Use lowercase sys.synonyms in mssql.

[nvoxland]

Description

The sys.synonyms view is actually lowercase in mssql. Case-insensitive database collations are fine with our use of SYS.SYNONYMS, but case sensitive ones like Latin1_General_BIN2 or any that end in BIN or BIN2 are not.

Fixes #2044

---

Dev Handoff Notes (Internal Use)

Links

• Fixed Issue: Generating changelog fails for MSSQL 2019 when sys.synonyms exist #2044
• Test Results: https://github.com/liquibase/liquibase/pull/2225/checks

Testing

• Steps to Reproduce: Fixed Issue: Generating changelog fails for MSSQL 2019 when sys.synonyms exist #2044
• Guidance:
  • Impact: impacts diff-changelog or generate-changelog on mssql only

Dev Verification

Started a database with a BIN collation: docker run ~~e ACCEPT_EULA=Y -e MSSQL_PID="Developer"~~ e MSSQL_SA_PASSWORD="LiquibasePass1" ~~e MSSQL_TCP_PORT=14333~~ e MSSQL_COLLATION=Latin1_General_BIN2 ~~p 14333:14333~~ d mcr.microsoft.com/mssql/server:20 19-GA-ubuntu-16.04

then ran liquibase generate-changelog.

Prior to the fix, it outputs a fine-level message of Cannot get object dependencies like [2021-11-18 10:52:55] FINE [liquibase.diff] Cannot get object dependencies: Error executing SQL select object_schema_name(referencing_id) as referencing_schema_name, object_name(referencing_id) as referencing_name, object_name(referenced_id) as referenced_name, object_schema_name(referenced_id) as referenced_schema_name from sys.sql_expression_dependencies depz where (object_schema_name(referenced_id)='dbo') UNION select object_schema_name(object_id) as referencing_schema_name, object_name(object_id) as referencing_name, object_name(parent_object_id) as referenced_name, object_schema_name(parent_object_id) as referenced_schema_name from sys.objects where parent_object_id > 0 and is_ms_shipped=0 and (object_schema_name(object_id)='dbo') UNION select object_schema_name(fk.object_id) as referencing_schema_name, fk.name as referencing_name, i.name as referenced_name, object_schema_name(i.object_id) as referenced_schema_name from sys.foreign_keys fk join sys.indexes i on fk.referenced_object_id=i.object_id and fk.key_index_id=i.index_id where fk.is_ms_shipped=0 and (object_schema_name(fk.object_id)='dbo') UNION select object_schema_name(i.object_id) as referencing_schema_name, object_name(i.object_id) as referencing_name, s.name as referenced_name, null as referenced_schema_name from sys.indexes i join sys.partition_schemes s on i.data_space_id = s.data_space_id UNION select null as referencing_schema_name, s.name as referencing_name, f.name as referenced_name, null as referenced_schema_name from sys.partition_functions f join sys.partition_schemes s on s.function_id=f.function_id UNION select null as referencing_schema_name, s.name as referencing_name, fg.name as referenced_name, null as referenced_schema_name from sys.partition_schemes s join sys.destination_data_spaces ds on s.data_space_id=ds.partition_scheme_id join sys.filegroups fg on ds.data_space_id=fg.data_space_id UNION select distinct null as referencing_schema_name, f.name as referencing_name, ds.name as referenced_name, null as referenced_schema_name from sys.database_files f join sys.data_spaces ds on f.data_space_id=ds.data_space_id where f.data_space_id > 1 UNION select object_schema_name(t.object_id) as referencing_schema_name, t.name as referencing_name, ds.name as referenced_name, null as referenced_schema_name from sys.tables t join sys.data_spaces ds on t.filestream_data_space_id=ds.data_space_id where t.filestream_data_space_id > 1 UNION select object_schema_name(t.object_id) as referencing_schema_name, t.name as referencing_name, ds.name as referenced_name, null as referenced_schema_name from sys.tables t join sys.data_spaces ds on t.lob_data_space_id=ds.data_space_id where t.lob_data_space_id > 1 UNION select object_schema_name(i.object_id) as referencing_schema_name, i.name as referencing_name, ds.name as referenced_name, null as referenced_schema_name from sys.indexes i join sys.data_spaces ds on i.data_space_id=ds.data_space_id where i.data_space_id > 1 UNION select object_schema_name(i.object_id) as referencing_schema_name, i.name as referencing_name, object_name(i.object_id) as referenced_name, object_schema_name(i.object_id) as referenced_schema_name from sys.indexes i where object_schema_name(i.object_id)='dbo' UNION SELECT SCHEMA_NAME(SCHEMA_ID) as referencing_schema_name, name as referencing_name, PARSENAME(BASE_OBJECT_NAME,1) AS referenced_name, (CASE WHEN PARSENAME(BASE_OBJECT_NAME,2) IS NULL THEN schema_name(schema_id) else PARSENAME(BASE_OBJECT_NAME,2) END) AS referenced_schema_name FROM SYS.SYNONYMS WHERE is_ms_shipped='false' AND SCHEMA_NAME(SCHEMA_ID)='dbo' UNION select object_schema_name(c.object_id) as referencing_schema_name, c.name as referencing_name, object_schema_name(nc.object_id) as referenced_schema_name, nc.name as referenced_name from sys.indexes c join sys.indexes nc on c.object_id=nc.object_id JOIN sys.objects o ON c.object_id = o.object_id where c.index_id != nc.index_id and c.type_desc='CLUSTERED' and c.is_unique='true' and (not(nc.type_desc='CLUSTERED') OR nc.is_unique='false') AND o.type_desc='VIEW' AND o.name='AR_DETAIL_OPEN': Invalid object name 'SYS.SYNONYMS'. .

After the fix, there is no "cannot get object dependencies" log messages

Test Requirements (Liquibase Internal QA)

This test requires:

• a SQL Server 2019 Docker with a case-sensitive collation enabled
• a SQL Server Docker with default collation (any SQL Server version; the container you typically use for testing SQL Server should have a default, case-insensitive collation.
• a Pro License (synonyms are a Pro-only object for snapshotting commands (generate/diff-changelog).

Start the MSSQL Docker container with case-sensitive collation by running:

docker run --name mssql_19 -e ACCEPT_EULA=Y -e MSSQL_PID="Developer" -e MSSQL_SA_PASSWORD="LiquibasePass1" -e MSSQL_TCP_PORT=14333 -e MSSQL_COLLATION=Latin1_General_BIN2 -p 14333:14333 -d mcr.microsoft.com/mssql/server:2019-GA-ubuntu-16.04

Connect to the database as sa with password LiquibasePass1; the port is 14333.

Create a user and a database to use for testing. You can use the default master database but if you run drop-all on that master, you drop a lot of dependencies SQL Server itself needs. It is best practice not to use the master database for testing.

To create a SQL Server login datical_login and database proCatalog:

/****** Object:  Database [proCatalog]    Script Date: 10/11/2019 4:50:57 PM ******/
CREATE DATABASE [proCatalog]
GO

ALTER DATABASE [proCatalog] SET  READ_WRITE;
GO

USE [proCatalog]
GO
/* For security reasons the login is created disabled and with a random password. */
/****** Object:  Login [datical_login]    Script Date: 10/11/2019 5:13:54 PM ******/
CREATE LOGIN [datical_login] WITH PASSWORD='D4tic4l01', DEFAULT_DATABASE=[proCatalog], DEFAULT_LANGUAGE=[us_english], CHECK_EXPIRATION=OFF, CHECK_POLICY=ON
GO

ALTER LOGIN [datical_login] ENABLE
GO

/****** Object:  User [dvone4145]    Script Date: 10/11/2019 5:43:19 PM ******/
CREATE USER [datical] FOR LOGIN [datical_login] WITH DEFAULT_SCHEMA=[dbo]
GO

USE [master]
GO
ALTER SERVER ROLE [dbcreator] ADD MEMBER [datical_login]
GO

GRANT VIEW ANY DEFINITION TO [datical_login]
GO
ALTER SERVER ROLE [dbcreator] ADD MEMBER [datical_login]
GO
EXEC master..sp_addsrvrolemember @loginame = N'datical_login', @rolename = N'sysadmin'
GO

USE [proCatalog]
GO

EXEC sp_addrolemember N'db_ddladmin', N'datical'
GO
EXEC sp_addrolemember N'db_datareader', N'datical'
GO
EXEC sp_addrolemember N'db_datawriter', N'datical'
GO
EXEC sp_addrolemember N'db_securityadmin', N'datical'
GO
GRANT VIEW DATABASE STATE to [datical]
GO

Create a liquibase.properties file that both url and referenceUrl defined:

### CONNECTION Properties ###
url: jdbc:sqlserver://localhost:14333;databaseName=proCatalog
username: datical_login
password: D4tic4l01
driver: com.microsoft.sqlserver.jdbc.SQLServerDriver

referenceUrl: jdbc:sqlserver://localhost:14333;databaseName=proCatalog
referenceUsername: sa
referencePassword: LiquibasePass1
referenceDriver: com.microsoft.sqlserver.jdbc.SQLServerDriver

### Liquibase Pro Key ###
liquibaseProKey: KEYHERE

Use the attached changelogs and run a liquibase update to create a private and a public synonym on the database.

liquibase update --changelog-file lb2175-changelog.xml --labels=setup,createPrivateSynonym,createPublicSynonym

Query to Determine SQL Server Collation (run directly on database)

SELECT CONVERT (varchar(256), SERVERPROPERTY('collation'));

• Latin1_General_BIN2 is a case-sensitive collation setting
• SQL_Latin1_General_CP1_CI_AS is a case-insensitive collation setting.

Manual Tests

Verify generatechangelog does not throw a database error when connecting to MSSQL with a casesensitive collation type.

• liquibase --log-level INFO generate-changelog --changelog-file gcl.mssql.sql
• generate-changelog is successful.
• There is no error that SYS.SYNONYMS is an invalid object name.
• The generated changelog has three changesets: a create table and two create synonyms.

Verify diffchangelog does not throw a database error when connecting to MSSQL with a casesensitive collation type.

• liquibase --log-level INFO diff-changelog --changelog-file dcl.mssql.sql
• diff-changelog is successful.
• There is no error that SYS.SYNONYMS is an invalid object name.
• The generated changelog has three changesets: a create table and two create synonyms.

Verify that generatechangelog does not throw a database error when there are no synonyms present when connecting to MSSQL with a casesensitive collation type.

• SETUP: Drop the synonyms from proCatalog (make sure the jdbc url is configured to use proCatalog, not master!)
• liquibase --log-level FINE generate-changelog --changelog-file gclEmpty.mssql.sql
• There is no log output that SYS.SYNONYMS is an invalid object name.

Verify generate-changelog does not throw an error when connecting to a MSSQL with default collation.

• SETUP: Your “normal” SQL Sever Docker will not have case-sensitive collation; you can use whatever Docker container you usually use to test for this. Don’t forget to update your liquibase.properties file to connect to this MSSQL instance.
• liquibase --log-level FINE generate-changelog --changelog-file gclEmpty2.mssql.sql
• There is no log output that SYS.SYNONYMS is an invalid object name.

Automated Tests

• No new automated tests are required for this fix.

┆Issue is synchronized with this Jira Bug by Unito