MLflow

The MLflow page tracks your machine learning experiment activity in Databricks. It aggregates MLflow experiments and runs with their metrics, giving ML teams visibility into experiment activity, success rates, run durations, and user engagement.

Experiment list

The Experiments tab shows MLflow experiments with their aggregated run statistics:

Column	What it shows
Experiment	Experiment name and ID
Runs	Total number of runs, with a badge showing how many are currently running
Success	Number of successful runs, with a badge showing failed runs if any
Users	Number of distinct users who ran experiments
Avg Duration	Average run duration across all runs
Last Run	When the most recent run completed

Sorting

• Runs (descending) — Find the most active experiments
• Experiment (ascending) — Browse experiments alphabetically
• Success (descending) — Find the most successful experiments

Headline stats

The page header shows aggregate statistics across all experiments:

Metric	What it shows
Experiments	Total number of MLflow experiments
Total Runs	Total number of runs across all experiments
Successful	Number of runs that completed successfully
Failed	Number of runs that failed

Runs by experiment chart

A bar chart showing the top 10 experiments by run count. Use this to quickly identify which experiments have the most activity.

Runs tab

The Runs tab shows individual MLflow runs across all experiments:

Column	What it shows
Run	Run name and ID
Status	FINISHED, FAILED, or RUNNING
User	Who initiated the run
Duration	How long the run took
Metrics	Number of distinct metrics logged in the run
Started	When the run started

Daily Activity tab

The Daily Activity tab shows per-day aggregated MLflow activity for a selected time range:

Column	What it shows
Date	The date of recorded activity
Experiment	Experiment name or ID
Runs	Number of completed runs, with failed count if any
Users	Number of distinct users active that day
Avg Duration	Average run duration for that day

Filtering

Filter	Options
Time range	Analysis period (applies to Daily Activity tab)

Work unit integration

LakeSentry maps MLflow data into the work unit model. Each MLflow experiment becomes a work unit of type mlflow_experiment, and each MLflow run becomes a work unit run. This enables MLflow experiments and runs to participate in the same attribution and tracking framework as jobs and pipelines.

Note

MLflow data is tracked through the system.mlflow.experiments_latest, system.mlflow.runs_latest, and system.mlflow.run_metrics_history system tables. These must be enabled in your Databricks account for LakeSentry to track ML experiment activity.

Common workflows

Finding high-activity experiments

1. Sort the experiment list by Runs (descending).
2. Look at the top experiments and their success/failure ratios.
3. Experiments with high failure rates may benefit from investigation.
4. Check average duration to spot experiments with unusually long runs.

Investigating ML activity trends

1. Switch to the Daily Activity tab and set the time range to the last 30 or 90 days.
2. Look for experiments with increasing daily run counts.
3. Use the Users column to understand whether activity is from one user or a team.
4. Cross-reference with Compute to see if the underlying clusters are efficiently utilized.

Monitoring experiment health

1. Check the page header stats for the overall Failed count.
2. Sort the experiment list by Success to see failure ratios.
3. Switch to the Runs tab to find specific failed runs and their users.

Next steps

• Work Units (Jobs & Pipelines) — Job-level tracking
• Compute (Clusters & Warehouses) — Cluster utilization for training workloads
• Model Serving — Tracking for deployed model endpoints
• Cost Attribution & Confidence Tiers — How costs are attributed across workloads