[v11,4/4] ceph: add metadata perf metric support

Commit Message

Xiubo Li March 19, 2020, 6 a.m. UTC

From: Xiubo Li <[e-mail]>

Add a new "r_ended" field to struct ceph_mds_request and use that to
maintain the average latency of MDS requests.

URL: https://tracker.ceph.com/issues/43215
Signed-off-by: Xiubo Li <[e-mail]>
---
 fs/ceph/debugfs.c    |  9 +++++++++
 fs/ceph/mds_client.c |  5 +++++
 fs/ceph/mds_client.h |  3 ++-
 fs/ceph/metric.c     | 31 +++++++++++++++++++++++++++++++
 fs/ceph/metric.h     | 11 +++++++++++
 5 files changed, 58 insertions(+), 1 deletion(-)

Patch

diff --git a/fs/ceph/debugfs.c b/fs/ceph/debugfs.c
index e50499c..657f2f1 100644
--- a/fs/ceph/debugfs.c
+++ b/fs/ceph/debugfs.c
@@ -189,6 +189,15 @@  static int metric_show(struct seq_file *s, void *p)
 	sq = percpu_counter_sum(&m->write_latency_sq_sum);
 	CEPH_METRIC_SHOW("write", total, avg, min, max, sq);
 
+	avg = get_avg(&m->total_metadatas,
+		      &m->metadata_latency_sum,
+		      &m->metadata_latency_lock,
+		      &total);
+	min = atomic64_read(&m->metadata_latency_min);
+	max = atomic64_read(&m->metadata_latency_max);
+	sq = percpu_counter_sum(&m->metadata_latency_sq_sum);
+	CEPH_METRIC_SHOW("metadata", total, avg, min, max, sq);
+
 	seq_printf(s, "\n");
 	seq_printf(s, "item          total           miss            hit\n");
 	seq_printf(s, "-------------------------------------------------\n");
diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
index 1e242d8..b3f985a 100644
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -2547,6 +2547,8 @@  static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
 static void complete_request(struct ceph_mds_client *mdsc,
 			     struct ceph_mds_request *req)
 {
+	req->r_ended = jiffies;
+
 	if (req->r_callback)
 		req->r_callback(mdsc, req);
 	complete_all(&req->r_completion);
@@ -3155,6 +3157,9 @@  static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
 
 	/* kick calling process */
 	complete_request(mdsc, req);
+
+	ceph_update_metadata_latency(&mdsc->metric, req->r_started,
+				     req->r_ended, err);
 out:
 	ceph_mdsc_put_request(req);
 	return;
diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
index ae1d01c..9018fa7 100644
--- a/fs/ceph/mds_client.h
+++ b/fs/ceph/mds_client.h
@@ -298,7 +298,8 @@  struct ceph_mds_request {
 	u32               r_readdir_offset;
 
 	unsigned long r_timeout;  /* optional.  jiffies, 0 is "wait forever" */
-	unsigned long r_started;  /* start time to measure timeout against */
+	unsigned long r_started;  /* start time to measure timeout against and latency */
+	unsigned long r_ended;    /* finish time to measure latency */
 	unsigned long r_request_started; /* start time for mds request only,
 					    used to measure lease durations */
 
diff --git a/fs/ceph/metric.c b/fs/ceph/metric.c
index 4bf461d..23ceeb2 100644
--- a/fs/ceph/metric.c
+++ b/fs/ceph/metric.c
@@ -50,8 +50,20 @@  int ceph_metric_init(struct ceph_client_metric *m)
 	atomic64_set(&m->write_latency_min, S64_MAX);
 	atomic64_set(&m->write_latency_max, 0);
 
+	ret = percpu_counter_init(&m->metadata_latency_sq_sum, 0, GFP_KERNEL);
+	if (ret)
+		goto err_metadata_latency_sq_sum;
+
+	spin_lock_init(&m->metadata_latency_lock);
+	atomic64_set(&m->total_metadatas, 0);
+	atomic64_set(&m->metadata_latency_sum, 0);
+	atomic64_set(&m->metadata_latency_min, S64_MAX);
+	atomic64_set(&m->metadata_latency_max, 0);
+
 	return 0;
 
+err_metadata_latency_sq_sum:
+	percpu_counter_destroy(&m->write_latency_sq_sum);
 err_write_latency_sq_sum:
 	percpu_counter_destroy(&m->read_latency_sq_sum);
 err_read_latency_sq_sum:
@@ -71,6 +83,7 @@  void ceph_metric_destroy(struct ceph_client_metric *m)
 	if (!m)
 		return;
 
+	percpu_counter_destroy(&m->metadata_latency_sq_sum);
 	percpu_counter_destroy(&m->write_latency_sq_sum);
 	percpu_counter_destroy(&m->read_latency_sq_sum);
 	percpu_counter_destroy(&m->i_caps_mis);
@@ -160,3 +173,21 @@  void ceph_update_write_latency(struct ceph_client_metric *m,
 			    &m->write_latency_lock,
 			    lat);
 }
+
+void ceph_update_metadata_latency(struct ceph_client_metric *m,
+				  unsigned long r_start,
+				  unsigned long r_end,
+				  int rc)
+{
+	unsigned long lat = r_end - r_start;
+
+	if (unlikely(rc && rc != -ENOENT))
+		return;
+
+	__update_min_latency(&m->metadata_latency_min, lat);
+	__update_max_latency(&m->metadata_latency_max, lat);
+	__update_avg_and_sq(&m->total_metadatas, &m->metadata_latency_sum,
+			    &m->metadata_latency_sq_sum,
+			    &m->metadata_latency_lock,
+			    lat);
+}
diff --git a/fs/ceph/metric.h b/fs/ceph/metric.h
index a24482e..fc1ffb1 100644
--- a/fs/ceph/metric.h
+++ b/fs/ceph/metric.h
@@ -27,6 +27,13 @@  struct ceph_client_metric {
 	atomic64_t write_latency_sum;
 	atomic64_t write_latency_min;
 	atomic64_t write_latency_max;
+
+	struct percpu_counter metadata_latency_sq_sum;
+	spinlock_t metadata_latency_lock;
+	atomic64_t total_metadatas;
+	atomic64_t metadata_latency_sum;
+	atomic64_t metadata_latency_min;
+	atomic64_t metadata_latency_max;
 };
 
 extern int ceph_metric_init(struct ceph_client_metric *m);
@@ -50,4 +57,8 @@  extern void ceph_update_write_latency(struct ceph_client_metric *m,
 				      unsigned long r_start,
 				      unsigned long r_end,
 				      int rc);
+extern void ceph_update_metadata_latency(struct ceph_client_metric *m,
+					 unsigned long r_start,
+					 unsigned long r_end,
+					 int rc);
 #endif /* _FS_CEPH_MDS_METRIC_H */