pgdiff compares the schema between two PostgreSQL 9 databases and generates alter statements to be manually run against the second database to make them match. The provided pgdiff.sh script helps automate the process.
pgdiff is transparent in what it does, so it never modifies a database directly. You alone are responsible for verifying the generated SQL before running it against your database. Go ahead and see what SQL gets generated.
pgdiff is written to be easy to expand and improve the accuracy of the diff.
pgdiff [options] <schemaType>
(where options and <schemaType> are listed below)
There seems to be an ideal order for running the different schema types. This order should minimize the problems you encounter. For example, you will always want to add new tables before you add new columns.
In addition, some types can have dependencies which are not in the right order. A classic case is views which depend on other views. The missing view SQL is generated in alphabetical order so if a view create fails due to a missing view, just run the views SQL file over again. The pgdiff.sh script will prompt you about running it again.
Schema type ordering:
- SCHEMA
- ROLE
- SEQUENCE
- TABLE
- COLUMN
- INDEX
- VIEW
- FOREIGN_KEY
- FUNCTION
- TRIGGER
- OWNER
- GRANT_RELATIONSHIP
- GRANT_ATTRIBUTE
- ALL (all above in one run)
I have found it helpful to take --schema-only dumps of the databases in question, load them into a local postgres, then do my sql generation and testing there before running the SQL against a more official database. Your local postgres instance will need the correct users/roles populated because db dumps do not copy that information.
pgdiff -U dbuser -H localhost -D refDB -O "sslmode=disable" -S public \
-u dbuser -h localhost -d compDB -o "sslmode=disable" -s public \
TABLE
| options | explanation |
|---|---|
| -V, --version | prints the version of pgdiff being used |
| -?, --help | displays helpful usage information |
| -U, --user1 | first postgres user |
| -u, --user2 | second postgres user |
| -W, --password1 | first db password |
| -w, --password2 | second db password |
| -H, --host1 | first db host. default is localhost |
| -h, --host2 | second db host. default is localhost |
| -P, --port1 | first db port number. default is 5432 |
| -p, --port2 | second db port number. default is 5432 |
| -D, --dbname1 | first db name |
| -d, --dbname2 | second db name |
| -S, --schema1 | first schema name. default is * (all non-system schemas) |
| -s, --schema2 | second schema name. default is * (all non-system schemas) |
| -O, --option1 | first db options. example: sslmode=disable |
| -o, --option2 | second db options. example: sslmode=disable |
linux and osx binaries are packaged with an extra, optional bash script and pgrun program that helps speed the diffing process.
- download the tgz file for your OS
- untar it:
tar -xzvf pgdiff.tgz - cd to the new pgdiff directory
- edit the db connection defaults in pgdiff.sh
- ...or manually run pgdiff for each schema type listed in the usage section above
- review the SQL output for each schema type and, if you want to make them match, run it against the second db
- download pgdiff.exe from the bin-win directory on github
- either install cygwin so you can run pgdiff.sh or...
- manually run pgdiff.exe for each schema type listed in the usage section above
- review the SQL output and, if you want to make them match, run it against the second db
This project works on Windows, just not as nicely as it does for Linux and Mac. If you are inclined to write a Windows complement to the pgdiff.sh script, feel free to contribute it or we can link to it. Even better would be a replacement written in Go.
- 0.9.0 - Implemented ROLE, SEQUENCE, TABLE, COLUMN, INDEX, FOREIGN_KEY, OWNER, GRANT_RELATIONSHIP, GRANT_ATTRIBUTE
- 0.9.1 - Added VIEW, FUNCTION, and TRIGGER (Thank you, Shawn Carroll AKA SparkeyG)
- 0.9.2 - Fixed bug when using the non-default port
- 0.9.3 - Fixed VARCHAR bug when no max length specified
- 1.0.0 - Adding support for comparing two different schemas (same or different db), one schema between databases, or all schemas between databases. (Also removed binaries from git repository)
If you think you found a bug, it might help replicate it if you find the appropriate test script (in the test directory) and modify it to show the problem. Attach the script to an Issue request.
- fix SQL for adding an array column
- create windows version of pgdiff.sh (or even better: re-write it all in Go)
- allow editing of individual SQL lines after failure (this would probably be done in the script pgdiff.sh)
- store failed SQL statements in an error file for later fixing and rerunning?