From: Willy Tarreau <w@1wt.eu>
Date: Wed, 24 Apr 2019 06:10:57 +0000 (+0200)
Subject: MINOR: tasks/activity: report the context switch and task wakeup rates
X-Git-Tag: v2.0-dev3~179
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=0212fadd658bfcade0342dea708e398b47e31188;p=thirdparty%2Fhaproxy.git

MINOR: tasks/activity: report the context switch and task wakeup rates

It's particularly useful to spot runaway tasks to see this. The context
switch rate covers all tasklet calls (tasks and I/O handlers) while the
task wakeups only covers tasks picked from the run queue to be executed.
High values there will indicate either an intense traffic or a bug that
mades a task go wild.
---

diff --git a/include/types/activity.h b/include/types/activity.h
index 329513a39e..0c23f76f63 100644
--- a/include/types/activity.h
+++ b/include/types/activity.h
@@ -50,6 +50,8 @@ struct activity {
 	/* one cache line */
 	struct freq_ctr cpust_1s;  // avg amount of half-ms stolen over last second
 	struct freq_ctr_period cpust_15s; // avg amount of half-ms stolen over last 15s
+	struct freq_ctr ctxsw_rate;// context switching rate over last second
+	struct freq_ctr tasks_rate;// task wakeup rate over last second
 	unsigned int avg_loop_us;  // average run time per loop over last 1024 runs
 	unsigned int accepted;     // accepted incoming connections
 	unsigned int accq_pushed;  // accept queue connections pushed
diff --git a/src/cli.c b/src/cli.c
index 843c3d043f..9581369c23 100644
--- a/src/cli.c
+++ b/src/cli.c
@@ -1087,6 +1087,8 @@ static int cli_io_handler_show_activity(struct appctx *appctx)
 	chunk_appendf(&trash, "\nstream:");       for (thr = 0; thr < global.nbthread; thr++) chunk_appendf(&trash, " %u", activity[thr].stream);
 	chunk_appendf(&trash, "\nempty_rq:");     for (thr = 0; thr < global.nbthread; thr++) chunk_appendf(&trash, " %u", activity[thr].empty_rq);
 	chunk_appendf(&trash, "\nlong_rq:");      for (thr = 0; thr < global.nbthread; thr++) chunk_appendf(&trash, " %u", activity[thr].long_rq);
+	chunk_appendf(&trash, "\nctxsw_rate:");   for (thr = 0; thr < global.nbthread; thr++) chunk_appendf(&trash, " %u", read_freq_ctr(&activity[thr].ctxsw_rate));
+	chunk_appendf(&trash, "\ntasks_rate:");   for (thr = 0; thr < global.nbthread; thr++) chunk_appendf(&trash, " %u", read_freq_ctr(&activity[thr].tasks_rate));
 	chunk_appendf(&trash, "\ncpust_ms_tot:"); for (thr = 0; thr < global.nbthread; thr++) chunk_appendf(&trash, " %u", activity[thr].cpust_total/2);
 	chunk_appendf(&trash, "\ncpust_ms_1s:");  for (thr = 0; thr < global.nbthread; thr++) chunk_appendf(&trash, " %u", read_freq_ctr(&activity[thr].cpust_1s)/2);
 	chunk_appendf(&trash, "\ncpust_ms_15s:"); for (thr = 0; thr < global.nbthread; thr++) chunk_appendf(&trash, " %u", read_freq_ctr_period(&activity[thr].cpust_15s, 15000)/2);
diff --git a/src/task.c b/src/task.c
index de6eda1e72..04476fe5aa 100644
--- a/src/task.c
+++ b/src/task.c
@@ -20,10 +20,11 @@
 #include <eb32sctree.h>
 #include <eb32tree.h>
 
+#include <proto/fd.h>
+#include <proto/freq_ctr.h>
 #include <proto/proxy.h>
 #include <proto/stream.h>
 #include <proto/task.h>
-#include <proto/fd.h>
 
 DECLARE_POOL(pool_head_task,    "task",    sizeof(struct task));
 DECLARE_POOL(pool_head_tasklet, "tasklet", sizeof(struct tasklet));
@@ -278,6 +279,8 @@ void process_runnable_tasks()
 	struct eb32sc_node *lrq = NULL; // next local run queue entry
 	struct eb32sc_node *grq = NULL; // next global run queue entry
 	struct task *t;
+	int to_process;
+	int wakeups;
 	int max_processed;
 
 	if (!(active_tasks_mask & tid_bit)) {
@@ -292,6 +295,9 @@ void process_runnable_tasks()
 	if (likely(niced_tasks))
 		max_processed = (max_processed + 3) / 4;
 
+	to_process = max_processed;
+	wakeups = 0;
+
 	/* Note: the grq lock is always held when grq is not null */
 
 	while (task_per_thread[tid].task_list_size < max_processed) {
@@ -344,6 +350,7 @@ void process_runnable_tasks()
 
 		/* And add it to the local task list */
 		task_insert_into_tasklet_list(t);
+		wakeups++;
 	}
 
 	/* release the rqueue lock */
@@ -419,6 +426,11 @@ void process_runnable_tasks()
 		_HA_ATOMIC_OR(&active_tasks_mask, tid_bit);
 		activity[tid].long_rq++;
 	}
+
+	if (wakeups)
+		update_freq_ctr(&activity[tid].tasks_rate, wakeups);
+	if (to_process - max_processed)
+		update_freq_ctr(&activity[tid].ctxsw_rate, to_process - max_processed);
 }
 
 /*