Authors and runs Flyway database migrations - versioned (`V1__add_users.sql`), repeatable (`R__refresh_views.sql`), and undo (`U1__remove_users.sql`) migration files in `db/migration/`; runs `flyway migrate` / `info` / `validate` / `clean` / `baseline` / `repair`; tracks state in the `flyway_schema_history` table; supports 50+ databases including Oracle / SQL Server / MySQL / PostgreSQL / MariaDB / Snowflake / BigQuery; integrates with Maven, Gradle, CLI, and Docker. Use when the user works with Flyway-managed schemas, asks about migration ordering, or needs CI gates on schema changes.
80
100%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
View guide
High
Do not use without reviewing
Flyway tracks applied migrations in a per-database flyway_schema_history
table and applies pending migrations in order by version number
(fw-how).
db/migration/ (or configured) directory of
V*.sql / R*.sql files.Follow Steps 1 - 7 below in order; each numbered step is the single source for that part of the workflow.
Flyway runs on Windows, macOS, Linux, and Docker, plus Maven and Gradle plugin distributions (fw-home). Common install paths:
# Docker (zero-install for CI)
docker run --rm flyway/flyway -url=jdbc:postgresql://host/db -user=usr -password=pwd migrate
# Homebrew (macOS / Linux)
brew install flyway
# Maven plugin (Spring Boot etc.)
# add to pom.xml under <build><plugins>Migrations may be written in SQL, Java, or other scripting languages
(fw-how). File naming places migrations in the configured
locations (default db/migration):
db/migration/
├── V1__create_users.sql
├── V2__add_email_index.sql
├── R__refresh_active_users_view.sql # repeatable, reruns on checksum change
└── U1__remove_users.sql # undo (Flyway Teams)The prefix scheme:
| Prefix | Type | Reruns? | Use |
|---|---|---|---|
V<n>__ | Versioned | Once | New schema changes; immutable after merge |
R__ | Repeatable | When checksum changes | Views / stored procs / seed data |
U<n>__ | Undo | Inverse of versioned | Rollback (Teams edition) |
__ (double underscore) separates version + description; .sql (or
configured suffix) marks the file as a migration.
The daily loop is flyway info (preview pending) -> flyway migrate (apply
pending) -> flyway validate (checksum-verify applied files before deploy).
After a failed migration, flyway repair fixes flyway_schema_history -
never edit an applied file. Full command reference (baseline, undo, clean,
and the rest): references/commands.md.
Migrations with a version lower than the history table's current version are
ignored by default; the rest are pending - available but not applied
(fw-how). Safety property: a developer who pulls main and runs
flyway migrate applies only the new migrations; those already in
flyway_schema_history are not re-run.
Configuration via flyway.conf file, env vars (FLYWAY_*), or CLI
flags. Key settings:
flyway.url=jdbc:postgresql://localhost:5432/mydb
flyway.user=myuser
flyway.password=mypass
flyway.locations=filesystem:db/migration,classpath:db/migration
flyway.baselineOnMigrate=true # auto-baseline empty schemas
flyway.cleanDisabled=true # CRITICAL for prod - disable destructive `clean`
flyway.outOfOrder=false # reject migrations with versions lower than max applied
flyway.validateOnMigrate=true # checksum-validate before applyingcleanDisabled=true is a mandatory production guard -
flyway clean drops every object in the schema. Always set this in
production config; only enable for ephemeral test databases.
Gate every PR on an ephemeral DB (Docker / Testcontainers): spin the DB,
apply migrations, run tests against the migrated schema. The full GitHub
Actions job and the Testcontainers @BeforeAll pattern are in
references/commands.md.
Before merge, apply adversarial review of new migrations - classify each as additive / breaking / data-loss / locking.
Add an index on users.email and ship it through CI:
db/migration/V2__add_email_index.sql with
CREATE INDEX idx_users_email ON users(email);.flyway info - it lists V2 as pending while V1
shows applied.flyway migrate applies only V2 and appends a row to
flyway_schema_history with its checksum.flyway ... migrate, then mvn test against the migrated schema.V2 file; on their next run
flyway validate fails with a checksum mismatch, so they add
V3__... instead of mutating V2.| Anti-pattern | Why it fails | Fix |
|---|---|---|
| Edit a previously-applied versioned migration | Checksum mismatch; validate fails on next run | Add a new V_n+1 migration that adjusts |
cleanDisabled=false in production config | One stray flyway clean drops the schema | Always cleanDisabled=true (Step 5) |
| Mixing versioned + repeatable migrations for the same object | Repeatable applies after every versioned change → race | Pick one per object class |
outOfOrder=true without team agreement | Lower-version migrations apply mid-stream; ordering breaks | Default false; enable per change with team review |
| Skip CI gating on per-PR ephemeral DB | Migrations break in production for the first time | Always run migrations in CI (Step 6) |
flyway clean is irreversible; the cleanDisabled=true guard is
the only protection.liquibase-migrations,
atlas-migrations,
sqlmesh-migrations - sister
tools (Liquibase = changelog-driven; Atlas = declarative HCL;
SQLMesh = data-pipeline + schema)