Server error stats (#581)

* Server message hook

* Report server errors

* Remove server_errors from show pools

* save

* accerate arm builds

Author: levkk

Dockerfile

@@ -12,6 +12,9 @@ WORKDIR /build
 
 RUN rm /bin/sh && ln -s /bin/bash /bin/sh
 RUN source ~/.cargo/env && \
+    if [ "$(uname -m)" = "aarch64" ] || [ "$(uname -m)" = "arm64" ]; then \
+        export RUSTFLAGS="-Ctarget-feature=+lse"; \
+    fi && \
     cd pgdog && \
     cargo build --release
 

pgdog/src/admin/show_stats.rs

@@ -42,6 +42,7 @@ impl Command for ShowStats {
                         Field::numeric(&format!("{}_server_parse_count", prefix)),
                         Field::numeric(&format!("{}_bind_count", prefix)),
                         Field::numeric(&format!("{}_close_count", prefix)),
+                        Field::numeric(&format!("{}_errors", prefix)),
                     ]
                 })
                 .collect::<Vec<Field>>(),
@@ -82,7 +83,8 @@ impl Command for ShowStats {
                             .add(stat.wait_time.as_millis() as u64)
                             .add(stat.parse_count)
                             .add(stat.bind_count)
-                            .add(stat.close);
+                            .add(stat.close)
+                            .add(stat.errors);
                     }
 
                     messages.push(dr.message()?);

pgdog/src/backend/pool/stats.rs

@@ -24,6 +24,7 @@ pub struct Counts {
     pub rollbacks: usize,
     pub healthchecks: usize,
     pub close: usize,
+    pub errors: usize,
 }
 
 impl Sub for Counts {
@@ -47,6 +48,7 @@ impl Sub for Counts {
             rollbacks: self.rollbacks.saturating_sub(rhs.rollbacks),
             healthchecks: self.healthchecks.saturating_add(rhs.healthchecks),
             close: self.close.saturating_add(rhs.close),
+            errors: self.errors.saturating_add(rhs.errors),
         }
     }
 }
@@ -70,6 +72,7 @@ impl Div<usize> for Counts {
             rollbacks: self.rollbacks.saturating_div(rhs),
             healthchecks: self.healthchecks.saturating_div(rhs),
             close: self.close.saturating_div(rhs),
+            errors: self.errors.saturating_div(rhs),
         }
     }
 }
@@ -93,6 +96,7 @@ impl Add<BackendCounts> for Counts {
             rollbacks: self.rollbacks + rhs.rollbacks,
             healthchecks: self.healthchecks + rhs.healthchecks,
             close: self.close + rhs.close,
+            errors: self.errors + rhs.errors,
         }
     }
 }
@@ -129,6 +133,7 @@ impl Add for Counts {
             rollbacks: self.rollbacks.saturating_add(rhs.rollbacks),
             healthchecks: self.healthchecks.saturating_add(rhs.healthchecks),
             close: self.close.saturating_add(rhs.close),
+            errors: self.errors.saturating_add(rhs.errors),
         }
     }
 }

pgdog/src/frontend/client/query_engine/hooks/mod.rs

@@ -37,4 +37,12 @@ impl QueryEngineHooks {
     ) -> Result<(), Error> {
         Ok(())
     }
+
+    pub(super) fn on_server_message(
+        &mut self,
+        context: &mut QueryEngineContext<'_>,
+        message: &Message,
+    ) -> Result<(), Error> {
+        Ok(())
+    }
 }

pgdog/src/frontend/client/query_engine/query.rs

@@ -225,6 +225,7 @@ impl QueryEngine {
         if code == 'Z' {
             self.pending_explain = None;
         }
+        self.hooks.on_server_message(context, &message)?;
 
         Ok(())
     }

pgdog/src/stats/pools.rs

@@ -65,6 +65,9 @@ impl Pools {
         let mut avg_query_time = vec![];
         let mut total_close = vec![];
         let mut avg_close = vec![];
+        let mut total_server_errors = vec![];
+        let mut avg_server_errors = vec![];
+
         for (user, cluster) in databases().all() {
             for (shard_num, shard) in cluster.shards().iter().enumerate() {
                 for (role, pool) in shard.pools_with_roles() {
@@ -191,6 +194,16 @@ impl Pools {
                         labels: labels.clone(),
                         measurement: averages.close.into(),
                     });
+
+                    total_server_errors.push(Measurement {
+                        labels: labels.clone(),
+                        measurement: totals.errors.into(),
+                    });
+
+                    avg_server_errors.push(Measurement {
+                        labels: labels.clone(),
+                        measurement: averages.errors.into(),
+                    });
                 }
             }
         }
@@ -372,6 +385,22 @@ impl Pools {
             metric_type: None,
         }));
 
+        metrics.push(Metric::new(PoolMetric {
+            name: "total_server_errors".into(),
+            measurements: total_server_errors,
+            help: "Total number of errors returned by server connections.".into(),
+            unit: None,
+            metric_type: Some("counter".into()),
+        }));
+
+        metrics.push(Metric::new(PoolMetric {
+            name: "avg_server_errors".into(),
+            measurements: avg_server_errors,
+            help: "Average number of errors returned by server connections.".into(),
+            unit: None,
+            metric_type: None,
+        }));
+
         Pools { metrics }
     }
 }