feat: allow metrics to be sampled at custom intervals (#192)

Adds a new config parameter to control the file-exposition interval. Previously, this functionality was limited to secondly sampling. Now the user may specify other values, including sub-secondly sampling.

configs/segcache.toml

@@ -13,6 +13,9 @@ duration = 300
 # optionally, specify the format for the output stats file (default is json)
 # possible values are (json, msgpack, parquet).
 #metrics_format = "parquet"
+# optionally, specify the sampling interval for metrics. Input is a string
+# with the unit attached; for example "100ms" or "1s". Default to 100ms.
+#metrics_interval = "1s"
 # run the admin thread with a HTTP listener at the address provided, this allows
 # stats exposition via HTTP
 admin = "127.0.0.1:9090"

src/config/general.rs

@@ -13,6 +13,10 @@ pub enum MetricsFormat {
     Parquet,
 }
 
+pub fn metrics_interval() -> String {
+    "100ms".into()
+}
+
 #[derive(Clone, Deserialize)]
 pub struct General {
     /// The protocol to be used for the test.
@@ -29,6 +33,9 @@ pub struct General {
     /// ignored if no metrics output is specified.
     #[serde(default)]
     metrics_format: MetricsFormat,
+    /// The reporting interval. Specify time along with unit; default to 100ms.
+    #[serde(default = "metrics_interval")]
+    metrics_interval: String,
     /// The admin listen address
     admin: String,
     /// The initial seed for initializing the PRNGs. This can be any string and
@@ -57,6 +64,13 @@ impl General {
         self.metrics_format
     }
 
+    pub fn metrics_interval(&self) -> Duration {
+        self.metrics_interval
+            .parse::<humantime::Duration>()
+            .unwrap()
+            .into()
+    }
+
     pub fn admin(&self) -> String {
         self.admin.clone()
     }
@@ -73,4 +87,18 @@ impl General {
             Seed512(seed)
         }
     }
+
+    pub fn validate(&self) {
+        let interval = self.metrics_interval.parse::<humantime::Duration>();
+        if let Err(e) = interval {
+            eprintln!("metrics_interval is not valid: {e}");
+            std::process::exit(1);
+        }
+
+        let interval = interval.unwrap().as_millis();
+        if interval < Duration::from_millis(10).as_millis() {
+            eprintln!("metrics_interval should be larger than 10ms");
+            std::process::exit(1);
+        }
+    }
 }

src/config/mod.rs

@@ -59,14 +59,16 @@ impl Config {
                 std::process::exit(1);
             }
         }
-        let toml = toml::from_str(&content);
-        match toml {
-            Ok(toml) => toml,
-            Err(error) => {
-                eprintln!("Failed to parse TOML config: {filename}\n{error}");
+        let config: Config = toml::from_str(&content)
+            .map_err(|e| {
+                eprintln!("Failed to parse TOML config: {filename}\n{e}");
                 std::process::exit(1);
-            }
-        }
+            })
+            .unwrap();
+
+        config.general.validate();
+        config.workload.ratelimit().validate();
+        config
     }
 
     pub fn general(&self) -> &General {

src/output/mod.rs

@@ -239,7 +239,7 @@ pub async fn metrics(config: Config) {
     // get the stop time
     let stop = start + config.general().duration();
 
-    let mut interval = tokio::time::interval_at(start, Duration::from_secs(1));
+    let mut interval = tokio::time::interval_at(start, config.general().metrics_interval());
 
     let snapshotter = SnapshotterBuilder::new()
         .metadata("source".to_string(), env!("CARGO_BIN_NAME").to_string())

src/workload/mod.rs

@@ -741,7 +741,6 @@ pub struct Ratelimit {
 impl Ratelimit {
     pub fn new(config: &Config) -> Option<Self> {
         let ratelimit_config = config.workload().ratelimit();
-        ratelimit_config.validate();
 
         if !ratelimit_config.is_dynamic() {
             return None;