|
| 1 | +//! #2119 watchdog tests: poll timeout and error-streak tracking. |
| 2 | +//! |
| 3 | +//! Tests use a real axum fake-server (no internal mocks) to verify that |
| 4 | +//! `run_forever` does not hang when `FetchTask` is slow or erroneous. |
| 5 | +
|
| 6 | +use std::sync::{Arc, Mutex}; |
| 7 | +use std::time::Duration; |
| 8 | + |
| 9 | +use axum::{Json, Router, extract::State, routing::post}; |
| 10 | +use serde_json::{Value, json}; |
| 11 | +use terraphim_gitea_runner::client::ReqwestRunnerClient; |
| 12 | +use terraphim_gitea_runner::config::RunnerConfig; |
| 13 | +use terraphim_gitea_runner::policy::DeterministicPlanner; |
| 14 | +use terraphim_gitea_runner::poller::Poller; |
| 15 | +use terraphim_gitea_runner::state::RunnerState; |
| 16 | +use tokio::time::Instant; |
| 17 | + |
| 18 | +fn runner_state() -> RunnerState { |
| 19 | + RunnerState { |
| 20 | + uuid: "uuid-1".into(), |
| 21 | + token: "tok-1".into(), |
| 22 | + name: "fake".into(), |
| 23 | + version: "0.1.0".into(), |
| 24 | + labels: vec!["terraphim-native".into()], |
| 25 | + ephemeral: false, |
| 26 | + } |
| 27 | +} |
| 28 | + |
| 29 | +async fn register(Json(_b): Json<Value>) -> Json<Value> { |
| 30 | + Json( |
| 31 | + json!({"runner": {"id": 1, "uuid": "uuid-1", "token": "tok-1", |
| 32 | + "name": "fake", "version": "0.1.0", "labels": ["terraphim-native"], "ephemeral": false}}), |
| 33 | + ) |
| 34 | +} |
| 35 | +async fn declare(Json(b): Json<Value>) -> Json<Value> { |
| 36 | + Json(json!({"version": b["version"], "labels": b["labels"]})) |
| 37 | +} |
| 38 | +async fn update_task(Json(_b): Json<Value>) -> Json<Value> { |
| 39 | + Json(json!({"tasksVersion": 1, "sentOutputs": {}})) |
| 40 | +} |
| 41 | +async fn update_log(Json(b): Json<Value>) -> Json<Value> { |
| 42 | + let ack = b["index"].as_i64().unwrap_or(0) |
| 43 | + + b["rows"].as_array().map(|a| a.len() as i64).unwrap_or(0); |
| 44 | + Json(json!({"ackIndex": ack})) |
| 45 | +} |
| 46 | + |
| 47 | +#[derive(Default, Clone)] |
| 48 | +struct Counters { |
| 49 | + fetch_calls: usize, |
| 50 | +} |
| 51 | +type Shared = Arc<Mutex<Counters>>; |
| 52 | + |
| 53 | +/// Serve a slow FetchTask that sleeps longer than the client timeout. |
| 54 | +async fn fetch_slow(State(s): State<Shared>, Json(_b): Json<Value>) -> Json<Value> { |
| 55 | + s.lock().unwrap().fetch_calls += 1; |
| 56 | + tokio::time::sleep(Duration::from_millis(500)).await; |
| 57 | + Json(json!({"tasksVersion": 0})) |
| 58 | +} |
| 59 | + |
| 60 | +/// Serve a fast FetchTask that immediately returns no task. |
| 61 | +async fn fetch_fast(State(s): State<Shared>, Json(_b): Json<Value>) -> Json<Value> { |
| 62 | + s.lock().unwrap().fetch_calls += 1; |
| 63 | + Json(json!({"tasksVersion": 0})) |
| 64 | +} |
| 65 | + |
| 66 | +async fn spawn_server(shared: Shared, fetch: axum::routing::MethodRouter<Shared>) -> String { |
| 67 | + let base = "/api/actions/runner.v1.RunnerService"; |
| 68 | + let app = Router::new() |
| 69 | + .route(&format!("{base}/Register"), post(register)) |
| 70 | + .route(&format!("{base}/Declare"), post(declare)) |
| 71 | + .route(&format!("{base}/FetchTask"), fetch) |
| 72 | + .route(&format!("{base}/UpdateTask"), post(update_task)) |
| 73 | + .route(&format!("{base}/UpdateLog"), post(update_log)) |
| 74 | + .with_state(shared); |
| 75 | + let listener = tokio::net::TcpListener::bind("127.0.0.1:0").await.unwrap(); |
| 76 | + let addr = listener.local_addr().unwrap(); |
| 77 | + tokio::spawn(async move { axum::serve(listener, app).await.unwrap() }); |
| 78 | + format!("http://{addr}") |
| 79 | +} |
| 80 | + |
| 81 | +/// When FetchTask takes longer than `http_request_timeout`, `poll_once` returns |
| 82 | +/// an error (reqwest timeout) and `run_forever` continues without hanging. |
| 83 | +#[tokio::test(flavor = "multi_thread")] |
| 84 | +async fn poll_timeout_does_not_hang_forever() { |
| 85 | + let shared: Shared = Arc::new(Mutex::new(Counters::default())); |
| 86 | + // Use a short HTTP timeout: slow server sleeps 500ms, we timeout at 100ms. |
| 87 | + let http_timeout = Duration::from_millis(100); |
| 88 | + let http = reqwest::Client::builder() |
| 89 | + .timeout(http_timeout) |
| 90 | + .build() |
| 91 | + .unwrap(); |
| 92 | + let url = spawn_server(shared.clone(), post(fetch_slow)).await; |
| 93 | + let config = RunnerConfig { |
| 94 | + instance_url: url.clone(), |
| 95 | + poll_interval: Duration::from_millis(10), |
| 96 | + poll_timeout: Duration::from_millis(200), |
| 97 | + http_request_timeout: http_timeout, |
| 98 | + ..RunnerConfig::default() |
| 99 | + }; |
| 100 | + let client = |
| 101 | + Arc::new(terraphim_gitea_runner::client::ReqwestRunnerClient::with_http(url, http)); |
| 102 | + let _tmpdir = tempfile::tempdir().unwrap(); |
| 103 | + let poller = Poller::new( |
| 104 | + client, |
| 105 | + Arc::new(DeterministicPlanner::default()), |
| 106 | + config, |
| 107 | + _tmpdir.path(), |
| 108 | + ); |
| 109 | + let st = runner_state(); |
| 110 | + |
| 111 | + // run_forever loops without end; we let it run for 600ms and verify it |
| 112 | + // made multiple attempts (not frozen waiting on the slow server). |
| 113 | + let start = Instant::now(); |
| 114 | + let _ = tokio::time::timeout(Duration::from_millis(600), poller.run_forever(&st)).await; |
| 115 | + let elapsed = start.elapsed(); |
| 116 | + |
| 117 | + // Must have returned within the outer timeout, not the server's 500ms sleep. |
| 118 | + assert!( |
| 119 | + elapsed < Duration::from_millis(700), |
| 120 | + "run_forever must not hang on slow FetchTask (elapsed={elapsed:?})" |
| 121 | + ); |
| 122 | + // Must have made at least 2 poll attempts, proving it recovered and retried. |
| 123 | + let calls = shared.lock().unwrap().fetch_calls; |
| 124 | + assert!( |
| 125 | + calls >= 2, |
| 126 | + "at least 2 FetchTask attempts expected (got {calls})" |
| 127 | + ); |
| 128 | +} |
| 129 | + |
| 130 | +/// `new_with_timeout` builds a client whose underlying reqwest client has the |
| 131 | +/// configured timeout — verified by checking that a slow server causes an error |
| 132 | +/// response from `poll_once` faster than the server sleeps. |
| 133 | +#[tokio::test] |
| 134 | +async fn new_with_timeout_sets_request_timeout() { |
| 135 | + let shared: Shared = Arc::new(Mutex::new(Counters::default())); |
| 136 | + let url = spawn_server(shared.clone(), post(fetch_slow)).await; |
| 137 | + |
| 138 | + // Client with 100ms timeout; server sleeps 500ms. |
| 139 | + let client = Arc::new(ReqwestRunnerClient::new_with_timeout( |
| 140 | + &url, |
| 141 | + Duration::from_millis(100), |
| 142 | + )); |
| 143 | + let config = RunnerConfig { |
| 144 | + instance_url: url, |
| 145 | + poll_timeout: Duration::from_secs(10), |
| 146 | + http_request_timeout: Duration::from_millis(100), |
| 147 | + ..RunnerConfig::default() |
| 148 | + }; |
| 149 | + let _tmpdir2 = tempfile::tempdir().unwrap(); |
| 150 | + let poller = Poller::new( |
| 151 | + client, |
| 152 | + Arc::new(DeterministicPlanner::default()), |
| 153 | + config, |
| 154 | + _tmpdir2.path(), |
| 155 | + ); |
| 156 | + let st = runner_state(); |
| 157 | + |
| 158 | + let start = Instant::now(); |
| 159 | + let result = poller.poll_once(&st, 0).await; |
| 160 | + let elapsed = start.elapsed(); |
| 161 | + |
| 162 | + assert!(result.is_err(), "poll_once must error when server is slow"); |
| 163 | + assert!( |
| 164 | + elapsed < Duration::from_millis(300), |
| 165 | + "poll_once must time out before the server's 500ms sleep (elapsed={elapsed:?})" |
| 166 | + ); |
| 167 | +} |
| 168 | + |
| 169 | +/// A fast server + quick poll_interval: `run_forever` should call FetchTask |
| 170 | +/// multiple times within a short window, demonstrating the normal loop works. |
| 171 | +#[tokio::test(flavor = "multi_thread")] |
| 172 | +async fn run_forever_polls_repeatedly_on_no_task() { |
| 173 | + let shared: Shared = Arc::new(Mutex::new(Counters::default())); |
| 174 | + let url = spawn_server(shared.clone(), post(fetch_fast)).await; |
| 175 | + let config = RunnerConfig { |
| 176 | + instance_url: url.clone(), |
| 177 | + poll_interval: Duration::from_millis(10), |
| 178 | + http_request_timeout: Duration::from_secs(5), |
| 179 | + poll_timeout: Duration::from_secs(10), |
| 180 | + ..RunnerConfig::default() |
| 181 | + }; |
| 182 | + let client = Arc::new(ReqwestRunnerClient::new(url)); |
| 183 | + let _tmpdir3 = tempfile::tempdir().unwrap(); |
| 184 | + let poller = Poller::new( |
| 185 | + client, |
| 186 | + Arc::new(DeterministicPlanner::default()), |
| 187 | + config, |
| 188 | + _tmpdir3.path(), |
| 189 | + ); |
| 190 | + let st = runner_state(); |
| 191 | + |
| 192 | + let _ = tokio::time::timeout(Duration::from_millis(120), poller.run_forever(&st)).await; |
| 193 | + |
| 194 | + let calls = shared.lock().unwrap().fetch_calls; |
| 195 | + assert!( |
| 196 | + calls >= 5, |
| 197 | + "at least 5 FetchTask polls in 120ms with 10ms interval (got {calls})" |
| 198 | + ); |
| 199 | +} |
0 commit comments