odahu / odahu-ui Goto Github PK

We have to create a feedback tab on the model deployment view to allow a user leaving of model feedback.

description
doesn't sort by:
Toolchain, VCS ID for Model Trainings
Integration for Model Packagings
and Image pull connection ID for Model Deployment

when you try to sort by any of these columns for the first time, most possibly, it would change the order of entities but not as you expect. And if you try again to change the order, it wouldn't work.

As a user, I expect that buttons for packaging/deployment of train/pack are inactive while training/packaging doesn't succeeded, because when "PACK"/"DEPLOY" button is active it implies that one already can be run.

default and min values on UI should be unified with odahu server default values for model deployment:

model deployment:

default value of Liveness probe initial delay is 10, should be 2
default value of Readiness probe initial delay is 10, should be 2
min value of Liveness probe initial delay is 1, should be 0
min value of Readiness probe initial delay is 1, should be 0

description
exactly during the edit of connection with long keySecret or privateKey

steps to reproduce

1. Create a new git connection
2. Start to edit the connection on EDIT tab
3. Proceed to Specification page of Edit Connection tab
4. Put standard (long) SSH private key and press NEXT

expected behavior
no changes in headers

actual behavior
VIEW, EDIT, YAML headers shifted far along right during Review
and after moving to VIEW tab after SUBMIT (also relevant if edit on YAML tab).

screenshots

When we define schema validation, for example https://git.epam.com/epmd-legn/odahu-ui/blob/develop/src/views/trainings/editable/schemas.ts#L4 , we duplicate some fields two times.

We have to find out a way how to fix it.

• User adds more than one ModelDeployment for ModelRoute and tune their weights (traffic shares)
• User adds one model as mirror (all traffic can be duplicated)
• Grafana Board with useful metrics

after pressing 'submit' (creating, updating) for conn, train, etc. the user should move from 'edit' tab to the 'view' tab of the one.

description
when you create a VCS connection, the reference field is optional. But when you create it through UI, the reference field is required

expected behavior
the reference field is optional for VCS connection

Description
the "404 Not Found" status code is returned on the "PLAY" tab when the deployment is not ready or ready but not for a long enough time.

Preconditions

• the packaging docker image is ready

Steps to Reproduce

1. Create a deployment.
2. Wait until deployment is Ready.
3. Click on the deployment name on the list.
4. Go from "VIEW" tab to the "PLAY" tab.

Actual result
the "404 Not Found" status code is returned, even though the Deployment is already ready.

Expected result
It should show that it's loading 🔄 until the "PLAY" page from the Service Catalog will be ready.

Attachments

Logs must be more fresh
We should add button "refresh" to manual forcing the refreshing logs or decrease pull period for logs

scroll bar is disappearing when you change rows per page option in table views.

We have to add the Create package button in the training view.

After click on the button, UI will be redirected to the "New Package" view. The training artefact will be used as default value for "artefactName" field.

Describe the bug
when you started to select (checkbox) entities and someone else deleted these entities the number of selected wouldn't change and checkboxes would work wrongly. more in gif, screenshots

Precondition
exists several trainings

To Reproduce

1. checkbox several (1-2) entities
2. Open a new tab with odahu UI
3. delete one entity that was selected in step (1)
4. return to the first tab
5. wait until this tab self-refreshes state (without the deleted entity in step (3))

Expected behavior
the number of selected should change to the real number of selected

Actual behavior
the number of selected didn't change and checkboxes behavior yourself as there is still exists deleted entity.

Screenshot

1. the diagrams shouldn't go out their frames
2. add a legend for pie charts
3. take out axes for pie charts
4. the diagrams at dashboard doesn't update after making changes at another browser tab.
   

• 'odahu-viewer' – can do any GET requests to all API resources
• 'odahu-flow-scientist' – can run, delete, change any trainings, packagers, deployment, but can only read
  connections and integrations
• 'odahu-admin' – can make any requests to API

For a viewer:

• no buttons as +new, clone, delete;
• take out checkboxes for Connections, Model Trainings, Model Packagings, Model Deployments
• no edit tab for Connections, Model Trainings, Model Packagings, Model Deployments

for a scientist:

• no buttons as +new, clone, delete for Connections
• take out checkboxes for Connections
• no edit tab for Connections

for an admin:

• add buttons as +new, clone, delete for Toolchains, Packagers
• add edit tab for Toolchains, Packagers
• and add checkboxes for Toolchains, Packagers

To Reproduce

1. go to Connections panel
2. press + New button
3. fill in required fields and press NEXT
4. on Specification page fill in the required fields and press NEXT

Expected behavior
you see maskes instead of actual values for Password or keySecret fields on REVIEW page

Actual behavior
you can see unmasked Password or keySecret fields on REVIEW page

Screenshots

We have to add a field on the table, which allow to sort entities by text(id, names, images and so on)

Tasks:

• Sort on tables - 3md
• "Package button" in a training view - 1md
• Training(MLFlow) metrics - 3md
• [ODAHU component versions view] - 1md
• Unit tests - 3 md
• [Default values from ODAHU config] - 1md
• asterisks (*) for the mandatory fields - 1md
• Feedback tab - 1 md
• ODAHU entities from file - 2md
• Model REST endpoints tab - 3md
• [Log out button] - 1 md

Sum: 20md

description
when you sort the entities at UI and they have names with different length, and don't fit on one page (more than 5 entities).
The table head starts to move quite drastically during sorting.

proposal
improve UX and appearance by stabilizing tables of entities

screenshot

'open link in ...' should be in context menu for 'ODAHU Components' but not open image.

As User I need to see deployment logs including container logs & knative sheduling logs in one window the same way as it works for training & packaging tabs.

This will require:

• add required knative logs to odahu index
• create proper kibana requests to display all stages of deployment lifecycle logs
• add logs tab that will embed kibana as it's done for training & packaging tabs

Description:
After a train/pack has been changed and you move to another tab (e.g. view, edit) and comes back. the old description is shown

Steps to Reproduce:

1. Make valid changes (for resources or artifactName) on YAML tab in desc. of train.
2. Submit button. ("Model Training submitted" shown).
3. Move on VIEW tab of this training.
4. Move back on YAML tab.

Expected Result:
The updated description of the training is shown.

Actual Result:
Shown description of training before changes.

can change in Docs.

As A User I would like to create packaging based on finished training, and deployment based on packaging

Using UI

ODAHU API doesn't accept values that start or end with spaces. This spaces should be removed on client side or|and on backend.
This should be applied to all string params.

Description
when you create a connection and not fill in the Reference field, you will get not a user-friendly hint

can set requested resources bigger than limits and submit training

Expected Result: Should give the error that requested resources are bigger than limits.

When you ticked one training/pack and then tick another one and 'clone' button disappears, the 'delete' button changes a little shape.

Env: Win10, Firefox Mozilla (better observed)

The list of forms and required fields for each form.

ODAHU Entities:

• Connections

• Model Trainings

• Model Packagings

• Model Deployments

• Connections:

  ID, Type fields for all connections.
  

  • git type:
    Git SSH URL, SSH private key
    

  • docker type:
    URI, Username, Password
    

  • gcs type:
    Project, URI, Service account secret
    

  • azureblob type:
    URI, SAS Token
    

  • s3 type:
    URI, Region, Access Key ID, Access Key Secret
    

  • ecr type
    URI, Access Key ID, Access Key Secret
    

• Model Training:

  • Metadata step:
    ID, Model name, Model version, Toolchain, Output connection
    
  • Specification step:
    Entrypoint
    if Data section is specified, for every data entry,
    the Connection ID and Target Path are required fields:
    
    if Environment variables or Hyper Parameters are specified,
    for every entry the Name and Value are required fields:
    
    

• Model Packaging:

  • Metadata step:
    ID, Output Connection ID, Integration
    
  • Specification step:
    Artifact
    if Target entry is specified, Name and Connection ID are required fields.
    if Arguments entry is specified, Name and Value are required fields.
    

• Model Deployment:

  • Metadata step:
    ID, Image pull connection ID
    
  • Specification step:
    Predictor, Docker image, Minimum replicas, Maximum replicas
    

Actual: If user edits any entity form will become empty
Expected: If user edits any entity backdrop form will become empty

Describtion
on UI, when you try to move from ODAHU components (bento menu) > Feedback storage you get this:

or when you try to follow the Web UI link:

To Reproduce

1. Open Web UI
2. Click on ODAHU components (bento menu, 9 dots)
3. Click on Feedback storage button

Expected behavior
you went to portal.azure.com and directory where the connection or feedback is pointing (where the data stored ).

Actual behavior
you get the error PublicAccessNotPermitted

Attachment

Environment

• cloud provider: Microsoft Azure

To Reproduce

1. go to Connections panel
2. press + New button
3. select Type as azureblob
4. fill in required fields and press NEXT

Expected behavior
you should be asked to pass keySecret which is in based64 format

Actual behavior
you are asked for SAS Token on Azure:

and for service account on GCP type: gcs:

and update comment for ecr and s3 types on AWS:

Description
When you create deploy (through + New button) pre-filled predictor field shows warning that Predictor is required field.

Precondition

• have succeeded packaging

Steps to Reproduce

1. Go to 'Deployments' on sidebar.
2. Click on '+ New' button.
3. Fill required fields on 'Metadata' step and click NEXT.
4. Check that Predictor is the pre-filled field.
5. Select docker image of succeeded packaging and click NEXT.
6. Pay attention what said under Predictor field

Actual result
Pre-filled Predictor value is not selected for field.

Expected result
Pre-filled Predictor value is placed into the field and you can proceed with Model Deployment creation.

Attachments

As User I need additional tab in deployment view that will display model info and allow to submit data to deployed model and see the response (maybe the same way as it works in service-registry)

We have to create a model REST endpoints tab to allow a user to send HTTP requests.

https://github.com/swagger-api/swagger-js

We have to allow a user to create ODAHU entity from manifest files.

description
when you're on Toolchain tab and try to open the view of the toolchain/packager by clicking on the ti name (link), it shows that you're trying to open the link <cluster name>/toolchains/item/<instance name> but at the end, it brings you to <cluster name>/toolchains/item/undefined/View

expected behavior
it should bring you to the view of the toolchain/packager instance view

actual behavior
it opens the tab <cluster name>/toolchains/item/undefined/View and if you try to move to YAML tab, forth and back, it continues to add undefined for example:

screenshot

description
when training (e.g. 'wine') has already succeeded and you open logs at LOGS tab. It shows only a few rows for training/packaging.

ENV
cloud provider: azure

screenshot

Description
when clicking on artefacts of training on VIEW tab under Results, it redirects to mlflow only to wine trainings and no matter which training I was viewing.

PreCondition

• exists succeeded training on Training panel
• exists several entries of experiment on mlflow, <cluster link>/mlflow/#/

To Reproduce
Steps to reproduce the behavior:

1. Click on the name of training (go to the training VIEW tab)
2. Click on the Run ID button

Expected behavior
you should see the <model training name> > Run...
the link should be <cluster link>/mlflow/#/experiments/5/runs/ffb06e2ec0df40f78a5e41b9a7a2279b

Actual behavior
you see the first training name that were run on cluster and your Run id
the link is <cluster link>/mlflow/#/experiments/0/runs/ffb06e2ec0df40f78a5e41b9a7a2279b

The scroller should end at the green line without part in the red rectangle.

check that model deployment ID must start with an alphabetic character during typing of the ID on New Deployment tab (during the creation of Model Deployment).

If model packaging was created through cli then this packaging cannot be updated through UI.
Stops at Specification of EDIT tab, cannot press NEXT button.

As ODAHU pipeline related logs will be stored in separate index, we should update hard coded links to new values.

If train, conn, etc. update and right after that move to the YAML tab, it will show the old yaml file (before changes were made).

Description
For the creation of model training in Specification, when you completely delete Reference for GIT Settings then develop is refilled back.

To Reproduce
Steps to reproduce the behavior:

1. On the sidebar, go to 'Trainings'
2. Click on '+ New',
3. Fill required fields and press 'NEXT'
4. At the GIT Settings delete the value for Reference field

Expected behavior
When you completely delete reference it should be empty.

Actual behavior
When you delete reference it returns 'develop' value.