max / makenotwork

3.4 KB · 96 lines History Blame Raw

1	//! SyncKit field-version task. Polls MNW's internal client-versions endpoint on a
2	//! per-target interval and records the distribution so `target_status` can answer
3	//! which SDK is actually in the field.
4	//!
5	//! Unlike every other task here, this one fires no alerts. There is no bad
6	//! distribution to alert on (see `pom::checks::synckit_fleet`), and paging on the
7	//! remaining case, a readout PoM cannot take, would mean an email every hour for
8	//! as long as a stale token went unnoticed. It surfaces as a degraded condition
9	//! on the target instead, which is where an operator already looks.
10
11	use tokio::task::JoinHandle;
12	use tracing::{info, warn};
13
14	use pom::checks::synckit_fleet;
15	use pom::config::Config;
16	use pom::db;
17
18	use super::{CheckInterval, configured_targets};
19
20	pub(crate) fn spawn_synckit_fleet_tasks(
21	config: &Config,
22	pool: &sqlx::SqlitePool,
23	cancel: &tokio_util::sync::CancellationToken,
24	) -> Vec<JoinHandle<()>> {
25	let mut handles = Vec::new();
26
27	// The endpoint is authed with the alerts ingest token, so an alert config that
28	// omits it leaves this check with no credential. Read it once here rather than
29	// per tick: a token that appears later arrives by config reload, which restarts
30	// these tasks anyway.
31	let token = config
32	.alerts
33	.as_ref()
34	.and_then(\|a\| a.alerts_ingest_token.clone());
35
36	for (name, target_config) in configured_targets(config) {
37	let Some(fleet_config) = target_config.synckit_fleet else {
38	continue;
39	};
40
41	let Some(token) = token.clone() else {
42	// Loud, because the alternative is a configured check that silently
43	// never runs and a fleet column that stays blank for weeks.
44	warn!(
45	"{name}: synckit fleet check configured but alerts.alerts_ingest_token is unset, \
46	not spawning"
47	);
48	continue;
49	};
50
51	let name = name.clone();
52	let pool = pool.clone();
53	let cancel = cancel.clone();
54
55	info!(
56	"{name}: synckit fleet check every {}s against {} ({}d window)",
57	fleet_config.interval_secs, fleet_config.base_url, fleet_config.window_days,
58	);
59
60	handles.push(tokio::spawn(async move {
61	let mut ticks = CheckInterval::new(fleet_config.interval_secs, cancel);
62
63	while ticks.next().await {
64	let result = synckit_fleet::check_synckit_fleet(
65	&name,
66	&fleet_config.base_url,
67	&token,
68	fleet_config.window_days,
69	fleet_config.timeout_secs,
70	)
71	.await;
72
73	match &result.error {
74	Some(error) => warn!(
75	target = %name,
76	%error,
77	"synckit fleet readout unavailable"
78	),
79	None => info!(
80	"{name}: synckit fleet {} device(s) over {}d across {} version(s)",
81	result.devices,
82	result.window_days,
83	result.versions.len()
84	),
85	}
86
87	if let Err(e) = db::insert_synckit_fleet_check(&pool, &result).await {
88	tracing::error!("{name}: failed to store synckit fleet check: {e}");
89	}
90	}
91	}));
92	}
93
94	handles
95	}
96