Skip to main content
Version: 1.7 (Latest)

Release Notes 1.7

Your Db2 data, in Parquet, with nothing to install. FastBCP 1.7 brings IBM Db2 into the ADBC FastDrivers: the embedded ArpeDb2 driver talks to Db2 directly, so the IBM client, its configuration and its DSNs are gone from your pipeline. Parallel exports out of the box, one thread per data partition on partitioned tables, and the same command on Windows and Linux. And because a fast export is only worth it when it is right, 1.7 also makes Db2 and Oracle queries, Db2 DECIMAL values and SQL Server backups come out exactly as your database has them.

0
IBM client to install

No Data Server Driver, no CLI, no DSN. The driver ships inside FastBCP.

7
parallel methods on Db2

Including DBPartition: each thread reads its own data partition.

1
command, Windows and Linux

The same command line on both, straight from Db2 to Parquet.

102 / 102
AdventureWorks tables from backups

Identical to SQL Server, ROW and PAGE compressed, typed XML included.

What's New: At a Glance​

Since
New source--connectiontype adbc_db2: IBM Db2 to Parquet through the embedded ArpeDb2 driver, no IBM client1.7.0
Db2 partitions--parallelmethod DBPartition on Db2 for LUW range-partitioned tables: each thread reads one data partition1.7.0
Db2 and Oracle queriesTheir own SQL rules in --query: CURRENT DATE - 30 DAYS on Db2, SERIAL# and 'C:\temp\' on Oracle1.7.0
Db2 through ODBCExact DECIMAL values with IBM's ODBC driver on a Windows with a decimal comma1.7.0
SQL Server backupsLarge PAGE dictionaries and typed xml read, and xml written as SQL Server writes it1.7.0
ParametersCase-insensitive values on the command line and in YAML; read this before upgrading1.7.0
ADBC FastDriversEvery driver updated1.7.0

IBM Db2 to Parquet, No IBM Client — --connectiontype adbc_db2​

Getting data out of Db2 used to start with a chore: install IBM's ODBC / CLI driver on every machine that runs an export, configure it, declare DSNs, keep them in sync across servers, containers and CI runners. Not anymore. With --connectiontype adbc_db2, FastBCP reads Db2 through the embedded ArpeDb2 ADBC driver, which speaks Db2's native protocol and ships inside FastBCP. Unzip, run, get Parquet.

Three parameters and you are connected:

  • --server is host[:port], port 50000 by default.
  • --database is the Db2 database name: it is required.
  • --user and --password, or a --sourceconnectstring (TLS on the server's SSL port, SSL Mode=verify-full;SSL Root Cert=...).
.\FastBCP.exe `
--connectiontype "adbc_db2" `
--server "db2server.company.corp:50000" `
--database "SAMPLE" `
--user "db2user" `
--password "password" `
--sourceschema "DB2ADMIN" `
--sourcetable "ORDERS" `
--directory "C:\temp" `
--fileoutput "orders.parquet" `
--parallelmethod "Ntile" `
--distributekeycolumn "ORDER_ID" `
--paralleldegree 8
TargetsDb2 for LUW (Linux, Unix, Windows); Db2 for i for queries. Db2 for z/OS is not supported yet
Parallel methodsNone, Random, DataDriven, RangeId, Ntile, Timepartition, DBPartition
Data typesPrecision and scale of DECIMAL kept; see the ArpeDb2 data types
Doc Wizardadbc_db2 in the ADBC (High Performance) connectors
Good to know
  • Output is Parquet only, as with every ADBC connector, on Windows x64 and Linux x64 (not Linux ARM64).
  • Trusted authentication (--trusted) is not supported: give --user and --password.
  • --parallelmethod Physical is not available on Db2: use Ntile, RangeId, Random or DBPartition.

See also: Connection Parameters, DB2 Examples

Partitioned Tables: One Thread per Partition​

Your largest Db2 tables are usually range partitioned — by month, by year, by region. FastBCP 1.7 puts that design to work: with --parallelmethod DBPartition, each thread takes one data partition, and Db2 reads only that partition. No full scan per thread, no distribute key column to pick and tune: the partitioning you already have becomes your parallelism.

./FastBCP \
--connectiontype "adbc_db2" \
--server "db2server.company.corp:50000" \
--database "SAMPLE" \
--user "db2user" \
--password "password" \
--sourceschema "DB2ADMIN" \
--sourcetable "SALES_HISTORY" \
--directory "/data/export/sales_history" \
--fileoutput "sales_history.parquet" \
--parallelmethod "DBPartition" \
--paralleldegree 4
Good to know
  • Partitions not yet visible to queries (attached and not yet validated, being detached) are skipped.
  • A table that is not range partitioned, and Db2 for i, fall back to None with a warning.
  • The output layout follows FASTBCP_DBPARTITION_TARGET_MODE (CHUNK or HIVE), as for the other sources.

Your Db2 and Oracle Queries, Sent as You Wrote Them​

A query that runs in your SQL client should run in FastBCP, unchanged. FastBCP reads your --query before sending it to the database, and 1.7 teaches it the SQL rules of Db2 and Oracle: the queries your database accepts are no longer refused, nor altered on the way.

SourceNow works
Db2 (adbc_db2, and odbc on a Db2 server)Special registers in several words: CURRENT DATE, CURRENT TIMESTAMP(6), CURRENT SCHEMA, …; labeled durations: CURRENT DATE - 30 DAYS, ORDER_DATE + 1 MONTH. These queries failed to parse up to 1.6
Oracle (oraodp, adbc_oracle, and odbc on an Oracle server)SELECT SID, SERIAL# FROM V$SESSION keeps SERIAL#; 'C:\temp\' parses; LIKE 'C:\temp\f1%' matches what Oracle matches. The (+) outer join, IGNORE NULLS and T@DBLINK keep working
Oracle queries with # or \

Up to 1.6, the end of a line after # was dropped, and \t in a string became a tab: such a query could return other rows, or none, without an error. Check the exports whose --query contains # or \ after upgrading.

Exact DECIMAL Values, Whatever the Windows Locale​

Amounts are the last place you want a surprise. With --connectiontype odbc and IBM's Db2 ODBC / CLI driver, on a Windows machine whose regional settings use a decimal comma (French, German, …), DECIMAL values were exported multiplied by 10^scale, with no error: 173665.47 became 17366547.

FastBCP 1.7 recognises IBM's driver on a Db2 server and sets it up so that every amount comes out exact, whatever the locale of the machine. Nothing to change in your commands: your own Patch2 settings are kept, and the log shows what was applied. Other ODBC drivers and adbc_db2 were not affected — and with adbc_db2, there is no ODBC driver to worry about at all.

SQL Server Backups: More Tables, Read Right​

Reading tables straight from .bak files, with no SQL Server and no restore, is only worth it if every table comes out right. 1.7 widens what --connectiontype backup_mssql reads (ArpeMSBackup 0.3.1) — tables that failed, or read wrong values, up to 1.6.8:

PAGE compressionTables with many distinct values per page failed with page symbol … out of the dictionary or prefix longer than the anchor value, or read the next column wrong. Fixed
PAGE compressionPages whose values share no common prefix (hashes, codes) failed with corrupt page dictionary, or lost rows. Fixed
Typed xmlxml columns bound to an XML schema collection failed with BINXML token 0xEA is not supported. They are read, their values written as SQL Server returns them (12.5, 2003-12-11Z, 1.0E21)
xml textA carriage return is written 
, and TAB / LF / CR in attribute values 	 / 
 / 
, as SQL Server writes them

Proven, not promised: every table and every column of AdventureWorks2025 and AdventureWorksDW2025 — as is, ROW compressed and PAGE compressed, from uncompressed, MS_XPRESS and ZSTD backups — comes out identical to SQL Server.

Type Parameter Values Your Way​

MSSQL or mssql, Physical or physical: it no longer matters. The values of --connectiontype, --parallelmethod, --parquetcompression, --boolformat, --loglevel and --applicationintent are now case-insensitive, on the command line and in a YAML configuration file — --connectiontype MSSQL, parallelmethod: physical and --parquetcompression ZSTD all work, and FastBCP uses the canonical spelling. And a typo in a YAML file is now caught right away instead of silently doing nothing (see below).

Upgrading to 1.7​

YAML values are now checked

Up to 1.6, a value of a YAML configuration file that is not in the list (parallelmethod: physycal) was silently ignored, and the export did nothing. It now fails with the message of the command line: Value physycal is not allowed for argument m(parallelmethod). Fix the value in the file.

The hanaexport connection type, refused by --connectiontype for a long time, is removed. Use --connectiontype hana.

Get Started with 1.7​

  • New to Db2 exports? Build your first adbc_db2 command in a minute with the Doc Wizard, or copy one from the DB2 Examples.
  • Already on ODBC? For your Parquet exports, switch --connectiontype odbc to --connectiontype adbc_db2, give --server, --database, --user and --password, and the IBM client is no longer needed for them.
  • Using an AI assistant? The FastBCP agent skill and MCP server help it write and check your commands.

Release History​

1.7.0​

New source--connectiontype adbc_db2: IBM Db2 through the embedded ArpeDb2 driver, no IBM client
adbc_db2DBPartition on Db2 for LUW range-partitioned tables
--queryDb2 and Oracle SQL rules
odbcFixed: DECIMAL values multiplied by 10^scale with IBM's Db2 ODBC driver on a Windows with a decimal comma
backup_mssqlFixed: PAGE-compressed tables with large page dictionaries, typed xml; xml escaping as SQL Server
ParametersCase-insensitive values; YAML values checked; hanaexport removed
Doc Wizardadbc_db2
ComponentVersion
ArpeMSSQL ADBC driver0.8.0
ArpePGSQL ADBC driver0.5.2
ArpeOracle ADBC driver0.4.5
ArpeNetezza ADBC driver0.3.9
ArpeTeradata ADBC driver0.3.6
ArpeDb2 ADBC driver0.4.0 (new)
ArpeMSBackup ADBC driver0.3.1
ArpeOraBackup ADBC driver0.3.1
Parquet.Net6.2.0
Copyright © 2026 Architecture & Performance.