From a72c4c374afea785929a0a0a98d7840faf3df34c Mon Sep 17 00:00:00 2001
From: Felix Leupold <felixleupold90@gmail.com>
Date: Tue, 28 Nov 2023 13:07:26 +0100
Subject: [PATCH] Add metrics for the shadow autopilot (#2079)

# Description
When changing the shadow infra over to the colocated architecture, we
need to also create a replacement for the shadow [solver
dashboard](https://g-0263500beb.grafana-workspace.eu-central-1.amazonaws.com/d/AGoBVqH7z/gpv2-shadow-traffic-staging?orgId=1).
This PR prepares this.

# Changes
<!-- List of detailed changes (how the change is accomplished) -->

- [ ] Implement dummy liveness check for shadow runloop
- [ ] Serve metrics in shadow mode
- [ ] Record a few more metrics (mainly the winner in each auction +
total orders in the book)

## How to test
Running locally and checking http://localhost:9589/metrics

Will update the dashboard once this is merged.
---
 crates/autopilot/src/run.rs    |  2 ++
 crates/autopilot/src/shadow.rs | 20 +++++++++++++++++++-
 2 files changed, 21 insertions(+), 1 deletion(-)

diff --git a/crates/autopilot/src/run.rs b/crates/autopilot/src/run.rs
index 6e90f8d3de..11bd9f4ad8 100644
--- a/crates/autopilot/src/run.rs
+++ b/crates/autopilot/src/run.rs
@@ -672,6 +672,8 @@ async fn shadow_mode(args: Arguments) -> ! {
         .await
     };
 
+    shared::metrics::serve_metrics(Arc::new(shadow::Liveness), args.metrics_address);
+
     let shadow = shadow::RunLoop::new(
         orderbook,
         drivers,
diff --git a/crates/autopilot/src/shadow.rs b/crates/autopilot/src/shadow.rs
index 8e38be329c..1c2c9c74f2 100644
--- a/crates/autopilot/src/shadow.rs
+++ b/crates/autopilot/src/shadow.rs
@@ -21,11 +21,20 @@ use {
     number::nonzero::U256 as NonZeroU256,
     primitive_types::{H160, U256},
     rand::seq::SliceRandom,
-    shared::token_list::AutoUpdatingTokenList,
+    shared::{metrics::LivenessChecking, token_list::AutoUpdatingTokenList},
     std::{cmp, time::Duration},
     tracing::Instrument,
 };
 
+pub struct Liveness;
+#[async_trait::async_trait]
+impl LivenessChecking for Liveness {
+    async fn is_alive(&self) -> bool {
+        // can we somehow check that we keep processing auctions?
+        true
+    }
+}
+
 pub struct RunLoop {
     orderbook: protocol::Orderbook,
     drivers: Vec<Driver>,
@@ -108,6 +117,7 @@ impl RunLoop {
     async fn single_run(&self, id: AuctionId, auction: Auction) {
         tracing::info!("solving");
         Metrics::get().auction.set(id);
+        Metrics::get().orders.set(auction.orders.len() as _);
 
         let mut participants = self.competition(id, &auction).await;
 
@@ -140,6 +150,7 @@ impl RunLoop {
                 .performance_rewards
                 .with_label_values(&[&driver.name])
                 .inc_by(reward.to_f64_lossy());
+            Metrics::get().wins.with_label_values(&[&driver.name]).inc();
         }
 
         let hex = |bytes: &[u8]| format!("0x{}", hex::encode(bytes));
@@ -292,6 +303,9 @@ struct Metrics {
     /// Tracks the last seen auction.
     auction: prometheus::IntGauge,
 
+    /// Tracks the number of orders in the auction.
+    orders: prometheus::IntGauge,
+
     /// Tracks the result of every driver.
     #[metric(labels("driver", "result"))]
     results: prometheus::IntCounterVec,
@@ -299,6 +313,10 @@ struct Metrics {
     /// Tracks the approximate performance rewards per driver
     #[metric(labels("driver"))]
     performance_rewards: prometheus::CounterVec,
+
+    /// Tracks the winner of every auction.
+    #[metric(labels("driver"))]
+    wins: prometheus::CounterVec,
 }
 
 impl Metrics {