| 1 |
|
| 2 |
|
| 3 |
|
| 4 |
|
| 5 |
|
| 6 |
|
| 7 |
|
| 8 |
|
| 9 |
|
| 10 |
|
| 11 |
|
| 12 |
|
| 13 |
use tokio::task::JoinHandle; |
| 14 |
use tracing::{error, info}; |
| 15 |
|
| 16 |
use pom::alerts::Alerter; |
| 17 |
use pom::checks::{drift, ssh}; |
| 18 |
use pom::config::{Config, TestsConfig}; |
| 19 |
use pom::db; |
| 20 |
use pom::types::TestRun; |
| 21 |
|
| 22 |
use super::{CheckInterval, configured_targets}; |
| 23 |
|
| 24 |
pub(crate) fn spawn_test_tasks( |
| 25 |
config: &Config, |
| 26 |
pool: &sqlx::SqlitePool, |
| 27 |
cancel: &tokio_util::sync::CancellationToken, |
| 28 |
alerter: Option<&Alerter>, |
| 29 |
) -> Vec<JoinHandle<()>> { |
| 30 |
let sweep_secs = config.serve.test_sweep_interval_secs; |
| 31 |
let mut handles = Vec::new(); |
| 32 |
|
| 33 |
for (name, target_config) in configured_targets(config) { |
| 34 |
let Some(tests_config) = target_config.tests else { |
| 35 |
continue; |
| 36 |
}; |
| 37 |
let label = target_config.label.clone(); |
| 38 |
let pool = pool.clone(); |
| 39 |
let alerter = alerter.cloned(); |
| 40 |
let cancel = cancel.clone(); |
| 41 |
|
| 42 |
info!( |
| 43 |
"{name}: test staleness sweep every {sweep_secs}s (threshold={}d, command={})", |
| 44 |
tests_config.staleness_days, tests_config.command |
| 45 |
); |
| 46 |
|
| 47 |
handles.push(tokio::spawn(async move { |
| 48 |
let mut ticks = CheckInterval::new(sweep_secs, cancel); |
| 49 |
|
| 50 |
while ticks.next().await { |
| 51 |
let Some(reason) = staleness_reason(&pool, &name, &tests_config).await else { |
| 52 |
continue; |
| 53 |
}; |
| 54 |
|
| 55 |
info!("{name}: running tests ({reason})"); |
| 56 |
let run = ssh::run_tests(&name, &tests_config, None).await; |
| 57 |
let previous_passed = db::get_latest_test_run(&pool, &name) |
| 58 |
.await |
| 59 |
.ok() |
| 60 |
.flatten() |
| 61 |
.map(|r| r.passed); |
| 62 |
|
| 63 |
if let Err(e) = store_run(&pool, &name, &run).await { |
| 64 |
error!("{name}: failed to store test run: {e}"); |
| 65 |
} |
| 66 |
|
| 67 |
info!( |
| 68 |
"{}: tests {} ({} passed, {} failed, {}s)", |
| 69 |
name, |
| 70 |
if run.passed { "passed" } else { "FAILED" }, |
| 71 |
run.summary.total_passed.unwrap_or(-1), |
| 72 |
run.summary.total_failed.unwrap_or(-1), |
| 73 |
run.duration_secs.unwrap_or(-1), |
| 74 |
); |
| 75 |
|
| 76 |
if let Some(ref alerter) = alerter { |
| 77 |
alert_on_transition(alerter, &name, &label, &run, previous_passed).await; |
| 78 |
} |
| 79 |
} |
| 80 |
})); |
| 81 |
} |
| 82 |
|
| 83 |
handles |
| 84 |
} |
| 85 |
|
| 86 |
|
| 87 |
async fn staleness_reason( |
| 88 |
pool: &sqlx::SqlitePool, |
| 89 |
name: &str, |
| 90 |
tests_config: &TestsConfig, |
| 91 |
) -> Option<String> { |
| 92 |
let current_version = db::get_health_history(pool, Some(name), 1) |
| 93 |
.await |
| 94 |
.unwrap_or_default() |
| 95 |
.first() |
| 96 |
.and_then(|s| s.details.as_ref()) |
| 97 |
.and_then(|d| d.version.clone()); |
| 98 |
|
| 99 |
let latest_test = db::get_latest_test_run(pool, name).await.unwrap_or(None); |
| 100 |
|
| 101 |
let tested_version = match latest_test { |
| 102 |
Some(ref test) => db::get_version_at_time(pool, name, &test.started_at) |
| 103 |
.await |
| 104 |
.unwrap_or(None), |
| 105 |
None => None, |
| 106 |
}; |
| 107 |
|
| 108 |
let staleness = drift::compute_test_staleness( |
| 109 |
current_version.as_deref(), |
| 110 |
tested_version.as_deref(), |
| 111 |
latest_test.as_ref().map(|t| t.started_at.as_str()), |
| 112 |
tests_config.staleness_days, |
| 113 |
); |
| 114 |
|
| 115 |
staleness |
| 116 |
.stale |
| 117 |
.then(|| staleness.reason.unwrap_or_else(|| "stale".to_string())) |
| 118 |
} |
| 119 |
|
| 120 |
|
| 121 |
|
| 122 |
async fn store_run( |
| 123 |
pool: &sqlx::SqlitePool, |
| 124 |
name: &str, |
| 125 |
run: &TestRun, |
| 126 |
) -> Result<(), pom::error::PomError> { |
| 127 |
let run_id = db::insert_test_run(pool, run).await?; |
| 128 |
|
| 129 |
if !run.summary.details.is_empty() |
| 130 |
&& let Err(e) = db::insert_test_details(pool, run_id, &run.summary.details).await |
| 131 |
{ |
| 132 |
error!("{name}: failed to store test details: {e}"); |
| 133 |
} |
| 134 |
|
| 135 |
Ok(()) |
| 136 |
} |
| 137 |
|
| 138 |
|
| 139 |
#[derive(Debug, Clone, Copy, PartialEq, Eq)] |
| 140 |
enum Transition { |
| 141 |
Broke, |
| 142 |
Recovered, |
| 143 |
} |
| 144 |
|
| 145 |
|
| 146 |
|
| 147 |
|
| 148 |
fn transition(previous_passed: Option<bool>, now_passed: bool) -> Option<Transition> { |
| 149 |
match (previous_passed.unwrap_or(true), now_passed) { |
| 150 |
(true, false) => Some(Transition::Broke), |
| 151 |
(false, true) => Some(Transition::Recovered), |
| 152 |
_ => None, |
| 153 |
} |
| 154 |
} |
| 155 |
|
| 156 |
async fn alert_on_transition( |
| 157 |
alerter: &Alerter, |
| 158 |
name: &str, |
| 159 |
label: &str, |
| 160 |
run: &TestRun, |
| 161 |
previous_passed: Option<bool>, |
| 162 |
) { |
| 163 |
match transition(previous_passed, run.passed) { |
| 164 |
Some(Transition::Broke) => { |
| 165 |
let detail = failure_detail(run); |
| 166 |
alerter |
| 167 |
.send_test_failure_alert( |
| 168 |
name, |
| 169 |
label, |
| 170 |
run.summary.total_failed, |
| 171 |
run.exit_code.map(i64::from), |
| 172 |
&detail, |
| 173 |
) |
| 174 |
.await; |
| 175 |
} |
| 176 |
Some(Transition::Recovered) => alerter.send_test_recovery(name, label).await, |
| 177 |
None => {} |
| 178 |
} |
| 179 |
} |
| 180 |
|
| 181 |
|
| 182 |
|
| 183 |
|
| 184 |
fn failure_detail(run: &TestRun) -> String { |
| 185 |
let failed: Vec<&str> = run |
| 186 |
.summary |
| 187 |
.details |
| 188 |
.iter() |
| 189 |
.filter(|d| !d.passed) |
| 190 |
.map(|d| d.test_name.as_str()) |
| 191 |
.take(10) |
| 192 |
.collect(); |
| 193 |
|
| 194 |
if !failed.is_empty() { |
| 195 |
return failed.join(", "); |
| 196 |
} |
| 197 |
|
| 198 |
let tail: Vec<&str> = run |
| 199 |
.raw_output |
| 200 |
.lines() |
| 201 |
.filter(|l| !l.trim().is_empty()) |
| 202 |
.rev() |
| 203 |
.take(5) |
| 204 |
.collect(); |
| 205 |
|
| 206 |
if tail.is_empty() { |
| 207 |
"no output captured".to_string() |
| 208 |
} else { |
| 209 |
tail.into_iter().rev().collect::<Vec<_>>().join(" | ") |
| 210 |
} |
| 211 |
} |
| 212 |
|
| 213 |
#[cfg(test)] |
| 214 |
mod tests { |
| 215 |
use super::*; |
| 216 |
use pom::types::{TestDetail, TestSummary}; |
| 217 |
|
| 218 |
fn run_with(passed: bool, details: Vec<TestDetail>, raw_output: &str) -> TestRun { |
| 219 |
TestRun { |
| 220 |
id: None, |
| 221 |
target: "mnw".to_string(), |
| 222 |
started_at: "2026-08-06T00:00:00Z".to_string(), |
| 223 |
finished_at: None, |
| 224 |
duration_secs: Some(1), |
| 225 |
exit_code: Some(if passed { 0 } else { 101 }), |
| 226 |
passed, |
| 227 |
summary: TestSummary { |
| 228 |
steps: vec![], |
| 229 |
total_passed: None, |
| 230 |
total_failed: None, |
| 231 |
details, |
| 232 |
}, |
| 233 |
raw_output: raw_output.to_string(), |
| 234 |
filter: None, |
| 235 |
} |
| 236 |
} |
| 237 |
|
| 238 |
fn detail(name: &str, passed: bool) -> TestDetail { |
| 239 |
TestDetail { |
| 240 |
test_name: name.to_string(), |
| 241 |
passed, |
| 242 |
} |
| 243 |
} |
| 244 |
|
| 245 |
#[test] |
| 246 |
fn failure_detail_prefers_failing_test_names() { |
| 247 |
let run = run_with( |
| 248 |
false, |
| 249 |
vec![ |
| 250 |
detail("checks::whois::ok", true), |
| 251 |
detail("checks::rdap::broken", false), |
| 252 |
detail("db::also_broken", false), |
| 253 |
], |
| 254 |
"irrelevant output", |
| 255 |
); |
| 256 |
assert_eq!( |
| 257 |
failure_detail(&run), |
| 258 |
"checks::rdap::broken, db::also_broken" |
| 259 |
); |
| 260 |
} |
| 261 |
|
| 262 |
#[test] |
| 263 |
fn failure_detail_caps_the_name_list() { |
| 264 |
let details: Vec<TestDetail> = (0..20).map(|i| detail(&format!("t{i}"), false)).collect(); |
| 265 |
let run = run_with(false, details, ""); |
| 266 |
assert_eq!(failure_detail(&run).split(", ").count(), 10); |
| 267 |
} |
| 268 |
|
| 269 |
#[test] |
| 270 |
fn failure_detail_falls_back_to_output_tail_in_order() { |
| 271 |
|
| 272 |
|
| 273 |
|
| 274 |
let run = run_with( |
| 275 |
false, |
| 276 |
vec![], |
| 277 |
"ssh: connect to host 100.106.221.39 port 22: Connection refused\n\ |
| 278 |
exit status 255\n", |
| 279 |
); |
| 280 |
let detail = failure_detail(&run); |
| 281 |
assert!(detail.starts_with("ssh: connect to host"), "got {detail}"); |
| 282 |
assert!(detail.ends_with("exit status 255"), "got {detail}"); |
| 283 |
} |
| 284 |
|
| 285 |
#[test] |
| 286 |
fn failure_detail_handles_no_output() { |
| 287 |
assert_eq!( |
| 288 |
failure_detail(&run_with(false, vec![], " \n\n")), |
| 289 |
"no output captured" |
| 290 |
); |
| 291 |
} |
| 292 |
|
| 293 |
#[test] |
| 294 |
fn transition_first_ever_run_failing_is_a_break() { |
| 295 |
|
| 296 |
|
| 297 |
|
| 298 |
assert_eq!(transition(None, false), Some(Transition::Broke)); |
| 299 |
} |
| 300 |
|
| 301 |
#[test] |
| 302 |
fn transition_first_ever_run_passing_is_silent() { |
| 303 |
assert_eq!(transition(None, true), None); |
| 304 |
} |
| 305 |
|
| 306 |
#[test] |
| 307 |
fn transition_still_red_does_not_repage() { |
| 308 |
assert_eq!(transition(Some(false), false), None); |
| 309 |
} |
| 310 |
|
| 311 |
#[test] |
| 312 |
fn transition_still_green_is_silent() { |
| 313 |
assert_eq!(transition(Some(true), true), None); |
| 314 |
} |
| 315 |
|
| 316 |
#[test] |
| 317 |
fn transition_red_to_green_recovers() { |
| 318 |
assert_eq!(transition(Some(false), true), Some(Transition::Recovered)); |
| 319 |
} |
| 320 |
|
| 321 |
#[test] |
| 322 |
fn transition_green_to_red_breaks() { |
| 323 |
assert_eq!(transition(Some(true), false), Some(Transition::Broke)); |
| 324 |
} |
| 325 |
} |
| 326 |
|