From fc18f28ff98aa42bb252bd1fb363a47d4398a64a Mon Sep 17 00:00:00 2001
From: "Brian L. Troutwine" <brian.troutwine@datadoghq.com>
Date: Fri, 13 Dec 2024 16:07:51 -0800
Subject: [PATCH] Allow users to configure capture metrics expiration

This commit adjusts the capture metrics expiration so that it is by
default Duration::MAX and is a configurable setting. In the future we
will bring this value down, but this preserves behavior between lading
releases while still allowing for new capability.

Signed-off-by: Brian L. Troutwine <brian.troutwine@datadoghq.com>
---
 CHANGELOG.md             |  2 +-
 lading/src/bin/lading.rs |  5 ++++-
 lading/src/captures.rs   |  3 ++-
 lading/src/config.rs     | 11 ++++++++++-
 4 files changed, 17 insertions(+), 4 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 5a92b91239..3f4ec3b352 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,13 +7,13 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 ## Added
 - Parse nearly the complete field list of smaps/smaps_rollup in the Linux observer. 
+- Metrics storage is now generational, expiring unwritten metrics by configuration parameter `expiration`. 
 ## Removed
 - Removed the unused target-rss-byte-limit from the command line, internal stub of. 
 ## Changed
 - logrotate_fs is now behind a feature flag and not enabled in the default
   build. It remains enabled in the release artifact.
 - The build now includes http1 and http2 support. Actual usage and availability may vary.
-- Metrics storage is now generational, expiring unwritten metrics every 3 seconds.
 - CPU data now sourced from cgroup v2 on Linux, memory data expanded significantly. 
 - CPU data now also includes kubernetes style 'millicore' calculations.
 
diff --git a/lading/src/bin/lading.rs b/lading/src/bin/lading.rs
index 7b48166669..8cb0af662b 100644
--- a/lading/src/bin/lading.rs
+++ b/lading/src/bin/lading.rs
@@ -11,7 +11,7 @@ use clap::{ArgGroup, Parser, Subcommand};
 use lading::{
     blackhole,
     captures::CaptureManager,
-    config::{Config, Telemetry},
+    config::{default_expiration, Config, Telemetry},
     generator::{self, process_tree},
     inspector, observer,
     target::{self, Behavior, Output},
@@ -301,6 +301,7 @@ fn get_config(ops: &Opts, config: Option<String>) -> Result<Config, Error> {
         config.telemetry = Telemetry::Log {
             path: capture_path.parse().map_err(|_| Error::CapturePath)?,
             global_labels: options_global_labels.inner,
+            expiration: default_expiration(),
         };
     } else {
         match config.telemetry {
@@ -380,12 +381,14 @@ async fn inner_main(
         Telemetry::Log {
             path,
             global_labels,
+            expiration,
         } => {
             let mut capture_manager = CaptureManager::new(
                 path,
                 shutdown_watcher.register()?,
                 experiment_started_watcher.clone(),
                 target_running_watcher.clone(),
+                expiration,
             )
             .await?;
             for (k, v) in global_labels {
diff --git a/lading/src/captures.rs b/lading/src/captures.rs
index 816520465f..57a1f6a09b 100644
--- a/lading/src/captures.rs
+++ b/lading/src/captures.rs
@@ -84,6 +84,7 @@ impl CaptureManager {
         shutdown: lading_signal::Watcher,
         experiment_started: lading_signal::Watcher,
         target_running: lading_signal::Watcher,
+        expiration: Duration,
     ) -> Result<Self, io::Error> {
         let fp = tokio::fs::File::create(&capture_path).await?;
         let fp = fp.into_std().await;
@@ -93,7 +94,7 @@ impl CaptureManager {
             recency: Recency::new(
                 quanta::Clock::new(),
                 MetricKindMask::GAUGE | MetricKindMask::COUNTER,
-                Some(Duration::from_secs(3)),
+                Some(expiration),
             ),
         };
 
diff --git a/lading/src/config.rs b/lading/src/config.rs
index 5f680bf3a3..43081c763c 100644
--- a/lading/src/config.rs
+++ b/lading/src/config.rs
@@ -1,7 +1,7 @@
 //! This module controls configuration parsing from the end user, providing a
 //! convenience mechanism for the rest of the program. Crashes are most likely
 //! to originate from this code, intentionally.
-use std::{net::SocketAddr, path::PathBuf};
+use std::{net::SocketAddr, path::PathBuf, time::Duration};
 
 use rustc_hash::FxHashMap;
 use serde::Deserialize;
@@ -51,6 +51,12 @@ pub struct Config {
     pub inspector: Option<inspector::Config>,
 }
 
+/// Default value for [`Telemetry::Log::expiration`]
+#[must_use]
+pub fn default_expiration() -> Duration {
+    Duration::MAX
+}
+
 #[derive(Debug, Deserialize, PartialEq, Eq)]
 #[serde(rename_all = "snake_case")]
 #[serde(deny_unknown_fields)]
@@ -80,6 +86,9 @@ pub enum Telemetry {
         path: PathBuf,
         /// Additional labels to include in every metric
         global_labels: FxHashMap<String, String>,
+        /// The time metrics that have not been written to will take to expire.
+        #[serde(default = "default_expiration")]
+        expiration: Duration,
     },
 }