2

This is in reference to rather comment (not answer), I added here: MLflow: INVALID_PARAMETER_VALUE: Unsupported URI './mlruns' for model registry store

I extracted files from here

train.py     MLproject     wine-quality.csv

These are in directory:feb24MLFLOW

I am in directory feb24MLFLOW with following contents

:memory          mlruns           train.py         wine-quality.csv

When I run following command

mlflow server --backend-store-uri sqlite:///:memory --default-artifact-root ./mlruns

The UI loads but does not show any data in it neigther does database as below. see screenshot. enter image description here I am using --default-artifact-root ./mlruns flag because, when I print print(mlflow.get_tracking_uri()), I get the current directory

file:///<mydirectorylocations>/feb24MLFLOW/mlruns

For some reason I see my database is not updating (or inserting). I checked that with in terminal.

$  sqlite3
sqlite> .open :memory 
sqlite> .tables
alembic_version        metrics                registered_model_tags
experiment_tags        model_version_tags     registered_models    
experiments            model_versions         runs                 
latest_metrics         params                 tags                 
sqlite> select * from runs;
sqlite>

As you can see there is no data after running select * from runs above.

Please note that I have following contents in

./mlruns

d6db5cf1443d49c19971a1b8b606d692 meta.yaml

Can somebody suggest I show results in the UI? or insert in databse? or what am I doing wrong?

Please note that when I run mlflow ui, I see data in the UI but I get:

error_code: "INVALID_PARAMETER_VALUE"
message: " Model registry functionality is unavailable; got unsupported URI './mlruns' for model registry data storage. Supported URI schemes are: ['postgresql', 'mysql', 'sqlite', 'mssql']. See https://www.mlflow.org/docs/latest/tracking.html#storage for how to run an MLflow server against one of the supported backend storage locations."
lpt
  • 931
  • 16
  • 35
  • I think, when specify a db in memory like `sqlite:///:memory`, it's empty every time you launch airflow server. The solution is not to use :memory? as it's not persistent – Artem Trunov Feb 26 '21 at 14:36
  • Right, I tried with sqlite://mydb.db. Still dont see results. – lpt Feb 26 '21 at 15:09
  • In that case if this server is live and now I execute eg python train.py, should not that update the databse? (It does not, it throws, No ID error) – lpt Feb 26 '21 at 15:15
  • remember to run server from the same directory your run projects? so that they both see ./mlruns in the same place/ – Artem Trunov Feb 26 '21 at 15:49
  • Of coursre, as outlined above, I am in the folder where mlruns is located. And from there when I rul mlflow ui, project runs fine. but I cannot serve it. I think I am missing some environment config. I am not sure I am I missing. But if I am missing env why does it run with mlflow ui. – lpt Feb 26 '21 at 19:40

0 Answers0