Merge pull request #37 from fpco/error-count

Track total error count
fpco · Jul 22, 2024 · 399afa1 · 399afa1
2 parents 1aa3998 + a7406e4
commit 399afa1
Show file tree

Hide file tree

Showing 2 changed files with 25 additions and 4 deletions.
diff --git a/packages/cosmos/src/client/node.rs b/packages/cosmos/src/client/node.rs
@@ -40,14 +40,18 @@ struct NodeInner {
 pub(crate) struct QueryCount {
     pub(crate) first_request: Option<DateTime<Utc>>,
     pub(crate) total_query_count: u64,
+    pub(crate) total_error_count: u64,
 }
 
 impl QueryCount {
-    pub(crate) fn incr(&mut self) {
+    pub(crate) fn incr(&mut self, is_error: bool) {
         if self.first_request.is_none() {
             self.first_request = Some(Utc::now());
         }
         self.total_query_count += 1;
+        if is_error {
+            self.total_error_count += 1;
+        }
     }
 }
 
@@ -188,7 +192,10 @@ impl Node {
     }
 
     pub(super) fn log_query_result(&self, res: QueryResult) {
-        self.node_inner.query_count.write().incr();
+        self.node_inner.query_count.write().incr(match res {
+            QueryResult::Success => false,
+            QueryResult::NetworkError { .. } | QueryResult::OtherError => true,
+        });
         let mut guard = self.node_inner.last_error.write();
         match res {
             QueryResult::Success | QueryResult::OtherError => {
@@ -227,6 +234,7 @@ impl Node {
         let QueryCount {
             first_request,
             total_query_count,
+            total_error_count,
         } = *self.node_inner.query_count.read();
         SingleNodeHealthReport {
             grpc_url: self.node_inner.grpc_url.clone(),
@@ -253,6 +261,7 @@ impl Node {
             }),
             first_request,
             total_query_count,
+            total_error_count,
         }
     }
 

diff --git a/packages/cosmos/src/error.rs b/packages/cosmos/src/error.rs
@@ -1059,6 +1059,7 @@ pub struct SingleNodeHealthReport {
     pub error_count: usize,
     pub first_request: Option<DateTime<Utc>>,
     pub total_query_count: u64,
+    pub total_error_count: u64,
 }
 
 /// Describes the health status of an individual node.
@@ -1133,11 +1134,22 @@ impl Display for SingleNodeHealthReport {
                 ConversionError::Overflow => "Overflow when converting since".to_owned(),
                 ConversionError::DivideByZero => "since is 0".to_owned(),
             });
+            let errors_per_minute = (|| {
+                let since = u64::try_from(since).map_err(|_| ConversionError::Overflow)?;
+                self.total_error_count
+                    .checked_div(since)
+                    .ok_or(ConversionError::DivideByZero)
+                    .map(|item| item.to_string())
+            })()
+            .unwrap_or_else(|err| match err {
+                ConversionError::Overflow => "Overflow when converting since".to_owned(),
+                ConversionError::DivideByZero => "since is 0".to_owned(),
+            });
 
             write!(
                 f,
-                ". First request: {} (Since {} minutes). Total queries: {} (RPM: {})",
-                first_request, since, self.total_query_count, rate_per_minute
+                ". First request: {} (Since {} minutes). Total queries: {} (RPM: {}). Total errors: {} (RPM: {})",
+                first_request, since, self.total_query_count, rate_per_minute, self.total_error_count, errors_per_minute
             )?;
         }
         Ok(())