Merge pull request #12808 from sreecha/cq_stats

Counter for number of trylocks (successes/failures) and transient pop-failures in cq

5 files changed, 49 insertions, 2 deletions

diff --git a/src/core/lib/debug/stats_data.cc b/src/core/lib/debug/stats_data.cc
index 5bd7884e28..b4ae8f312c 100644
--- a/src/core/lib/debug/stats_data.cc
+++ b/src/core/lib/debug/stats_data.cc
@@ -112,6 +112,9 @@ const char *grpc_stats_counter_name[GRPC_STATS_COUNTER_COUNT] = {
     "executor_push_retries",
     "server_requested_calls",
     "server_slowpath_requests_queued",
+    "cq_ev_queue_trylock_failures",
+    "cq_ev_queue_trylock_successes",
+    "cq_ev_queue_transient_pop_failures",
 };
 const char *grpc_stats_counter_doc[GRPC_STATS_COUNTER_COUNT] = {
     "Number of client side calls created by this process",
@@ -222,6 +225,12 @@ const char *grpc_stats_counter_doc[GRPC_STATS_COUNTER_COUNT] = {
     "How many calls were requested (not necessarily received) by the server",
     "How many times was the server slow path taken (indicates too few "
     "outstanding requests)",
+    "Number of lock (trylock) acquisition failures on completion queue event "
+    "queue. High value here indicates high contention on completion queues",
+    "Number of lock (trylock) acquisition successes on completion queue event "
+    "queue.",
+    "Number of times NULL was popped out of completion queue's event queue "
+    "even though the event queue was not empty",
 };
 const char *grpc_stats_histogram_name[GRPC_STATS_HISTOGRAM_COUNT] = {
     "call_initial_size",
diff --git a/src/core/lib/debug/stats_data.h b/src/core/lib/debug/stats_data.h
index d8e4e7d264..d4c4437ca0 100644
--- a/src/core/lib/debug/stats_data.h
+++ b/src/core/lib/debug/stats_data.h
@@ -118,6 +118,9 @@ typedef enum {
   GRPC_STATS_COUNTER_EXECUTOR_PUSH_RETRIES,
   GRPC_STATS_COUNTER_SERVER_REQUESTED_CALLS,
   GRPC_STATS_COUNTER_SERVER_SLOWPATH_REQUESTS_QUEUED,
+  GRPC_STATS_COUNTER_CQ_EV_QUEUE_TRYLOCK_FAILURES,
+  GRPC_STATS_COUNTER_CQ_EV_QUEUE_TRYLOCK_SUCCESSES,
+  GRPC_STATS_COUNTER_CQ_EV_QUEUE_TRANSIENT_POP_FAILURES,
   GRPC_STATS_COUNTER_COUNT
 } grpc_stats_counters;
 extern const char *grpc_stats_counter_name[GRPC_STATS_COUNTER_COUNT];
@@ -425,6 +428,15 @@ typedef enum {
 #define GRPC_STATS_INC_SERVER_SLOWPATH_REQUESTS_QUEUED(exec_ctx) \
   GRPC_STATS_INC_COUNTER((exec_ctx),                             \
                          GRPC_STATS_COUNTER_SERVER_SLOWPATH_REQUESTS_QUEUED)
+#define GRPC_STATS_INC_CQ_EV_QUEUE_TRYLOCK_FAILURES(exec_ctx) \
+  GRPC_STATS_INC_COUNTER((exec_ctx),                          \
+                         GRPC_STATS_COUNTER_CQ_EV_QUEUE_TRYLOCK_FAILURES)
+#define GRPC_STATS_INC_CQ_EV_QUEUE_TRYLOCK_SUCCESSES(exec_ctx) \
+  GRPC_STATS_INC_COUNTER((exec_ctx),                           \
+                         GRPC_STATS_COUNTER_CQ_EV_QUEUE_TRYLOCK_SUCCESSES)
+#define GRPC_STATS_INC_CQ_EV_QUEUE_TRANSIENT_POP_FAILURES(exec_ctx) \
+  GRPC_STATS_INC_COUNTER(                                           \
+      (exec_ctx), GRPC_STATS_COUNTER_CQ_EV_QUEUE_TRANSIENT_POP_FAILURES)
 #define GRPC_STATS_INC_CALL_INITIAL_SIZE(exec_ctx, value) \
   grpc_stats_inc_call_initial_size((exec_ctx), (int)(value))
 void grpc_stats_inc_call_initial_size(grpc_exec_ctx *exec_ctx, int x);
diff --git a/src/core/lib/debug/stats_data.yaml b/src/core/lib/debug/stats_data.yaml
index 5c0ab2262e..00e81e74e2 100644
--- a/src/core/lib/debug/stats_data.yaml
+++ b/src/core/lib/debug/stats_data.yaml
@@ -272,4 +272,14 @@
 - counter: server_slowpath_requests_queued
   doc: How many times was the server slow path taken (indicates too few
        outstanding requests)
+# cq
+- counter: cq_ev_queue_trylock_failures
+  doc: Number of lock (trylock) acquisition failures on completion queue event
+       queue. High value here indicates high contention on completion queues
+- counter: cq_ev_queue_trylock_successes
+  doc: Number of lock (trylock) acquisition successes on completion queue event
+       queue.
+- counter: cq_ev_queue_transient_pop_failures
+  doc: Number of times NULL was popped out of completion queue's event queue
+       even though the event queue was not empty
 
diff --git a/src/core/lib/debug/stats_data_bq_schema.sql b/src/core/lib/debug/stats_data_bq_schema.sql
index 54869977b0..f34b0c25a0 100644
--- a/src/core/lib/debug/stats_data_bq_schema.sql
+++ b/src/core/lib/debug/stats_data_bq_schema.sql
@@ -86,4 +86,7 @@ executor_wakeup_initiated_per_iteration:FLOAT,
 executor_queue_drained_per_iteration:FLOAT,
 executor_push_retries_per_iteration:FLOAT,
 server_requested_calls_per_iteration:FLOAT,
-server_slowpath_requests_queued_per_iteration:FLOAT
+server_slowpath_requests_queued_per_iteration:FLOAT,
+cq_ev_queue_trylock_failures_per_iteration:FLOAT,
+cq_ev_queue_trylock_successes_per_iteration:FLOAT,
+cq_ev_queue_transient_pop_failures_per_iteration:FLOAT
diff --git a/src/core/lib/surface/completion_queue.cc b/src/core/lib/surface/completion_queue.cc
index 36b4b835f8..21664f03c8 100644
--- a/src/core/lib/surface/completion_queue.cc
+++ b/src/core/lib/surface/completion_queue.cc
@@ -362,11 +362,24 @@ static bool cq_event_queue_push(grpc_cq_event_queue *q, grpc_cq_completion *c) {
 
 static grpc_cq_completion *cq_event_queue_pop(grpc_cq_event_queue *q) {
   grpc_cq_completion *c = NULL;
+  grpc_exec_ctx exec_ctx = GRPC_EXEC_CTX_INIT;
+
   if (gpr_spinlock_trylock(&q->queue_lock)) {
-    c = (grpc_cq_completion *)gpr_mpscq_pop(&q->queue);
+    GRPC_STATS_INC_CQ_EV_QUEUE_TRYLOCK_SUCCESSES(&exec_ctx);
+
+    bool is_empty = false;
+    c = (grpc_cq_completion *)gpr_mpscq_pop_and_check_end(&q->queue, &is_empty);
     gpr_spinlock_unlock(&q->queue_lock);
+
+    if (c == NULL && !is_empty) {
+      GRPC_STATS_INC_CQ_EV_QUEUE_TRANSIENT_POP_FAILURES(&exec_ctx);
+    }
+  } else {
+    GRPC_STATS_INC_CQ_EV_QUEUE_TRYLOCK_FAILURES(&exec_ctx);
   }
 
+  grpc_exec_ctx_finish(&exec_ctx);
+
   if (c) {
     gpr_atm_no_barrier_fetch_add(&q->num_queue_items, -1);
   }