Struct ServerState 

pub struct ServerState {
    pub sites: HashMap<String, SiteBackend>,
    pub console_inactivity_timeout: Duration,
    pub audit: AuditLog,
    pub auth_rate_limit_per_minute: Option<u32>,
    pub request_timeout: Duration,
    pub shutdown_grace_period: Duration,
    pub migrate_backup_root: Option<PathBuf>,
}

Shared state for all HTTP handlers.

Holds one SiteBackend per configured site so a single server can serve multiple clusters. Each request supplies the target site via the X-Manta-Site header; handlers call ServerState::infra_context (or, via the handlers::RequestCtx extractor, the cached RequestCtx::infra() shortcut) to retrieve the per-site data.

Plumbed through Axum’s State<Arc<ServerState>> extractor. Owned by start_server and cloned (cheaply, since it’s an Arc) into every spawned task.

Fields

sites: HashMap<String, SiteBackend>

Per-site connection data, keyed by site name.

console_inactivity_timeout: Duration

How long a WebSocket console session may be idle before the server closes it. Protects against leaked Kubernetes pod attachments.

audit: AuditLog

Audit trail every mutating request is recorded to (always the log; Kafka too when [auditor.kafka] is configured).

auth_rate_limit_per_minute: Option<u32>

Per-source-IP rate limit on /v2/auth/* (requests/minute). None disables in-process rate limiting.

request_timeout: Duration

Global request timeout applied to every HTTP route (router-level TimeoutLayer). All long-running work (power transitions, SAT dispatch) runs CLI-side, so this is the only request-timeout knob the server has.

shutdown_grace_period: Duration

Drain window for axum_server::Handle::graceful_shutdown on SIGTERM / Ctrl+C. Sourced from server.toml’s [server] shutdown_grace_period_secs.

migrate_backup_root: Option<PathBuf>

Filesystem root that confines POST /migrate/{backup,restore} file access. None disables both endpoints — even admin callers must wait for an operator to opt in via [server] migrate_backup_root. The path is stored already-canonicalised so per-request validation is a single starts_with against this.

Implementations

impl ServerState

pub fn infra_context<'a>( &'a self, site_name: &'a str, ) -> Result<InfraContext<'a>, Error>

Build a borrowed InfraContext for the named site.

Called per-request so the service layer can work with its existing &InfraContext<'_> API without taking ownership of the underlying SiteBackend.

Errors

Returns [Error::NotFound] when site_name is not in Self::sites.