Performance specifications
These are the Executor's performance specifications: the design targets the Executor is built to, each with the exact conditions it is stated for. They are targets from the Executor's requirements, not measured results.
The specifications
| Area | Design target | Conditions |
|---|---|---|
| Supervisor binary overhead | Less than 3.0 MB (core variant); less than 6.0 MB (extended variant) | Stripped static Rust supervisor overhead per wrapped executable. The extended variant adds the message brokers, the webhook, the auto-scaler and the load balancer. |
| Warm startup overhead | Less than 15 ms | Wrapped native Go or Rust binaries, and warm-cached Python or JVM workloads. |
| Cold startup overhead | Under 350 ms | The first run, with cold runtime-cache extraction. |
| Supervisor footprint | Less than 6 MB RSS and less than 0.1% CPU | At a one-minute HTTP exit-event polling interval, for the core variant with no load-balancer connections. |
| Immediate stop latency | Under 5 ms | From exit-event evaluation to complete process-tree termination in IMMEDIATE mode, across 50 concurrent child worker processes. |
| Strict CPU cap accuracy | Never more than 2% over the limit | CPU usage of the capped tree over any one-second window, with kernel enforcement. |
| Strict memory cap accuracy | Never over the limit | Memory usage of the capped tree, with kernel enforcement. Percentage limits resolve exactly: 10% of 10 cores is 1.0 core. |
| Load-balancer latency | Less than 1 ms added median latency | Per proxied connection on loopback. |
| Load-balancer memory | Less than 20 KB per idle connection | Supervisor memory per idle proxied connection in tcp mode. |
| Scale-out reaction | New replica launched within 1 s; traffic within 100 ms | Launch counted from the end of the scale-out window; traffic counted from the replica passing its readiness check. |
| Integrity | Any byte change fails verification | Any modification to the supervisor, the embedded policy or the bundled runtime archive fails BLAKE3 and Ed25519 verification before the guest starts. |
How the design meets the targets
- Small binaries. The supervisor is compiled with
lto = "fat",codegen-units = 1andpanic = "abort"againstmuslon Linux and the static CRT on Windows. Unlike runtimes that bundle a garbage collector, it adds only the supervisor itself, and it runs insideFROM scratchimages. The smallest variant that satisfies the manifest is always chosen; see Supervisor variants. - Fast starts. Interpreted runtimes are extracted once into a content-addressed, read-only runtime cache and reused on every later start.
- A quiet supervisor. A single-threaded asynchronous event loop multiplexes polling, monotonic timers, the checkpoint socket and the child wait queues (
pidfdon Linux, I/O completion ports on Windows,kqueueon macOS), with negligible idle CPU. - Fast, complete stops. The whole process tree is terminated through one kernel object:
cgroup.killon Linux,TerminateJobObjecton Windows. - Accurate caps. Caps are enforced by kernel primitives on the cgroup or Job Object that holds the whole tree:
cpu.maxandmemory.maxwith swap disabled on Linux, Job Object limits on Windows. - Lightweight balancing. The load balancer runs on the supervisor's event loop, not as a separate process.
Related pages
- CPU and memory capping for cap enforcement on each platform.
- Auto-scaling and load balancer for the scaling model.
- Stopping and exit codes for the stop pipeline.