The open source alternative to Oracle Tuxedo
Fuxedo is an open source implementation of the X/Open XATMI specification and the Oracle Tuxedo extensions that applications actually use. If you knew the product in an earlier life as BEA Tuxedo or AT&T Tuxedo, it is the same API; the short history explains the naming. It is compatible at three levels: the API (the same C calls, the same FML32 buffers, the same MIB administration classes), the commands (tmloadcf, tmunloadcf, tmboot, tmshutdown, tmadmin, qmadmin, ud32 and friends, same flags and formats), and the System V IPC mechanics (the same message queues, shared memory, and semaphores, used the way Tuxedo uses them). If your application runs on Tuxedo, migration is a recompile away, and no new behavior surprises come with it.
The name is a wordplay on “free” and Tuxedo, and on the fuxedo itself: a one-piece suit that combines the elegance of a tuxedo with the convenience of a jumpsuit. That is roughly the engineering goal too.
API level. tpinit, tpcall, tpacall, tpbegin, FML32 and its boolean expressions, tpadvertise, tpenqueue: the same functions with the same semantics, so application source compiles and behaves the same. Internal details that programs cannot observe, such as buffer indexes, are free to differ.
Command level. The administrator’s toolbox keeps its manuals: tmloadcf loads the configuration, tmboot boots it, tmshutdown stops it, tmadmin inspects the bulletin board, qmadmin runs the queues, and tmunloadcf prints the configuration back. Scripts written for Tuxedo keep working.
System V IPC level. Servers serve from blocking msgrcv() on System V queues, MSSQ sets share one queue among several servers, the bulletin board lives in shared memory, and locks are semaphores. The mechanics you have tuned kernels and runbooks around are the mechanics you get.
No surprises, one improvement. Where a behavior is observable, it matches; where internals can differ without being seen, they may (queue storage is folders and files, wire bytes inside the IPC queues are our own). The one deliberate deviation: calls do not degrade the way a full queue under a one-second retry would let them, so a busy system keeps answering.
The configuration is a file, not a database. TUXCONFIG is a small header in front of the plain UBBCONFIG text. tmloadcf parses and validates the text, then builds the bulletin board in shared memory. tmunloadcf prints the text back. Administration goes through the same MIB classes Tuxedo documents (T_DOMAIN, T_GROUP, T_SERVER), and tpadmcall works on the configuration file directly, which is how the configuration can be created programmatically from scratch, no running system needed.
FML32, familiar down to the field ids. The API matches Tuxedo, field numbers and types follow the same tables, and what crosses process or system boundaries stays compatible: tpimport/tpexport payloads and compiled boolean expressions (Fboolco32 trees, memory prefixes and all) work on both sides. Nested sub-buffers included.
Queues you can ls. The /Q component stores a queue device as a directory, queues as folders, and messages as files, committed with an atomic rename. Everything is persistent, everything is visible, and qmadmin speaks the same command language as the original.
Same tests, two implementations. The compatibility suite runs unmodified against both Fuxedo and Oracle Tuxedo, so behavior matches the documentation on either side, and stays matched as the project grows.
Scriptable from Python. The tuxedo-python bindings run unchanged on top: servers, clients, the MIB administration calls, and the demo applications from the book Modernizing Oracle Tuxedo Applications with Python. Write a Tuxedo server as a Python class, advertise it with tpadvertise, boot it with tmboot.
Python bindings. tuxedo-python wraps the whole XATMI API for Python 3: tpcall, tpacall, tpgetrply, tpbegin, tpenqueue, tpdequeue, tpadvertise, tpadmcall, FML32 buffers as dicts, /Q queues, and multithreaded servers. The same module runs against Oracle Tuxedo 12.x and 22.1, so code written on it is portable across both. It is the basis of the book Modernizing Oracle Tuxedo Applications with Python and its demonstration applications.
Training. The author of the bindings and the book runs hands-on Oracle Tuxedo training for developers and administrators: two or three days, from building and operating a domain to C++ internals and debugging, taught from system-call level mechanics rather than manual recitals. Day three can be tailored to your own systems, Fuxedo migration questions included.
Monitoring. Because the queues are System V, the kernel can be asked what /proc and ipcs will not tell: eBPF and /proc based queue monitoring, including the caller to callee wait view and ipcqmon, works against both Fuxedo and Oracle Tuxedo.
Fuxedo is built and maintained by Aivars Kalvans, author of the Packt book Modernizing Oracle Tuxedo Applications with Python and of the tuxedo-python bindings, technical reviewer of Packt's Getting Started with Oracle Tuxedo, with close to two decades of running C++ Tuxedo applications in production, and years on the Oracle Tuxedo customer advisory board. The compatibility suite runs unmodified against Oracle Tuxedo 12.2.2 and 22.1 in continuous integration, on every commit, so the API claims stay verified rather than aspirational.
The honest status, the gap list, and what is next live on the status page.
Fuxedo itself is free software, MIT licensed, commercial use included, no copyleft and no per-CPU counting. Around it there is a shelf for teams that want a human on call:
For any of it, write to hello@fuxedo.io.
Source, build instructions, and the current status live in the GitHub repository. For bugs and feature requests, open an issue. For anything else, write to hello@fuxedo.io.