10

I've run into an issue using MLflow server. When I first ran the command to start an mlflow server on an ec2 instance, everything worked fine. Now, although logs and artifacts are being stored to postgres and s3, the UI is not listing the artifacts. Instead, the artifact section of the UI shows:

Loading Artifacts Failed
Unable to list artifacts stored under <s3-location> for the current run. Please contact your tracking server administrator to notify them of this error, which can happen when the tracking server lacks permission to list artifacts under the current run's root artifact directory.

But when I check in s3, I see the artifact in the s3 location that the error shows. What could possibly have started causing this as it used to work not too long ago and nothing was changed on the ec2 that is hosting mlflow?

JMV12
  • 965
  • 1
  • 20
  • 52

1 Answers1

14

I found the answer. The error was that mlflow could not find boto3, so a conda installation of that worked. The logs for this were buried and hard to find in stdout.

JMV12
  • 965
  • 1
  • 20
  • 52
  • 1
    Had the same issue. I just installed the `boto3` package via `pip install` and then everything worked. Thanks @JMV12. – Andrei Madalin Butnaru Aug 12 '20 at 08:54
  • 1
    what does that mean mlflow could not find boto3? I have boto3 installed on the pod where mlflow is running, but this error is still happening. Could you please advise on the fix that you applied. – frany Jun 10 '21 at 15:09
  • same as me - also wasnt able to list and boto3 is already installed – Adam Nov 25 '21 at 04:07
  • 1
    Installing `boto3` in my MLflow container hadn't resolved this error for me. I've `pip list` to confirm the `boto3` had been installed and also restarted the container. Checking the MLflow UI, the error in the Artifacts section persists. – Ray Jan 08 '22 at 13:41