Databricks Upgrade Migration

Databricks Upgrade & Migration

Overview

Upgrade Databricks Runtime versions and migrate between platform features.

Prerequisites

• Admin access to workspace
• Test environment for validation
• Understanding of current workload dependencies

Instructions

Step 1: Runtime Version Upgrade

Version Compatibility Matrix

Upgrade Process

# scripts/upgrade_clusters.py
from databricks.sdk import WorkspaceClient
from databricks.sdk.service.compute import ClusterSpec

def upgrade_cluster_dbr(
    w: WorkspaceClient,
    cluster_id: str,
    target_version: str = "14.3.x-scala2.12",
    dry_run: bool = True,
) -> dict:
    """
    Upgrade cluster to new DBR version.

    Args:
        w: WorkspaceClient
        cluster_id: Cluster to upgrade
        target_version: Target Spark version
        dry_run: If True, only validate without applying

    Returns:
        Upgrade plan or result
    """
    cluster = w.clusters.get(cluster_id)

    upgrade_plan = {
        "cluster_id": cluster_id,
        "cluster_name": cluster.cluster_name,
        "current_version": cluster.spark_version,
        "target_version": target_version,
        "changes": [],
    }

    # Check for deprecated configs
    if cluster.spark_conf:
        deprecated_configs = [
            "spark.databricks.delta.preview.enabled",
            "spark.sql.legacy.createHiveTableByDefault",
        ]
        for config in deprecated_configs:
            if config in cluster.spark_conf:
                upgrade_plan["changes"].append({
                    "type": "remove_config",
                    "config": config,
                    "reason": "Deprecated in target version",
                })

    # Check for library updates
    if cluster.cluster_libraries:
        for lib in cluster.cluster_libraries:
            # Check for incompatible versions
            pass

    if not dry_run:
        # Apply upgrade
        w.clusters.edit(
            cluster_id=cluster_id,
            spark_version=target_version,
            # Remove deprecated configs
            spark_conf={
                k: v for k, v in (cluster.spark_conf or {}).items()
                if k not in deprecated_configs
            }
        )
        upgrade_plan["status"] = "APPLIED"
    else:
        upgrade_plan["status"] = "DRY_RUN"

    return upgrade_plan

Step 2: Unity Catalog Migration

Migration Steps

-- Step 1: Create Unity Catalog objects
CREATE CATALOG IF NOT EXISTS main;
CREATE SCHEMA IF NOT EXISTS main.migrated;

-- Step 2: Migrate tables from Hive Metastore
-- Option A: SYNC (keeps data in place)
SYNC SCHEMA main.migrated
FROM hive_metastore.old_schema;

-- Option B: CTAS (copies data)
CREATE TABLE main.migrated.customers AS
SELECT * FROM hive_metastore.old_schema.customers;

-- Step 3: Migrate views
CREATE VIEW main.migrated.customer_summary AS
SELECT * FROM hive_metastore.old_schema.customer_summary;

-- Step 4: Set up permissions
GRANT USAGE ON CATALOG main TO `data-team`;
GRANT SELECT ON SCHEMA main.migrated TO `data-team`;

-- Step 5: Verify migration
SHOW TABLES IN main.migrated;
DESCRIBE TABLE EXTENDED main.migrated.customers;

Python Migration Script

# scripts/migrate_to_unity_catalog.py
from databricks.sdk import WorkspaceClient
from pyspark.sql import SparkSession

def migrate_schema_to_unity(
    spark: SparkSession,
    source_schema: str,
    target_catalog: str,
    target_schema: str,
    tables: list[str] = None,
    method: str = "sync",  # "sync" or "copy"
) -> list[dict]:
    """
    Migrate Hive Metastore schema to Unity Catalog.

    Args:
        spark: SparkSession
        source_schema: Hive metastore schema (e.g., "hive_metastore.old_db")
        target_catalog: Unity Catalog catalog name
        target_schema: Target schema name
        tables: Specific tables to migrate (None = all)
        method: "sync" (in-place) or "copy" (duplicate data)

    Returns:
        List of migration results
    """
    results = []

    # Get tables to migrate
    if tables is None:
        tables_df = spark.sql(f"SHOW TABLES IN {source_schema}")
        tables = [row.tableName for row in tables_df.collect()]

    # Create target schema
    spark.sql(f"CREATE SCHEMA IF NOT EXISTS {target_catalog}.{target_schema}")

    for table in tables:
        source_table = f"{source_schema}.{table}"
        target_table = f"{target_catalog}.{target_schema}.{table}"

        try:
            if method == "sync":
                # SYNC keeps data in original location
                spark.sql(f"""
                    CREATE TABLE IF NOT EXISTS {target_table}
                    USING DELTA
                    LOCATION (SELECT location FROM
                        (DESCRIBE DETAIL {source_table}))
                """)
            else:
                # Copy creates new data
                spark.sql(f"""
                    CREATE TABLE {target_table}
                    DEEP CLONE {source_table}
                """)

            results.append({
                "table": table,
                "status": "SUCCESS",
                "method": method,
            })

        except Exception as e:
            results.append({
                "table": table,
                "status": "FAILED",
                "error": str(e),
            })

    return results

Step 3: API Migration (v2.0 to v2.1)

# Migrate deprecated API calls
from databricks.sdk import WorkspaceClient

def migrate_api_calls(w: WorkspaceClient):
    """Update deprecated API usage patterns."""

    # Old: clusters/create with deprecated params
    # New: Use instance pools and policies

    # Old: jobs/create with existing_cluster_id
    # New: Use job_cluster_key for better isolation

    # Old: dbfs/put for large files
    # New: Use Volumes or cloud storage

    # Old: Workspace API for notebooks
    # New: Use Repos API for version control

    pass

Step 4: Delta Lake Upgrade

# Upgrade Delta Lake protocol version
def upgrade_delta_tables(
    spark: SparkSession,
    catalog: str,
    schema: str,
    min_reader: int = 3,
    min_writer: int = 7,
) -> list[dict]:
    """
    Upgrade Delta Lake protocol for tables.

    Protocol version benefits:
    - Reader 2+: Column mapping
    - Reader 3+: Deletion vectors
    - Writer 5+: Change Data Feed
    - Writer 7+: Deletion vectors, liquid clustering
    """
    results = []

    tables = spark.sql(f"SHOW TABLES IN {catalog}.{schema}").collect()

    for table_row in tables:
        table = f"{catalog}.{schema}.{table_row.tableName}"

        try:
            # Check current protocol
            detail = spark.sql(f"DESCRIBE DETAIL {table}").first()
            current_reader = detail.minReaderVersion
            current_writer = detail.minWriterVersion

            if current_reader < min_reader or current_writer < min_writer:
                # Upgrade protocol
                spark.sql(f"""
                    ALTER TABLE {table}
                    SET TBLPROPERTIES (
                        'delta.minReaderVersion' = '{min_reader}',
                        'delta.minWriterVersion' = '{min_writer}'
                    )
                """)

                results.append({
                    "table": table,
                    "status": "UPGRADED",
                    "from": f"r{current_reader}/w{current_writer}",
                    "to": f"r{min_reader}/w{min_writer}",
                })
            else:
                results.append({
                    "table": table,
                    "status": "ALREADY_CURRENT",
                })

        except Exception as e:
            results.append({
                "table": table,
                "status": "FAILED",
                "error": str(e),
            })

    return results

Output

• Upgraded DBR version
• Unity Catalog migration complete
• Updated API calls
• Delta Lake protocol upgraded

Error Handling

Examples

Complete Migration Runbook

#!/bin/bash
# migrate_workspace.sh

# 1. Pre-migration backup
echo "Creating backup..."
databricks workspace export-dir /production /tmp/backup --overwrite

# 2. Test migration on staging
echo "Testing on staging..."
databricks bundle deploy -t staging
databricks bundle run -t staging migration-test-job

# 3. Run migration
echo "Running migration..."
python scripts/migrate_to_unity_catalog.py

# 4. Validate migration
echo "Validating..."
databricks bundle run -t staging validation-job

# 5. Update jobs to use new tables
echo "Updating jobs..."
databricks bundle deploy -t prod

echo "Migration complete!"

Resources

• Databricks Runtime Release Notes
• Unity Catalog Migration
• Delta Lake Protocol Versions

Next Steps

For CI/CD integration, see databricks-ci-integration.