From 02b805976a29afb115885904d8324a90edf459f7 Mon Sep 17 00:00:00 2001
From: ilya <ilya@cow.fi>
Date: Thu, 9 Jan 2025 17:18:10 +0000
Subject: [PATCH] Bad token detector metrics (#3228)

# Description
Adds Prometheus metrics for driver's bad token detector strategies, so
it will be possible to show how many tokens were marked as `bad` by each
solver and strategy. It would be helpful to identify solvers that
perform as not expected and to analyze the work of each detection
strategy individually.

## How to test
Create 2 grafana panels: the one that shows detected tokens by solver
and another - by strategy.

Added `hotfix` label to deploy it asap.

(cherry picked from commit b46fc728dde8fc648108f117c3e13c84481f4dd5)
---
 .../domain/competition/bad_tokens/metrics.rs  | 20 +++++++++++++++++--
 .../competition/bad_tokens/simulation.rs      |  4 +++-
 crates/driver/src/infra/api/mod.rs            |  1 +
 crates/driver/src/infra/observe/metrics.rs    |  3 +++
 crates/driver/src/infra/observe/mod.rs        |  2 +-
 5 files changed, 26 insertions(+), 4 deletions(-)

diff --git a/crates/driver/src/domain/competition/bad_tokens/metrics.rs b/crates/driver/src/domain/competition/bad_tokens/metrics.rs
index c6951fbea6..daaad1ee7c 100644
--- a/crates/driver/src/domain/competition/bad_tokens/metrics.rs
+++ b/crates/driver/src/domain/competition/bad_tokens/metrics.rs
@@ -1,6 +1,9 @@
 use {
     super::Quality,
-    crate::domain::eth,
+    crate::{
+        domain::eth,
+        infra::{observe::metrics, solver},
+    },
     dashmap::DashMap,
     std::{
         sync::Arc,
@@ -27,6 +30,7 @@ pub struct Detector {
     counter: Arc<DashMap<eth::TokenAddress, TokenStatistics>>,
     log_only: bool,
     token_freeze_time: Duration,
+    solver: solver::Name,
 }
 
 impl Detector {
@@ -35,6 +39,7 @@ impl Detector {
         required_measurements: u32,
         log_only: bool,
         token_freeze_time: Duration,
+        solver: solver::Name,
     ) -> Self {
         Self {
             failure_ratio,
@@ -42,6 +47,7 @@ impl Detector {
             counter: Default::default(),
             log_only,
             token_freeze_time,
+            solver,
         }
     }
 
@@ -120,6 +126,10 @@ impl Detector {
                 tokens = ?new_unsupported_tokens,
                 "mark tokens as unsupported"
             );
+            metrics::get()
+                .bad_tokens_detected
+                .with_label_values(&[&self.solver.0, "metrics"])
+                .inc_by(new_unsupported_tokens.len() as u64);
         }
     }
 }
@@ -133,7 +143,13 @@ mod tests {
     #[tokio::test]
     async fn unfreeze_bad_tokens() {
         const FREEZE_DURATION: Duration = Duration::from_millis(50);
-        let detector = Detector::new(0.5, 2, false, FREEZE_DURATION);
+        let detector = Detector::new(
+            0.5,
+            2,
+            false,
+            FREEZE_DURATION,
+            solver::Name("mysolver".to_string()),
+        );
 
         let token_a = eth::TokenAddress(eth::ContractAddress(H160([1; 20])));
         let token_b = eth::TokenAddress(eth::ContractAddress(H160([2; 20])));
diff --git a/crates/driver/src/domain/competition/bad_tokens/simulation.rs b/crates/driver/src/domain/competition/bad_tokens/simulation.rs
index dd8eb5e0f4..9b593b8726 100644
--- a/crates/driver/src/domain/competition/bad_tokens/simulation.rs
+++ b/crates/driver/src/domain/competition/bad_tokens/simulation.rs
@@ -8,7 +8,7 @@ use {
             },
             eth,
         },
-        infra,
+        infra::{self, observe::metrics},
     },
     futures::FutureExt,
     model::interaction::InteractionData,
@@ -100,6 +100,8 @@ impl Detector {
                         }
                         Ok(TokenQuality::Bad { reason }) => {
                             tracing::debug!(reason, token=?sell_token.0, "cache token as unsupported");
+                            // All solvers share the same cache for the simulation detector, so there is no need to specify the solver name here.
+                            metrics::get().bad_tokens_detected.with_label_values(&["any", "simulation"]).inc();
                             inner
                                 .cache
                                 .update_quality(sell_token, false, now);
diff --git a/crates/driver/src/infra/api/mod.rs b/crates/driver/src/infra/api/mod.rs
index de7897cfa8..ccbdc5a178 100644
--- a/crates/driver/src/infra/api/mod.rs
+++ b/crates/driver/src/infra/api/mod.rs
@@ -84,6 +84,7 @@ impl Api {
                     bad_token_config.metrics_strategy_required_measurements,
                     bad_token_config.metrics_strategy_log_only,
                     bad_token_config.metrics_strategy_token_freeze_time,
+                    name.clone(),
                 ));
             }
 
diff --git a/crates/driver/src/infra/observe/metrics.rs b/crates/driver/src/infra/observe/metrics.rs
index 330e250867..237206c668 100644
--- a/crates/driver/src/infra/observe/metrics.rs
+++ b/crates/driver/src/infra/observe/metrics.rs
@@ -19,6 +19,9 @@ pub struct Metrics {
     /// The results of the mempool submission.
     #[metric(labels("mempool", "result"))]
     pub mempool_submission: prometheus::IntCounterVec,
+    /// How many tokens detected by specific solver and strategy.
+    #[metric(labels("solver", "strategy"))]
+    pub bad_tokens_detected: prometheus::IntCounterVec,
 }
 
 /// Setup the metrics registry.
diff --git a/crates/driver/src/infra/observe/mod.rs b/crates/driver/src/infra/observe/mod.rs
index f8c1394e54..c3c0e34a7b 100644
--- a/crates/driver/src/infra/observe/mod.rs
+++ b/crates/driver/src/infra/observe/mod.rs
@@ -28,7 +28,7 @@ use {
     url::Url,
 };
 
-mod metrics;
+pub mod metrics;
 
 /// Setup the observability. The log argument configures the tokio tracing
 /// framework.