Default repo VertaAI/modeldb.git does not responds

[adhikari23]

Cannot access the repo VertaAI/modeldb.git . I have used this code to have my metadata stored in the modelDB
repo = client.set_repository('Census Income Local')
commit = repo.get_commit(branch='master').new_branch("log-reg")

However in webpage of verta.ai, the following repo shows that 404 page not found.

when clicked on the repo , following page opens up.

Also, can I have my own repo for storing this meta data?

[convoliution]

Hi @adhikari23, could you let me know what version of git you are using (git --version in terminal)? Your client version would be helpful as well (verta.__version__ in Python).

Also, can I have my own repo for storing this meta data?

This functionality can be used to capture your code version metadata in your own git repository, yes! Repo will be the URL for a remote repository (usually GitHub) that you are using.

[adhikari23]

Hi @convoliution
Git version : 2.27.0.windows.1
Client version : 0.15.4

Also I tried to use my own git repo.

This is the error that I get.
Also I want to use gerrit repository in future...is it possible? If yes, how?

[convoliution]

Hello @adhikari23!

Regarding your git version: I see that you are using Windows. Our client development has been focused on macOS/Linux, so I will work on confirming this behavior in Windows. For now, the link indeed may not be valid.

As for using your own git repo, I should clarify that client.set_repository() is for a ModelDB repository (not a git repository), which can track metadata and versions for many things: code, data, environment, etc. For example,

repo = client.set_repository("My Repository")

will create a ModelDB repository called "My Repository" that you can view at
http://localhost:3000/personal/repositories/My%20Repository/data

To capture information about a git repository, you would use a util class:

from verta.code import Notebook
code_version = Notebook()
# Notebook Version
#     client/workflows/demos/census.ipynb
#         11560 bytes
#         last modified: 2020-10-07 17:45:06.726000
#         MD5 checksum: 57f6ab14005c1cd083f8df9cba137721
#     Git Version
#         commit be4f88e173ef1d1fb09dd3b7bfbc733bb9512b49
#         on branch master
#         in repo [email protected]:VertaAI/modeldb.git

commit.update('code', code_version)
commit.save("Capture git details")

and this will automatically capture the git repository that contains the running notebook, and save the metadata to the ModelDB repository. Or you can use Git(), which also allows you to manually pass in git details.

Finally, in regards to gerrit, I am not too familiar with that software but I expect this to work with anything that git recognizes as valid remote repositories. Please let me know if you see any issues in this regard. Thank you!

[adhikari23]

Hi @convoliution
Thanks for your clarification regarding the above issue.
As you have mentioned :
This functionality can be used to capture your code version metadata in your own git repository, yes! Repo will be the URL for a remote repository (usually GitHub) that you are using.

Where should I configure my own repository so that the code version metadata can be captured in that repo instead of vertaAI/modeldb.git?

[convoliution]

Hello, @adhikari23!

Where should I configure my own repository so that the code version metadata can be captured in that repo instead of vertaAI/modeldb.git?

The Python file that you use to run our client (.py or .ipynb) has to be inside your repository, and your local repository must have a remote repository set using git remote for the URL to be captured..

[roivant-alan-menaged]

I see that this bug report is a bit old, but I think I am encountering a related issue..

When I log a GitHub repo with ExperimentRun.log_code(), the dashboard link to the repo works. However, the links to the source code and the hash do not because the .git suffix is included in the base repo url. When I manually take out .git from within the url, or when I pass in the repo url explicitly without the .git suffix, GitHub is able to find the page.

[convoliution]

@roivant-alan-menaged This does seem potentially related indeed.

A couple questions to clarify:

1. Are you passing any arguments to log_code()?
2. What version of git are you using on your machine?

[roivant-alan-menaged]

I am not passing any arguments. (When I do pass repo_url it works as expected!)

git version 2.24.3 (Apple Git-128)

I think maybe the .git suffix needs to be removed somewhere leading up to here:

modeldb/webapp/client/src/shared/view/domain/CodeVersion/GitCodeVersionButton/GitCodeVersionButton.tsx

Line 60 in 51bbd74

data.remoteRepoUrl.value,

[convoliution]

@roivant-alan-menaged Thank you for the info!

Just a couple more asks (I'm working mainly on reproducing this situation to see if there's other places this comes up):

1. What version of the Python client are you using? i.e. import verta; print(verta.__version__)
2. What version of the ModelDB frontend are you running? modeldb-frontend should be tagged vertaaiofficial/modeldb-frontend:2.0.8.2
3. If you're comfortable with sharing details, or if you could replace sensitive info with xs, could you do me a favor and let me know what this outputs:

   run._fetch_with_no_cache()
   print(run._msg.code_version_snapshot.git_snapshot)

[roivant-alan-menaged]

Verta version: 0.17.2
Frontend version: 2.0.8.2

filepaths: "scripts/modeldb_example.py"
repo: "https://github.com/<ORGANIZATION>/<REPO>.git"
hash: "b2c2123bc0b6edf85d5ee3c9560ff898c25a64f9"
is_dirty: FALSE

[convoliution]

Thanks @roivant-alan-menaged! We'll get this patched as soon as we can.