Skip to content
IntoSQL home
Convert a file

Migration guides

Oracle to SQL Server migration guide

Convert an Oracle export from SQL Developer or SQLcl into a SQL Server script with the same tables, keys, indexes and data, and every rule the converter applies.

View as Markdown

IntoSQL reads the .sql file SQL Developer's Database Export or SQLcl writes, with DBMS_METADATA's table definitions and Insert into statements, and turns it into one SQL Server script with the same tables, columns, keys, indexes, foreign keys, defaults, counters and rows. It evaluates what those exports write for values: to_date and to_timestamp with their format, TO_CLOB and HEXTORAW pieces, decimal commas from German or Spanish sessions. Every value SQL Server can't hold as it is, and every object it doesn't carry, is listed in the report. The script loads on SQL Server 2017 and later.

Convert in the browser: Oracle to SQL Server opens the converter set up for this pair.

Export for the best result

In SQL Developer, choose Tools, Database Export, with DDL and the Insert data format, as one file. SQLcl writes the same data with SET SQLFORMAT INSERT. Before exporting, set a date format that keeps the time and the century:

SQL
ALTER SESSION SET NLS_DATE_FORMAT = 'SYYYY-MM-DD HH24:MI:SS';
ALTER SESSION SET NLS_TIMESTAMP_FORMAT = 'SYYYY-MM-DD HH24:MI:SS.FF9';
  • The default format, DD-MON-RR, writes each date without its time and with two digits of the year; the report says so (oracle_date_format) and each date is read as its day, between 1950 and 2049.
  • A .dmp file from exp or expdp is binary. Import it into an Oracle database, Oracle Database Free runs in Docker, and export it with SQL Developer; the converter says so when it gets one (dump_oracle_binary).
  • Every test export comes from Oracle 21c and 23ai, in American, German and ISO sessions, and is compared with its source cell by cell.

Convert

In the browser, open Oracle to SQL Server or the converter on the home page, drop the export and choose SQL Server. It reads the file from your disk in pieces, twice, and writes the script as it goes, so the file is never uploaded and its size isn't limited by memory. The desktop app does the same with the file on disk.

With the command line:

Terminal
intosql inspect shop.sql
intosql convert shop.sql --to sql-server --output shop.sql_server.sql --report shop.report.json

inspect lists the tables, their rows and what won't be converted, without writing anything. convert writes the script and, with --report, the report as JSON.

Through the API, an export converts like any other file: open a conversion with its name and "output": {"dialect": "sql-server"}.

Load into SQL Server

Terminal
sqlcmd -S localhost -Q "CREATE DATABASE shop"
sqlcmd -b -f 65001 -S localhost -d shop -i shop.sql_server.sql

-b stops at the first error and -f 65001 reads the script as UTF-8. Text compares exactly, as in Oracle (Latin1_General_100_BIN2). DATE becomes datetime2(0), keeping its time, and TIMESTAMP WITH TIME ZONE becomes datetimeoffset. Identity columns become IDENTITY, sequences become SQL Server sequences, and defaults that draw from a sequence keep drawing from it. Foreign keys are added WITH NOCHECK and then checked.

What isn't converted

Views, packages, procedures, functions, triggers other than one that fills a column from a sequence, types and synonyms are written in PL/SQL. They are listed in the report by kind and name, never dropped silently, so you can recreate them in SQL Server.

Data types

Oracle to SQL Server data types lists each type that changes in the script: what it becomes, why, and the report line.

Every other rule

Generated from the engine, where each rule is decided.

In Oracle In SQL Server Why Report line
collations that ignore case (BINARY_CI) the same collation on each column Columns compare and sort as they did, and unique keys keep the same rows apart. None
CREATE SEQUENCE CREATE SEQUENCE with the same options The export writes where a sequence has reached as its START WITH, so it continues from there. None
virtual columns (GENERATED ALWAYS AS ... VIRTUAL) computed columns SQL Server computes them from the same formula; the values the export wrote for them are left out. None
virtual columns with functions IntoSQL doesn't write, and the || operator a plain column A formula that can't be written for SQL Server leaves a plain column and a report line; the values the export wrote for virtual columns are left out, and SQL Server has none to compute. generated_not_translated
defaults with functions IntoSQL doesn't write (SYS_GUID(), ||) no default A default that can't be written is left out and reported. default_not_translated
CHECK constraints translated The same checks, tested as each row loads. None
CHECK constraints with functions IntoSQL doesn't write not created A check that can't be written is left out and reported. check_skipped
CHECK constraints that are DISABLE or ENABLE NOVALIDATE not created Oracle doesn't apply these checks to the rows already there, so the rows may break them. check_not_enforced
function-based, bitmap and domain indexes not created IntoSQL writes b-tree indexes on whole columns; create the others again after loading. index_skipped
foreign keys after the data WITH NOCHECK, then checked with a warning on orphans The key is created after the rows and checked last; rows without a parent leave it unchecked with a warning instead of stopping the load. None
ON DELETE CASCADE that cascades in a cycle or along a second path NO ACTION SQL Server refuses cascades that loop or take a second path, so the key is created without that action and the report names it. foreign_key_action_dropped
foreign keys to tables the export doesn't create, and DISABLE or NOVALIDATE ones not created A key can't name a table that isn't there, and one Oracle doesn't check may have rows without a parent. foreign_key_skipped
partitioned tables one table with every partition's rows The rows are the same; partition the table again in SQL Server if it needs it. partitions_dropped
NOT NULL columns with values that become NULL NOT NULL dropped A column whose values had to become NULL can't keep NOT NULL, or the load would stop. not_null_relaxed
views, synonyms, types, packages, procedures, functions and other triggers listed in the report Their code is written in PL/SQL, so they are listed for you to recreate. not_converted_objects
exports without CREATE TABLE (data only) types read from the values A data-only export has only rows, so the types come from the values. no_schema
expressions in INSERT values other than the ones exports write NULL A converter that doesn't run a database can't evaluate them; to_date, to_timestamp, TO_CLOB, HEXTORAW and the others exports write are read. expression_value
text decoded twice (ñ) copied as it is The script holds exactly these characters, so they are copied; the report points them out. mojibake_suspected