Storage

The Storage page tracks costs from Databricks’ predictive optimization system — the automated operations that compact, vacuum, analyze, and cluster your Delta tables. These system-initiated operations consume DBUs and generate cost, often without visibility into what triggered them or how much they cost.

Overview

The top of the page shows headline metrics:

Metric	What it shows
Total Cost	Aggregate estimated cost of predictive optimization operations for the selected period
Operations	Total number of optimization operations performed
DBUs Used	Total DBU consumption across all operations
Failed	Count of operations that did not complete successfully

Two bar charts provide a visual breakdown:

• Operations by Type — How many operations of each type were performed
• Cost by Operation Type — How much each operation type cost

Overview tab

The Overview tab shows a daily operations table with the following columns:

Column	What it shows
Date	Day the operations were performed
Workspace	Which Databricks workspace
Operation	Operation type (COMPACT, VACUUM, ANALYZE, CLUSTERING)
Count	Number of operations on that day
Success	Count of successful operations
Failed	Count of failed operations
DBUs	Total DBU consumption
Est. Cost	Estimated cost in USD

The table is sorted by estimated cost (descending) by default.

By Table tab

The By Table tab shows optimization activity grouped by individual table:

Column	What it shows
Date	Day the operations were performed
Table	Fully qualified table name (catalog.schema.table)
Workspace	Which Databricks workspace
Total Ops	Total number of operations on this table
Compact	Number of compaction operations
Vacuum	Number of vacuum operations
Analyze	Number of analyze operations
Cluster	Number of clustering operations
Est. Cost	Estimated cost in USD

This view helps identify which tables are most expensive to maintain through predictive optimization.

Predictive optimization operation types

LakeSentry tracks operations performed by Databricks’ predictive optimization system:

Operation type	What it does
COMPACTION	Compacts small files into larger ones for better read performance. Reduces the number of files scanned during queries.
VACUUM	Removes old file versions that are no longer needed. Reduces storage cost by cleaning up files left behind by Delta operations.
ANALYZE	Collects statistics on table data to improve query planning and reduce staleness.
CLUSTERING	Re-clusters data by frequently filtered columns. Improves partition pruning and reduces bytes scanned. Includes AUTO_CLUSTERING_COLUMN_SELECTION operations.

Cost is estimated using DBU consumption and a standard DBU rate. The raw data comes from the Databricks system.storage.predictive_optimization_operations_history system table.

Note

Predictive optimization data requires Databricks Unity Catalog. If your tables aren’t registered in Unity Catalog, these operations won’t appear.

Filtering

The Storage page respects the global workspace filter and time range selector. Operations can also be filtered by operation type via the API.

Storage cost categories

Databricks storage costs appear in billing data under several usage types:

Usage type	What it covers
STORAGE_SPACE	Cloud storage for Delta tables and volumes
NETWORK_BYTE	Network transfer costs for cross-region reads

These usage types are visible in the Cost Explorer compute types breakdown, not on the Storage page itself. The Storage page focuses specifically on predictive optimization operation costs (which are DBU-based), not on the underlying cloud storage charges.

Table cost attribution

Table-level cost attribution (which tables cost the most to write to, which are unused) is available in the Cost Explorer Tables tab, not on this page. That view shows cost attributed through work units that write to each table, along with unused table detection.

Common investigation workflows

Identifying expensive optimization operations

1. Check the Cost by Operation Type chart to see which operation types drive the most spend.
2. Switch to the By Table tab and sort by Est. Cost (descending).
3. Look for tables with disproportionately high optimization costs relative to their value.

Monitoring optimization health

1. Check the Failed metric in the page header. Any non-zero value warrants investigation.
2. In the Overview tab, look for rows with high failed counts.
3. Failed operations may indicate table configuration issues or resource constraints.

Reviewing per-table optimization activity

1. Switch to the By Table tab.
2. Look for tables with unusually high operation counts — frequent compaction may indicate a write pattern that produces many small files.
3. Review CLUSTERING operations on frequently queried tables to verify they are improving query performance (cross-reference with SQL Analysis).

Next steps

• SQL Analysis — Query-level cost investigation
• Cost Explorer — Explore storage costs alongside compute, including table cost attribution
• Waste Detection & Insights — How unused data and optimization opportunities are identified
• Metrics & Aggregations — How storage metrics are computed