max / makenotwork

2.8 KB · 90 lines History Blame Raw

1	//! Test-suite alert/recovery messages. Domain half of the Alerter split;
2	//! shared dispatch/cooldown plumbing lives in the parent module.
3	//!
4	//! Distinct from [`super::test_duration`], which fires on a suite that still
5	//! passes but has slowed down. These fire on pass/fail transitions.
6
7	use tracing::instrument;
8
9	use super::{AlertCategory, AlertMeta, Alerter};
10
11	impl Alerter {
12	#[instrument(skip_all)]
13	pub async fn send_test_failure_alert(
14	&self,
15	target: &str,
16	label: &str,
17	failed: Option<i64>,
18	exit_code: Option<i64>,
19	detail: &str,
20	) {
21	let alert_key = format!("tests:{target}");
22	let subject = match failed {
23	Some(n) => format!("[PoM] {label}: {n} test(s) failing"),
24	None => format!("[PoM] {label}: test suite failed to complete"),
25	};
26	let body = format!(
27	"Target: {label} ({target})\n\
28	Failed: {}\n\
29	Exit code: {}\n\
30	Detail: {detail}\n\
31	Instance: {}\n\
32	Time: {}\n\n\
33	- PoM",
34	failed.map_or_else(\|\| "unknown".to_string(), \|n\| n.to_string()),
35	exit_code.map_or_else(\|\| "none".to_string(), \|c\| c.to_string()),
36	self.instance_name,
37	chrono::Utc::now().to_rfc3339(),
38	);
39
40	// A suite that never produced a count did not run at all (SSH refused,
41	// missing checkout, timeout), which is worse than a known count of
42	// failures: it means the target is unverified rather than broken.
43	let priority = if failed.is_some() { "high" } else { "critical" };
44
45	self.fire_failure(
46	&subject,
47	&body,
48	priority,
49	"pom-tests",
50	Some(target),
51	AlertMeta {
52	key: &alert_key,
53	category: AlertCategory::TestFailure,
54	from: None,
55	to: None,
56	error: Some(detail),
57	},
58	)
59	.await;
60	}
61
62	#[instrument(skip_all)]
63	pub async fn send_test_recovery(&self, target: &str, label: &str) {
64	let alert_key = format!("tests:{target}");
65	let subject = format!("[PoM] {label}: test suite passing again");
66	let body = format!(
67	"Target: {label} ({target})\n\
68	The test suite passed on its latest scheduled run.\n\
69	Instance: {}\n\
70	Time: {}\n\n\
71	- PoM",
72	self.instance_name,
73	chrono::Utc::now().to_rfc3339(),
74	);
75
76	self.fire_recovery(
77	&subject,
78	&body,
79	AlertMeta {
80	key: &alert_key,
81	category: AlertCategory::TestRecovery,
82	from: None,
83	to: None,
84	error: None,
85	},
86	)
87	.await;
88	}
89	}
90