faq.md

FAQ

Is Gula a replacement for systemd?

No. systemd is a full system init plus a service manager plus a journal plus a dozen other things. Gula is a single-purpose process supervisor that you typically run under systemd, not in place of it. A common deployment is a single systemd unit that launches gula run --config /etc/gula/gula.yaml, and that unit owns the application stack.

Does Gula understand cgroups?

Not yet. Memory limits are enforced from user space by polling RSS via sysinfo and reacting via the configured memory_threshold_action. This is sufficient for "kill the misbehaving Python service" but it is not the same as a cgroup memory.max hard limit.

A cgroup v2 backend is a possible future addition, but it is not part of the 0.1.x behavior.

Gula does work correctly when it is launched inside an externally managed CPU boundary, such as a systemd service or slice with AllowedCPUs=3-6. In that mode, gula validate checks process affinity values against the inherited Cpus_allowed_list, and workers can be pinned to individual CPUs inside that range. Gula does not yet create or manage those cgroups itself.

Why Linux-only? Will macOS or Windows be supported?

Gula relies on /proc, pidfd_open(2), ioprio_set(2), and POSIX process groups with negative-PID kill(2) semantics. macOS lacks /proc and pidfd; Windows lacks all of the above. Porting would mean rewriting most of src/utils/ and src/process/execution/ against per-OS abstractions, which is out of scope for the 0.1.x line.

How does Gula compare to ROS 2 launch?

ROS 2 launch describes how to start a graph of nodes. It does not monitor them, restart them, or enforce resource limits. Gula complements launch: use launch to build the command lines (with parameter files, remapping, etc.), then put those commands in a gula.yaml so Gula owns the lifecycle. The optional iceoryx2 heartbeat feature is specifically designed to detect when a ROS-style middleware node has wedged.

Why YAML? I prefer TOML / JSON / KDL

YAML is the only supported config format in 0.1.x. It works well for nested, human-edited files with comments, and it maps directly onto the serde configuration structs used by gula validate.

Other formats would need explicit parser support, schema updates, tests, and docs before they could be treated as supported.

Is there a supported Rust embedding API?

Not in the 0.1.x line.

Gula is intentionally CLI-first right now. The supported user-facing interface is the gula binary plus YAML configuration. The 0.1.x package does not expose a Rust library target.

If you need programmatic integration today, prefer one of these approaches:

1. Generate YAML and call gula validate / gula run as a subprocess.
2. Vendor against the current source and accept that internals may change.

There is no embedder hook you need to wire up for scheduling. Commands are run as sh -c <command>, and when affinity, niceness, or ionice are set, Gula applies those settings inside its own child spawn path immediately before exec.

How is Gula tested?

Gula has an extensive Rust test suite split between in-source unit tests, integration tests under tests/, and benchmark regression tests under benchmarks/tests/. The integration and example-config tests spawn the actual gula binary against fixtures in config/, templates/, and examples/.

What does "Gula" mean?

The name comes from Gula, the Sumerian goddess associated with healing and restoration.