mirror of
https://github.com/rustfs/rustfs.git
synced 2026-08-27 15:37:02 +00:00
fix(health): keep liveness peer independent (#6576)
Keep liveness probes local by avoiding readiness collection and omitting readiness-only fields from liveness payloads. Readiness and MinIO cluster probes continue to report dependency and quorum state. Co-authored-by: heihutu <heihutu@gmail.com>
This commit is contained in:
@@ -961,6 +961,44 @@ mod tests {
|
||||
.await;
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
#[serial]
|
||||
async fn console_liveness_omits_readiness_state() {
|
||||
temp_env::async_with_vars([(rustfs_config::ENV_HEALTH_MINIMAL_RESPONSE_ENABLE, Some("false"))], async {
|
||||
let object_traffic_health =
|
||||
Arc::new(crate::app::object_traffic_health::ObjectTrafficHealth::enabled_for_test(Duration::ZERO));
|
||||
let _stalled = object_traffic_health
|
||||
.track_write_storage()
|
||||
.expect("write tracking must be enabled");
|
||||
let app_context =
|
||||
crate::app::gating_test_env::app_context_with_object_traffic_health(Arc::clone(&object_traffic_health)).await;
|
||||
let server_ctx = crate::runtime_sources::ServerContextSlot::new();
|
||||
assert!(server_ctx.install(app_context));
|
||||
|
||||
let response = health_check(
|
||||
Method::GET,
|
||||
format!("{CONSOLE_PREFIX}{HEALTH_PREFIX}")
|
||||
.parse()
|
||||
.expect("console liveness URI"),
|
||||
Some(Extension(server_ctx)),
|
||||
)
|
||||
.await;
|
||||
|
||||
assert_eq!(response.status(), StatusCode::OK);
|
||||
let body = response
|
||||
.into_body()
|
||||
.collect()
|
||||
.await
|
||||
.expect("console liveness body")
|
||||
.to_bytes();
|
||||
let payload: serde_json::Value = serde_json::from_slice(&body).expect("console liveness JSON");
|
||||
assert_eq!(payload["status"], "ok");
|
||||
assert!(payload.get("ready").is_none());
|
||||
assert!(payload.get("degradedReasons").is_none());
|
||||
})
|
||||
.await;
|
||||
}
|
||||
|
||||
// setup_console_middleware_stack reads ENV_HEALTH_ENDPOINT_ENABLE (see above).
|
||||
#[tokio::test]
|
||||
#[serial]
|
||||
|
||||
@@ -113,8 +113,8 @@ mod tests {
|
||||
fn test_liveness_state_iam_not_ready() {
|
||||
let state = health_check_state(true, false, true, true, HealthProbe::Liveness);
|
||||
assert_eq!(state.status_code, StatusCode::OK);
|
||||
assert_eq!(state.status, "degraded");
|
||||
assert!(!state.ready);
|
||||
assert_eq!(state.status, "ok");
|
||||
assert!(state.ready);
|
||||
}
|
||||
|
||||
#[test]
|
||||
@@ -137,8 +137,8 @@ mod tests {
|
||||
fn test_liveness_state_lock_not_ready() {
|
||||
let state = health_check_state(true, true, false, true, HealthProbe::Liveness);
|
||||
assert_eq!(state.status_code, StatusCode::OK);
|
||||
assert_eq!(state.status, "degraded");
|
||||
assert!(!state.ready);
|
||||
assert_eq!(state.status, "ok");
|
||||
assert!(state.ready);
|
||||
}
|
||||
|
||||
#[test]
|
||||
@@ -180,7 +180,7 @@ mod tests {
|
||||
#[test]
|
||||
fn test_readiness_probe_uses_node_collector_only() {
|
||||
assert_eq!(readiness_source_for_probe(HealthProbe::Readiness), Some(HealthReadinessSource::Node));
|
||||
assert_eq!(readiness_source_for_probe(HealthProbe::Liveness), Some(HealthReadinessSource::Node));
|
||||
assert_eq!(readiness_source_for_probe(HealthProbe::Liveness), None);
|
||||
}
|
||||
|
||||
#[test]
|
||||
@@ -246,7 +246,7 @@ mod tests {
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_build_health_response_liveness_returns_200_when_deps_not_ready() {
|
||||
fn test_build_health_response_liveness_omits_readiness_state_when_deps_not_ready() {
|
||||
let readiness_report = crate::shared_types::DependencyReadinessReport {
|
||||
readiness: crate::shared_types::DependencyReadiness {
|
||||
storage_ready: false,
|
||||
@@ -264,20 +264,17 @@ mod tests {
|
||||
None,
|
||||
None,
|
||||
);
|
||||
// Liveness HTTP status remains 200 (process is alive).
|
||||
assert_eq!(parts.status_code, StatusCode::OK);
|
||||
let payload = parts.payload.expect("GET should include payload");
|
||||
// But `ready` now reflects actual readiness state.
|
||||
assert_eq!(payload["status"], "degraded");
|
||||
assert_eq!(payload["ready"], false);
|
||||
// Dependency details are included when readiness report is present.
|
||||
assert!(payload.get("details").is_some());
|
||||
assert!(payload.get("degradedReasons").is_some());
|
||||
assert_eq!(payload["status"], "ok");
|
||||
assert!(payload.get("ready").is_none());
|
||||
assert!(payload.get("details").is_none());
|
||||
assert!(payload.get("degradedReasons").is_none());
|
||||
}
|
||||
|
||||
#[test]
|
||||
#[serial]
|
||||
fn test_health_and_readiness_body_agree_when_only_lock_quorum_is_unavailable() {
|
||||
fn test_liveness_body_stays_peer_independent_when_only_lock_quorum_is_unavailable() {
|
||||
with_var(rustfs_config::ENV_HEALTH_MINIMAL_RESPONSE_ENABLE, Some("false"), || {
|
||||
let readiness_report = crate::shared_types::DependencyReadinessReport {
|
||||
readiness: crate::shared_types::DependencyReadiness {
|
||||
@@ -310,13 +307,13 @@ mod tests {
|
||||
assert_eq!(readiness.status_code, StatusCode::SERVICE_UNAVAILABLE);
|
||||
let liveness_payload = liveness.payload.expect("GET should include liveness payload");
|
||||
let readiness_payload = readiness.payload.expect("GET should include readiness payload");
|
||||
assert_eq!(liveness_payload["status"], "degraded");
|
||||
assert_eq!(liveness_payload["status"], "ok");
|
||||
assert_eq!(readiness_payload["status"], "degraded");
|
||||
assert_eq!(liveness_payload["ready"], false);
|
||||
assert!(liveness_payload.get("ready").is_none());
|
||||
assert_eq!(readiness_payload["ready"], false);
|
||||
assert_eq!(liveness_payload["details"]["lock"]["ready"], false);
|
||||
assert!(liveness_payload.get("details").is_none());
|
||||
assert_eq!(readiness_payload["details"]["lock"]["ready"], false);
|
||||
assert_eq!(liveness_payload["degradedReasons"][0], "lock_quorum_unavailable");
|
||||
assert!(liveness_payload.get("degradedReasons").is_none());
|
||||
assert_eq!(readiness_payload["degradedReasons"][0], "lock_quorum_unavailable");
|
||||
});
|
||||
}
|
||||
@@ -350,6 +347,7 @@ mod tests {
|
||||
let health = health_check_state(true, false, true, true, HealthProbe::Readiness);
|
||||
with_var(rustfs_config::ENV_HEALTH_MINIMAL_RESPONSE_ENABLE, Some("true"), || {
|
||||
let payload = build_health_payload(HealthPayloadContext {
|
||||
probe: HealthProbe::Readiness,
|
||||
health,
|
||||
storage_ready: true,
|
||||
iam_ready: false,
|
||||
@@ -375,6 +373,7 @@ mod tests {
|
||||
with_var(rustfs_config::ENV_HEALTH_MINIMAL_RESPONSE_ENABLE, Some("false"), || {
|
||||
let health = health_check_state(false, false, false, true, HealthProbe::Readiness);
|
||||
let payload = build_health_payload(HealthPayloadContext {
|
||||
probe: HealthProbe::Readiness,
|
||||
health,
|
||||
storage_ready: false,
|
||||
iam_ready: false,
|
||||
|
||||
+28
-16
@@ -54,6 +54,7 @@ pub(crate) struct HealthResponseParts {
|
||||
|
||||
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
|
||||
pub(crate) struct HealthPayloadContext<'a> {
|
||||
pub(crate) probe: HealthProbe,
|
||||
pub(crate) health: HealthCheckState,
|
||||
pub(crate) storage_ready: bool,
|
||||
pub(crate) iam_ready: bool,
|
||||
@@ -97,7 +98,8 @@ fn apply_object_traffic_snapshot(report: &mut DependencyReadinessReport, snapsho
|
||||
|
||||
pub(crate) fn readiness_source_for_probe(probe: HealthProbe) -> Option<HealthReadinessSource> {
|
||||
match probe {
|
||||
HealthProbe::Liveness | HealthProbe::Readiness => Some(HealthReadinessSource::Node),
|
||||
HealthProbe::Liveness => None,
|
||||
HealthProbe::Readiness => Some(HealthReadinessSource::Node),
|
||||
HealthProbe::ClusterWrite => Some(HealthReadinessSource::ClusterWrite),
|
||||
HealthProbe::ClusterRead => Some(HealthReadinessSource::ClusterRead),
|
||||
}
|
||||
@@ -113,13 +115,12 @@ pub(crate) fn health_check_state(
|
||||
let ready = storage_ready && iam_ready && lock_quorum_ready && peer_health_ready;
|
||||
|
||||
if probe == HealthProbe::Liveness {
|
||||
// Liveness always returns HTTP 200 (process is alive), but the `ready`
|
||||
// field now reflects actual node readiness so that callers who inspect
|
||||
// the body get a truthful signal instead of a hardcoded `true`.
|
||||
// Liveness is intentionally local and peer-independent. Dependency
|
||||
// readiness belongs to `/health/ready` and the MinIO cluster probes.
|
||||
return HealthCheckState {
|
||||
status_code: StatusCode::OK,
|
||||
status: if ready { "ok" } else { "degraded" },
|
||||
ready,
|
||||
status: "ok",
|
||||
ready: true,
|
||||
};
|
||||
}
|
||||
|
||||
@@ -295,7 +296,7 @@ pub(crate) fn build_health_response_parts(
|
||||
lock_quorum_ready,
|
||||
health_check_state(storage_ready, iam_ready, lock_quorum_ready, peer_health_ready, probe),
|
||||
readiness_report.degraded_reasons.clone(),
|
||||
true,
|
||||
probe != HealthProbe::Liveness,
|
||||
)
|
||||
}
|
||||
(HealthProbe::Readiness | HealthProbe::ClusterWrite | HealthProbe::ClusterRead, None) => (
|
||||
@@ -341,6 +342,7 @@ pub(crate) fn build_health_response_parts(
|
||||
None
|
||||
} else {
|
||||
Some(build_health_payload(HealthPayloadContext {
|
||||
probe,
|
||||
health,
|
||||
storage_ready,
|
||||
iam_ready,
|
||||
@@ -361,20 +363,29 @@ pub(crate) fn build_health_response_parts(
|
||||
|
||||
pub(crate) fn build_health_payload(ctx: HealthPayloadContext<'_>) -> Value {
|
||||
if health_minimal_response_enabled() {
|
||||
return json!({
|
||||
"status": ctx.health.status,
|
||||
"ready": ctx.health.ready,
|
||||
});
|
||||
return if ctx.probe == HealthProbe::Liveness {
|
||||
json!({
|
||||
"status": ctx.health.status,
|
||||
})
|
||||
} else {
|
||||
json!({
|
||||
"status": ctx.health.status,
|
||||
"ready": ctx.health.ready,
|
||||
})
|
||||
};
|
||||
}
|
||||
|
||||
let mut payload = json!({
|
||||
"status": ctx.health.status,
|
||||
"ready": ctx.health.ready,
|
||||
"service": ctx.service,
|
||||
"timestamp": jiff::Zoned::now().to_string(),
|
||||
"version": env!("CARGO_PKG_VERSION"),
|
||||
});
|
||||
|
||||
if ctx.probe != HealthProbe::Liveness {
|
||||
payload["ready"] = json!(ctx.health.ready);
|
||||
}
|
||||
|
||||
if ctx.include_dependency_details {
|
||||
payload["details"] = build_component_details(ctx.storage_ready, ctx.iam_ready, ctx.lock_quorum_ready, ctx.kms_ready);
|
||||
payload["degradedReasons"] = build_degraded_reasons(ctx.degraded_reasons);
|
||||
@@ -491,7 +502,7 @@ mod tests {
|
||||
|
||||
#[test]
|
||||
#[serial]
|
||||
fn liveness_and_readiness_payloads_share_lock_quorum_readiness() {
|
||||
fn liveness_payload_omits_lock_quorum_readiness() {
|
||||
with_var(rustfs_config::ENV_HEALTH_MINIMAL_RESPONSE_ENABLE, Some("false"), || {
|
||||
let mut report = ready_report();
|
||||
report.readiness.lock_quorum_ready = false;
|
||||
@@ -506,11 +517,12 @@ mod tests {
|
||||
assert_eq!(readiness.status_code, StatusCode::SERVICE_UNAVAILABLE);
|
||||
let liveness_payload = liveness.payload.expect("liveness GET should include payload");
|
||||
let readiness_payload = readiness.payload.expect("readiness GET should include payload");
|
||||
assert_eq!(liveness_payload["ready"], false);
|
||||
assert_eq!(liveness_payload["status"], "ok");
|
||||
assert!(liveness_payload.get("ready").is_none());
|
||||
assert!(liveness_payload.get("details").is_none());
|
||||
assert!(liveness_payload.get("degradedReasons").is_none());
|
||||
assert_eq!(readiness_payload["ready"], false);
|
||||
assert_eq!(liveness_payload["details"]["lock"]["ready"], false);
|
||||
assert_eq!(readiness_payload["details"]["lock"]["ready"], false);
|
||||
assert_eq!(liveness_payload["degradedReasons"], json!(["lock_quorum_unavailable"]));
|
||||
assert_eq!(readiness_payload["degradedReasons"], json!(["lock_quorum_unavailable"]));
|
||||
});
|
||||
}
|
||||
|
||||
@@ -2718,13 +2718,10 @@ mod tests {
|
||||
let body = BodyExt::collect(response.into_body()).await.expect("body").to_bytes();
|
||||
let payload: serde_json::Value =
|
||||
serde_json::from_slice(&body).expect("public liveness health response should be valid JSON");
|
||||
assert!(matches!(payload["status"].as_str(), Some("ok" | "degraded")));
|
||||
assert!(payload["ready"].is_boolean());
|
||||
assert!(payload["details"].is_object());
|
||||
assert!(payload["details"]["storage"]["ready"].is_boolean());
|
||||
assert!(payload["details"]["iam"]["ready"].is_boolean());
|
||||
assert!(payload["details"]["lock"]["ready"].is_boolean());
|
||||
assert!(payload["degradedReasons"].is_array());
|
||||
assert_eq!(payload["status"], "ok");
|
||||
assert!(payload.get("ready").is_none());
|
||||
assert!(payload.get("details").is_none());
|
||||
assert!(payload.get("degradedReasons").is_none());
|
||||
},
|
||||
)
|
||||
.await;
|
||||
@@ -3058,6 +3055,14 @@ mod tests {
|
||||
.await
|
||||
.expect("liveness response");
|
||||
assert_eq!(response.status(), StatusCode::OK);
|
||||
let body = BodyExt::collect(response.into_body())
|
||||
.await
|
||||
.expect("liveness body")
|
||||
.to_bytes();
|
||||
let payload: serde_json::Value = serde_json::from_slice(&body).expect("liveness JSON");
|
||||
assert_eq!(payload["status"], "ok");
|
||||
assert!(payload.get("ready").is_none());
|
||||
assert!(payload.get("degradedReasons").is_none());
|
||||
assert_eq!(calls.load(Ordering::SeqCst), 0);
|
||||
|
||||
drop(stalled);
|
||||
|
||||
Reference in New Issue
Block a user