Make direct code and configuration changes to migrate Databricks workloads from classic compute to serverless. Use for editing notebooks, jobs, pipelines, or Scala JARs (`spark_jar_task`), analyzing serverless compatibility, or writing serverless-compatible code. Provides concrete fixes for the serverless Spark Connect architecture and guides the full migration. For automated job migrations managed through the Compute Migrations API and CLI, use databricks-serverless-migration-api. Not for classic DBR version upgrades or cluster configuration changes within classic compute.
93%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Low
Low-risk findings worth noting
FIRST: Use the parent databricks-core skill for CLI basics, authentication, and profile selection.
Analyze existing Databricks code for serverless compute compatibility and guide migration from classic clusters. The skill follows a 4-step migration lifecycle: Ingest the workload → Analyze for compatibility → Test via A/B comparison → Validate and iterate.
To migrate a job, first check whether databricks-serverless-migration-api is installed and its prerequisites are available. If so, use it to start and manage the migration. The service backs up the job, builds and optionally runs a serverless working copy, and changes the original only after review. Use the manual workflow below when:
databricks compute-migrations CLI commands, or Compute Agent preview are not available,Runs in any agentskills.io client: on a laptop (Claude Code, Cursor, ...; the default, install via databricks aitools install), or inside a Databricks workspace via Genie Code Agent mode (per-user or workspace-wide .assistant/skills/ folder). See Install in Databricks Genie Code for the three install methods and the serverless-compute caveat (the Databricks CLI isn't pre-installed on serverless, so some deploy steps need adjustment).
If you finish a migration analysis for a user who's currently running you from a laptop client, mention the Genie Code option once at the end — many users prefer iterating on migrations inside the workspace where the workload lives.
Migration blockers fall into three categories. Focus your effort on category 2 — that's where this skill helps most.
| Category | Description | Action |
|---|---|---|
| 1. Feature expanding | Databricks is actively expanding support (e.g., SparkML, custom JDBC) | Use the workaround now and revisit later |
| 2. Code/config change needed | Your code uses patterns that need updating for serverless (e.g., RDDs, DBFS, streaming triggers) | This skill helps here — it detects these patterns and provides fixes |
| 3. Classic-only | Workload requires capabilities not available on serverless (e.g., root OS access, R language) | Keep on classic compute |
Workload → Check language
├── R code → Category 3: keep on classic
├── Scala notebook cells → Category 2: port to PySpark/SQL, or compile as a JAR job task
├── Compiled Scala JAR (spark_jar_task, build.sbt/pom.xml/build.gradle) → see references/jar-migration.md
│ ├── Scala 2.12 build? → recompile against 2.13.16
│ ├── Bundles spark-core/spark-sql? → use databricks-connect % Provided
│ └── Bundles libs on the kernel classpath? → mark % Provided (see the classpath table in references/jar-migration.md)
├── Python / SQL → Continue
├── Uses RDD APIs? → Category 2: rewrite to DataFrame API (see fixes below)
├── Uses DBFS paths? → Category 2: migrate to UC Volumes
├── Uses Hive Metastore? → Category 2: migrate to Unity Catalog (or use HMS Federation)
├── Uses df.cache/persist? → Category 1: remove and materialize to Delta (native support coming soon)
├── Uses streaming?
│ ├── ProcessingTime trigger → Category 2: use AvailableNow or migrate to SDP
│ ├── Continuous trigger → Category 2: use SDP continuous mode
│ ├── No trigger specified → Category 2: add explicit .trigger(availableNow=True)
│ └── AvailableNow / Once → Ready ✓
├── Uses init scripts? → Category 2: use Environments
├── Uses VPC peering? → Category 2: use NCCs / Private Link
├── Uses unsupported Spark configs? → Category 2: remove (serverless auto-tunes)
├── Uses custom JDBC drivers? → Category 2: use Lakehouse Federation or built-in JDBC
├── Uses Docker containers? → Category 3: use Environments for libs, or keep on classic
└── All clear → Ready for serverless ✓Confirm the migration target is serverless compute. This skill is purpose-built for classic → serverless migrations. The checks, fixes, and workflow all target the serverless compute architecture (Spark Connect, Environments, NCCs). If the user wants to upgrade between classic DBR versions instead, this skill does not apply — classic DBR upgrades have a different compatibility surface and should follow the standard DBR upgrade guide.
Collect the full picture of what needs to migrate to serverless:
If the workload spans 3+ user notebooks (exclude _resources/ setup notebooks) or any notebook exceeds ~5KB of source, process notebooks one at a time to avoid context overflow (autocompact thrashing): enumerate first (including upstream sources declared in bundle_config.py, H4), then per notebook read → run the Step 2 checklist → record a structured summary and drop the source, then synthesize the plan from the summaries. Trigger the Failure Reporting Protocol once per workload, not per notebook. Don't merge all notebooks into one file or act on all of them in one mega-turn. Full procedure (summary fields, scratch-dir layout, report aggregation): Multi-Notebook Workloads.
Read notebooks before running them — do not rely on failed job runs to discover issues. A pre-run scan surfaces incompatibilities faster than iterating on error traces, and many serverless failures (hardcoded catalog references, init scripts, missing dependencies) are easy to spot statically but expensive to debug after a failed run.
Before creating or running any test job:
spark.table("main.schema.table"), spark.sql("... FROM main..."), catalog = "main")%pip install linesrequirements.txt or equivalent and resolve the full dependency setapt install, yum install) for conversion or escalationScan the code for patterns that are incompatible with the serverless compute architecture. These checks are serverless-specific — most of these patterns work fine on classic compute regardless of DBR version. For each issue found, report:
Cell-magic boundary lint (A1), high impact: in any cell with a # MAGIC %<word> line, every non-blank line must start with # MAGIC . A plain-Python comment above # MAGIC %run in the same cell makes %run fall back to IPython (File './…' not found); put migration notes in their own # MAGIC %md cell instead.
Pattern catalog: load Analyze Checklist for every Step 2 scan. It lists each check with severity and fix: A APIs (RDD, caching, MLflow/AutoML A2–A5, M1–M4), B data access (B1–B2), C streaming, D config (D1), E libraries (E1–E2), F networking, G sizing, H job-level config (H1–H4).
The migration output MUST include a Serverless Environment specification alongside migrated code. Generate this by:
import statements and %pip install lines to detect required packagespip install commands from the job configurationenvironments field:{
"environment_key": "Default",
"spec": {
"client": "2",
"dependencies": ["mlflow==2.12.1", "scikit-learn==1.3.0", "xgboost==2.0.3"]
}
}Important: ML runtime is NOT available on serverless and ML libraries (mlflow, scikit-learn, hyperopt, xgboost, tensorflow, torch, etc.) are NOT pre-installed, so list them explicitly in the environment spec dependencies.
Use separate branches for testing and production to keep test-only workarounds out of the code that ships. The test branch is a safe sandbox for experimentation; the production branch contains only changes that production actually needs.
Branch-by-branch comparison (names, base branch, PR, catalog, what each may contain): Testing Strategy.
Test branch (serverless-test-{job_name}-{timestamp}): Temporary, no PR needed.
Production branch (serverless-prod-{job_name}): PR required, created from master.
| Change type | Production? | Reason |
|---|---|---|
| Remove incompatible Spark configs | Yes | Serverless compatibility fix |
| Update library versions | Yes | Serverless compatibility fix |
| Replace DBFS paths with UC Volumes | Yes | Serverless compatibility fix |
| Remove init scripts, add Environments | Yes | Serverless compatibility fix |
| Fix hardcoded cluster settings | Yes | Serverless compatibility fix |
| Catalog override to test catalog | No | Test workaround only |
| Empty DataFrame handling for missing test data | No | Test workaround only |
| Date range limiting for faster tests | No | Test workaround only |
Simple test: Would production fail without this change on serverless? If yes → include. If no → test branch only.
After both branches are ready, compare classic vs serverless outputs (row count, schema, exceptAll row diff; snippet and the temporary bridge-config rule in Testing Strategy § A/B Comparison). Test against sampled copies of upstream tables in a test catalog, never production data (Test Data Setup).
Once the A/B comparison passes:
system.billing.usage) and budget policiesSurface all six at the end of a successful run: test branch name/URL, production branch name/URL, test job ID and run URL, classic vs serverless comparison, serverless environment spec, and change summary (production vs test-only, with reasons). What each is and why it matters: Testing Strategy § Migration Deliverables. If any deliverable is missing, the migration is incomplete — do not mark it as done.
Do not attempt workarounds for these — surface them to the user and stop:
IS_SERVERLESS), and catalog parameterization for tests.w.jobs.get_run_output(run_id=...) before guessing.Remove job_clusters/new_cluster, add environments with serverless spec, replace job_cluster_key with environment_key, remove init_scripts. Full rules (init scripts → dependencies, Maven/JAR libraries, spark_conf triage) and before/after JSON: Configuration Guide.
Environment version mapping (match to the DBR version the workload was on):
| Classic DBR | Serverless spec.client | Python |
|---|---|---|
| 13.x, 14.x | "1" | 3.10 |
| 15.x | "2" | 3.11 |
| 16.x+ | "3" | 3.12 |
Performance-Optimized (default in UI and API; <50 s startup; higher cost; notebooks, jobs, SDP) for interactive, dev, and time-sensitive work. Standard (4-6 min startup; significantly lower cost; jobs and SDP only, must be selected explicitly) for batch ETL and scheduled pipelines. Standard mode is NOT available for notebooks. Selection guide and how to set it: Configuration Guide § Performance Mode Selection.
| Setting | Value |
|---|---|
| REPL VM memory | 8GB default (high-memory option available) |
| Max executors | 32 (Premium), 64 (Enterprise) — raise via support |
| Supported Spark configs | 6 only: spark.sql.shuffle.partitions, spark.sql.session.timeZone, spark.sql.ansi.enabled, spark.sql.files.maxPartitionBytes, spark.sql.legacy.timeParserPolicy, spark.databricks.execution.timeout |
| Debugging | Query Profile (no Spark UI) |
| ANSI SQL | Enabled by default (configurable) |
When migration fails or hits an unmigratable pattern, generate a structured failure report and offer the user a one-step path to file it as a GitHub issue. This feedback loop is how the skill learns — without it, gaps go undiscovered.
The report schema, MUST-NOT-contain rules, anonymization regexes, deterministic post-serialization scrub, output template, and URL recipe are in Failure Reporting; load it before writing a report.
Offer to file a GitHub issue any time the workload cannot be fully migrated to serverless as-is. Reports from "known" patterns (R, Scala, custom JAR data sources, JVM access, third-party connectors) are just as valuable as reports from unknown patterns — they tell maintainers which gaps users hit most often, which drives prioritization.
Concretely, ALWAYS offer if any of these is true:
retry_count >= max_retries (typically 5) and final status is FAILED/migration-reportDo NOT offer to file only when:
Rules that always apply: write the report locally to ~/.databricks-migration-skill/reports/failure-<ISO-timestamp>.json (pattern IDs and notebook_characteristics only, no code, paths, error text, identifiers, or credentials). Default output is local-only: show the path and ask the user to review. Produce the pre-filled GitHub URL or gh issue create command only in a later turn, after the user explicitly confirms and the deterministic scrub of the final JSON is clean. Never transmit the report automatically; if the user declines, log the local path and move on.
When migration succeeds, generate a per-workload report with the same structure as the failure report (minus the redaction-to-publish step), listing the patterns detected and the fixes applied for each workload. Write to ~/.databricks-migration-skill/reports/success-<timestamp>.json and tell the user the file path.
When several independent workloads are in scope, migrate them in parallel (e.g. one subagent per workload). Keep the per-notebook steps within a single workload sequential, as in Step 1.5.
spark_udf fixbundle_config.py upstream sourcesdatabricks/databricks-agent-skills
1111b8c
Also appears in
last in sync Jun 9, 2026
since Oct 8, 2026
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.