|
| 1 | +// Copyright The OpenTelemetry Authors |
| 2 | +// SPDX-License-Identifier: Apache-2.0 |
| 3 | + |
| 4 | +//! Global health and status endpoints. |
| 5 | +//! |
| 6 | +//! - GET `/status` - list all pipelines and their status |
| 7 | +//! - GET `/livez` - liveness probe |
| 8 | +//! - GET `/readyz` - readiness probe |
| 9 | +
|
| 10 | +use crate::AppState; |
| 11 | +use axum::extract::State; |
| 12 | +use axum::http::StatusCode; |
| 13 | +use axum::routing::get; |
| 14 | +use axum::{Json, Router}; |
| 15 | +use chrono::Utc; |
| 16 | +use otap_df_state::PipelineKey; |
| 17 | +use otap_df_state::conditions::{Condition, ConditionKind, ConditionReason, ConditionStatus}; |
| 18 | +use otap_df_state::pipeline_status::PipelineStatus; |
| 19 | +use serde::Serialize; |
| 20 | +use std::collections::HashMap; |
| 21 | + |
| 22 | +/// All the routes for health and status endpoints. |
| 23 | +pub(crate) fn routes() -> Router<AppState> { |
| 24 | + Router::new() |
| 25 | + // Returns a summary of all pipelines and their statuses. |
| 26 | + .route("/status", get(show_status)) |
| 27 | + // Returns liveness status. |
| 28 | + .route("/livez", get(livez)) |
| 29 | + // Returns readiness status. |
| 30 | + .route("/readyz", get(readyz)) |
| 31 | +} |
| 32 | + |
| 33 | +#[derive(Serialize)] |
| 34 | +#[serde(rename_all = "camelCase")] |
| 35 | +pub(crate) struct StatusResponse { |
| 36 | + generated_at: String, |
| 37 | + pipelines: HashMap<PipelineKey, PipelineStatus>, |
| 38 | +} |
| 39 | + |
| 40 | +#[derive(Serialize)] |
| 41 | +#[serde(rename_all = "camelCase")] |
| 42 | +pub(crate) struct ProbeResponse { |
| 43 | + probe: &'static str, |
| 44 | + status: &'static str, |
| 45 | + generated_at: String, |
| 46 | + #[serde(skip_serializing_if = "Vec::is_empty")] |
| 47 | + failing: Vec<PipelineConditionFailure>, |
| 48 | +} |
| 49 | + |
| 50 | +#[derive(Serialize)] |
| 51 | +#[serde(rename_all = "camelCase")] |
| 52 | +struct PipelineConditionFailure { |
| 53 | + pipeline: PipelineKey, |
| 54 | + condition: Condition, |
| 55 | +} |
| 56 | + |
| 57 | +pub async fn show_status( |
| 58 | + State(state): State<AppState>, |
| 59 | +) -> Result<Json<StatusResponse>, StatusCode> { |
| 60 | + Ok(Json(StatusResponse { |
| 61 | + generated_at: Utc::now().to_rfc3339(), |
| 62 | + pipelines: state.observed_state_store.snapshot(), |
| 63 | + })) |
| 64 | +} |
| 65 | + |
| 66 | +pub(crate) async fn livez(State(state): State<AppState>) -> (StatusCode, Json<ProbeResponse>) { |
| 67 | + let snapshot = state.observed_state_store.snapshot(); |
| 68 | + let failing = collect_condition_failures( |
| 69 | + &snapshot, |
| 70 | + ConditionKind::Accepted, |
| 71 | + skip_pipelines_without_runtimes, |
| 72 | + acceptance_failure, |
| 73 | + ); |
| 74 | + |
| 75 | + if failing.is_empty() { |
| 76 | + (StatusCode::OK, Json(ProbeResponse::ok("livez"))) |
| 77 | + } else { |
| 78 | + ( |
| 79 | + StatusCode::INTERNAL_SERVER_ERROR, |
| 80 | + Json(ProbeResponse::fail("livez", failing)), |
| 81 | + ) |
| 82 | + } |
| 83 | +} |
| 84 | + |
| 85 | +pub(crate) async fn readyz(State(state): State<AppState>) -> (StatusCode, Json<ProbeResponse>) { |
| 86 | + let snapshot = state.observed_state_store.snapshot(); |
| 87 | + let failing = collect_condition_failures( |
| 88 | + &snapshot, |
| 89 | + ConditionKind::Ready, |
| 90 | + skip_pipelines_without_runtimes, |
| 91 | + |cond| cond.status != ConditionStatus::True, |
| 92 | + ); |
| 93 | + |
| 94 | + if failing.is_empty() { |
| 95 | + (StatusCode::OK, Json(ProbeResponse::ok("readyz"))) |
| 96 | + } else { |
| 97 | + ( |
| 98 | + StatusCode::SERVICE_UNAVAILABLE, |
| 99 | + Json(ProbeResponse::fail("readyz", failing)), |
| 100 | + ) |
| 101 | + } |
| 102 | +} |
| 103 | + |
| 104 | +fn collect_condition_failures<FSkip, FFail>( |
| 105 | + pipelines: &HashMap<PipelineKey, PipelineStatus>, |
| 106 | + condition_kind: ConditionKind, |
| 107 | + skip: FSkip, |
| 108 | + failure_predicate: FFail, |
| 109 | +) -> Vec<PipelineConditionFailure> |
| 110 | +where |
| 111 | + FSkip: Fn(&PipelineStatus) -> bool, |
| 112 | + FFail: Fn(&Condition) -> bool, |
| 113 | +{ |
| 114 | + pipelines |
| 115 | + .iter() |
| 116 | + .filter(|(_, status)| !skip(status)) |
| 117 | + .filter_map(|(key, status)| { |
| 118 | + let condition = status |
| 119 | + .conditions() |
| 120 | + .into_iter() |
| 121 | + .find(|c| c.kind == condition_kind)?; |
| 122 | + failure_predicate(&condition).then(|| PipelineConditionFailure { |
| 123 | + pipeline: key.clone(), |
| 124 | + condition, |
| 125 | + }) |
| 126 | + }) |
| 127 | + .collect() |
| 128 | +} |
| 129 | + |
| 130 | +fn acceptance_failure(condition: &Condition) -> bool { |
| 131 | + match condition.status { |
| 132 | + ConditionStatus::True => false, |
| 133 | + ConditionStatus::Unknown => { |
| 134 | + !matches!(condition.reason, Some(ConditionReason::NoPipelineRuntime)) |
| 135 | + } |
| 136 | + ConditionStatus::False => { |
| 137 | + let benign = matches!( |
| 138 | + condition.reason, |
| 139 | + Some( |
| 140 | + ConditionReason::Pending |
| 141 | + | ConditionReason::StartRequested |
| 142 | + | ConditionReason::Deleting |
| 143 | + | ConditionReason::ForceDeleting |
| 144 | + | ConditionReason::Deleted |
| 145 | + | ConditionReason::NoPipelineRuntime |
| 146 | + ) |
| 147 | + ); |
| 148 | + !benign |
| 149 | + } |
| 150 | + } |
| 151 | +} |
| 152 | + |
| 153 | +fn skip_pipelines_without_runtimes(status: &PipelineStatus) -> bool { |
| 154 | + status.total_cores() == 0 |
| 155 | +} |
| 156 | + |
| 157 | +impl ProbeResponse { |
| 158 | + fn ok(probe: &'static str) -> Self { |
| 159 | + Self { |
| 160 | + probe, |
| 161 | + status: "ok", |
| 162 | + generated_at: Utc::now().to_rfc3339(), |
| 163 | + failing: Vec::new(), |
| 164 | + } |
| 165 | + } |
| 166 | + |
| 167 | + fn fail(probe: &'static str, failing: Vec<PipelineConditionFailure>) -> Self { |
| 168 | + Self { |
| 169 | + probe, |
| 170 | + status: "failed", |
| 171 | + generated_at: Utc::now().to_rfc3339(), |
| 172 | + failing, |
| 173 | + } |
| 174 | + } |
| 175 | +} |
| 176 | + |
| 177 | +#[cfg(test)] |
| 178 | +mod tests { |
| 179 | + use super::*; |
| 180 | + |
| 181 | + fn cond(status: ConditionStatus, reason: Option<ConditionReason>) -> Condition { |
| 182 | + Condition { |
| 183 | + kind: ConditionKind::Accepted, |
| 184 | + status, |
| 185 | + reason, |
| 186 | + message: None, |
| 187 | + last_transition_time: None, |
| 188 | + } |
| 189 | + } |
| 190 | + |
| 191 | + #[test] |
| 192 | + fn acceptance_failure_ignores_benign_reasons() { |
| 193 | + assert!(!acceptance_failure(&cond( |
| 194 | + ConditionStatus::False, |
| 195 | + Some(ConditionReason::Pending) |
| 196 | + ))); |
| 197 | + assert!(!acceptance_failure(&cond( |
| 198 | + ConditionStatus::False, |
| 199 | + Some(ConditionReason::Deleted) |
| 200 | + ))); |
| 201 | + assert!(!acceptance_failure(&cond( |
| 202 | + ConditionStatus::Unknown, |
| 203 | + Some(ConditionReason::NoPipelineRuntime) |
| 204 | + ))); |
| 205 | + } |
| 206 | + |
| 207 | + #[test] |
| 208 | + fn acceptance_failure_flags_errors() { |
| 209 | + assert!(acceptance_failure(&cond( |
| 210 | + ConditionStatus::False, |
| 211 | + Some(ConditionReason::AdmissionError) |
| 212 | + ))); |
| 213 | + assert!(acceptance_failure(&cond(ConditionStatus::Unknown, None))); |
| 214 | + } |
| 215 | +} |
0 commit comments