From 6a95fb6b74b34ade1f9ca97d35e3da90643648ce Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Marcel=20M=C3=A4rtens?= <marcel.cochem@googlemail.com>
Date: Sat, 27 Mar 2021 16:27:59 +0100
Subject: [PATCH] extend network of incomming TCP metrics and failed handshake
 metric

---
 network/src/metrics.rs   | 22 ++++++++++++++++++++++
 network/src/scheduler.rs | 14 +++++++++++---
 2 files changed, 33 insertions(+), 3 deletions(-)

diff --git a/network/src/metrics.rs b/network/src/metrics.rs
index 9733b477a1..c46fe16bda 100644
--- a/network/src/metrics.rs
+++ b/network/src/metrics.rs
@@ -9,6 +9,8 @@ use std::error::Error;
 pub struct NetworkMetrics {
     pub listen_requests_total: IntCounterVec,
     pub connect_requests_total: IntCounterVec,
+    pub incoming_connections_total: IntCounterVec,
+    pub failed_handshakes_total: IntCounter,
     pub participants_connected_total: IntCounter,
     pub participants_disconnected_total: IntCounter,
     // channel id's, seperated by PARTICIPANT, max 5
@@ -44,6 +46,17 @@ impl NetworkMetrics {
             ),
             &["protocol"],
         )?;
+        let incoming_connections_total = IntCounterVec::new(
+            Opts::new(
+                "incoming_connections_total",
+                "Shows the number of external requests to the scheduler",
+            ),
+            &["protocol"],
+        )?;
+        let failed_handshakes_total = IntCounter::with_opts(Opts::new(
+            "failed_handshakes_total",
+            "Shows the number of failed handshakes",
+        ))?;
         let participants_connected_total = IntCounter::with_opts(Opts::new(
             "participants_connected_total",
             "Shows the number of participants connected to the network",
@@ -110,6 +123,8 @@ impl NetworkMetrics {
         Ok(Self {
             listen_requests_total,
             connect_requests_total,
+            incoming_connections_total,
+            failed_handshakes_total,
             participants_connected_total,
             participants_disconnected_total,
             participants_channel_ids,
@@ -125,6 +140,8 @@ impl NetworkMetrics {
     pub fn register(&self, registry: &Registry) -> Result<(), Box<dyn Error>> {
         registry.register(Box::new(self.listen_requests_total.clone()))?;
         registry.register(Box::new(self.connect_requests_total.clone()))?;
+        registry.register(Box::new(self.incoming_connections_total.clone()))?;
+        registry.register(Box::new(self.failed_handshakes_total.clone()))?;
         registry.register(Box::new(self.participants_connected_total.clone()))?;
         registry.register(Box::new(self.participants_disconnected_total.clone()))?;
         registry.register(Box::new(self.participants_channel_ids.clone()))?;
@@ -137,6 +154,11 @@ impl NetworkMetrics {
         Ok(())
     }
 
+    pub(crate) fn connect_requests_cache(&self, protocol: &ProtocolAddr) -> prometheus::IntCounter {
+        self.incoming_connections_total
+            .with_label_values(&[protocol_name(protocol)])
+    }
+
     pub(crate) fn channels_connected(&self, remote_p: &str, no: usize, cid: Cid) {
         self.channels_connected_total
             .with_label_values(&[remote_p])
diff --git a/network/src/scheduler.rs b/network/src/scheduler.rs
index abb54b6578..5d02c24876 100644
--- a/network/src/scheduler.rs
+++ b/network/src/scheduler.rs
@@ -384,6 +384,8 @@ impl Scheduler {
         s2a_listen_result_s: oneshot::Sender<io::Result<()>>,
     ) {
         trace!(?addr, "Start up channel creator");
+        #[cfg(feature = "metrics")]
+        let mcache = self.metrics.connect_requests_cache(&addr);
         match addr {
             ProtocolAddr::Tcp(addr) => {
                 let listener = match net::TcpListener::bind(addr).await {
@@ -395,7 +397,7 @@ impl Scheduler {
                         info!(
                             ?addr,
                             ?e,
-                            "Tcp bind error durin listener startup"
+                            "Tcp bind error during listener startup"
                         );
                         s2a_listen_result_s.send(Err(e)).unwrap();
                         return;
@@ -414,8 +416,10 @@ impl Scheduler {
                             continue;
                         },
                     };
-                    info!("Accepting Tcp from: {}", remote_addr);
+                    #[cfg(feature = "metrics")]
+                    mcache.inc();
                     let cid = self.channel_ids.fetch_add(1, Ordering::Relaxed);
+                    info!(?remote_addr, ?cid, "Accepting Tcp from");
                     self.init_protocol(Protocols::new_tcp(stream, cid, Arc::clone(&self.protocol_metrics)), cid, None, true)
                         .await;
                 }
@@ -433,8 +437,10 @@ impl Scheduler {
                 } {
                     let (remote_to_local_s, remote_to_local_r) = mpsc::channel(Self::MPSC_CHANNEL_BOUND);
                     local_remote_to_local_s.send(remote_to_local_s).unwrap();
-                    info!(?addr, "Accepting Mpsc from");
+                    #[cfg(feature = "metrics")]
+                    mcache.inc();
                     let cid = self.channel_ids.fetch_add(1, Ordering::Relaxed);
+                    info!(?addr, ?cid, "Accepting Mpsc from");
                     self.init_protocol(Protocols::new_mpsc(local_to_remote_s, remote_to_local_r, cid, Arc::clone(&self.protocol_metrics)), cid, None, true)
                         .await;
                 }
@@ -650,6 +656,8 @@ impl Scheduler {
                     },
                     Err(e) => {
                         debug!(?cid, ?e, "Handshake from a new connection failed");
+                        #[cfg(feature = "metrics")]
+                        metrics.failed_handshakes_total.inc();
                         if let Some(pid_oneshot) = s2a_return_pid_s {
                             // someone is waiting with `connect`, so give them their Error
                             trace!(?cid, "returning the Err to api who requested the connect");