From cd5fbac13f68d2e054632570bb3520f9a3ff8e42 Mon Sep 17 00:00:00 2001
From: Hazel Atkinson <yellowsink@riseup.net>
Date: Mon, 7 Apr 2025 13:43:12 +0100
Subject: cleanups, make workers close neatly on container exit

---
 src/config.rs     |  2 +-
 src/main.rs       |  7 ++++---
 src/stats_task.rs | 15 ++++++++++++++-
 3 files changed, 19 insertions(+), 5 deletions(-)

diff --git a/src/config.rs b/src/config.rs
index 52cb69c..212ca64 100644
--- a/src/config.rs
+++ b/src/config.rs
@@ -35,7 +35,7 @@ pub static CONFIG: LazyLock<CspyConfig> = LazyLock::new(|| {
 struct ProtoDeserVisitor;
 
 /// deserialization boilerplate
-impl<'de> confique::serde::de::Visitor<'de> for ProtoDeserVisitor {
+impl confique::serde::de::Visitor<'_> for ProtoDeserVisitor {
 	type Value = Protocol;
 
 	fn expecting(&self, formatter: &mut std::fmt::Formatter) -> std::fmt::Result {
diff --git a/src/main.rs b/src/main.rs
index dd8f6a2..c703fc2 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -4,9 +4,9 @@ use anyhow::Result;
 use bollard::Docker;
 use config::CONFIG;
 use opentelemetry_otlp::{MetricExporter, Protocol, WithExportConfig};
-use opentelemetry_sdk::metrics::{PeriodicReader, PeriodicReaderBuilder, SdkMeterProvider};
+use opentelemetry_sdk::metrics::{PeriodicReader, SdkMeterProvider};
 use tokio::task::JoinHandle;
-use tokio::time::interval;
+use tokio::time::MissedTickBehavior;
 use tokio_util::sync::CancellationToken;
 
 mod config;
@@ -80,7 +80,8 @@ async fn main() -> Result<()> {
 		st2.cancel();
 	});
 
-	let mut container_search_interval = tokio::time::interval(Duration::from_secs(1));
+	let mut container_search_interval = tokio::time::interval(Duration::from_millis(CONFIG.otlp_export_interval.unwrap_or(6000)) / 2);
+	container_search_interval.set_missed_tick_behavior(MissedTickBehavior::Skip);
 
 	let mut tasks: BTreeMap<String, JoinHandle<()>> = BTreeMap::new();
 
diff --git a/src/stats_task.rs b/src/stats_task.rs
index 4240a63..8b82373 100644
--- a/src/stats_task.rs
+++ b/src/stats_task.rs
@@ -1,5 +1,5 @@
 use std::mem::MaybeUninit;
-use bollard::container::{BlkioStats, BlkioStatsEntry, StatsOptions};
+use bollard::container::{BlkioStatsEntry, StatsOptions};
 use bollard::models::ContainerSummary;
 use bollard::Docker;
 use opentelemetry::metrics::MeterProvider;
@@ -136,6 +136,19 @@ pub fn launch_stats_task(
 
 		while let Some(val) = stats_stream.next().await {
 			if let Ok(stats) = val {
+
+				// when a container exits, instead of a None we get sent Ok()s with zeroes in it forever, horror
+				if stats.cpu_stats.cpu_usage.total_usage == 0 {
+					if stats.precpu_stats.cpu_usage.total_usage != 0 { break; }
+					else {
+						// last time was ALSO a zero, so this MIGHT actually be (SOMEHOW?) legit,
+						// so just loop around again, and wait for the main task to abort() this worker task instead!
+						// which it will if this container died, or if we are gonna get real stats later, it won't...
+						// man i dont know i should probably just break lol
+						continue;
+					}
+				};
+
 				meter_container_cpu_usage_seconds_total.add(
 					cpu_delta_from_docker(
 						stats.cpu_stats.cpu_usage.total_usage,
-- 
cgit