Track dropped spans and logs due to full buffer

opentelemetry-sdk/src/logs/log_processor.rs

@@ -13,7 +13,7 @@
 use opentelemetry::logs::Severity;
 use opentelemetry::{otel_debug, otel_error, otel_warn, InstrumentationScope};
 
-use std::sync::atomic::AtomicBool;
+use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
 use std::{cmp::min, env, sync::Mutex};
 use std::{
     fmt::{self, Debug, Formatter},
@@ -154,6 +154,13 @@
 /// them at a pre-configured interval.
 pub struct BatchLogProcessor<R: RuntimeChannel> {
     message_sender: R::Sender<BatchMessage>,
+
+    // Track dropped logs. We'll log this at shutdown and also emit
+    // as a metric.
+    dropped_logs_count: AtomicUsize,
+
+    // Track the maximum queue size that was configured for this processor
+    max_queue_size: usize,
 }
 
 impl<R: RuntimeChannel> Debug for BatchLogProcessor<R> {
@@ -172,11 +179,13 @@
         )));
 
         // TODO - Implement throttling to prevent error flooding when the queue is full or closed.
-        if let Err(err) = result {
-            otel_error!(
-                name: "BatchLogProcessor.Export.Error",
-                error = format!("{}", err)
-            );
+        if result.is_err() {
+            // Increment dropped logs counter and metric. The first time we have top drop a log,
+            // emit a warning.
+            if self.dropped_logs_count.fetch_add(1, Ordering::Relaxed) == 0 {
+                otel_warn!(name: "BatchLogProcessor.LogDroppingStarted",
+                    message = "Beginning to drop log messages due to full exporter queue.");
+            }
         }
     }
 
@@ -192,6 +201,17 @@
     }
 
     fn shutdown(&self) -> LogResult<()> {
+        let dropped_logs = self.dropped_logs_count.load(Ordering::Relaxed);
+        let max_queue_size = self.max_queue_size;
+        if dropped_logs > 0 {
+            otel_warn!(
+                name: "BatchLogProcessor.LogsDropped",
+                dropped_logs_count = dropped_logs,
+                max_queue_size = max_queue_size,
+                message = "Logs were dropped due to a full or closed queue. The count represents the total count of lost logs in the lifetime of the BatchLogProcessor. Consider increasing the queue size and/or decrease delay between intervals."
+            );
+        }
+
         let (res_sender, res_receiver) = oneshot::channel();
         self.message_sender
             .try_send(BatchMessage::Shutdown(res_sender))
@@ -215,6 +235,7 @@
         let (message_sender, message_receiver) =
             runtime.batch_message_channel(config.max_queue_size);
         let inner_runtime = runtime.clone();
+        let max_queue_size = config.max_queue_size;
 
         // Spawn worker process via user-defined spawn function.
         runtime.spawn(Box::pin(async move {
@@ -296,8 +317,13 @@
                 }
             }
         }));
+
         // Return batch processor with link to worker
-        BatchLogProcessor { message_sender }
+        BatchLogProcessor {
+            message_sender,
+            dropped_logs_count: AtomicUsize::new(0),
+            max_queue_size,
+        }
     }
 
     /// Create a new batch processor builder

opentelemetry-sdk/src/trace/span_processor.rs

@@ -45,12 +45,13 @@
     stream::{self, FusedStream, FuturesUnordered},
     StreamExt as _,
 };
-use opentelemetry::{otel_debug, otel_error};
+use opentelemetry::{otel_debug, otel_error, otel_warn};
 use opentelemetry::{
     trace::{TraceError, TraceResult},
     Context,
 };
 use std::cmp::min;
+use std::sync::atomic::{AtomicUsize, Ordering};
 use std::sync::{Arc, Mutex};
 use std::{env, fmt, str::FromStr, time::Duration};
 
@@ -227,6 +228,12 @@
 /// [`async-std`]: https://async.rs
 pub struct BatchSpanProcessor<R: RuntimeChannel> {
     message_sender: R::Sender<BatchMessage>,
+
+    // Track dropped spans
+    dropped_spans_count: AtomicUsize,
+
+    // Track the maximum queue size that was configured for this processor
+    max_queue_size: usize,
 }
 
 impl<R: RuntimeChannel> fmt::Debug for BatchSpanProcessor<R> {
@@ -249,11 +256,14 @@
 
         let result = self.message_sender.try_send(BatchMessage::ExportSpan(span));
 
-        if let Err(err) = result {
-            otel_debug!(
-                name: "BatchSpanProcessor.OnEnd.ExportQueueingFailed",
-                reason = format!("{:?}", TraceError::Other(err.into()))
-            );
+        // If the queue is full, and we can't buffer a span
+        if result.is_err() {
+            // Increment the number of dropped spans. If this is the first time we've had to drop,
+            // emit a warning.
+            if self.dropped_spans_count.fetch_add(1, Ordering::Relaxed) == 0 {
+                otel_warn!(name: "BatchSpanProcessor.SpanDroppingStarted",
+                    message = "Beginning to drop span messages due to full exporter queue.");
+            }
         }
     }
 
@@ -269,6 +279,17 @@
     }
 
     fn shutdown(&self) -> TraceResult<()> {
+        let dropped_spans = self.dropped_spans_count.load(Ordering::Relaxed);
+        let max_queue_size = self.max_queue_size;
+        if dropped_spans > 0 {
+            otel_warn!(
+                name: "BatchSpanProcessor.Shutdown",
+                dropped_spans = dropped_spans,
+                max_queue_size = max_queue_size,
+                message = "Spans were dropped due to a full or closed queue. The count represents the total count of lost logs in the lifetime of the BatchLogProcessor. Consider increasing the queue size and/or decrease delay between intervals."
+            );
+        }
+
         let (res_sender, res_receiver) = oneshot::channel();
         self.message_sender
             .try_send(BatchMessage::Shutdown(res_sender))
@@ -469,6 +490,8 @@
         let (message_sender, message_receiver) =
             runtime.batch_message_channel(config.max_queue_size);
 
+        let max_queue_size = config.max_queue_size;
+
         let inner_runtime = runtime.clone();
         // Spawn worker process via user-defined spawn function.
         runtime.spawn(Box::pin(async move {
@@ -493,7 +516,11 @@
         }));
 
         // Return batch processor with link to worker
-        BatchSpanProcessor { message_sender }
+        BatchSpanProcessor {
+            message_sender,
+            dropped_spans_count: AtomicUsize::new(0),
+            max_queue_size,
+        }
     }
 
     /// Create a new batch processor builder