Crate mesofact 

mesofact — the Rust-native web framework facade.

This is the one crate consumers depend on. Subsystems are selected by feature rather than by picking crate names out of a 7-crate workspace (bevy-style). Two distribution strategies, one crate:

• Prebuilt binary / container — we build the mesofact bin with the deploy preset, arch-native per libc, and kamaji fetches it. The consumer’s project compiles zero Rust, only their TypeScript — the Node.js model.
• Crate dependency — the same crate from crates.io, where a consumer picks features and builds a bespoke binary. What Node can’t offer, because it’s Rust-native all the way down instead of C++ addons.

They are the same crate at two lifecycle stages; the shared deploy feature preset is what keeps them honest. See W225 §2a.

Feature tiers

• default — the lean, V8-free Rust-handler harness: serve_app / wrap + the standard stack (HEALTH_PATH, TraceLayer, graceful shutdown). For services whose handlers are Rust functions; the dogfood is yah’s cloud-admin dashboard. Continues the “replacing bun with Rust-native SSR” arc (R448 rolldown → R449 deno_core → R450 default-flip).
• ssr — the SSR serving tier: the Server engine’s V8 dispatch plus the revalidate receiver. Links the prebuilt librusty_v8.a (a CDN download, not a from-source compile).
• build — adds the bundler (rolldown + lightningcss), the one genuinely uncached from-source compile. Its own crate so a stray feature flip can’t drag it into a lean consumer.

What lives here vs. mesofact-dev

The prod serving engine (server, proxy, and under ssr the ssr, revalidate and tenants modules) lives here. mesofact-dev holds only the dev affordances — the file watcher and the local S3 surface — and depends on this crate. That direction is load-bearing: it is what keeps a prod binary from linking dev code (W225 §2), and it is enforced by the dependency graph rather than by dead-stripping. Do not add a dev affordance to this crate, and do not add a prod-serving bin to mesofact-dev.

Cache / session / resilience layers live in mesofact_core::proxy today and are bundle-shaped; lifting them here as caller-composable tower::Layers is a follow-up once a second Rust-handler service needs them.

@yah:relay(R445, “mesofact-app: lean Rust-native app harness for Rust-handler services (continues R448/R449/R450 ‘replacing bun with rust-native SSR’ arc; dogfooded by yah parent R568-T4)”) @yah:at(2026-06-30T07:22:46Z) @yah:status(review) @yah:assignee(agent:bundle-anthropic-ashguard) @arch:see(.yah/docs/working/W174-mesofact-rust-native-pipeline.md) @yah:next(“yah parent camp R568-T4 consumes this via root [patch.crates-io] mesofact-app = { path = "oss/mesofact/crates/mesofact-app" } + a path-deferred version dep in crates/yah/cloud-admin.”) @yah:next(“Once a 2nd Rust-handler mesofact service exists, lift cache/session/resilience layers from mesofact::core::proxy::* into the mesofact facade as caller-composable tower::Layers (deferred per lib doc until 2nd consumer appears).”) @yah:handoff(“Landed mesofact-app crate (oss/mesofact/crates/mesofact-app, publish=false). Lean Rust-handler harness: pub HEALTH_PATH const + pub fn wrap(Router) -> Router (adds /__mesofact/health + tower-http TraceLayer) + pub async fn serve_app(Router, SocketAddr) -> Result<()> (binds, wraps, axum::serve with graceful Ctrl-C/SIGTERM) + pub async fn shutdown_signal. Companion to mesofact-dev: no mesofact-ssr/deno_core/V8 dep so pure-Rust services don’t inherit the ~75MB V8 binary. Registered in oss/mesofact workspace members. 3 tests pass (health auto-add, serve_app round-trip, documented panic guard on duplicate health route). Continues R448/R449/R450 arc (replacing bun with Rust-native SSR) – this is the next milestone after R449 swapped the engine, taking handlers from JS to Rust.”) @yah:verify(“cargo test -p mesofact-app # 3 passed”) @yah:gotcha(“Tier: Cleric – discovery+replicate. Mirrored mesofact-dev’s health/shutdown_signal shape so probes are drop-in compatible across JS-bundle and Rust-handler services.”) @yah:gotcha(“wrap() panics if the caller already registered HEALTH_PATH (axum::Router::merge rejects overlapping method routes regardless of order). Constraint is documented + pinned by a should_panic test; richer-probe services must bypass wrap.”) @yah:gotcha(“RESOLVED 2026-07-23 (W225 §2a) – was: ‘mesofact-dev refactor to delegate its bind/serve to mesofact-app is deferred’. The whole serving engine moved INTO this crate and mesofact-dev now depends on it, so the delegation is structural rather than deferred. Server’s router now uses this crate’s HEALTH_PATH + default_health + shutdown_signal instead of the byte-identical copies it carried in mesofact-dev.”)

Re-exports

pub use health::Health;

pub use health::LEGACY_HEALTH_PATH;

pub use health::LIVE_PATH;

pub use health::READY_PATH;

pub use proxy::ProxyMap;

pub use proxy::ProxyState;

pub use route_headers::RouteHeaderTable;

pub use server::declared_cache_policy;

pub use server::read_manifest_bytes;

pub use server::routes_declaring_ssr;

pub use server::routes_requiring_user;

pub use server::DistPointer;

pub use server::Identity;

pub use server::Server;

pub use server::DEFAULT_PORT;

pub use mesofact_core as core;

Modules

cache_headers

cache_policy enforcement for the sovereign serving path (R749-T1).

cli

The mesofact CLI — one subcommand-driven prod binary.

health

Kubernetes-conventional probe endpoints.

proxy

Same-origin reverse proxy for the dev server (R513-F10, W207 Gap #1).

route_headers

Per-route response headers for the sovereign serving path (R749-F3, W334) — the Rust half of what the @mesofact/edge Worker does with its ROUTE_HEADERS binding.

server

mesofact-dev — axum static-file server for mesofact-static workload artifacts, with optional file-watch + auto-rebuild + atomic pointer swap.

Structs

CachePolicyTable

Per-route cache headers derived from the built manifest. Empty = no route declared a non-inert policy, and every apply is a no-op.

Constants

DEFAULT_DRAIN_GRACE

How long /readyz reports failure before the listener stops accepting, on SIGTERM. Covers endpoint-removal propagation (kube-proxy, ingress, sidecar caches), which is concurrent with — not ordered before — the signal.

HEALTH_PATH

The historical probe path, predating the /livez + /readyz split.

Functions

serve_app

Bind addr, wrap app with the standard stack via wrap, and serve until Ctrl+C or SIGTERM — failing readiness for DEFAULT_DRAIN_GRACE before the listener closes.

shutdown_signal

Resolve when Ctrl+C or (on unix) SIGTERM is received.

shutdown_signal_for

shutdown_signal with the readiness half of a graceful shutdown: flip /readyz to 503, hold for the drain grace, then resolve so axum closes the listener.

wrap

Wrap a caller’s Router with the standard mesofact middleware stack — the health probe routes, the tower-http trace layer, and nothing else magical. The caller keeps full control of the route table.

wrap_with

wrap against a caller-owned Health, for services that need to hold the handle — to flip gates as subsystems come up, or to drain on their own shutdown path.