Fail closed when SpacetimeDB auth restore is unavailable
This commit is contained in:
@@ -2,11 +2,12 @@ use axum::{
|
||||
Router,
|
||||
body::Body,
|
||||
extract::{Extension, FromRef},
|
||||
http::Request,
|
||||
http::{Request, StatusCode},
|
||||
middleware,
|
||||
response::Response,
|
||||
routing::{get, post},
|
||||
};
|
||||
use serde_json::json;
|
||||
use tower_http::{
|
||||
classify::ServerErrorsFailureClass,
|
||||
trace::{DefaultOnRequest, TraceLayer},
|
||||
@@ -18,6 +19,7 @@ use crate::{
|
||||
backpressure::limit_concurrent_requests,
|
||||
creation_entry_config::require_creation_entry_route_enabled,
|
||||
error_middleware::normalize_error_response,
|
||||
http_error::AppError,
|
||||
modules,
|
||||
request_context::{RequestContext, attach_request_context, resolve_request_id},
|
||||
response_headers::propagate_request_id_header,
|
||||
@@ -164,6 +166,96 @@ pub fn build_router(state: AppState) -> Router {
|
||||
.with_state(state)
|
||||
}
|
||||
|
||||
pub fn build_spacetime_unavailable_router(message: String) -> Router {
|
||||
Router::new()
|
||||
.fallback(spacetime_unavailable_handler)
|
||||
.layer(Extension(SpacetimeUnavailableState {
|
||||
message: message.into(),
|
||||
}))
|
||||
// 依赖不可用模式不挂业务 state,统一返回 503,并继续保留 request_id / API 版本 / 耗时响应头。
|
||||
.layer(middleware::from_fn(normalize_error_response))
|
||||
.layer(middleware::from_fn(propagate_request_id_header))
|
||||
.layer(
|
||||
TraceLayer::new_for_http()
|
||||
.make_span_with(|request: &Request<Body>| {
|
||||
let request_id =
|
||||
resolve_request_id(request).unwrap_or_else(|| "unknown".to_string());
|
||||
let route = crate::telemetry::observability_route(request.uri().path());
|
||||
let scheme = crate::telemetry::resolve_request_scheme(request.headers());
|
||||
let span_name = format!("{} {}", request.method(), route);
|
||||
|
||||
info_span!(
|
||||
"http.request",
|
||||
otel.kind = "server",
|
||||
otel.name = %span_name,
|
||||
otel.status_code = tracing::field::Empty,
|
||||
http.response.status_code = tracing::field::Empty,
|
||||
method = %request.method(),
|
||||
http.request.method = %request.method(),
|
||||
http.route = %route,
|
||||
url.scheme = %scheme,
|
||||
url.path = %request.uri().path(),
|
||||
request_id = %request_id,
|
||||
status = tracing::field::Empty,
|
||||
latency_ms = tracing::field::Empty,
|
||||
)
|
||||
})
|
||||
.on_request(DefaultOnRequest::new().level(Level::INFO))
|
||||
.on_response(
|
||||
|response: &axum::response::Response,
|
||||
latency: std::time::Duration,
|
||||
span: &Span| {
|
||||
let latency_ms = latency.as_millis().min(u64::MAX as u128) as u64;
|
||||
let status = response.status().as_u16();
|
||||
span.record("status", status);
|
||||
span.record("http.response.status_code", status);
|
||||
span.record(
|
||||
"otel.status_code",
|
||||
if response.status().is_server_error() {
|
||||
"ERROR"
|
||||
} else {
|
||||
"OK"
|
||||
},
|
||||
);
|
||||
span.record("latency_ms", latency_ms);
|
||||
},
|
||||
)
|
||||
.on_failure(
|
||||
|failure: ServerErrorsFailureClass,
|
||||
latency: std::time::Duration,
|
||||
span: &Span| {
|
||||
let latency_ms = latency.as_millis().min(u64::MAX as u128) as u64;
|
||||
error!(
|
||||
parent: span,
|
||||
latency_ms,
|
||||
failure = %failure,
|
||||
"http request failed"
|
||||
);
|
||||
},
|
||||
),
|
||||
)
|
||||
.layer(middleware::from_fn(attach_request_context))
|
||||
}
|
||||
|
||||
#[derive(Clone, Debug)]
|
||||
struct SpacetimeUnavailableState {
|
||||
message: std::sync::Arc<str>,
|
||||
}
|
||||
|
||||
async fn spacetime_unavailable_handler(
|
||||
Extension(state): Extension<SpacetimeUnavailableState>,
|
||||
Extension(request_context): Extension<RequestContext>,
|
||||
) -> Response {
|
||||
AppError::from_status(StatusCode::SERVICE_UNAVAILABLE)
|
||||
.with_message("SpacetimeDB 暂不可用,api-server 正在等待数据库恢复")
|
||||
.with_details(json!({
|
||||
"provider": "spacetimedb",
|
||||
"reason": "spacetime_startup_unavailable",
|
||||
"message": state.message.as_ref(),
|
||||
}))
|
||||
.into_response_with_context(Some(&request_context))
|
||||
}
|
||||
|
||||
async fn record_api_tracking_after_success(
|
||||
axum::extract::State(state): axum::extract::State<AppState>,
|
||||
Extension(request_context): Extension<RequestContext>,
|
||||
@@ -368,7 +460,7 @@ mod tests {
|
||||
|
||||
use crate::{config::AppConfig, state::AppState};
|
||||
|
||||
use super::build_router;
|
||||
use super::{build_router, build_spacetime_unavailable_router};
|
||||
|
||||
const TEST_PASSWORD: &str = "secret123";
|
||||
const INTERNAL_TEST_SECRET: &str = "test-internal-secret";
|
||||
@@ -564,6 +656,38 @@ mod tests {
|
||||
);
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn spacetime_unavailable_router_returns_service_unavailable_for_requests() {
|
||||
let app = build_spacetime_unavailable_router("SpacetimeDB 启动恢复认证快照超时".to_string());
|
||||
|
||||
let response = app
|
||||
.oneshot(
|
||||
Request::builder()
|
||||
.uri("/api/auth/login-options")
|
||||
.header("x-request-id", "req-spacetime-unavailable")
|
||||
.body(Body::empty())
|
||||
.expect("request should build"),
|
||||
)
|
||||
.await
|
||||
.expect("request should succeed");
|
||||
|
||||
assert_eq!(response.status(), StatusCode::SERVICE_UNAVAILABLE);
|
||||
assert_eq!(
|
||||
response
|
||||
.headers()
|
||||
.get("x-request-id")
|
||||
.and_then(|value| value.to_str().ok()),
|
||||
Some("req-spacetime-unavailable")
|
||||
);
|
||||
let body = read_json_response(response).await;
|
||||
assert_eq!(body["error"]["code"], "SERVICE_UNAVAILABLE");
|
||||
assert_eq!(
|
||||
body["error"]["details"]["reason"],
|
||||
"spacetime_startup_unavailable"
|
||||
);
|
||||
assert_eq!(body["error"]["details"]["provider"], "spacetimedb");
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn creation_entry_route_disabled_returns_service_unavailable() {
|
||||
let state = AppState::new(AppConfig::default()).expect("state should build");
|
||||
|
||||
Reference in New Issue
Block a user