diff options
| -rw-r--r-- | src/core/lib/debug/stats_data.c | 3 | ||||
| -rw-r--r-- | src/core/lib/debug/stats_data.h | 4 | ||||
| -rw-r--r-- | src/core/lib/debug/stats_data.yaml | 4 | ||||
| -rw-r--r-- | src/core/lib/debug/stats_data_bq_schema.sql | 3 | ||||
| -rw-r--r-- | src/core/lib/surface/completion_queue.c | 4 | ||||
| -rw-r--r-- | tools/run_tests/performance/massage_qps_stats.py | 1 | ||||
| -rw-r--r-- | tools/run_tests/performance/scenario_result_schema.json | 10 |
7 files changed, 28 insertions, 1 deletions
diff --git a/src/core/lib/debug/stats_data.c b/src/core/lib/debug/stats_data.c index c0aec63c1d..41d9e95e29 100644 --- a/src/core/lib/debug/stats_data.c +++ b/src/core/lib/debug/stats_data.c @@ -111,6 +111,7 @@ const char *grpc_stats_counter_name[GRPC_STATS_COUNTER_COUNT] = { "executor_push_retries", "server_requested_calls", "server_slowpath_requests_queued", + "cq_failed_queue_trylocks", }; const char *grpc_stats_counter_doc[GRPC_STATS_COUNTER_COUNT] = { "Number of client side calls created by this process", @@ -220,6 +221,8 @@ 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", }; 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 28dab00117..0a2b323bdf 100644 --- a/src/core/lib/debug/stats_data.h +++ b/src/core/lib/debug/stats_data.h @@ -113,6 +113,7 @@ 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_FAILED_QUEUE_TRYLOCKS, GRPC_STATS_COUNTER_COUNT } grpc_stats_counters; extern const char *grpc_stats_counter_name[GRPC_STATS_COUNTER_COUNT]; @@ -417,6 +418,9 @@ 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_FAILED_QUEUE_TRYLOCKS(exec_ctx) \ + GRPC_STATS_INC_COUNTER((exec_ctx), \ + GRPC_STATS_COUNTER_CQ_FAILED_QUEUE_TRYLOCKS) #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 b5c15ff55c..5efe353200 100644 --- a/src/core/lib/debug/stats_data.yaml +++ b/src/core/lib/debug/stats_data.yaml @@ -270,3 +270,7 @@ - counter: server_slowpath_requests_queued doc: How many times was the server slow path taken (indicates too few outstanding requests) +# cq +- counter: cq_failed_queue_trylocks + doc: Number of lock (trylock) acquisition failures on completion queue event + queue. High value here indicates high contention on completion queues diff --git a/src/core/lib/debug/stats_data_bq_schema.sql b/src/core/lib/debug/stats_data_bq_schema.sql index f96e40c00e..6ad58e2530 100644 --- a/src/core/lib/debug/stats_data_bq_schema.sql +++ b/src/core/lib/debug/stats_data_bq_schema.sql @@ -85,4 +85,5 @@ 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_failed_queue_trylocks_per_iteration:FLOAT diff --git a/src/core/lib/surface/completion_queue.c b/src/core/lib/surface/completion_queue.c index fed66e3a20..037b10e848 100644 --- a/src/core/lib/surface/completion_queue.c +++ b/src/core/lib/surface/completion_queue.c @@ -378,6 +378,10 @@ static grpc_cq_completion *cq_event_queue_pop(grpc_cq_event_queue *q) { if (gpr_spinlock_trylock(&q->queue_lock)) { c = (grpc_cq_completion *)gpr_mpscq_pop(&q->queue); gpr_spinlock_unlock(&q->queue_lock); + } else { + grpc_exec_ctx exec_ctx = GRPC_EXEC_CTX_INIT; + GRPC_STATS_INC_CQ_FAILED_QUEUE_TRYLOCKS(&exec_ctx); + grpc_exec_ctx_finish(&exec_ctx); } if (c) { diff --git a/tools/run_tests/performance/massage_qps_stats.py b/tools/run_tests/performance/massage_qps_stats.py index 9b9355308a..49d165d94f 100644 --- a/tools/run_tests/performance/massage_qps_stats.py +++ b/tools/run_tests/performance/massage_qps_stats.py @@ -108,6 +108,7 @@ def massage_qps_stats(scenario_result): stats["core_executor_push_retries"] = massage_qps_stats_helpers.counter(core_stats, "executor_push_retries") stats["core_server_requested_calls"] = massage_qps_stats_helpers.counter(core_stats, "server_requested_calls") stats["core_server_slowpath_requests_queued"] = massage_qps_stats_helpers.counter(core_stats, "server_slowpath_requests_queued") + stats["core_cq_failed_queue_trylocks"] = massage_qps_stats_helpers.counter(core_stats, "cq_failed_queue_trylocks") h = massage_qps_stats_helpers.histogram(core_stats, "call_initial_size") stats["core_call_initial_size"] = ",".join("%f" % x for x in h.buckets) stats["core_call_initial_size_bkts"] = ",".join("%f" % x for x in h.boundaries) diff --git a/tools/run_tests/performance/scenario_result_schema.json b/tools/run_tests/performance/scenario_result_schema.json index 2f0fd916d4..11d8dba5d3 100644 --- a/tools/run_tests/performance/scenario_result_schema.json +++ b/tools/run_tests/performance/scenario_result_schema.json @@ -552,6 +552,11 @@ }, { "mode": "NULLABLE", + "name": "core_cq_failed_queue_trylocks", + "type": "INTEGER" + }, + { + "mode": "NULLABLE", "name": "core_call_initial_size", "type": "STRING" }, @@ -1344,6 +1349,11 @@ }, { "mode": "NULLABLE", + "name": "core_cq_failed_queue_trylocks", + "type": "INTEGER" + }, + { + "mode": "NULLABLE", "name": "core_call_initial_size", "type": "STRING" }, |