Re: [PATCH 4/4] ceph: add IO size metrics support

[Xiubo Li <xiubli@redhat.com>]

On 2021/3/23 20:29, Jeff Layton wrote:
> On Mon, 2021-03-22 at 20:28 +0800, xiubli@redhat.com wrote:
>> From: Xiubo Li <xiubli@redhat.com>
>>
>> This will collect IO's total size and then calculate the average
>> size, and also will collect the min/max IO sizes.
>>
>> The debugfs will show the size metrics in byte and will let the
>> userspace applications to switch to what they need.
>>
>> URL: https://tracker.ceph.com/issues/49913
>> Signed-off-by: Xiubo Li <xiubli@redhat.com>
>> ---
>>   fs/ceph/addr.c    | 14 ++++++++------
>>   fs/ceph/debugfs.c | 37 +++++++++++++++++++++++++++++++++----
>>   fs/ceph/file.c    | 23 +++++++++++------------
>>   fs/ceph/metric.c  | 44 ++++++++++++++++++++++++++++++++++++++++++--
>>   fs/ceph/metric.h  | 10 ++++++++--
>>   5 files changed, 102 insertions(+), 26 deletions(-)
>>
>> diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
>> index 7c2802758d0e..d8a3624bc81d 100644
>> --- a/fs/ceph/addr.c
>> +++ b/fs/ceph/addr.c
>> @@ -227,7 +227,7 @@ static void finish_netfs_read(struct ceph_osd_request *req)
>>   	int err = req->r_result;
>>   
>>
>>
>>
>>   	ceph_update_read_metrics(&fsc->mdsc->metric, req->r_start_latency,
>> -				 req->r_end_latency, err);
>> +				 req->r_end_latency, osd_data->length, err);
>>   
>>
>>
>>
>>   	dout("%s: result %d subreq->len=%zu i_size=%lld\n", __func__, req->r_result,
>>   	     subreq->len, i_size_read(req->r_inode));
>> @@ -561,7 +561,7 @@ static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
>>   		err = ceph_osdc_wait_request(osdc, req);
>>   
>>
>>
>>
>>   	ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
>> -				  req->r_end_latency, err);
>> +				  req->r_end_latency, len, err);
>>   
>>
>>
>>
>>   	ceph_osdc_put_request(req);
>>   	if (err == 0)
>> @@ -636,6 +636,7 @@ static void writepages_finish(struct ceph_osd_request *req)
>>   	struct ceph_snap_context *snapc = req->r_snapc;
>>   	struct address_space *mapping = inode->i_mapping;
>>   	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
>> +	unsigned int len = 0;
>>   	bool remove_page;
>>   
>>
>>
>>
>>   	dout("writepages_finish %p rc %d\n", inode, rc);
>> @@ -648,9 +649,6 @@ static void writepages_finish(struct ceph_osd_request *req)
>>   		ceph_clear_error_write(ci);
>>   	}
>>   
>>
>>
>>
>> -	ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
>> -				  req->r_end_latency, rc);
>> -
>>   	/*
>>   	 * We lost the cache cap, need to truncate the page before
>>   	 * it is unlocked, otherwise we'd truncate it later in the
>> @@ -667,6 +665,7 @@ static void writepages_finish(struct ceph_osd_request *req)
>>   
>>
>>
>>
>>   		osd_data = osd_req_op_extent_osd_data(req, i);
>>   		BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
>> +		len += osd_data->length;
>>   		num_pages = calc_pages_for((u64)osd_data->alignment,
>>   					   (u64)osd_data->length);
>>   		total_pages += num_pages;
>> @@ -699,6 +698,9 @@ static void writepages_finish(struct ceph_osd_request *req)
>>   		release_pages(osd_data->pages, num_pages);
>>   	}
>>   
>>
>>
>>
>> +	ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
>> +				  req->r_end_latency, len, rc);
>> +
>>   	ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
>>   
>>
>>
>>
>>   	osd_data = osd_req_op_extent_osd_data(req, 0);
>> @@ -1720,7 +1722,7 @@ int ceph_uninline_data(struct file *filp, struct page *locked_page)
>>   		err = ceph_osdc_wait_request(&fsc->client->osdc, req);
>>   
>>
>>
>>
>>   	ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
>> -				  req->r_end_latency, err);
>> +				  req->r_end_latency, len, err);
>>   
>>
>>
>>
>>   out_put:
>>   	ceph_osdc_put_request(req);
>> diff --git a/fs/ceph/debugfs.c b/fs/ceph/debugfs.c
>> index 425f3356332a..38b78b45811f 100644
>> --- a/fs/ceph/debugfs.c
>> +++ b/fs/ceph/debugfs.c
>> @@ -127,7 +127,7 @@ static int mdsc_show(struct seq_file *s, void *p)
>>   	return 0;
>>   }
>>   
>>
>>
>>
>> -#define CEPH_METRIC_SHOW(name, total, avg, min, max, sq) {		\
>> +#define CEPH_LAT_METRIC_SHOW(name, total, avg, min, max, sq) {		\
>>   	s64 _total, _avg, _min, _max, _sq, _st;				\
>>   	_avg = ktime_to_us(avg);					\
>>   	_min = ktime_to_us(min == KTIME_MAX ? 0 : min);			\
>> @@ -140,6 +140,12 @@ static int mdsc_show(struct seq_file *s, void *p)
>>   		   name, total, _avg, _min, _max, _st);			\
>>   }
>>   
>>
>>
>>
>> +#define CEPH_SZ_METRIC_SHOW(name, total, avg, min, max, sum) {		\
>> +	u64 _min = min == U64_MAX ? 0 : min;				\
>> +	seq_printf(s, "%-14s%-12lld%-16llu%-16llu%-16llu%llu\n",	\
>> +		   name, total, avg, _min, max, sum);			\
>> +}
>> +
>>   static int metric_show(struct seq_file *s, void *p)
>>   {
>>   	struct ceph_fs_client *fsc = s->private;
>> @@ -147,6 +153,7 @@ static int metric_show(struct seq_file *s, void *p)
>>   	struct ceph_client_metric *m = &mdsc->metric;
>>   	int nr_caps = 0;
>>   	s64 total, sum, avg, min, max, sq;
>> +	u64 sum_sz, avg_sz, min_sz, max_sz;
>>   
>>
>>
>>
>>   	sum = percpu_counter_sum(&m->total_inodes);
>>   	seq_printf(s, "item                               total\n");
>> @@ -170,7 +177,7 @@ static int metric_show(struct seq_file *s, void *p)
>>   	max = m->read_latency_max;
>>   	sq = m->read_latency_sq_sum;
>>   	spin_unlock(&m->read_metric_lock);
>> -	CEPH_METRIC_SHOW("read", total, avg, min, max, sq);
>> +	CEPH_LAT_METRIC_SHOW("read", total, avg, min, max, sq);
>>   
>>
>>
>>
>>   	spin_lock(&m->write_metric_lock);
>>   	total = m->total_writes;
>> @@ -180,7 +187,7 @@ static int metric_show(struct seq_file *s, void *p)
>>   	max = m->write_latency_max;
>>   	sq = m->write_latency_sq_sum;
>>   	spin_unlock(&m->write_metric_lock);
>> -	CEPH_METRIC_SHOW("write", total, avg, min, max, sq);
>> +	CEPH_LAT_METRIC_SHOW("write", total, avg, min, max, sq);
>>   
>>
>>
>>
>>   	spin_lock(&m->metadata_metric_lock);
>>   	total = m->total_metadatas;
>> @@ -190,7 +197,29 @@ static int metric_show(struct seq_file *s, void *p)
>>   	max = m->metadata_latency_max;
>>   	sq = m->metadata_latency_sq_sum;
>>   	spin_unlock(&m->metadata_metric_lock);
>> -	CEPH_METRIC_SHOW("metadata", total, avg, min, max, sq);
>> +	CEPH_LAT_METRIC_SHOW("metadata", total, avg, min, max, sq);
>> +
>> +	seq_printf(s, "\n");
>> +	seq_printf(s, "item          total       avg_sz(bytes)   min_sz(bytes)   max_sz(bytes)  total_sz(bytes)\n");
>> +	seq_printf(s, "----------------------------------------------------------------------------------------\n");
>> +
>> +	spin_lock(&m->read_metric_lock);
>> +	total = m->total_reads;
>> +	sum_sz = m->read_size_sum;
>> +	avg_sz = total > 0 ? DIV64_U64_ROUND_CLOSEST(sum_sz, total) : 0;
>> +	min_sz = m->read_size_min;
>> +	max_sz = m->read_size_max;
>> +	spin_unlock(&m->read_metric_lock);
>> +	CEPH_SZ_METRIC_SHOW("read", total, avg_sz, min_sz, max_sz, sum_sz);
>> +
>> +	spin_lock(&m->write_metric_lock);
>> +	total = m->total_writes;
>> +	sum_sz = m->write_size_sum;
>> +	avg_sz = total > 0 ? DIV64_U64_ROUND_CLOSEST(sum_sz, total) : 0;
>> +	min_sz = m->write_size_min;
>> +	max_sz = m->write_size_max;
>> +	spin_unlock(&m->write_metric_lock);
>> +	CEPH_SZ_METRIC_SHOW("write", total, avg_sz, min_sz, max_sz, sum_sz);
>>   
>>
>>
>>
>>   	seq_printf(s, "\n");
>>   	seq_printf(s, "item          total           miss            hit\n");
>> diff --git a/fs/ceph/file.c b/fs/ceph/file.c
>> index 31542eac7e59..db43d2d013b9 100644
>> --- a/fs/ceph/file.c
>> +++ b/fs/ceph/file.c
>> @@ -898,7 +898,7 @@ static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
>>   		ceph_update_read_metrics(&fsc->mdsc->metric,
>>   					 req->r_start_latency,
>>   					 req->r_end_latency,
>> -					 ret);
>> +					 len, ret);
>>   
>>
>>
>>
>>   		ceph_osdc_put_request(req);
>>   
>>
>>
>>
>> @@ -1030,12 +1030,12 @@ static void ceph_aio_complete_req(struct ceph_osd_request *req)
>>   	struct ceph_aio_request *aio_req = req->r_priv;
>>   	struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
>>   	struct ceph_client_metric *metric = &ceph_sb_to_mdsc(inode->i_sb)->metric;
>> +	unsigned int len = osd_data->bvec_pos.iter.bi_size;
>>   
>>
>>
>>
>>   	BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_BVECS);
>>   	BUG_ON(!osd_data->num_bvecs);
>>   
>>
>>
>>
>> -	dout("ceph_aio_complete_req %p rc %d bytes %u\n",
>> -	     inode, rc, osd_data->bvec_pos.iter.bi_size);
>> +	dout("ceph_aio_complete_req %p rc %d bytes %u\n", inode, rc, len);
>>   
>>
>>
>>
>>   	if (rc == -EOLDSNAPC) {
>>   		struct ceph_aio_work *aio_work;
>> @@ -1053,9 +1053,9 @@ static void ceph_aio_complete_req(struct ceph_osd_request *req)
>>   	} else if (!aio_req->write) {
>>   		if (rc == -ENOENT)
>>   			rc = 0;
>> -		if (rc >= 0 && osd_data->bvec_pos.iter.bi_size > rc) {
>> +		if (rc >= 0 && len > rc) {
>>   			struct iov_iter i;
>> -			int zlen = osd_data->bvec_pos.iter.bi_size - rc;
>> +			int zlen = len - rc;
>>   
>>
>>
>>
>>   			/*
>>   			 * If read is satisfied by single OSD request,
>> @@ -1072,8 +1072,7 @@ static void ceph_aio_complete_req(struct ceph_osd_request *req)
>>   			}
>>   
>>
>>
>>
>>   			iov_iter_bvec(&i, READ, osd_data->bvec_pos.bvecs,
>> -				      osd_data->num_bvecs,
>> -				      osd_data->bvec_pos.iter.bi_size);
>> +				      osd_data->num_bvecs, len);
>>   			iov_iter_advance(&i, rc);
>>   			iov_iter_zero(zlen, &i);
>>   		}
>> @@ -1083,10 +1082,10 @@ static void ceph_aio_complete_req(struct ceph_osd_request *req)
>>   	if (req->r_start_latency) {
>>   		if (aio_req->write)
>>   			ceph_update_write_metrics(metric, req->r_start_latency,
>> -						  req->r_end_latency, rc);
>> +						  req->r_end_latency, len, rc);
>>   		else
>>   			ceph_update_read_metrics(metric, req->r_start_latency,
>> -						 req->r_end_latency, rc);
>> +						 req->r_end_latency, len, rc);
>>   	}
>>   
>>
>>
>>
>>   	put_bvecs(osd_data->bvec_pos.bvecs, osd_data->num_bvecs,
>> @@ -1294,10 +1293,10 @@ ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
>>   
>>
>>
>>
>>   		if (write)
>>   			ceph_update_write_metrics(metric, req->r_start_latency,
>> -						  req->r_end_latency, ret);
>> +						  req->r_end_latency, len, ret);
>>   		else
>>   			ceph_update_read_metrics(metric, req->r_start_latency,
>> -						 req->r_end_latency, ret);
>> +						 req->r_end_latency, len, ret);
>>   
>>
>>
>>
>>   		size = i_size_read(inode);
>>   		if (!write) {
>> @@ -1471,7 +1470,7 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
>>   			ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
>>   
>>
>>
>>
>>   		ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
>> -					  req->r_end_latency, ret);
>> +					  req->r_end_latency, len, ret);
>>   out:
>>   		ceph_osdc_put_request(req);
>>   		if (ret != 0) {
>> diff --git a/fs/ceph/metric.c b/fs/ceph/metric.c
>> index d5560ff99a9d..ff3c9d5cf9ff 100644
>> --- a/fs/ceph/metric.c
>> +++ b/fs/ceph/metric.c
>> @@ -189,6 +189,9 @@ int ceph_metric_init(struct ceph_client_metric *m)
>>   	m->read_latency_max = 0;
>>   	m->total_reads = 0;
>>   	m->read_latency_sum = 0;
>> +	m->read_size_min = U64_MAX;
>> +	m->read_size_max = 0;
>> +	m->read_size_sum = 0;
>>   
>>
>>
>>
>>   	spin_lock_init(&m->write_metric_lock);
>>   	m->write_latency_sq_sum = 0;
>> @@ -196,6 +199,9 @@ int ceph_metric_init(struct ceph_client_metric *m)
>>   	m->write_latency_max = 0;
>>   	m->total_writes = 0;
>>   	m->write_latency_sum = 0;
>> +	m->write_size_min = U64_MAX;
>> +	m->write_size_max = 0;
>> +	m->write_size_sum = 0;
>>   
>>
>>
>>
>>   	spin_lock_init(&m->metadata_metric_lock);
>>   	m->metadata_latency_sq_sum = 0;
>> @@ -306,9 +312,41 @@ static inline void __update_latency(struct ceph_client_metric *m,
>>   	*sq_sump += sq;
>>   }
>>   
>>
>>
>>
>> +static inline void __update_size(struct ceph_client_metric *m,
>> +				 metric_type type, unsigned int size)
>> +{
>> +	ktime_t total;
>> +	u64 *minp, *maxp, *sump;
>> +
>> +	switch (type) {
>> +	case CEPH_METRIC_READ:
>> +		total = m->total_reads;
>> +		sump = &m->read_size_sum;
>> +		minp = &m->read_size_min;
>> +		maxp = &m->read_size_max;
>> +		break;
>> +	case CEPH_METRIC_WRITE:
>> +		total = m->total_writes;
> "total" and "sump" are unused in this function, aside from the
> assignment.

Will fix it.

Just assuming to add the read/write IO speeds here, will it make sense ?


>> +		sump = &m->write_size_sum;
>> +		minp = &m->write_size_min;
>> +		maxp = &m->write_size_max;
>> +		break;
>> +	case CEPH_METRIC_METADATA:
>> +	default:
>> +		return;
>> +	}
>> +
>> +	*sump += size;
>> +
>> +	if (unlikely(size < *minp))
>> +		*minp = size;
>> +	if (unlikely(size > *maxp))
>> +		*maxp = size;
>> +}
>> +
>>   void ceph_update_read_metrics(struct ceph_client_metric *m,
>>   			      ktime_t r_start, ktime_t r_end,
>> -			      int rc)
>> +			      unsigned int size, int rc)
>>   {
>>   	ktime_t lat = ktime_sub(r_end, r_start);
>>   
>>
>>
>>
>> @@ -317,12 +355,13 @@ void ceph_update_read_metrics(struct ceph_client_metric *m,
>>   
>>
>>
>>
>>   	spin_lock(&m->read_metric_lock);
>>   	__update_latency(m, CEPH_METRIC_READ, lat);
>> +	__update_size(m, CEPH_METRIC_READ, size);
>>   	spin_unlock(&m->read_metric_lock);
>>   }
>>   
>>
>>
>>
>>   void ceph_update_write_metrics(struct ceph_client_metric *m,
>>   			       ktime_t r_start, ktime_t r_end,
>> -			       int rc)
>> +			       unsigned int size, int rc)
>>   {
>>   	ktime_t lat = ktime_sub(r_end, r_start);
>>   
>>
>>
>>
>> @@ -331,6 +370,7 @@ void ceph_update_write_metrics(struct ceph_client_metric *m,
>>   
>>
>>
>>
>>   	spin_lock(&m->write_metric_lock);
>>   	__update_latency(m, CEPH_METRIC_WRITE, lat);
>> +	__update_size(m, CEPH_METRIC_WRITE, size);
>>   	spin_unlock(&m->write_metric_lock);
>>   }
>>   
>>
>>
>>
>> diff --git a/fs/ceph/metric.h b/fs/ceph/metric.h
>> index 57b5f0ec38be..64651b6ac886 100644
>> --- a/fs/ceph/metric.h
>> +++ b/fs/ceph/metric.h
>> @@ -110,6 +110,9 @@ struct ceph_client_metric {
>>   
>>
>>
>>
>>   	spinlock_t read_metric_lock;
>>   	u64 total_reads;
>> +	u64 read_size_sum;
>> +	u64 read_size_min;
>> +	u64 read_size_max;
>>   	ktime_t read_latency_sum;
>>   	ktime_t read_latency_sq_sum;
>>   	ktime_t read_latency_min;
>> @@ -117,6 +120,9 @@ struct ceph_client_metric {
>>   
>>
>>
>>
>>   	spinlock_t write_metric_lock;
>>   	u64 total_writes;
>> +	u64 write_size_sum;
>> +	u64 write_size_min;
>> +	u64 write_size_max;
>>   	ktime_t write_latency_sum;
>>   	ktime_t write_latency_sq_sum;
>>   	ktime_t write_latency_min;
>> @@ -164,10 +170,10 @@ static inline void ceph_update_cap_mis(struct ceph_client_metric *m)
>>   
>>
>>
>>
>>   extern void ceph_update_read_metrics(struct ceph_client_metric *m,
>>   				     ktime_t r_start, ktime_t r_end,
>> -				     int rc);
>> +				     unsigned int size, int rc);
>>   extern void ceph_update_write_metrics(struct ceph_client_metric *m,
>>   				      ktime_t r_start, ktime_t r_end,
>> -				      int rc);
>> +				      unsigned int size, int rc);
>>   extern void ceph_update_metadata_metrics(struct ceph_client_metric *m,
>>   				         ktime_t r_start, ktime_t r_end,
>>   					 int rc);