gsqlcmd | Synchronizing Data Across Databases on Multiple Platforms

gsqlcmd allows synchronizing data across databases on different database server platforms.

Below we discuss several techniques.

You may find the source code in the Examples\Sync Databases folder of the downloaded package.

Also, you may try any batch file with a demo sample hosted in Microsoft Azure SQL Database.

Contents:

Basics
Two-way synchronization using GUID identifiers
Using SQL Server rowversion/timestamp columns to synchronize new and changed rows
Using last update columns to synchronize new and changed rows
Complete synchronization including deletes
Complete synchronization across multiple editable databases with integer key columns
Improving and unifying solutions
Final Notes

Basics

Suppose, we have two databases, db1 and db2, with tables that contain customers: table11 on db1 and table12 on db2.

CREATE TABLE [s61].[table11] (
      [id] int IDENTITY(1,1) NOT NULL
    , [name] nvarchar(50) NOT NULL
    , CONSTRAINT [PK_table11] PRIMARY KEY ([id])
);

CREATE TABLE [s61].[table12] (
      [id] int IDENTITY(1,1) NOT NULL
    , [name] nvarchar(50) NOT NULL
    , CONSTRAINT [PK_table12] PRIMARY KEY ([id])
);

The table11 contains the following records:

id	name
1	Customer C1
2	Customer C2
3	Customer C3
4	Customer C4
5	Customer C5

To merge data from table11 to table12, we may use the following batch file:

gsqlcmd exec db1 s61.table11 table11.csv

gsqlcmd make db2 table11.csv table12.sql /table=s61.table12 /merge /insertIdentity

gsqlcmd exec db2 table12.sql

The first exec command exports data from table11 to a CSV file:

id;"name"
1;"Customer C1"
2;"Customer C2"
3;"Customer C3"
4;"Customer C4"
5;"Customer C5"

The second make command produces merge commands for table12:

UPDATE [s61].[table12] SET [name] = N'Customer C1' WHERE [id] = 1;
UPDATE [s61].[table12] SET [name] = N'Customer C2' WHERE [id] = 2;
UPDATE [s61].[table12] SET [name] = N'Customer C3' WHERE [id] = 3;
UPDATE [s61].[table12] SET [name] = N'Customer C4' WHERE [id] = 4;
UPDATE [s61].[table12] SET [name] = N'Customer C5' WHERE [id] = 5;
GO
SET IDENTITY_INSERT [s61].[table12] ON;
INSERT INTO [s61].[table12] ([id], [name]) SELECT s.[id], s.[name] FROM (SELECT 1 AS [id], N'Customer C1' AS [name]) s LEFT OUTER JOIN [s61].[table12] t ON t.[id] = s.[id] WHERE t.[id] IS NULL;
INSERT INTO [s61].[table12] ([id], [name]) SELECT s.[id], s.[name] FROM (SELECT 2 AS [id], N'Customer C2' AS [name]) s LEFT OUTER JOIN [s61].[table12] t ON t.[id] = s.[id] WHERE t.[id] IS NULL;
INSERT INTO [s61].[table12] ([id], [name]) SELECT s.[id], s.[name] FROM (SELECT 3 AS [id], N'Customer C3' AS [name]) s LEFT OUTER JOIN [s61].[table12] t ON t.[id] = s.[id] WHERE t.[id] IS NULL;
INSERT INTO [s61].[table12] ([id], [name]) SELECT s.[id], s.[name] FROM (SELECT 4 AS [id], N'Customer C4' AS [name]) s LEFT OUTER JOIN [s61].[table12] t ON t.[id] = s.[id] WHERE t.[id] IS NULL;
INSERT INTO [s61].[table12] ([id], [name]) SELECT s.[id], s.[name] FROM (SELECT 5 AS [id], N'Customer C5' AS [name]) s LEFT OUTER JOIN [s61].[table12] t ON t.[id] = s.[id] WHERE t.[id] IS NULL;
SET IDENTITY_INSERT [s61].[table12] OFF;
GO
print 'Processed 5 total records';
GO

These SQL commands update names on the target table and insert new rows.

The make command uses the /merge option to generate such commands and the /insertIdentity option to keep identity values.

The third exec command executes the commands against db2.

Solving Issues with SET IDENTITY_INSERT

To execute the script with SET IDENTITY_INSERT on Microsoft SQL Server, a user of the database connection must have the ALTER ON TABLE permission.

Otherwise, it returns an error: Cannot find the object "s61.table12" because it does not exist or you do not have permissions.

See details at https://docs.microsoft.com/en-us/sql/t-sql/statements/set-identity-insert-transact-sql?view=sql-server-2017.

To solve this issue, you may use a separate user to synchronize data or create the target table with the primary key column without the IDENTITY option:

CREATE TABLE [s61].[table13] (
      [id] int NOT NULL
    , [name] nvarchar(50) NOT NULL
    , CONSTRAINT [PK_table13] PRIMARY KEY ([id])
);

The technique remains the same, but the merge SQL commands do not contain SET IDENTITY_INSERT.

If you cannot change the identity column, you may use the solution below.

Using Different ID Values in Source and Target Tables

If you do not need to keep ID values the same in both tables, you may omit the identity columns in synchronization.

Here is a sample:

gsqlcmd exec db1 "SELECT name FROM s61.table11" table11.csv

gsqlcmd make db2 table11.csv table14.sql /table=s61.table14 /keys=name /merge

gsqlcmd exec db2 table14.sql

The first exec command exports data without the ID column:

"name"
"Customer C1"
"Customer C2"
"Customer C3"
"Customer C4"
"Customer C5"

Pay attention the command uses an inline select: SELECT name FROM s61.table11

You may use this feature to export the required columns and change their names according to names in the target table.

Also, you may calculate values for the required columns of the target table.

The second command produces the following SQL commands to insert absent names into the table12:

INSERT INTO [s61].[table14] ([name]) SELECT s.[name] FROM (SELECT N'Customer C1' AS [name]) s LEFT OUTER JOIN [s61].[table14] t ON t.[name] = s.[name] WHERE t.[name] IS NULL;
INSERT INTO [s61].[table14] ([name]) SELECT s.[name] FROM (SELECT N'Customer C2' AS [name]) s LEFT OUTER JOIN [s61].[table14] t ON t.[name] = s.[name] WHERE t.[name] IS NULL;
INSERT INTO [s61].[table14] ([name]) SELECT s.[name] FROM (SELECT N'Customer C3' AS [name]) s LEFT OUTER JOIN [s61].[table14] t ON t.[name] = s.[name] WHERE t.[name] IS NULL;
INSERT INTO [s61].[table14] ([name]) SELECT s.[name] FROM (SELECT N'Customer C4' AS [name]) s LEFT OUTER JOIN [s61].[table14] t ON t.[name] = s.[name] WHERE t.[name] IS NULL;
INSERT INTO [s61].[table14] ([name]) SELECT s.[name] FROM (SELECT N'Customer C5' AS [name]) s LEFT OUTER JOIN [s61].[table14] t ON t.[name] = s.[name] WHERE t.[name] IS NULL;
GO
print 'Processed 5 total records';
GO

Pay attention that the make command uses the /keys=name option to specify columns used in the JOIN conditions.

Using Import Mode

The one-way synchronization, discussed above, always include three parts:

Exporting data from the source table
Creating SQL commands to update the target table
Executing the SQL commands

You may create batch files using CSV files in the first and second steps. This feature is available in all paid editions.

The gsqlcmd Enterprise edition allows using the import mode that combines these three commands into a single one.

For example, the following command merges data from table11 to table13:

gsqlcmd import db2 db1::s61.table11 /table=s61.table13 /merge

The main command is executed against the db2 connection.

However, it loads the source data from the s61.table11 table of the db1 database into the memory but not into a CSV file like the command

gsqlcmd exec db1 s61.table11 table11.csv

The import command, with an inline select, looks like this

gsqlcmd import db2 "db1::SELECT name FROM s61.table11" /table=s61.table14 /keys=name /merge

You may easily change the initial make to the import mode.

The next samples will use the import mode.

Two-way Synchronization using GUID Identifiers

In the previous samples, both tables have the primary key columns of the integer type.

The one-way synchronization, in this case, is simple and reliable.

The easiest way to have two-way synchronization is to use the primary key columns of the GUID type.

Here is a table declaration for Microsoft SQL Server:

CREATE TABLE [s61].[table21] (
      [id] uniqueidentifier NOT NULL DEFAULT(newid())
    , [name] nvarchar(50) NOT NULL
    , CONSTRAINT [PK_table21] PRIMARY KEY ([id])
);

CREATE TABLE [s61].[table22] (
      [id] uniqueidentifier NOT NULL DEFAULT(newid())
    , [name] nvarchar(50) NOT NULL
    , CONSTRAINT [PK_table22] PRIMARY KEY ([id])
);

The id column has the uniqueidentifier data type with the newid() default value.

In this case, Microsoft SQL Server creates unique id values for new rows.

Here are commands to merge data from the table21 to table22 and backward:

gsqlcmd import db2 db1::s61.table21 /table=s61.table22 /merge

gsqlcmd import db1 db2::s61.table22 /table=s61.table21 /merge

This technique is easy. It does not require additional tables or fields.

However, it is suitable for small and mid-size tables as all the data are loaded and updated every time.

Using SQL Server Rowversion/Timestamp Columns to Synchronize New and Changed Rows

Microsoft SQL Server and Microsoft SQL Server Compact have the rowversion data type with the timestamp synonym.

It is generally used as a mechanism for version-stamping table rows.

The server increments the server-level value on every INSERT and UPDATE, and updates the field value.

See details here:

So, the idea is simple:

Import all rows once and save the maximum rowversion value.
Synchronize data after the last saved rowversion value and save the maximum rowversion value again.

Here is a table declaration with the rowversion (timestamp) field:

CREATE TABLE [s61].[table31] (
      [id] uniqueidentifier NOT NULL DEFAULT(newid())
    , [name] nvarchar(50) NOT NULL
    , [ts] timestamp NOT NULL
    , CONSTRAINT [PK_table31] PRIMARY KEY ([id])
);

CREATE TABLE [s61].[table32] (
      [id] uniqueidentifier NOT NULL DEFAULT(newid())
    , [name] nvarchar(50) NOT NULL
    , [ts] timestamp NOT NULL
    , CONSTRAINT [PK_table32] PRIMARY KEY ([id])
);

Here is a sample table that contains saved timestamps:

CREATE TABLE [s61].[sync_timestamps] (
      [source] nvarchar(128) NOT NULL
    , [target] nvarchar(128) NOT NULL
    , [last_ts] binary(8) NOT NULL
    , [prev_ts] binary(8) NULL
    , CONSTRAINT [PK_sync_timestamps] PRIMARY KEY ([source], [target])
);

You may use a single table for all database tables and synchronization pairs due to the source and target columns.

Here is a batch file to merge new and updated rows only from the table31 to table32:

gsqlcmd exec db1 update-sync-table31-to-table32.sql

if errorlevel 1 goto :EOF

gsqlcmd import db2 db1::select-table31-to-table32.sql /table=s61.table32 /merge

if errorlevel 1 goto :EOF

gsqlcmd exec db1 commit-sync-table31-to-table32.sql

update-sync-table31-to-table32.sql has the following code:

MERGE s61.sync_timestamps AS t
USING (SELECT 'table31' AS [source], 'table32' AS [target], COALESCE(MAX(ts), 0x) AS ts FROM s61.table31) AS s ([source], [target], ts)
ON (t.[source] = s.[source] AND t.[target] = s.[target])
WHEN MATCHED AND last_ts = prev_ts THEN
    UPDATE SET last_ts = s.ts
WHEN NOT MATCHED THEN
    INSERT ([source], [target], last_ts) VALUES (s.[source], s.[target], s.ts);

The merge command inserts or updates the row for the specified synchronization pair.

The sync_timestamps table has values like these:

source	target	last_ts	prev_ts
table31	table32	0x00000000000084E9	0x00000000000084E8

The code updates the last_ts field only if the last_ts and prev_ts values are equal.

This technique prevents changes if the previous synchronization failed.

To commit changes, the batch uses the last command:

gsqlcmd exec db1 commit-sync-table31-to-table32.sql

commit-sync-table31-to-table32.sql has the following simple code:

UPDATE s61.sync_timestamps SET prev_ts = last_ts WHERE [source] = 'table31' AND [target] = 'table32'

Here is the code of the select-table31-to-table32.sql used to select new and changed rows only:

SELECT
    id, name
FROM
    s61.table31
WHERE
    ts > COALESCE((
        SELECT
            prev_ts
        FROM
            s61.sync_timestamps
        WHERE
            [source] = 'table31' AND [target] = 'table32'
        ), 0x)

Pay attention that it selects rows with timestamp values greater than the prev_ts field committed in the previous synchronization.

This topic showed the way to synchronize new and changed rows only using columns of the rowversion data type specific to SQL Server.

The main benefit of this way is a minimal and non-destructive change of the underlying tables. You need to add to tables a single column of the rowversion data type.

Also, this is only the way to synchronize data across Microsoft SQL Server Compact databases as they do not support triggers discussed below.

Using Last Update Columns to Synchronize New and Changed Rows

Another way to select new and updated rows only is to use columns that contain the creation or last update time.

The challenge is to create such fields and triggers for all tables to be synchronized.

Here is a sample of the table and its triggers for Microsoft SQL Server:

CREATE TABLE [s61].[table41] (
      [id] uniqueidentifier NOT NULL DEFAULT(newid())
    , [name] nvarchar(50) NOT NULL
    , [last_update] datetime NOT NULL DEFAULT(getutcdate())
    , CONSTRAINT [PK_table41] PRIMARY KEY ([id])
);

CREATE TRIGGER [s61].[trigger_table41_after_insert]
    ON [s61].[table41] AFTER INSERT
AS
BEGIN
SET NOCOUNT ON

UPDATE s61.table41
SET
    last_update = GETUTCDATE()
FROM
    s61.table41 t
    INNER JOIN inserted ON inserted.id = t.id
END;

CREATE TRIGGER [s61].[trigger_table41_after_update]
    ON [s61].[table41] AFTER UPDATE
AS
BEGIN
SET NOCOUNT ON

UPDATE s61.table41
SET
    last_update = GETUTCDATE()
FROM
    s61.table41 t
    INNER JOIN deleted ON deleted.id = t.id
END;

To keep the last time of synchronized data for each synchronization pair, we may use a table like this:

CREATE TABLE [s61].[sync_last_updates] (
      [source] nvarchar(128) NOT NULL
    , [target] nvarchar(128) NOT NULL
    , [last_update] datetime NOT NULL
    , [prev_update] datetime NULL
    , CONSTRAINT [PK_sync_last_updates] PRIMARY KEY ([source], [target])
);

Here is a batch file used to synchronize data from the table41 in db1 to table42 in db2:

gsqlcmd exec db1 update-sync-table41-to-table42.sql

if errorlevel 1 goto :EOF

gsqlcmd import db2 db1::select-table41-to-table42.sql /table=s61.table42 /merge

if errorlevel 1 goto :EOF

gsqlcmd exec db1 commit-sync-table41-to-table42.sql

The first command saves the last_update value to be committed in the last batch command if there are no errors occurred. The update-sync-table41-to-table42.sql has the following code:

MERGE s61.sync_last_updates AS t
USING (SELECT 'table41' AS [source], 'table42' AS [target], COALESCE(MAX(last_update), 0) AS last_update FROM s61.table41) AS s ([source], [target], last_update)
ON (t.[source] = s.[source] AND t.[target] = s.[target])
WHEN MATCHED AND t.last_update = t.prev_update THEN
    UPDATE SET last_update = s.last_update
WHEN NOT MATCHED THEN
    INSERT ([source], [target], last_update) VALUES (s.[source], s.[target], s.last_update);

The commit-sync-table41-to-table42.sql, used in the last batch command, has the following simple code:

UPDATE s61.sync_last_updates SET prev_update = last_update WHERE [source] = 'table41' AND [target] = 'table42'

And, here is the code of the select-table41-to-table42.sql used to select rows added or updated after the last synchronization time saved in the prev_update field:

SELECT
    id, name
FROM
    s61.table41
WHERE
    last_update > COALESCE((
        SELECT
            prev_update
        FROM
            s61.sync_last_updates
        WHERE
            [source] = 'table41' AND [target] = 'table42'
        ), 0)

Complete Synchronization Including Deletes

Topics above discuss adding and updating data from the source to target tables only.

Such synchronization may solve your tasks if you hide rows instead of deleting them.

To replicate deleting rows, you need to have data about the deletion.

A simple approach is to create a single table that contains the deleted identifiers and to add triggers that save such identifiers in the delete operations.

Here is a sample table that contains the deleted identifiers:

CREATE TABLE [s61].[deleted_guids] (
      [id] uniqueidentifier NOT NULL
    , [source] nvarchar(128) NOT NULL
    , [last_update] datetime NOT NULL
    , CONSTRAINT [PK_deleted_guids] PRIMARY KEY ([id])
);

As the GUID identifiers are globally unique, the table uses them as a primary key.

The source column contains the table of the deleted row.

The last_update column contains the deletion time used to synchronize the deletion once.

Here is a trigger sample:

CREATE TRIGGER [s61].[trigger_table51_after_delete]
    ON [s61].[table51] AFTER DELETE
AS
BEGIN
SET NOCOUNT ON

INSERT INTO deleted_guids (id, [source], last_update)
SELECT
    deleted.id
    , 'table51' AS [source]
    , GETUTCDATE() AS last_update
FROM
    deleted
END;

Here is a modified batch file that replicates deleting rows in the target table:

gsqlcmd exec db1 update-sync-table51-to-table52.sql

if errorlevel 1 goto :EOF

gsqlcmd import db2 db1::select-table51-to-table52.sql /table=s61.table52 /merge

if errorlevel 1 goto :EOF

gsqlcmd import db2 db1::select-table51-to-table52-deleted.sql /table=s61.table52 /delete

if errorlevel 1 goto :EOF

gsqlcmd exec db1 commit-sync-table51-to-table52.sql

The batch is similar to the discussed above. However, it includes additional actions for the deletions.

The update-sync-table51-to-table52.sql must save the last_update time to be committed for both tables, including deleted_guids:

MERGE s61.sync_last_updates AS t
USING (SELECT 'table51' AS [source], 'table52' AS [target], COALESCE(MAX(last_update), 0) AS last_update FROM s61.table51) AS s ([source], [target], last_update)
ON (t.[source] = s.[source] AND t.[target] = s.[target])
WHEN MATCHED AND t.last_update = t.prev_update THEN
    UPDATE SET last_update = s.last_update
WHEN NOT MATCHED THEN
    INSERT ([source], [target], last_update) VALUES (s.[source], s.[target], s.last_update);

MERGE s61.sync_last_updates AS t
USING (SELECT 'deleted_guids' AS [source], 'table52' AS [target], COALESCE(MAX(last_update), 0) AS last_update FROM s61.deleted_guids) AS s ([source], [target], last_update)
ON (t.[source] = s.[source] AND t.[target] = s.[target])
WHEN MATCHED AND t.last_update = t.prev_update THEN
    UPDATE SET last_update = s.last_update
WHEN NOT MATCHED THEN
    INSERT ([source], [target], last_update) VALUES (s.[source], s.[target], s.last_update);

The final command must commit the last time for both tables also. The commit-sync-table51-to-table52.sql has the following code:

UPDATE s61.sync_last_updates SET prev_update = last_update WHERE [source] = 'table51' AND [target] = 'table52';

UPDATE s61.sync_last_updates SET prev_update = last_update WHERE [source] = 'deleted_guids' AND [target] = 'table52';

And, here is the code of the select-table51-to-table52-deleted.sql used to select identifiers deleted in the table51 after the last synchronization:

SELECT
    id
FROM
    s61.deleted_guids
WHERE
    last_update > COALESCE((
        SELECT
            prev_update
        FROM
            s61.sync_last_updates
        WHERE
            [source] = 'deleted_guids' AND [target] = 'table52'
        ), 0)
    AND [source] = 'table51'

The batch uses this file in the following command:

gsqlcmd import db2 db1::select-table51-to-table52-deleted.sql /table=s61.table52 /delete

This command uses the /delete option instead of the /merge.

Complete Synchronization Across Multiple Editable Databases with Integer Key Columns

The most complicated case is synchronizing data across multiple editable databases that contain tables with integer key columns.

For example, you have a disconnected desktop application for your salespersons that should be synced with a central database.

In this case, you cannot guarantee unique identifiers across databases.

The possible solution is to add columns for the source table and its identifiers.

For example:

CREATE TABLE [s61].[table71] (
      [id] int IDENTITY(1,1) NOT NULL
    , [name] nvarchar(50) NOT NULL
    , [source_table] nvarchar(128) NULL
    , [source_row_id] int NULL
    , [last_update] datetime NULL
    , CONSTRAINT [PK_table71] PRIMARY KEY ([id])
);

The source_table column contains a table name where the row was initially added. The source_row_id column contains the row id.

The source_table column may include any additional data like a database name or a salesperson application computer name.

The insert trigger must update these fields automatically. Here is a sample:

CREATE TRIGGER [s61].[trigger_table71_after_insert]
    ON [s61].[table71] AFTER INSERT
AS
BEGIN
SET NOCOUNT ON

IF USER_NAME() IN ('sample61_user2')
    RETURN

UPDATE s61.table71
SET
    source_table = 'table71'
    , source_row_id = t.id
    , last_update = GETUTCDATE()
FROM
    s61.table71 t
    INNER JOIN inserted ON inserted.id = t.id
END;

The trigger updates the source_row_id column with the initial id generated by a database.

To prevent changes of the source_table, source_row_id, and last_update values, the trigger uses the following code:

IF USER_NAME() IN ('sample61_user2')
    RETURN

The sample uses the sample61_user2 username in the synchronization operations. So, it imports data as is.

For other users, the triggers update the fields regardless of user values.

The update trigger also must be modified to keep initial sync values of the row regardless of possible changes by a user:

CREATE TRIGGER [s61].[trigger_table71_after_update]
    ON s61.table71 AFTER UPDATE
AS
BEGIN
SET NOCOUNT ON

IF @@NESTLEVEL > 1
    RETURN;

IF USER_NAME() IN ('sample61_user2')
    RETURN

UPDATE s61.table71
SET
    source_table = deleted.source_table
    , source_row_id = deleted.source_row_id
    , last_update = GETUTCDATE()
FROM
    s61.table71 t
    INNER JOIN deleted ON deleted.id = t.id
END;

This trigger also skips updates in the synchronization operations executed under the sample61_user2 username that allows updating the last_update column values of the imported rows.

Deleting rows also is a challenge as we need to delete records in other databases using source_table and source_row_id values.

Here is a sample of the table that keeps identifiers of the deleted rows:

CREATE TABLE [s61].[deleted_ints] (
      [id] int NOT NULL
    , [source] nvarchar(128) NOT NULL
    , [source_table] nvarchar(128) NOT NULL
    , [source_row_id] int NOT NULL
    , [last_update] datetime NOT NULL
    , CONSTRAINT [PK_deleted_ints] PRIMARY KEY ([id], [source])
);

The table uses id and source fields as a primary key to store values for any table in a database.

Also, it contains the source_table and source_row_id columns that hold values used in the synchronization.

The last_update column contains the deletion time used to synchronize the deletion once.

The delete trigger has the following code:

CREATE TRIGGER [s61].[trigger_table71_after_delete]
    ON s61.table71 AFTER DELETE
AS
BEGIN
SET NOCOUNT ON

INSERT INTO deleted_ints (id, [source], source_table, source_row_id, last_update)
SELECT
    deleted.id
    , 'table71' AS [source]
    , source_table
    , source_row_id
    , GETUTCDATE() AS last_update
FROM
    deleted
END;

The synchronization batch file has the following code:

gsqlcmd exec db1 update-sync-table71-to-table72.sql

if errorlevel 1 goto :EOF

gsqlcmd import db2 db1::select-table71-to-table72.sql /table=s61.table72 /merge /keys=source_table,source_row_id

if errorlevel 1 goto :EOF

gsqlcmd import db2 db1::select-table71-to-table72-deleted.sql /table=s61.table72 /delete /keys=source_table,source_row_id

if errorlevel 1 goto :EOF

gsqlcmd exec db1 commit-sync-table71-to-table72.sql

The commands are similar to the discussed in the previous topic.

However, the import commands use the /keys=source_table,source_row_id option to use the source_table and source_row_id columns in the JOIN and WHERE conditions.

You have to create indexes for these columns to improve the performance.

The update-sync-table71-to-table72.sql and commit-sync-table71-to-table72.sql files are completely the same as in the previous topic except for the table names.

The select-table71-to-table72.sql is similar to the previous one except for the SELECT clause:

SELECT
    name, source_table, source_row_id, last_update

It does not contain the local id column but contains the global source_table and source_row_id.

The select-table71-to-table72-deleted.sql contains changed SELECT and FROM clauses:

SELECT
    source_table, source_row_id
FROM
    s61.deleted_ints
WHERE
    last_update > COALESCE((
        SELECT
            prev_update
        FROM
            s61.sync_last_updates
        WHERE
            [source] = 'deleted_ints' AND [target] = 'table72'
        ), 0)
    AND [source] = 'table71'

Improving and Unifying Solutions

The solutions above use batch files and SQL script files with hard-coded table names.

We did this for educational purposes to increase the complexity step by step and show the difference of the approaches.

Of course, you may place the code into stored procedures and unify the synchronization batch file.

For example, the batch file may look like this:

@echo off

set schema=s61
set source=%1
set target=%2

gsqlcmd exec   db1      "exec %schema%.usp_update_last_updates @source='%source%', @target='%target%'"

if errorlevel 1 goto :EOF

gsqlcmd import db2 "db1::exec %schema%.usp_select_upsert_rows  @source='%source%', @target='%target%'" /table=%schema%.%source% /merge  /keys=source_table,source_row_id

if errorlevel 1 goto :EOF

gsqlcmd import db2 "db1::exec %schema%.usp_select_delete_rows  @source='%source%', @target='%target%'" /table=%schema%.%source% /delete /keys=source_table,source_row_id

if errorlevel 1 goto :EOF

gsqlcmd exec   db1      "exec %schema%.usp_commit_last_updates @source='%source%', @target='%target%'"

All the procedures accept the source and target tables as parameters. So, you may call a single unified batch file with parameters to sync tables.

The procedures generate SQL codes to select the actual column names of the source tables.

You may find the source code in the Examples\Sync Databases folder of the gsqlcmd download package.

Final Notes

Any sync process must include at least three steps:

Getting data from the source
Producing SQL commands to change the target
Executing the SQL commands

gsqlcmd allows getting data from databases, files, and web resources.

Also, it allows creating and executing commands for database servers, including SQL Server, Oracle, DB2, MySQL, PostgreSQL, NuoDB, SQLite, and SQLCE.

So, you may synchronize data across any supported databases.

You may find the source code in the Examples\Sync Databases folder of the downloaded package.

Also, you may try any batch file with a demo sample hosted in Microsoft Azure SQL Database.