BerandaComputers and TechnologyShow HN: Python3 bindings for writing Oracle Tuxedo clients and servers

Show HN: Python3 bindings for writing Oracle Tuxedo clients and servers

Python3 bindings for writing Oracle Tuxedo clients and servers.


I’m a fan of the way tuxmodule enables you to interact with Oracle Tuxedo. Unfortunately, it’s out-dated and somehow limited. So I cloned tuxmodule and started to clean up compiler warnings and work on some features I had in mind:

  • A multi-threaded server
  • Support nested FML32 buffers and more types
  • Support newest Oracle Tuxedo features like tpadvertisex() and tpappthrinit()
  • Receive response even when the service returns TPFAIL (instead of exception)

But I realized that’s too much of C for me, so I decided to write my own Python module for Oracle Tuxedo in C++ and pybind11 focusing on the parts I find most important first.

Windows runtime requirements

On Windows, the Visual C++ redistributable packages are a runtime requirement for this project. It can be found here.

I have successfully built the module with Python 3.7.7.

Python 3.8.3 fails to import the module similar to but I have no solution yet.

ImportError:  DLL load failed while importing tuxedo:  The specified module could not be found.

Alternatives to Oracle Tuxedo

Tuxedo-Python can also be used with Open Source alternative to Oracle Tuxedo called Fuxedo. Just export TUXDIR pointing to the folder where Fuxedo is installed everything should work.

All demo code provided with the module works with both Oracle Tuxedo and Fuxedo and you can avoid vendor lock-in by using Python and Tuxedo-Python module.


tuxedo module supports only STRING and FML32 buffer types at the moment.

STRING is mapped to/from Python str type.

FML32 is mapped to/from Python dict type with field names (str) as keys and lists (list) of different types (int, str, float or dict) as values. dict to FML32 conversion also treats types int, str, float or dict as lists with a single element:

{'TA_CLASS':  'Single value'}

converted to FML32 and then back to dict becomes

{'TA_CLASS':  ['Single value']}

All XATMI functions that take buffer and length arguments in C take only buffer argument in Python.

Calling a service

tuxedo.tpcall() and tuxedo.tpgetrply() functions return a tuple with 3 elements or throw an exception when no data is received. This is the part I believe tuxmodule got wrong: a service may return a response
both when it succeeds (TPSUCCESS) and fails (TPFAIL) and often the failure response contains some important information.

  • tpurcode (the second argument to tpreturn)
  • data buffer
rval, rcode, data = t.tpcall('.TMIB', {'TA_CLASS':  'T_SVCGRP', 'TA_OPERATION':  'GET'})
if rval == 0: 
  # Service returned TPSUCCESS
  # rval == tuxedo.TPESVCFAIL
  # Service returned TPFAIL

Writing servers

Tuxedo servers are written as Python classes. tpsvrinit method of object will be called when Tuxedo calls tpsvrinit(3c) function and it must return 0 on success or -1 on error. A common task for tpsvrinit is to advertise services the server provides by calling tuxedo.tpadvertise() with a service name. A method with the same name must exist. tpsvrdone, tpsvrthrinit and tpsvrthrdone will be called when Tuxedo calls corresponding functions. All of these 4 methods are optional and tuxedo module always calls tpopen() and tpclose() functions before calling user-supplied methods.

Each service method receives a single argument with incoming buffer and service must end with either call to tuxedo.tpreturn() or tuxedo.tpforward(). Unlike in C tuxedo.tpreturn() and tuxedo.tpforward() do not perform longjmp but set up arguments for those calls once service method will return. Following two code fragments are equivalent but I believe the first one is less error-prone.

def ECHO(self, args): 
    return t.tpreturn(t.TPSUCCESS, 0, args)
def ECHO(self, args): 
    t.tpreturn(t.TPSUCCESS, 0, args)

After that must be called with an instance of the class and command-line arguments to start Tuxedo server’s main loop.

#!/usr/bin/env python3
import sys
import tuxedo as t

class Server: 
    def tpsvrinit(self, args): 
        return 0

    def tpsvrthrinit(self, args): 
        return 0

    def tpsvrthrdone(self): 

    def tpsvrdone(self): 

    def ECHO(self, args): 
        return t.tpreturn(t.TPSUCCESS, 0, args)

if __name__ == '__main__':, sys.argv)


To use Python code as Tuxedo server the file itself must be executable (chmod +x *.py) and it must contain shebang line with Python:

#!/usr/bin/env python3

After that you can use the *.py file as server executable in UBBCONFIG:


Writing clients

Nothing special is needed to implement Tuxedo clients, just import the module and start calling XATMI functions.

#!/usr/bin/env python3
import sys
import tuxedo as t

rval, rcode, data = t.tpcall('.TMIB', {'TA_CLASS':  'T_SVCGRP', 'TA_OPERATION':  'GET'})

Using Oracle Database

You can access Oracle database with cx_Oracle library and local transactions by just following the documentation of cx_Oracle.

If you want a server written in Python to participate in the global transaction first specify a resource manager name to use (similar to buidserver). tuxedo module currently supports:

  • NONE default “null” resource manager
  • Oracle_XA for Oracle Database, sys.argv, 'Oracle_XA')

After that you should create a database connection in tpsvrinit by using tuxedo.xaoSvcCtx() function:

def tpsvrinit(self, args): 
    self.db = cx_Oracle.connect(handle=t.xaoSvcCtx())

That is the only difference from standard cx_Oracle use case. Here is a complete example for a single-threaded server:

#!/usr/bin/env python3

import sys
import tuxedo as t
import cx_Oracle

class Server: 
    def tpsvrinit(self, args): 
        t.userlog('Server startup')
        self.db = cx_Oracle.connect(handle=t.xaoSvcCtx())
        return 0

    def DB(self, args): 
        dbc = self.db.cursor()
        dbc.execute('insert into pymsg(msg) values (:1)', ['Hello from python'])
        return t.tpreturn(t.TPSUCCESS, 0, args)

if __name__ == '__main__':, sys.argv, 'Oracle_XA')

For a multi-threaded server new connections for each thread must be created in tpsvrthrinit() (instead of tpsvrinit()) and stored in thread-local storage of threading.local().

Server must belong to a group with Oracle_XA as resource manager, something like this in UBBCONFIG

GROUP2 LMID=tuxapp GRPNO=2 TMSNAME=ORACLETMS OPENINFO="Oracle_XA:Oracle_XA+Objects=true+Acc=P/scott/tiger+SqlNet=ORCL+SesTm=60+LogDir=/tmp+Threads=true"


tpadmcall is made available for application administration even while application is down. It also has no service call overhead compared to calling .TMIB service. The Python function looks and behaves similary to tpcall except rcode (2nd element in result tuple) is always a constant 0.

#!/usr/bin/env python3
import tuxedo as t

rval, _, data = t.tpadmcall({'TA_CLASS':  'T_DOMAIN', 'TA_OPERATION':  'GET'})

Global transactions

Transactions can be started and committed or aborted by using tuxedo.tpbegin(), tuxedo.tpcommit(), tuxedo.tpabort(). These functions take the same arguments as their corresponding C functions.

Buffer export and import

FML32 identifiers

Fname32 and Fldid32 are available to find map from field identifier to name or the other way.

Functions to determine field number and type from identifier:

assert t.Fldtype32(t.Fmkfldid32(t.FLD_STRING, 10)) == t.FLD_STRING
assert t.Fldno32(t.Fmkfldid32(t.FLD_STRING, 10)) == 10


demo/ folder has some proof-of-concept code:

  • Oracle Tuxedo client
  • HTTP+JSON server running inside Oracle Tuxedo server
  • HTTP+XML client running inside Oracle Tuxedo server
  • multi-threaded in-memory cache
  • Access Oracle Database using cx_Oracle module within global transaction
  • Demo of tpimport/tpexport and FML32 identifiers


  • Implementing few more useful APIs

Read More



Please enter your comment!
Please enter your name here

Most Popular

Recent Comments