to use update
for existing ones. Intent of using create
is to have complete set of information needed to create one (using YAML), so that this YAML can be used as the single file representing the endpoint/deployment. This way you can fully leverage GitOps - you can do diff/compare and manage versions of YAML configuration and trigger automatic update if/when needed.
The confusion is on the documentation - sorry for your inconvenience. We're currently fixing the documentation issue.
Hi, we're sending this friendly reminder because we haven't heard back from you in a while. We need more information about this issue to help address it. Please be sure to give us your input within the next 7 days. If we don't hear back from you within 14 days of this comment the issue will be automatically closed. Thank you!
Hello @bachlean, thanks for your feedback. Sorry for the confusion, we had chosen to use
create
only for new endpoints/deployments, and we're guiding to useupdate
for existing ones. Intent of usingcreate
is to have complete set of information needed to create one (using YAML), so that this YAML can be used as the single file representing the endpoint/deployment. This way you can fully leverage GitOps - you can do diff/compare and manage versions of YAML configuration and trigger automatic update if/when needed.The confusion is on the documentation - sorry for your inconvenience. We're currently fixing the documentation issue.
Hello @dem108 , thanks for the reply. I understand. although, i have to say i find this kind of inconvenient and definitely inconsistent with the other "az ml XXX create" commands for workspace and compute. For workspace and compute, the "create" command does not fail and abort the pipeline run if the workspace/compute already exists.
+1 for this please - I raised this also at the Q+A as I was initially directed there. Idempotency across the az ml v2 command set needs a review
Could a new apply
subcommand be added that supports this desired more declaritive approach? Similar to kubectl
.
Related command az ml online-endpoint create ... az ml online-deployment create ...
Describe the bug Command fails (and with it, corresponding Tasks in AzureDevops Pipelines) despite the documentation stating:
[online-endpoint] To create an endpoint, provide a YAML file with batch endpoint configuration. If the endpoint already exists, it will be over-written with the new settings. [online-deployment] Create a deployment. If the deployment already exists, it will be over-written with the new settings.
This is extremely annoying, because it forces you to include a separate check and an if/else construct for create/update when trying to automate.
To Reproduce create the same endpoint/deployment twice in a row az ml online-endpoint create -w -g -n dummy-endpoint
az ml online-endpoint create -w -g -n dummy-endpoint
=> cli.azure.cli.core.azclierror: Met error <class 'Exception'>:Endpoint already exists
Expected behavior I expect the command to overwrite the existing endpoint (with potentially new settings) if it already exists (as would be the intended behavior according to the documentation)
Environment summary Cloudshell bash in azure portal azure-cli 2.39.0 core 2.39.0 telemetry 1.0.6 * Extensions: ml 2.6.1 ai-examples 0.2.5 ssh 1.1.2
Dependencies: msal 1.18.0b1 azure-mgmt-resource 21.1.0b1