stripe/pg-schema-diffPublic

NotificationsYou must be signed in to change notification settings
Fork36
Star527

Go library for diffing Postgres schemas and generating SQL migrations

License

MIT license

527 stars 36 forks Branches Tags Activity

Star

Notifications

You must be signed in to change notification settings

Branches Tags

Folders and files

Name		Name	Last commit message	Last commit date
Latest commit History 83 Commits
.github		.github
build		build
cmd/pg-schema-diff		cmd/pg-schema-diff
internal		internal
pkg		pkg
scripts		scripts
.editorconfig		.editorconfig
.gitignore		.gitignore
.golangci.yaml		.golangci.yaml
.sqlfluff		.sqlfluff
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE.md		LICENSE.md
Makefile		Makefile
README.md		README.md
go.mod		go.mod
go.sum		go.sum

Repository files navigation

pg-schema-diff

Computes the diff(erences) between Postgres database schemas and generates the SQL required to get your database schema from point A to B withminimal downtime & locks. This enables you to take your database and migrate it to any desired schema defined in plain DDL.

The tooling attempts to use native postgres migration operations to perform online migrations and avoid locking wherever possible. Not all migrations willbe lock-free and some might require downtime, but the hazards system will warn you ahead of time when that's the case.Stateful online migration techniques, like shadow tables, aren't yet supported.

Online index Replacement

Your project's diff:

$ git diffdiff --git a/schema/schema.sql b/schema/schema.sqlindex cc3a14b..cf4b32d 100644--- a/schema/schema.sql+++ b/schema/schema.sql@@ -2,5 +2,5 @@ CREATE TABLE foobar(  created_at timestamp, message text );-CREATE INDEX message_idx ON foobar(message);+CREATE INDEX message_idx ON foobar(message, created_at);

The generated plan (queries usingmessage_idx will always have an index backing them, even while the new index is being built):

$ pg-schema-diff plan --from-dsn "postgres://postgres:postgres@localhost:5432/postgres" --to-dir ./schema################################ Generated plan ################################1. ALTER INDEX "message_idx" RENAME TO "pgschemadiff_tmpidx_message_idx_IiaKzkvPQtyA7ob9piVqiQ";        -- Statement Timeout: 3s2. CREATE INDEX CONCURRENTLY message_idx ON public.foobar USING btree (message, created_at);        -- Statement Timeout: 20m0s        -- Lock Timeout: 3s        -- Hazard INDEX_BUILD: This might affect database performance. Concurrent index builds require a non-trivial amount of CPU, potentially affecting database performance. They also can take a while but do not lock out writes.3. DROP INDEX CONCURRENTLY "pgschemadiff_tmpidx_message_idx_IiaKzkvPQtyA7ob9piVqiQ";        -- Statement Timeout: 20m0s        -- Lock Timeout: 3s        -- Hazard INDEX_DROPPED: Dropping this index means queries that use this index might perform worse because they will no longer will be able to leverage it.

Online`NOT NULL` constraint creation

Your project's diff:

diff --git a/schema/schema.sql b/schema/schema.sqlindex cc3a14b..5a1cec2 100644--- a/schema/schema.sql+++ b/schema/schema.sql@@ -1,5 +1,5 @@ CREATE TABLE foobar(- created_at timestamp,+ created_at timestamp NOT NULL, message text ); CREATE INDEX message_idx ON foobar(message);

The generated plan (leverages check constraints to eliminate the need for a long-lived access-exclusive lock on the table):

$ pg-schema-diff plan --from-dsn "postgres://postgres:postgres@localhost:5432/postgres" --to-dir ./schema################################ Generated plan ################################1. ALTER TABLE "public"."foobar" ADD CONSTRAINT "pgschemadiff_tmpnn_BCOxMXqAQwaXlKPCRXoMMg" CHECK("created_at" IS NOT NULL) NOT VALID;        -- Statement Timeout: 3s2. ALTER TABLE "public"."foobar" VALIDATE CONSTRAINT "pgschemadiff_tmpnn_BCOxMXqAQwaXlKPCRXoMMg";        -- Statement Timeout: 3s3. ALTER TABLE "public"."foobar" ALTER COLUMN "created_at" SET NOT NULL;        -- Statement Timeout: 3s4. ALTER TABLE "public"."foobar" DROP CONSTRAINT "pgschemadiff_tmpnn_BCOxMXqAQwaXlKPCRXoMMg";        -- Statement Timeout: 3s

Key features

Declarative schema migrations
The use of postgres native operations for zero-downtime migrations wherever possible:
- Concurrent index builds
- Online index replacement: If some index is changed, the new version will be built before the old version is dropped, preventing a window where no index is backing queries
- Online constraint builds: Constraints (check, foreign key) are added asINVALID before being validated, eliminating the needfor a long access-exclusive lock on the table
- OnlineNOT NULL constraint creation using check constraints to eliminate the need for an access-exclusive lock on the table
- Prioritized index builds: Building new indexes is always prioritized over deleting old indexes
A comprehensive set of features to ensure the safety of planned migrations:
- Operators warned of dangerous operations.
- Migration plans are validated first against a temporary database exactly as they would be performed against the real database.
Strong support of partitions

Install

CLI

go install github.com/stripe/pg-schema-diff/cmd/pg-schema-diff@latest

Library

go get -u github.com/stripe/pg-schema-diff@latest

Using CLI

1. Apply schema to fresh database

Create a directory to hold your schema files. Then, generate sql files and place them into a schema dir.

mkdir schemaecho"CREATE TABLE foobar (id int);"> schema/foobar.sqlecho"CREATE TABLE bar (id varchar(255), message TEXT NOT NULL);"> schema/bar.sql

Apply the schema to a fresh database.The connection string spec can be found here.Setting thePGPASSWORD env var will override any password set in the connection string and is recommended.

pg-schema-diff apply --from-dsn"postgres://postgres:postgres@localhost:5432/postgres" --to-dir schema

2. Updating schema

Update the SQL file(s)

echo"CREATE INDEX message_idx ON bar(message)">> schema/bar.sql

Apply the schema. Any hazards in the generated plan must be approved

pg-schema-diff apply --from-dsn"postgres://postgres:postgres@localhost:5432/postgres" --to-dir schema --allow-hazards INDEX_BUILD

Using Library

Docs to use the library can be foundhere. Check outthe CLIfor an example implementation with the library

1. Generating plan

// The tempDbFactory is used in plan generation to extract the new schema and validate the plantempDbFactory,err:=tempdb.NewOnInstanceFactory(ctx,func(ctx context.Context,dbNamestring) (*sql.DB,error) {copiedConfig:=connConfig.Copy()copiedConfig.Database=dbNamereturnopenDbWithPgxConfig(copiedConfig)})iferr!=nil {panic("Generating the TempDbFactory failed")}defertempDbFactory.Close()// Generate the migration planplan,err:=diff.Generate(ctx,diff.DBSchemaSource(connPool),diff.DDLSchemaSource(ddl),diff.WithTempDbFactory(tempDbFactory),diff.WithDataPackNewTables(),)iferr!=nil {panic("Generating the plan failed")}

2. Applying plan

We leave plan application up to the user. For example, you might want to take out a session-level advisory lock if you areconcerned about concurrent migrations on your database. You might also want a second user to approve the planbefore applying it.

Example apply:

for_,stmt:=rangeplan.Statements {if_,err:=conn.ExecContext(ctx,fmt.Sprintf("SET SESSION statement_timeout = %d",stmt.Timeout.Milliseconds()));err!=nil {panic(fmt.Sprintf("setting statement timeout: %s",err))}if_,err:=conn.ExecContext(ctx,fmt.Sprintf("SET SESSION lock_timeout = %d",stmt.LockTimeout.Milliseconds()));err!=nil {panic(fmt.Sprintf("setting lock timeout: %s",err))}if_,err:=conn.ExecContext(ctx,stmt.ToSQL());err!=nil {panic(fmt.Sprintf("executing migration statement. the database maybe be in a dirty state: %s: %s",stmt,err))}}

Supported Postgres versions

Supported: 14, 15, 16, 17
Unsupported: <= 13 are not supported. Use at your own risk.

Unsupported migrations

An abridged list of unsupported migrations:

Views (Planned)
Privileges (Planned)
Types (Only enums are currently supported)
Renaming. The diffing library relies on names to identify the old and new versions of a table, index, etc. If you renamean object, it will be treated as a drop and an add

Contributing

This project is in its early stages. We appreciate all the feature/bug requests we receive, but we have limited cyclesto review direct code contributions at this time. SeeContributing to learn more.