Ada Spawn Manager

Spawn Manager lets a multitasking Ada program execute child processes without calling fork(2) from the application's tasking runtime. The application starts a pool of dedicated manager processes early; later requests cross a Unix-domain socket, and only a selected manager performs fork and execve.

Motivation

If the parent is using tasking, and needs to spawn subprocesses at arbitrary times, one technique is for the parent to spawn (very early) a particular spawn-manager subprocess whose job is to spawn other processes. The spawn-manager avoids tasking. The parent sends messages to the spawn-manager requesting it to spawn processes, using whatever inter-process communication mechanism you like, such as sockets.

— GNAT Compiler Components, System.OS_Lib specification

The project provides two request styles:

• a compatible command-string API that executes /bin/bash -o pipefail -c COMMAND; and
• a structured API with an explicit executable, ordered argument vector, complete replacement environment, working directory, independent standard output and error destinations, and timeout.

Both styles use the same bounded, versioned protocol and the same supervised execution core. There is no intermediate spawn_wrapper process.

Status and platform

This revision defines release 0.2.0. It introduces structured execution while retaining the existing shell API. Its contents are collected in the changelog; accepted later work is tracked in the work queue.

The current execution core is Linux-specific. It relies on Unix-domain sockets, process groups, prctl(2), /proc and POSIX process and signal operations. Version 1 sets no_new_privs for every child, so shell and structured requests cannot gain privileges through set-user-ID, set-group-ID or file-capability metadata during execve.

Execution model

Spawn.Pool.Init starts one or more long-lived managers from an explicit absolute path. Concurrent callers lease different managers; one manager handles one request at a time. Each request receives its own process group, and the manager owns timeout handling, termination and reaping.

The shell API inherits the manager environment and preserves Bash syntax. The structured API performs no shell parsing or quoting: arguments and environment entries arrive as exact protocol fields. It returns distinct results for a normal exit, signal termination, timeout, spawn failure, rejected request and protocol failure.

Pool lifecycle failures are fail-closed. If setup, transport, reset or process supervision leaves a manager's state uncertain, the complete pool is poisoned instead of making that manager available to another caller.

For the exact contracts, see:

• Protocol version 1 for the shell and structured wire formats, bounds and result encoding.
• Lifecycle version 1 for process groups, signals, timeouts, reaping, no_new_privs, cgroup callbacks and failure handling.
• Spawn.Pool for the public Ada API.

Anet boundary

Anet remains the socket-lifecycle abstraction. It creates and owns the Unix stream sockets, validates addresses, connects clients, accepts manager connections, switches descriptors to nonblocking mode and closes them. Spawn Manager also uses Anet's operating-system constants and filesystem utilities.

Frame serialization and exact frame transfer deliberately belong to Spawn Manager. Version 1 does not use Anet.Streams, Ada record streaming or the generic Anet.Sockets.Send and Receive operations for protocol data. Those socket operations each represent one system call, whereas this protocol must resume partial reads and writes, distinguish first-byte and completion deadlines, handle interruption and backpressure, reject trailing bytes and classify peer closure precisely. Spawn.Transport therefore operates on the descriptor owned by the Anet socket and implements those protocol-specific rules with poll(2), recv(2) and send(2).

This boundary is intentional rather than a replacement for Anet. A future Anet primitive offering the same exact, deadline-bounded nonblocking transfer semantics could replace the generic I/O part without changing the version 1 wire format. Frame layout, validation and protocol failure classification would remain Spawn Manager responsibilities.

Requirements

Building Spawn Manager requires:

• a GNAT Ada compiler and gnatmake;
• a C11 compiler and POSIX threads;
• Anet; and
• GNU Make.

The tests additionally require Ahven, Bash and standard Linux process utilities. Coverage needs lcov and genhtml. Building the optional HTML documentation requires Pandoc.

Build and test

Build the static Ada library and the manager executable:

make -j8

Run the complete Ada, protocol, transport, direct C and integration test suite. The test runner intentionally runs without Make-level parallelism so its process and log ordering remains deterministic:

make tests

Run the comparative execution benchmarks:

make perf

The performance target covers the legacy shell path, structured requests, small and larger output streams, manager startup and cleanup, and parallel throughput. Its wall-clock results are meaningful as paired measurements on the same host, not as portable absolute limits.

Useful build controls are:

• BUILD_TYPE=debug to enable assertions and additional warnings in the manager;
• ADAFLAGS="..." to append Ada compiler switches; and
• CC=... to select the compiler used by the direct C test fixtures.

Generate the HTML documentation under doc/html/ with:

make doc

Installation

Install the library, GNAT project file and manager executable with:

make PREFIX=/usr/local install

The default prefix is $HOME/ada. The manager is installed as $PREFIX/spawn_manager; the static library, Ada sources and project file are installed below the prefix's lib, include and lib/gnat directories. Installation removes a stale $PREFIX/spawn_wrapper from earlier versions.

Versioning

Release versions use annotated vMAJOR.MINOR.PATCH Git tags. The generated Spawn.Version value comes from git describe, with the leading v removed.

The historical 0.1.x tags reconstruct the distinct Spawn Manager states selected by Abuild's maintained Gitlink history. A Gitlink update that changed only tests or documentation does not create a component version. Beginning with 0.2.0, releases follow normal semantic versioning: compatible fixes increment the patch version, new capabilities increment the minor version, and an incompatible stable public contract requires a major version change.

Source repository and history

The canonical upstream is github.com/mweidle73/spawn-manager. The project was initially developed at Codelabs; that former repository location is retained as a redirect and historical point of reference.

master is the protected development and release branch. The former abuild integration branch and its abuild-gh GitHub CI overlay are frozen at the 0.2.0 transition and retained so their Gitlinks, reviews and merge history remain inspectable.

The imported add_remove_managers and cgroups branches preserve unmerged Codelabs development. archive/cgroups-caller-placement-wip preserves the later caller-side cgroup-placement prototype which led to the supported Pid_Setup and Pid_Reset callbacks. These branches are historical evidence, not maintained APIs or alternatives to master.

Authors and licence

Copyright (C) 2011-2026 secunet Security Networks AG
Copyright (C) 2012-2016 Reto Buerki reet@codelabs.ch

Authors and contributors include Reto Buerki, Adrian-Ken Rueegsegger, Matthias Weidle and Markus Vogt. See AUTHORS.

Spawn Manager is distributed under the GNU General Public License, version 2 or later, with the GNAT Modified GPL linking exception stated in the source headers. See COPYING for the GPL text.