ceph: add perf metrics support
Subsystems (2)
v6 cover ceph: add perf metrics support
From: Xiubo Li <[email protected]>
Changed in V6:
- fold r_end_stamp patch to its first user
- remove some parameters' declartion which are only used once
- switch debugfs sending_metric UI to a module parameter
- make the cap hit/mis metric as global per superblock
- some other small fixes
It will send the metrics to ceph cluster per metric_send_interval seconds if enabled, metric_send_interval is a module parameter and default value is 0, 0 also means disabled.
We can get the metrics from the debugfs:
$ cat /sys/kernel/debug/ceph/0c93a60d-5645-4c46-8568-4c8f63db4c7f.client4267/metrics item total sum_lat(us) avg_lat(us) ----------------------------------------------------- read 13 417000 32076 write 42 131205000 3123928 metadata 104 493000 4740
item total miss hit
-------------------------------------------------
d_lease 204 0 918 caps 204 213 368218
In the MDS side, we can get the metrics(NOTE: the latency is in nanosecond):
$ ./bin/ceph fs perf stats | python -m json.tool
{
"client_metadata": {
"client.4267": {
"IP": "v1:192.168.195.165",
"hostname": "fedora1",
"mount_point": "N/A",
"root": "/"
}
},
"counters": [
"cap_hit"
],
"global_counters": [
"read_latency",
"write_latency",
"metadata_latency",
"dentry_lease_hit"
],
"global_metrics": {
"client.4267": [
[
0,
32076923
],
[
3,
123928571
],
[
0,
4740384
],
[
918,
0
]
]
},
"metrics": {
"delayed_ranks": [],
"mds.0": {
"client.4267": [
[
368218,
213
]
]
}
}
}The provided metric flags in client metadata
$./bin/cephfs-journal-tool --rank=1:0 event get --type=SESSION json Wrote output to JSON file 'dump'
$ cat dump
[
{
"client instance": "client.4275 v1:192.168.195.165:0/461391971",
"open": "true",
"client map version": 1,
"inos": "[]",
"inotable version": 0,
"client_metadata": {
"client_features": {
"feature_bits": "0000000000001bff"
},
"metric_spec": {
"metric_flags": {
"feature_bits": "000000000000001f"
}
},
"entity_id": "",
"hostname": "fedora1",
"kernel_version": "5.5.0-rc2+",
"root": "/"
}
},
[...]Xiubo Li (9):
ceph: add global dentry lease metric support
ceph: add caps perf metric for each session
ceph: add global read latency metric support
ceph: add global write latency metric support
ceph: add global metadata perf metric support
ceph: periodically send perf metrics to ceph
ceph: add CEPH_DEFINE_RW_FUNC helper support
ceph: add reset metrics support
ceph: send client provided metric flags in client metadata
fs/ceph/acl.c | 2 + fs/ceph/addr.c | 13 ++ fs/ceph/caps.c | 29 +++ fs/ceph/debugfs.c | 107 ++++++++- fs/ceph/dir.c | 25 ++- fs/ceph/file.c | 22 ++ fs/ceph/mds_client.c | 381 +++++++++++++++++++++++++++++--- fs/ceph/mds_client.h | 6 + fs/ceph/metric.h | 155 +++++++++++++ fs/ceph/quota.c | 9 +- fs/ceph/super.c | 4 + fs/ceph/super.h | 11 + fs/ceph/xattr.c | 17 +- include/linux/ceph/ceph_fs.h | 1 + include/linux/ceph/debugfs.h | 14 ++ include/linux/ceph/osd_client.h | 1 + net/ceph/osd_client.c | 2 + 17 files changed, 759 insertions(+), 40 deletions(-) create mode 100644 fs/ceph/metric.h
Signature · 1 line
2.21.0
PATCH v6 1/9 ceph: add global dentry lease metric support
From: Xiubo Li <[email protected]>
For the dentry lease we will only count the hit/miss info triggered from the vfs calls, for the cases like request reply handling and perodically ceph_trim_dentries() we will ignore them.
Currently only the debugfs is support:
The output will be:
item total miss hit
-------------------------------------------------d_lease 11 7 141
URL: https://tracker.ceph.com/issues/43215
--- fs/ceph/debugfs.c | 32 ++++++++++++++++++++++++++++---- fs/ceph/dir.c | 16 ++++++++++++++-- fs/ceph/mds_client.c | 37 +++++++++++++++++++++++++++++++++++-- fs/ceph/mds_client.h | 4 ++++ fs/ceph/metric.h | 11 +++++++++++ fs/ceph/super.h | 1 + 6 files changed, 93 insertions(+), 8 deletions(-) create mode 100644 fs/ceph/metric.h
fs/ceph/debugfs.c +28 -4
@@ -124,6 +124,22 @@ static int mdsc_show(struct seq_file *s, void *p)return 0 ; } +static int metric_show (struct seq_file *s,void *p) +{ +struct ceph_fs_client *fsc = s->private; +struct ceph_mds_client *mdsc = fsc->mdsc; + +seq_printf (s,"item total miss hit\n" ); +seq_printf (s,"-------------------------------------------------\n" ); + +seq_printf (s,"%-14s%-16lld%-16lld%lld\n" ,"d_lease" , +atomic64_read (&mdsc->metric.total_dentries), +percpu_counter_sum (&mdsc->metric.d_lease_mis), +percpu_counter_sum (&mdsc->metric.d_lease_hit)); + +return 0 ; +} +static int caps_show_cb (struct inode *inode,struct ceph_cap *cap,void *p) {struct seq_file *s = p;
@@ -222,6 +238,7 @@ DEFINE_SHOW_ATTRIBUTE(mdsmap);DEFINE_SHOW_ATTRIBUTE (mdsc);DEFINE_SHOW_ATTRIBUTE (caps);DEFINE_SHOW_ATTRIBUTE (mds_sessions); +DEFINE_SHOW_ATTRIBUTE (metric);/*
@@ -255,6 +272,7 @@ void ceph_fs_debugfs_cleanup(struct ceph_fs_client *fsc)debugfs_remove (fsc->debugfs_mdsmap);debugfs_remove (fsc->debugfs_mds_sessions);debugfs_remove (fsc->debugfs_caps); +debugfs_remove (fsc->debugfs_metric);debugfs_remove (fsc->debugfs_mdsc); }
@@ -295,11 +313,17 @@ void ceph_fs_debugfs_init(struct ceph_fs_client *fsc) fsc, &mdsc_fops); + fsc->debugfs_metric =debugfs_create_file ("metrics" , +0400 , + fsc->client->debugfs_dir, + fsc, + &metric_fops); + fsc->debugfs_caps =debugfs_create_file ("caps" , -0400 , - fsc->client->debugfs_dir, - fsc, - &caps_fops); +0400 , + fsc->client->debugfs_dir, + fsc, + &caps_fops); }
fs/ceph/dir.c +14 -2
@@ -38,6 +38,8 @@ static int __dir_lease_try_check(const struct dentry *dentry);static int ceph_d_init (struct dentry *dentry) {struct ceph_dentry_info *di; +struct ceph_fs_client *fsc =ceph_sb_to_client (dentry->d_sb); +struct ceph_mds_client *mdsc = fsc->mdsc; di =kmem_cache_zalloc (ceph_dentry_cachep,GFP_KERNEL );if (!di)
@@ -48,6 +50,9 @@ static int ceph_d_init(struct dentry *dentry) di->time = jiffies; dentry->d_fsdata = di;INIT_LIST_HEAD (&di->lease_list); + +atomic64_inc (&mdsc->metric.total_dentries); +return 0 ; }
@@ -1613,6 +1618,7 @@ static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)*/ static int ceph_d_revalidate (struct dentry *dentry,unsigned int flags) { +struct ceph_mds_client *mdsc;int valid =0 ;struct dentry *parent;struct inode *dir, *inode;
@@ -1651,9 +1657,8 @@ static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags) } } + mdsc =ceph_sb_to_client (dir->i_sb)->mdsc;if (!valid) { -struct ceph_mds_client *mdsc = -ceph_sb_to_client (dir->i_sb)->mdsc;struct ceph_mds_request *req;int op, err;u32 mask;
@@ -1661,6 +1666,8 @@ static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)if (flags &LOOKUP_RCU )return -ECHILD ; +percpu_counter_inc (&mdsc->metric.d_lease_mis); + op =ceph_snap (dir) ==CEPH_SNAPDIR ?CEPH_MDS_OP_LOOKUPSNAP :CEPH_MDS_OP_LOOKUP ; req =ceph_mdsc_create_request (mdsc, op,USE_ANY_MDS );
@@ -1692,6 +1699,8 @@ static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)dout ("d_revalidate %p lookup result=%d\n" , dentry, err); } + }else { +percpu_counter_inc (&mdsc->metric.d_lease_hit); }dout ("d_revalidate %p %s\n" , dentry, valid ?"valid" :"invalid" );
@@ -1734,9 +1743,12 @@ static int ceph_d_delete(const struct dentry *dentry)static void ceph_d_release (struct dentry *dentry) {struct ceph_dentry_info *di =ceph_dentry (dentry); +struct ceph_fs_client *fsc =ceph_sb_to_client (dentry->d_sb);dout ("d_release %p\n" , dentry); +atomic64_dec (&fsc->mdsc->metric.total_dentries); +spin_lock (&dentry->d_lock);__dentry_lease_unlist (di); dentry->d_fsdata =NULL ;
fs/ceph/mds_client.c +35 -2
@@ -4158,10 +4158,31 @@ static void delayed_work(struct work_struct *work)schedule_delayed (mdsc); } +static int ceph_mdsc_metric_init (struct ceph_client_metric *metric) +{ +int ret; + +if (!metric) +return -EINVAL ; + +atomic64_set (&metric->total_dentries,0 ); + ret =percpu_counter_init (&metric->d_lease_hit,0 ,GFP_KERNEL ); +if (ret) +return ret; + ret =percpu_counter_init (&metric->d_lease_mis,0 ,GFP_KERNEL ); +if (ret) { +percpu_counter_destroy (&metric->d_lease_hit); +return ret; + } + +return 0 ; +} +int ceph_mdsc_init (struct ceph_fs_client *fsc) {struct ceph_mds_client *mdsc; +int err; mdsc =kzalloc (sizeof (struct ceph_mds_client ),GFP_NOFS );if (!mdsc)
@@ -4170,8 +4191,8 @@ int ceph_mdsc_init(struct ceph_fs_client *fsc)mutex_init (&mdsc->mutex); mdsc->mdsmap =kzalloc (sizeof (*mdsc->mdsmap),GFP_NOFS );if (!mdsc->mdsmap) { -kfree (mdsc); -return -ENOMEM ; + err = -ENOMEM ; + goto err_mdsc; } fsc->mdsc = mdsc;
@@ -4210,6 +4231,9 @@ int ceph_mdsc_init(struct ceph_fs_client *fsc)init_waitqueue_head (&mdsc->cap_flushing_wq);INIT_WORK (&mdsc->cap_reclaim_work, ceph_cap_reclaim_work);atomic_set (&mdsc->cap_reclaim_pending,0 ); + err =ceph_mdsc_metric_init (&mdsc->metric); +if (err) + goto err_mdsmap;spin_lock_init (&mdsc->dentry_list_lock);INIT_LIST_HEAD (&mdsc->dentry_leases);
@@ -4228,6 +4252,12 @@ int ceph_mdsc_init(struct ceph_fs_client *fsc)strscpy (mdsc->nodename,utsname ()->nodename,sizeof (mdsc->nodename));return 0 ; + +err_mdsmap: +kfree (mdsc->mdsmap); +err_mdsc: +kfree (mdsc); +return err; }/*
@@ -4485,6 +4515,9 @@ void ceph_mdsc_destroy(struct ceph_fs_client *fsc)ceph_mdsc_stop (mdsc); +percpu_counter_destroy (&mdsc->metric.d_lease_mis); +percpu_counter_destroy (&mdsc->metric.d_lease_hit); + fsc->mdsc =NULL ;kfree (mdsc);dout ("mdsc_destroy %p done\n" , mdsc);
fs/ceph/mds_client.h +4 -0
@@ -16,6 +16,8 @@#include <linux/ceph/mdsmap.h> #include <linux/ceph/auth.h> +#include "metric.h" +/* The first 8 bits are reserved for old ceph releases */ enum ceph_feature_type {CEPHFS_FEATURE_MIMIC =8 ,
@@ -446,6 +448,8 @@ struct ceph_mds_client {struct list_head dentry_leases;/* fifo list */ struct list_head dentry_dir_leases;/* lru list */ +struct ceph_client_metric metric; +spinlock_t snapid_map_lock;struct rb_root snapid_map_tree;struct list_head snapid_map_lru;
fs/ceph/metric.h new file +11 -0
@@ -0,0 +1,11 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _FS_CEPH_MDS_METRIC_H +#define _FS_CEPH_MDS_METRIC_H + +/* This is the global metrics */ +struct ceph_client_metric { +atomic64_t total_dentries; +struct percpu_counter d_lease_hit; +struct percpu_counter d_lease_mis; +}; +#endif /* _FS_CEPH_MDS_METRIC_H */
fs/ceph/super.h +1 -0
@@ -128,6 +128,7 @@ struct ceph_fs_client {struct dentry *debugfs_congestion_kb;struct dentry *debugfs_bdi;struct dentry *debugfs_mdsc, *debugfs_mdsmap; +struct dentry *debugfs_metric;struct dentry *debugfs_mds_sessions;#endif
Signature · 1 line
2.21.0
PATCH v6 2/9 ceph: add caps perf metric for each session
In reply to [email protected]From: Xiubo Li <[email protected]>
This will fulfill the cap hit/mis metric stuff per-superblock, it will count the hit/mis counters based each inode, and if one inode's 'issued & ~revoking == mask' will mean a hit, or a miss.
item total miss hit
-------------------------------------------------
caps 295 107 4119
URL: https://tracker.ceph.com/issues/43215
--- fs/ceph/acl.c | 2 ++ fs/ceph/caps.c | 29 +++++++++++++++++++++++++++++ fs/ceph/debugfs.c | 16 ++++++++++++++++ fs/ceph/dir.c | 9 +++++++-- fs/ceph/file.c | 2 ++ fs/ceph/mds_client.c | 26 ++++++++++++++++++++++---- fs/ceph/metric.h | 3 +++ fs/ceph/quota.c | 9 +++++++-- fs/ceph/super.h | 9 +++++++++ fs/ceph/xattr.c | 17 ++++++++++++++--- 10 files changed, 111 insertions(+), 11 deletions(-)
fs/ceph/acl.c +2 -0
@@ -22,6 +22,8 @@ static inline void ceph_set_cached_acl(struct inode *inode,struct ceph_inode_info *ci =ceph_inode (inode);spin_lock (&ci->i_ceph_lock); +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED ); +if (__ceph_caps_issued_mask (ci,CEPH_CAP_XATTR_SHARED ,0 ))set_cached_acl (inode, type, acl);else
fs/ceph/caps.c +29 -0
@@ -818,6 +818,32 @@ int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)return have; } +/* +* Counts the cap metric. +* +* This will try to traverse all the ci->i_caps, if we can +* get all the cap 'mask' it will count the hit, or the mis. +*/ +void __ceph_caps_metric (struct ceph_inode_info *ci,int mask) +{ +struct ceph_mds_client *mdsc = +ceph_sb_to_client (ci->vfs_inode.i_sb)->mdsc; +struct ceph_client_metric *metric = &mdsc->metric; +int issued; + +lockdep_assert_held (&ci->i_ceph_lock); + +if (mask <=0 ) +return ; + + issued =__ceph_caps_issued (ci,NULL ); + +if ((mask & issued) == mask) +percpu_counter_inc (&metric->i_caps_hit); +else +percpu_counter_inc (&metric->i_caps_mis); +} +/* * Get cap bits issued by caps other than @ocap */
@@ -2758,6 +2784,7 @@ int ceph_try_get_caps(struct inode *inode, int need, int want,BUG_ON (want & ~(CEPH_CAP_FILE_CACHE |CEPH_CAP_FILE_LAZYIO |CEPH_CAP_FILE_SHARED |CEPH_CAP_FILE_EXCL |CEPH_CAP_ANY_DIR_OPS )); +ceph_caps_metric (ceph_inode (inode), need | want); ret =try_get_cap_refs (inode, need, want,0 , nonblock, got);return ret == -EAGAIN ?0 : ret; }
@@ -2784,6 +2811,8 @@ int ceph_get_caps(struct file *filp, int need, int want, fi->filp_gen != READ_ONCE(fsc->filp_gen))return -EBADF ; +ceph_caps_metric (ci, need | want); +while (true) {if (endoff >0 )check_max_size (inode, endoff);
fs/ceph/debugfs.c +16 -0
@@ -128,6 +128,7 @@ static int metric_show(struct seq_file *s, void *p) {struct ceph_fs_client *fsc = s->private;struct ceph_mds_client *mdsc = fsc->mdsc; +int i, nr_caps =0 ;seq_printf (s,"item total miss hit\n" );seq_printf (s,"-------------------------------------------------\n" );
@@ -137,6 +138,21 @@ static int metric_show(struct seq_file *s, void *p)percpu_counter_sum (&mdsc->metric.d_lease_mis), percpu_counter_sum(&mdsc->metric.d_lease_hit)); +mutex_lock (&mdsc->mutex); +for (i =0 ; i < mdsc->max_sessions; i++) { +struct ceph_mds_session *s; + + s =__ceph_lookup_mds_session (mdsc, i); +if (!s) +continue ; + nr_caps += s->s_nr_caps; +ceph_put_mds_session (s); + } +mutex_unlock (&mdsc->mutex); +seq_printf (s,"%-14s%-16d%-16lld%lld\n" ,"caps" , nr_caps, +percpu_counter_sum (&mdsc->metric.i_caps_mis), +percpu_counter_sum (&mdsc->metric.i_caps_hit)); +return 0 ; }
fs/ceph/dir.c +7 -2
@@ -313,7 +313,7 @@ static int ceph_readdir(struct file *file, struct dir_context *ctx)struct ceph_fs_client *fsc =ceph_inode_to_client (inode);struct ceph_mds_client *mdsc = fsc->mdsc;int i; -int err; +int err, ret =-1 ;unsigned frag =-1 ;struct ceph_mds_reply_info_parsed *rinfo;
@@ -346,13 +346,16 @@ static int ceph_readdir(struct file *file, struct dir_context *ctx) !ceph_test_mount_opt(fsc, NOASYNCREADDIR) && ceph_snap(inode) != CEPH_SNAPDIR && __ceph_dir_is_complete_ordered(ci) && - __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) { + (ret = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {int shared_gen =atomic_read (&ci->i_shared_gen); +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED );spin_unlock (&ci->i_ceph_lock); err =__dcache_readdir (file, ctx, shared_gen);if (err != -EAGAIN )return err; } else { +if (ret !=-1 ) +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED );spin_unlock (&ci->i_ceph_lock); }
@@ -757,6 +760,8 @@ static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,struct ceph_dentry_info *di =ceph_dentry (dentry);spin_lock (&ci->i_ceph_lock); +__ceph_caps_metric (ci,CEPH_CAP_FILE_SHARED ); +dout (" dir %p flags are %d\n" , dir, ci->i_ceph_flags); if (strncmp(dentry->d_name.name, fsc->mount_options->snapdir_name,
fs/ceph/file.c +2 -0
@@ -384,6 +384,8 @@ int ceph_open(struct inode *inode, struct file *file)* asynchronously. */ spin_lock (&ci->i_ceph_lock); +__ceph_caps_metric (ci, wanted); +if (__ceph_is_any_real_caps (ci) && (((fmode &CEPH_FILE_MODE_WR ) ==0 ) || ci->i_auth_cap)) {int mds_wanted =__ceph_caps_mds_wanted (ci, true);
fs/ceph/mds_client.c +22 -4
@@ -4169,13 +4169,29 @@ static int ceph_mdsc_metric_init(struct ceph_client_metric *metric) ret =percpu_counter_init (&metric->d_lease_hit,0 ,GFP_KERNEL );if (ret)return ret; + ret =percpu_counter_init (&metric->d_lease_mis,0 ,GFP_KERNEL ); -if (ret) { -percpu_counter_destroy (&metric->d_lease_hit); -return ret; - } +if (ret) + goto err_d_lease_mis; + + ret =percpu_counter_init (&metric->i_caps_hit,0 ,GFP_KERNEL ); +if (ret) + goto err_i_caps_hit; + + ret =percpu_counter_init (&metric->i_caps_mis,0 ,GFP_KERNEL ); +if (ret) + goto err_i_caps_mis;return 0 ; + +err_i_caps_mis: +percpu_counter_destroy (&metric->i_caps_hit); +err_i_caps_hit: +percpu_counter_destroy (&metric->d_lease_mis); +err_d_lease_mis: +percpu_counter_destroy (&metric->d_lease_hit); + +return ret; }int ceph_mdsc_init (struct ceph_fs_client *fsc)
@@ -4515,6 +4531,8 @@ void ceph_mdsc_destroy(struct ceph_fs_client *fsc)ceph_mdsc_stop (mdsc); +percpu_counter_destroy (&mdsc->metric.i_caps_mis); +percpu_counter_destroy (&mdsc->metric.i_caps_hit);percpu_counter_destroy (&mdsc->metric.d_lease_mis);percpu_counter_destroy (&mdsc->metric.d_lease_hit);
fs/ceph/metric.h +3 -0
@@ -7,5 +7,8 @@ struct ceph_client_metric {atomic64_t total_dentries;struct percpu_counter d_lease_hit;struct percpu_counter d_lease_mis; + +struct percpu_counter i_caps_hit; + struct percpu_counter i_caps_mis; };#endif /* _FS_CEPH_MDS_METRIC_H */
fs/ceph/quota.c +7 -2
@@ -147,9 +147,14 @@ static struct inode *lookup_quotarealm_inode(struct ceph_mds_client *mdsc,return NULL ; }if (qri->inode) { +struct ceph_inode_info *ci =ceph_inode (qri->inode); +int ret; + +ceph_caps_metric (ci,CEPH_STAT_CAP_INODE ); +/* get caps */ -int ret =__ceph_do_getattr (qri->inode,NULL , -CEPH_STAT_CAP_INODE , true); + ret =__ceph_do_getattr (qri->inode,NULL , +CEPH_STAT_CAP_INODE , true);if (ret >=0 ) in = qri->inode;else
fs/ceph/super.h +9 -0
@@ -641,6 +641,14 @@ static inline bool __ceph_is_any_real_caps(struct ceph_inode_info *ci)return !RB_EMPTY_ROOT (&ci->i_caps); } +extern void __ceph_caps_metric (struct ceph_inode_info *ci,int mask); +static inline void ceph_caps_metric (struct ceph_inode_info *ci,int mask) +{ +spin_lock (&ci->i_ceph_lock); +__ceph_caps_metric (ci, mask); +spin_unlock (&ci->i_ceph_lock); +} +extern int __ceph_caps_issued (struct ceph_inode_info *ci,int *implemented);extern int __ceph_caps_issued_mask (struct ceph_inode_info *ci,int mask,int t); extern int __ceph_caps_issued_other(struct ceph_inode_info *ci,
@@ -927,6 +935,7 @@ extern int __ceph_do_getattr(struct inode *inode, struct page *locked_page, int mask, bool force);static inline int ceph_do_getattr (struct inode *inode,int mask,bool force) { +ceph_caps_metric (ceph_inode (inode), mask);return __ceph_do_getattr (inode,NULL , mask, force); }extern int ceph_permission (struct inode *inode,int mask);
fs/ceph/xattr.c +14 -3
@@ -829,6 +829,7 @@ ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,struct ceph_vxattr *vxattr =NULL ;int req_mask;ssize_t err; +int ret =-1 ;/* let's see if a virtual xattr was requested */ vxattr =ceph_match_vxattr (inode, name);
@@ -856,7 +857,9 @@ ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,if (ci->i_xattrs.version ==0 || !((req_mask &CEPH_CAP_XATTR_SHARED ) || -__ceph_caps_issued_mask (ci,CEPH_CAP_XATTR_SHARED ,1 ))) { + (ret =__ceph_caps_issued_mask (ci,CEPH_CAP_XATTR_SHARED ,1 )))) { +if (ret !=-1 ) +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED );spin_unlock (&ci->i_ceph_lock);/* security module gets xattr while filling trace */
@@ -871,6 +874,9 @@ ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,if (err)return err;spin_lock (&ci->i_ceph_lock); + }else { +if (ret !=-1 ) +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED ); } err =__build_xattrs (inode);
@@ -907,19 +913,24 @@ ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)struct ceph_inode_info *ci =ceph_inode (inode);bool len_only = (size ==0 );u32 namelen; -int err; +int err, ret =-1 ;spin_lock (&ci->i_ceph_lock);dout ("listxattr %p ver=%lld index_ver=%lld\n" , inode, ci->i_xattrs.version, ci->i_xattrs.index_version);if (ci->i_xattrs.version ==0 || - !__ceph_caps_issued_mask (ci,CEPH_CAP_XATTR_SHARED ,1 )) { + !(ret =__ceph_caps_issued_mask (ci,CEPH_CAP_XATTR_SHARED ,1 ))) { +if (ret !=-1 ) +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED );spin_unlock (&ci->i_ceph_lock); err =ceph_do_getattr (inode,CEPH_STAT_CAP_XATTR , true);if (err)return err;spin_lock (&ci->i_ceph_lock); + }else { +if (ret !=-1 ) +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED ); } err =__build_xattrs (inode);
Signature · 1 line
2.21.0
Re: [PATCH v6 2/9] ceph: add caps perf metric for each session
Quoted 140 lines · On Mon, 2020-02-10 at 00:34 -0500, [email protected] wrote:
From: Xiubo Li <[email protected]>
This will fulfill the cap hit/mis metric stuff per-superblock, it will count the hit/mis counters based each inode, and if one inode's 'issued & ~revoking == mask' will mean a hit, or a miss.
item total miss hit
-------------------------------------------------
caps 295 107 4119URL: https://tracker.ceph.com/issues/43215
Signed-off-by Xiubo Li <[email protected]>--- fs/ceph/acl.c | 2 ++ fs/ceph/caps.c | 29 +++++++++++++++++++++++++++++ fs/ceph/debugfs.c | 16 ++++++++++++++++ fs/ceph/dir.c | 9 +++++++-- fs/ceph/file.c | 2 ++ fs/ceph/mds_client.c | 26 ++++++++++++++++++++++---- fs/ceph/metric.h | 3 +++ fs/ceph/quota.c | 9 +++++++-- fs/ceph/super.h | 9 +++++++++ fs/ceph/xattr.c | 17 ++++++++++++++--- 10 files changed, 111 insertions(+), 11 deletions(-)fs/ceph/acl.c +2
-0@@ -22,6 +22,8 @@ static inline void ceph_set_cached_acl(struct inode *inode,struct ceph_inode_info *ci =ceph_inode (inode);spin_lock (&ci->i_ceph_lock); +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED ); +if (__ceph_caps_issued_mask (ci,CEPH_CAP_XATTR_SHARED ,0 ))set_cached_acl (inode, type, acl);else fs/ceph/caps.c +29
-0@@ -818,6 +818,32 @@ int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)return have; } +/* +* Counts the cap metric. +* +* This will try to traverse all the ci->i_caps, if we can +* get all the cap 'mask' it will count the hit, or the mis. +*/ +void __ceph_caps_metric (struct ceph_inode_info *ci,int mask) +{ +struct ceph_mds_client *mdsc = +ceph_sb_to_client (ci->vfs_inode.i_sb)->mdsc; +struct ceph_client_metric *metric = &mdsc->metric; +int issued; + +lockdep_assert_held (&ci->i_ceph_lock); + +if (mask <=0 ) +return ; + + issued =__ceph_caps_issued (ci,NULL ); + +if ((mask & issued) == mask) +percpu_counter_inc (&metric->i_caps_hit); +else +percpu_counter_inc (&metric->i_caps_mis); +} +/* * Get cap bits issued by caps other than @ocap */ @@ -2758,6 +2784,7 @@ int ceph_try_get_caps(struct inode *inode, int need, int want,BUG_ON (want & ~(CEPH_CAP_FILE_CACHE |CEPH_CAP_FILE_LAZYIO |CEPH_CAP_FILE_SHARED |CEPH_CAP_FILE_EXCL |CEPH_CAP_ANY_DIR_OPS )); +ceph_caps_metric (ceph_inode (inode), need | want); ret =try_get_cap_refs (inode, need, want,0 , nonblock, got);return ret == -EAGAIN ?0 : ret; }@@ -2784,6 +2811,8 @@ int ceph_get_caps(struct file *filp, int need, int want, fi->filp_gen != READ_ONCE(fsc->filp_gen))return -EBADF ; +ceph_caps_metric (ci, need | want); +while (true) {if (endoff >0 )check_max_size (inode, endoff);fs/ceph/debugfs.c +16
-0@@ -128,6 +128,7 @@ static int metric_show(struct seq_file *s, void *p) {struct ceph_fs_client *fsc = s->private;struct ceph_mds_client *mdsc = fsc->mdsc; +int i, nr_caps =0 ;seq_printf (s,"item total miss hit\n" );seq_printf (s,"-------------------------------------------------\n" );@@ -137,6 +138,21 @@ static int metric_show(struct seq_file *s, void *p)percpu_counter_sum (&mdsc->metric.d_lease_mis), percpu_counter_sum(&mdsc->metric.d_lease_hit)); +mutex_lock (&mdsc->mutex); +for (i =0 ; i < mdsc->max_sessions; i++) { +struct ceph_mds_session *s; + + s =__ceph_lookup_mds_session (mdsc, i); +if (!s) +continue ; + nr_caps += s->s_nr_caps; +ceph_put_mds_session (s); + } +mutex_unlock (&mdsc->mutex); +seq_printf (s,"%-14s%-16d%-16lld%lld\n" ,"caps" , nr_caps, +percpu_counter_sum (&mdsc->metric.i_caps_mis), +percpu_counter_sum (&mdsc->metric.i_caps_hit)); +return 0 ; }fs/ceph/dir.c +1
-1@@ -313,7 +313,7 @@ static int ceph_readdir(struct file *file, struct dir_context *ctx)struct ceph_fs_client *fsc =ceph_inode_to_client (inode);struct ceph_mds_client *mdsc = fsc->mdsc;int i; -int err; +int err, ret =-1 ;unsigned frag =-1 ;struct ceph_mds_reply_info_parsed *rinfo;
fs/ceph/dir.c static int ceph_readdir(struct file *file, struct dir_context *ctx) · line 351
Quoted 7 lines
fs/ceph/dir.c +2
-1@@ -346,13 +346,16 @@ static int ceph_readdir(struct file *file, struct dir_context *ctx) !ceph_test_mount_opt(fsc, NOASYNCREADDIR) && ceph_snap(inode) != CEPH_SNAPDIR && __ceph_dir_is_complete_ordered(ci) && - __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) { + (ret = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {int shared_gen =atomic_read (&ci->i_shared_gen); +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED );
Why is this dealing with Xs caps when you've checked Fs?
fs/ceph/dir.c static int ceph_readdir(struct file *file, struct dir_context *ctx) · line 358
Quoted 7 lines
fs/ceph/dir.c +2
-0spin_unlock (&ci->i_ceph_lock); err =__dcache_readdir (file, ctx, shared_gen);if (err != -EAGAIN )return err; }else { +if (ret !=-1 ) +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED );
Ditto.
Quoted 193 lines
fs/ceph/dir.c +2
-0spin_unlock (&ci->i_ceph_lock); }@@ -757,6 +760,8 @@ static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,struct ceph_dentry_info *di =ceph_dentry (dentry);spin_lock (&ci->i_ceph_lock); +__ceph_caps_metric (ci,CEPH_CAP_FILE_SHARED ); +dout (" dir %p flags are %d\n" , dir, ci->i_ceph_flags); if (strncmp(dentry->d_name.name, fsc->mount_options->snapdir_name,fs/ceph/file.c +2
-0@@ -384,6 +384,8 @@ int ceph_open(struct inode *inode, struct file *file)* asynchronously. */ spin_lock (&ci->i_ceph_lock); +__ceph_caps_metric (ci, wanted); +if (__ceph_is_any_real_caps (ci) && (((fmode &CEPH_FILE_MODE_WR ) ==0 ) || ci->i_auth_cap)) {int mds_wanted =__ceph_caps_mds_wanted (ci, true);fs/ceph/mds_client.c +22
-4@@ -4169,13 +4169,29 @@ static int ceph_mdsc_metric_init(struct ceph_client_metric *metric) ret =percpu_counter_init (&metric->d_lease_hit,0 ,GFP_KERNEL );if (ret)return ret; + ret =percpu_counter_init (&metric->d_lease_mis,0 ,GFP_KERNEL ); -if (ret) { -percpu_counter_destroy (&metric->d_lease_hit); -return ret; - } +if (ret) + goto err_d_lease_mis; + + ret =percpu_counter_init (&metric->i_caps_hit,0 ,GFP_KERNEL ); +if (ret) + goto err_i_caps_hit; + + ret =percpu_counter_init (&metric->i_caps_mis,0 ,GFP_KERNEL ); +if (ret) + goto err_i_caps_mis;return 0 ; + +err_i_caps_mis: +percpu_counter_destroy (&metric->i_caps_hit); +err_i_caps_hit: +percpu_counter_destroy (&metric->d_lease_mis); +err_d_lease_mis: +percpu_counter_destroy (&metric->d_lease_hit); + +return ret; }int ceph_mdsc_init (struct ceph_fs_client *fsc)@@ -4515,6 +4531,8 @@ void ceph_mdsc_destroy(struct ceph_fs_client *fsc)ceph_mdsc_stop (mdsc); +percpu_counter_destroy (&mdsc->metric.i_caps_mis); +percpu_counter_destroy (&mdsc->metric.i_caps_hit);percpu_counter_destroy (&mdsc->metric.d_lease_mis);percpu_counter_destroy (&mdsc->metric.d_lease_hit);fs/ceph/metric.h +3
-0@@ -7,5 +7,8 @@ struct ceph_client_metric {atomic64_t total_dentries;struct percpu_counter d_lease_hit;struct percpu_counter d_lease_mis; + +struct percpu_counter i_caps_hit; + struct percpu_counter i_caps_mis; };#endif /* _FS_CEPH_MDS_METRIC_H */ fs/ceph/quota.c +7
-2@@ -147,9 +147,14 @@ static struct inode *lookup_quotarealm_inode(struct ceph_mds_client *mdsc,return NULL ; }if (qri->inode) { +struct ceph_inode_info *ci =ceph_inode (qri->inode); +int ret; + +ceph_caps_metric (ci,CEPH_STAT_CAP_INODE ); +/* get caps */ -int ret =__ceph_do_getattr (qri->inode,NULL , -CEPH_STAT_CAP_INODE , true); + ret =__ceph_do_getattr (qri->inode,NULL , +CEPH_STAT_CAP_INODE , true);if (ret >=0 ) in = qri->inode;else fs/ceph/super.h +9
-0@@ -641,6 +641,14 @@ static inline bool __ceph_is_any_real_caps(struct ceph_inode_info *ci)return !RB_EMPTY_ROOT (&ci->i_caps); } +extern void __ceph_caps_metric (struct ceph_inode_info *ci,int mask); +static inline void ceph_caps_metric (struct ceph_inode_info *ci,int mask) +{ +spin_lock (&ci->i_ceph_lock); +__ceph_caps_metric (ci, mask); +spin_unlock (&ci->i_ceph_lock); +} +extern int __ceph_caps_issued (struct ceph_inode_info *ci,int *implemented);extern int __ceph_caps_issued_mask (struct ceph_inode_info *ci,int mask,int t); extern int __ceph_caps_issued_other(struct ceph_inode_info *ci,@@ -927,6 +935,7 @@ extern int __ceph_do_getattr(struct inode *inode, struct page *locked_page, int mask, bool force);static inline int ceph_do_getattr (struct inode *inode,int mask,bool force) { +ceph_caps_metric (ceph_inode (inode), mask);return __ceph_do_getattr (inode,NULL , mask, force); }extern int ceph_permission (struct inode *inode,int mask);fs/ceph/xattr.c +14
-3@@ -829,6 +829,7 @@ ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,struct ceph_vxattr *vxattr =NULL ;int req_mask;ssize_t err; +int ret =-1 ;/* let's see if a virtual xattr was requested */ vxattr =ceph_match_vxattr (inode, name);@@ -856,7 +857,9 @@ ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,if (ci->i_xattrs.version ==0 || !((req_mask &CEPH_CAP_XATTR_SHARED ) || -__ceph_caps_issued_mask (ci,CEPH_CAP_XATTR_SHARED ,1 ))) { + (ret =__ceph_caps_issued_mask (ci,CEPH_CAP_XATTR_SHARED ,1 )))) { +if (ret !=-1 ) +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED );spin_unlock (&ci->i_ceph_lock);/* security module gets xattr while filling trace */ @@ -871,6 +874,9 @@ ssize_t __ceph_getxattr(struct inode *inode, const char *name, void *value,if (err)return err;spin_lock (&ci->i_ceph_lock); + }else { +if (ret !=-1 ) +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED ); } err =__build_xattrs (inode);@@ -907,19 +913,24 @@ ssize_t ceph_listxattr(struct dentry *dentry, char *names, size_t size)struct ceph_inode_info *ci =ceph_inode (inode);bool len_only = (size ==0 );u32 namelen; -int err; +int err, ret =-1 ;spin_lock (&ci->i_ceph_lock);dout ("listxattr %p ver=%lld index_ver=%lld\n" , inode, ci->i_xattrs.version, ci->i_xattrs.index_version);if (ci->i_xattrs.version ==0 || - !__ceph_caps_issued_mask (ci,CEPH_CAP_XATTR_SHARED ,1 )) { + !(ret =__ceph_caps_issued_mask (ci,CEPH_CAP_XATTR_SHARED ,1 ))) { +if (ret !=-1 ) +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED );spin_unlock (&ci->i_ceph_lock); err =ceph_do_getattr (inode,CEPH_STAT_CAP_XATTR , true);if (err)return err;spin_lock (&ci->i_ceph_lock); + }else { +if (ret !=-1 ) +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED ); } err =__build_xattrs (inode);
Signature · 1 line
Jeff Layton <[email protected]>
Re: [PATCH v6 2/9] ceph: add caps perf metric for each session
Quoted 12 lines · Jeff Layton
Quoted 11 lines · On Mon, 2020-02-10 at 00:34 -0500, [email protected] wrote:
From: Xiubo Li <[email protected]>
This will fulfill the cap hit/mis metric stuff per-superblock, it will count the hit/mis counters based each inode, and if one inode's 'issued & ~revoking == mask' will mean a hit, or a miss.
item total miss hit
-------------------------------------------------
caps 295 107 4119[]
[...]
Quoted 10 lines
fs/ceph/dir.c static int ceph_readdir(struct file *file, struct dir_context *ctx) · line 351
Quoted 9 lines
fs/ceph/dir.c +2
-1@@ -346,13 +346,16 @@ static int ceph_readdir(struct file *file, struct dir_context *ctx) !ceph_test_mount_opt(fsc, NOASYNCREADDIR) && ceph_snap(inode) != CEPH_SNAPDIR && __ceph_dir_is_complete_ordered(ci) && - __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) { + (ret = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {int shared_gen =atomic_read (&ci->i_shared_gen); +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED );Why is this dealing with Xs caps when you've checked Fs?
Good catch.
This might just copied from some where and forgot to change it. Will fix it.
Quoted 9 lines
fs/ceph/dir.c static int ceph_readdir(struct file *file, struct dir_context *ctx) · line 358
Quoted 8 lines
fs/ceph/dir.c +2
-0spin_unlock (&ci->i_ceph_lock); err =__dcache_readdir (file, ctx, shared_gen);if (err != -EAGAIN )return err; }else { +if (ret !=-1 ) +__ceph_caps_metric (ci,CEPH_CAP_XATTR_SHARED );Ditto.
Here too.
Thanks.
Xiubo
PATCH v6 3/9 ceph: add global read latency metric support
In reply to [email protected]From: Xiubo Li <[email protected]>
It will calculate the latency for the read osd requests, which only include the time cousumed by network and the ceph osd.
item total sum_lat(us) avg_lat(us)
-----------------------------------------------------
read 1036 848000 818
URL: https://tracker.ceph.com/issues/43215
--- fs/ceph/addr.c | 6 ++++++ fs/ceph/debugfs.c | 11 +++++++++++ fs/ceph/file.c | 13 +++++++++++++ fs/ceph/mds_client.c | 14 ++++++++++++++ fs/ceph/metric.h | 20 ++++++++++++++++++++ include/linux/ceph/osd_client.h | 1 + net/ceph/osd_client.c | 2 ++ 7 files changed, 67 insertions(+)
fs/ceph/addr.c +6 -0
@@ -216,6 +216,8 @@ static int ceph_sync_readpages(struct ceph_fs_client *fsc,if (!rc) rc =ceph_osdc_wait_request (osdc, req); +ceph_update_read_latency (&fsc->mdsc->metric, req, rc); +ceph_osdc_put_request (req);dout ("readpages result %d\n" , rc);return rc;
@@ -299,6 +301,7 @@ static int ceph_readpage(struct file *filp, struct page *page)static void finish_read (struct ceph_osd_request *req) {struct inode *inode = req->r_inode; +struct ceph_fs_client *fsc =ceph_inode_to_client (inode);struct ceph_osd_data *osd_data;int rc = req->r_result <=0 ? req->r_result :0 ;int bytes = req->r_result >=0 ? req->r_result :0 ;
@@ -336,6 +339,9 @@ static void finish_read(struct ceph_osd_request *req)put_page (page); bytes -=PAGE_SIZE ; } + +ceph_update_read_latency (&fsc->mdsc->metric, req, rc); +kfree (osd_data->pages); }
fs/ceph/debugfs.c +11 -0
@@ -129,7 +129,18 @@ static int metric_show(struct seq_file *s, void *p)struct ceph_fs_client *fsc = s->private;struct ceph_mds_client *mdsc = fsc->mdsc;int i, nr_caps =0 ; +s64 total, sum, avg =0 ; +seq_printf (s,"item total sum_lat(us) avg_lat(us)\n" ); +seq_printf (s,"-----------------------------------------------------\n" ); + + total =percpu_counter_sum (&mdsc->metric.total_reads); + sum =percpu_counter_sum (&mdsc->metric.read_latency_sum); + sum =jiffies_to_usecs (sum); + avg = total ? sum / total :0 ; +seq_printf (s,"%-14s%-12lld%-16lld%lld\n" ,"read" , total, sum, avg); + +seq_printf (s,"\n" );seq_printf (s,"item total miss hit\n" );seq_printf (s,"-------------------------------------------------\n" );
fs/ceph/file.c +13 -0
@@ -660,6 +660,9 @@ static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to, ret =ceph_osdc_start_request (osdc, req, false);if (!ret) ret =ceph_osdc_wait_request (osdc, req); + +ceph_update_read_latency (&fsc->mdsc->metric, req, ret); +ceph_osdc_put_request (req); i_size =i_size_read (inode);
@@ -798,6 +801,8 @@ static void ceph_aio_complete_req(struct ceph_osd_request *req)struct inode *inode = req->r_inode;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_fs_client *fsc =ceph_inode_to_client (inode); +struct ceph_client_metric *metric = &fsc->mdsc->metric;BUG_ON (osd_data->type !=CEPH_OSD_DATA_TYPE_BVECS );BUG_ON (!osd_data->num_bvecs);
@@ -805,6 +810,10 @@ static void ceph_aio_complete_req(struct ceph_osd_request *req)dout ("ceph_aio_complete_req %p rc %d bytes %u\n" , inode, rc, osd_data->bvec_pos.iter.bi_size); +/* r_start_stamp == 0 means the request was not submitted */ +if (req->r_start_stamp && !aio_req->write) +ceph_update_read_latency (metric, req, rc); +if (rc == -EOLDSNAPC ) {struct ceph_aio_work *aio_work;BUG_ON (!aio_req->write);
@@ -933,6 +942,7 @@ ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,struct inode *inode =file_inode (file);struct ceph_inode_info *ci =ceph_inode (inode);struct ceph_fs_client *fsc =ceph_inode_to_client (inode); +struct ceph_client_metric *metric = &fsc->mdsc->metric;struct ceph_vino vino;struct ceph_osd_request *req;struct bio_vec *bvecs;
@@ -1049,6 +1059,9 @@ ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,if (!ret) ret =ceph_osdc_wait_request (&fsc->client->osdc, req); +if (!write) +ceph_update_read_latency (metric, req, ret); + size =i_size_read (inode);if (!write) {if (ret == -ENOENT )
fs/ceph/mds_client.c +14 -0
@@ -4182,8 +4182,20 @@ static int ceph_mdsc_metric_init(struct ceph_client_metric *metric)if (ret) goto err_i_caps_mis; + ret =percpu_counter_init (&metric->total_reads,0 ,GFP_KERNEL ); +if (ret) + goto err_total_reads; + + ret =percpu_counter_init (&metric->read_latency_sum,0 ,GFP_KERNEL ); +if (ret) + goto err_read_latency_sum; +return 0 ; +err_read_latency_sum: +percpu_counter_destroy (&metric->total_reads); +err_total_reads: +percpu_counter_destroy (&metric->i_caps_mis); err_i_caps_mis:percpu_counter_destroy (&metric->i_caps_hit); err_i_caps_hit:
@@ -4531,6 +4543,8 @@ void ceph_mdsc_destroy(struct ceph_fs_client *fsc)ceph_mdsc_stop (mdsc); +percpu_counter_destroy (&mdsc->metric.read_latency_sum); +percpu_counter_destroy (&mdsc->metric.total_reads);percpu_counter_destroy (&mdsc->metric.i_caps_mis);percpu_counter_destroy (&mdsc->metric.i_caps_hit);percpu_counter_destroy (&mdsc->metric.d_lease_mis);
fs/ceph/metric.h +20 -0
@@ -2,6 +2,8 @@ #ifndef _FS_CEPH_MDS_METRIC_H #define _FS_CEPH_MDS_METRIC_H +#include <linux/ceph/osd_client.h> + /* This is the global metrics */ struct ceph_client_metric { atomic64_t total_dentries;
@@ -10,5 +12,23 @@ struct ceph_client_metric {struct percpu_counter i_caps_hit;struct percpu_counter i_caps_mis; + +struct percpu_counter total_reads; + struct percpu_counter read_latency_sum; }; + +static inline void ceph_update_read_latency (struct ceph_client_metric *m, +struct ceph_osd_request *req, +int rc) +{ +if (!m || !req) +return ; + +if (rc >=0 || rc == -ENOENT || rc == -ETIMEDOUT ) { +s64 latency = req->r_end_stamp - req->r_start_stamp; + +percpu_counter_inc (&m->total_reads); +percpu_counter_add (&m->read_latency_sum, latency); + } +}#endif /* _FS_CEPH_MDS_METRIC_H */
include/linux/ceph/osd_client.h +1 -0
@@ -213,6 +213,7 @@ struct ceph_osd_request {/* internal */ unsigned long r_stamp;/* jiffies, send or check time */ unsigned long r_start_stamp;/* jiffies */ +unsigned long r_end_stamp;/* jiffies */ int r_attempts;u32 r_map_dne_bound;
net/ceph/osd_client.c +2 -0
@@ -2389,6 +2389,8 @@ static void finish_request(struct ceph_osd_request *req)WARN_ON (lookup_request_mc (&osdc->map_checks, req->r_tid));dout ("%s req %p tid %llu\n" , __func__, req, req->r_tid); + req->r_end_stamp = jiffies; +if (req->r_osd)unlink_request (req->r_osd, req);atomic_dec (&osdc->num_requests);
Signature · 1 line
2.21.0
PATCH v6 4/9 ceph: add global write latency metric support
In reply to [email protected]From: Xiubo Li <[email protected]>
It will calculate the latency for the write osd requests, which only include the time cousumed by network and the ceph osd.
item total sum_lat(us) avg_lat(us)
-----------------------------------------------------
write 1048 8778000 8375
URL: https://tracker.ceph.com/issues/43215
--- fs/ceph/addr.c | 7 +++++++ fs/ceph/debugfs.c | 6 ++++++ fs/ceph/file.c | 13 ++++++++++--- fs/ceph/mds_client.c | 14 ++++++++++++++ fs/ceph/metric.h | 18 ++++++++++++++++++ 5 files changed, 55 insertions(+), 3 deletions(-)
fs/ceph/addr.c +7 -0
@@ -649,6 +649,8 @@ static int ceph_sync_writepages(struct ceph_fs_client *fsc,if (!rc) rc =ceph_osdc_wait_request (osdc, req); +ceph_update_write_latency (&fsc->mdsc->metric, req, rc); +ceph_osdc_put_request (req);if (rc ==0 ) rc = len;
@@ -800,6 +802,8 @@ static void writepages_finish(struct ceph_osd_request *req)ceph_clear_error_write (ci); } +ceph_update_write_latency (&fsc->mdsc->metric, req, rc); +/* * We lost the cache cap, need to truncate the page before * it is unlocked, otherwise we'd truncate it later in the
@@ -1858,6 +1862,9 @@ int ceph_uninline_data(struct file *filp, struct page *locked_page) err =ceph_osdc_start_request (&fsc->client->osdc, req, false);if (!err) err =ceph_osdc_wait_request (&fsc->client->osdc, req); + +ceph_update_write_latency (&fsc->mdsc->metric, req, err); + out_put:ceph_osdc_put_request (req);if (err == -ECANCELED )
fs/ceph/debugfs.c +6 -0
@@ -140,6 +140,12 @@ static int metric_show(struct seq_file *s, void *p) avg = total ? sum / total :0 ;seq_printf (s,"%-14s%-12lld%-16lld%lld\n" ,"read" , total, sum, avg); + total =percpu_counter_sum (&mdsc->metric.total_writes); + sum =percpu_counter_sum (&mdsc->metric.write_latency_sum); + sum =jiffies_to_usecs (sum); + avg = total ? sum / total :0 ; +seq_printf (s,"%-14s%-12lld%-16lld%lld\n" ,"write" , total, sum, avg); +seq_printf (s,"\n" );seq_printf (s,"item total miss hit\n" );seq_printf (s,"-------------------------------------------------\n" );
fs/ceph/file.c +10 -3
@@ -811,8 +811,12 @@ static void ceph_aio_complete_req(struct ceph_osd_request *req) inode, rc, osd_data->bvec_pos.iter.bi_size);/* r_start_stamp == 0 means the request was not submitted */ -if (req->r_start_stamp && !aio_req->write) -ceph_update_read_latency (metric, req, rc); +if (req->r_start_stamp) { +if (aio_req->write) +ceph_update_write_latency (metric, req, rc); +else +ceph_update_read_latency (metric, req, rc); + }if (rc == -EOLDSNAPC ) {struct ceph_aio_work *aio_work;
@@ -1059,7 +1063,9 @@ ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,if (!ret) ret =ceph_osdc_wait_request (&fsc->client->osdc, req); -if (!write) +if (write) +ceph_update_write_latency (metric, req, ret); +else ceph_update_read_latency (metric, req, ret); size =i_size_read (inode);
@@ -1233,6 +1239,7 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,if (!ret) ret =ceph_osdc_wait_request (&fsc->client->osdc, req); +ceph_update_write_latency (&fsc->mdsc->metric, req, ret); out:ceph_osdc_put_request (req);if (ret !=0 ) {
fs/ceph/mds_client.c +14 -0
@@ -4190,8 +4190,20 @@ static int ceph_mdsc_metric_init(struct ceph_client_metric *metric)if (ret) goto err_read_latency_sum; + ret =percpu_counter_init (&metric->total_writes,0 ,GFP_KERNEL ); +if (ret) + goto err_total_writes; + + ret =percpu_counter_init (&metric->write_latency_sum,0 ,GFP_KERNEL ); +if (ret) + goto err_write_latency_sum; +return 0 ; +err_write_latency_sum: +percpu_counter_destroy (&metric->total_writes); +err_total_writes: +percpu_counter_destroy (&metric->read_latency_sum); err_read_latency_sum:percpu_counter_destroy (&metric->total_reads); err_total_reads:
@@ -4543,6 +4555,8 @@ void ceph_mdsc_destroy(struct ceph_fs_client *fsc)ceph_mdsc_stop (mdsc); +percpu_counter_destroy (&mdsc->metric.write_latency_sum); +percpu_counter_destroy (&mdsc->metric.total_writes);percpu_counter_destroy (&mdsc->metric.read_latency_sum);percpu_counter_destroy (&mdsc->metric.total_reads);percpu_counter_destroy (&mdsc->metric.i_caps_mis);
fs/ceph/metric.h +18 -0
@@ -15,6 +15,9 @@ struct ceph_client_metric {struct percpu_counter total_reads;struct percpu_counter read_latency_sum; + +struct percpu_counter total_writes; +struct percpu_counter write_latency_sum; }; static inline void ceph_update_read_latency(struct ceph_client_metric *m,
@@ -31,4 +34,19 @@ static inline void ceph_update_read_latency(struct ceph_client_metric *m, percpu_counter_add(&m->read_latency_sum, latency); } } + +static inline void ceph_update_write_latency (struct ceph_client_metric *m, +struct ceph_osd_request *req, +int rc) +{ +if (!m || !req) +return ; + +if (!rc || rc == -ETIMEDOUT ) { +s64 latency = req->r_end_stamp - req->r_start_stamp; + +percpu_counter_inc (&m->total_writes); +percpu_counter_add (&m->write_latency_sum, latency); + } +}#endif /* _FS_CEPH_MDS_METRIC_H */
Signature · 1 line
2.21.0
PATCH v6 5/9 ceph: add global metadata perf metric support
In reply to [email protected]From: Xiubo Li <[email protected]>
It will calculate the latency for the metedata requests, which only include the time cousumed by network and the ceph.
item total sum_lat(us) avg_lat(us)
-----------------------------------------------------
metadata 113 220000 1946
URL: https://tracker.ceph.com/issues/43215
--- fs/ceph/debugfs.c | 6 ++++++ fs/ceph/mds_client.c | 20 ++++++++++++++++++++ fs/ceph/metric.h | 13 +++++++++++++ 3 files changed, 39 insertions(+)
fs/ceph/debugfs.c +6 -0
@@ -146,6 +146,12 @@ static int metric_show(struct seq_file *s, void *p) avg = total ? sum / total :0 ;seq_printf (s,"%-14s%-12lld%-16lld%lld\n" ,"write" , total, sum, avg); + total =percpu_counter_sum (&mdsc->metric.total_metadatas); + sum =percpu_counter_sum (&mdsc->metric.metadata_latency_sum); + sum =jiffies_to_usecs (sum); + avg = total ? sum / total :0 ; +seq_printf (s,"%-14s%-12lld%-16lld%lld\n" ,"metadata" , total, sum, avg); +seq_printf (s,"\n" );seq_printf (s,"item total miss hit\n" );seq_printf (s,"-------------------------------------------------\n" );
fs/ceph/mds_client.c +20 -0
@@ -3019,6 +3019,12 @@ static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)/* kick calling process */ complete_request (mdsc, req); + +if (!result || result == -ENOENT ) { +s64 latency = jiffies - req->r_started; + +ceph_update_metadata_latency (&mdsc->metric, latency); + } out:ceph_mdsc_put_request (req);return ;
@@ -4198,8 +4204,20 @@ static int ceph_mdsc_metric_init(struct ceph_client_metric *metric)if (ret) goto err_write_latency_sum; + ret =percpu_counter_init (&metric->total_metadatas,0 ,GFP_KERNEL ); +if (ret) + goto err_total_metadatas; + + ret =percpu_counter_init (&metric->metadata_latency_sum,0 ,GFP_KERNEL ); +if (ret) + goto err_metadata_latency_sum; +return 0 ; +err_metadata_latency_sum: +percpu_counter_destroy (&metric->total_metadatas); +err_total_metadatas: +percpu_counter_destroy (&metric->write_latency_sum); err_write_latency_sum:percpu_counter_destroy (&metric->total_writes); err_total_writes:
@@ -4555,6 +4573,8 @@ void ceph_mdsc_destroy(struct ceph_fs_client *fsc)ceph_mdsc_stop (mdsc); +percpu_counter_destroy (&mdsc->metric.metadata_latency_sum); +percpu_counter_destroy (&mdsc->metric.total_metadatas);percpu_counter_destroy (&mdsc->metric.write_latency_sum);percpu_counter_destroy (&mdsc->metric.total_writes);percpu_counter_destroy (&mdsc->metric.read_latency_sum);
fs/ceph/metric.h +13 -0
@@ -18,6 +18,9 @@ struct ceph_client_metric {struct percpu_counter total_writes;struct percpu_counter write_latency_sum; + +struct percpu_counter total_metadatas; +struct percpu_counter metadata_latency_sum; }; static inline void ceph_update_read_latency(struct ceph_client_metric *m,
@@ -49,4 +52,14 @@ static inline void ceph_update_write_latency(struct ceph_client_metric *m, percpu_counter_add(&m->write_latency_sum, latency); } } + +static inline void ceph_update_metadata_latency (struct ceph_client_metric *m, +s64 latency) +{ +if (!m) +return ; + +percpu_counter_inc (&m->total_metadatas); +percpu_counter_add (&m->metadata_latency_sum, latency); +}#endif /* _FS_CEPH_MDS_METRIC_H */
Signature · 1 line
2.21.0
PATCH v6 6/9 ceph: periodically send perf metrics to ceph
In reply to [email protected]From: Xiubo Li <[email protected]>
Add metric_send_interval module parameter support, the default valume is 0, means disabled. If none zero it will enable the transmission of the metrics to the ceph cluster periodically per metric_send_interval seconds.
This will send the caps, dentry lease and read/write/metadata perf metrics to any available MDS only once per metric_send_interval seconds.
URL: https://tracker.ceph.com/issues/43215
--- fs/ceph/mds_client.c | 235 +++++++++++++++++++++++++++++++---- fs/ceph/mds_client.h | 2 + fs/ceph/metric.h | 76 +++++++++++ fs/ceph/super.c | 4 + fs/ceph/super.h | 1 + include/linux/ceph/ceph_fs.h | 1 + 6 files changed, 294 insertions(+), 25 deletions(-)
fs/ceph/mds_client.c +210 -25
@@ -4085,16 +4085,167 @@ static void maybe_recover_session(struct ceph_mds_client *mdsc)ceph_force_reconnect (fsc->sb); } -/* -* delayed work -- periodically trim expired leases, renew caps with mds -*/ +static bool ceph_mdsc_send_metrics (struct ceph_mds_client *mdsc, +struct ceph_mds_session *s, +u64 nr_caps) +{ +struct ceph_metric_head *head; +struct ceph_metric_cap *cap; +struct ceph_metric_dentry_lease *lease; +struct ceph_metric_read_latency *read; +struct ceph_metric_write_latency *write; +struct ceph_metric_metadata_latency *meta; +struct ceph_msg *msg; +struct timespec64 ts; +s64 sum, total; +s32 items =0 ; +s32 len; + +if (!mdsc || !s) +return false; + + len =sizeof (*head) +sizeof (*cap) +sizeof (*lease) +sizeof (*read) + +sizeof (*write) +sizeof (*meta); + + msg =ceph_msg_new (CEPH_MSG_CLIENT_METRICS , len,GFP_NOFS , true); +if (!msg) { +pr_err ("send metrics to mds%d, failed to allocate message\n" , + s->s_mds); +return false; + } + + head = msg->front.iov_base; + +/* encode the cap metric */ + cap = (struct ceph_metric_cap *)(head +1 ); + cap->type =cpu_to_le32 (CLIENT_METRIC_TYPE_CAP_INFO ); + cap->ver =1 ; + cap->compat =1 ; + cap->data_len =cpu_to_le32 (sizeof (*cap) -10 ); + cap->hit =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.i_caps_hit)); + cap->mis =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.i_caps_mis)); + cap->total =cpu_to_le64 (nr_caps); + items++; + +dout ("cap metric hit %lld, mis %lld, total caps %lld" , +le64_to_cpu (cap->hit),le64_to_cpu (cap->mis), +le64_to_cpu (cap->total)); + +/* encode the read latency metric */ + read = (struct ceph_metric_read_latency *)(cap +1 ); + read->type =cpu_to_le32 (CLIENT_METRIC_TYPE_READ_LATENCY ); + read->ver =1 ; + read->compat =1 ; + read->data_len =cpu_to_le32 (sizeof (*read) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_reads), + sum =percpu_counter_sum (&mdsc->metric.read_latency_sum); +jiffies_to_timespec64 (sum, &ts); + read->sec =cpu_to_le32 (ts.tv_sec); + read->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("read latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the write latency metric */ + write = (struct ceph_metric_write_latency *)(read +1 ); + write->type =cpu_to_le32 (CLIENT_METRIC_TYPE_WRITE_LATENCY ); + write->ver =1 ; + write->compat =1 ; + write->data_len =cpu_to_le32 (sizeof (*write) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_writes), + sum =percpu_counter_sum (&mdsc->metric.write_latency_sum); +jiffies_to_timespec64 (sum, &ts); + write->sec =cpu_to_le32 (ts.tv_sec); + write->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("write latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the metadata latency metric */ + meta = (struct ceph_metric_metadata_latency *)(write +1 ); + meta->type =cpu_to_le32 (CLIENT_METRIC_TYPE_METADATA_LATENCY ); + meta->ver =1 ; + meta->compat =1 ; + meta->data_len =cpu_to_le32 (sizeof (*meta) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_metadatas), + sum =percpu_counter_sum (&mdsc->metric.metadata_latency_sum); +jiffies_to_timespec64 (sum, &ts); + meta->sec =cpu_to_le32 (ts.tv_sec); + meta->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("metadata latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the dentry lease metric */ + lease = (struct ceph_metric_dentry_lease *)(meta +1 ); + lease->type =cpu_to_le32 (CLIENT_METRIC_TYPE_DENTRY_LEASE ); + lease->ver =1 ; + lease->compat =1 ; + lease->data_len =cpu_to_le32 (sizeof (*lease) -10 ); + lease->hit =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.d_lease_hit)); + lease->mis =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.d_lease_mis)); + lease->total =cpu_to_le64 (atomic64_read (&mdsc->metric.total_dentries)); + items++; +dout ("dentry lease metric hit %lld, mis %lld, total dentries %lld" , +le64_to_cpu (lease->hit),le64_to_cpu (lease->mis), +le64_to_cpu (lease->total)); + +put_unaligned_le32 (items, &head->num); + msg->front.iov_len =cpu_to_le32 (len); + msg->hdr.version =cpu_to_le16 (1 ); + msg->hdr.compat_version =cpu_to_le16 (1 ); + msg->hdr.front_len =cpu_to_le32 (msg->front.iov_len); +dout ("send metrics to mds%d %p\n" , s->s_mds, msg); +ceph_con_send (&s->s_con, msg); + +return true; +} + +#define CEPH_WORK_DELAY_DEF 5 +static void __schedule_delayed (struct delayed_work *work,int delay) +{ +unsigned int hz =round_jiffies_relative (HZ * delay); + +schedule_delayed_work (work, hz); +} +static void schedule_delayed (struct ceph_mds_client *mdsc) { -int delay =5 ; -unsigned hz =round_jiffies_relative (HZ * delay); -schedule_delayed_work (&mdsc->delayed_work, hz); +__schedule_delayed (&mdsc->delayed_work,CEPH_WORK_DELAY_DEF ); +} + +static void metric_schedule_delayed (struct ceph_mds_client *mdsc) +{ +/* delay CEPH_WORK_DELAY_DEF seconds when idle */ +int delay = metric_send_interval ? :CEPH_WORK_DELAY_DEF ; + +__schedule_delayed (&mdsc->metric_delayed_work, delay); +} + +static bool check_session_state (struct ceph_mds_client *mdsc, +struct ceph_mds_session *s) +{ +if (s->s_state ==CEPH_MDS_SESSION_CLOSING ) { +dout ("resending session close request for mds%d\n" , + s->s_mds); +request_close_session (mdsc, s); +return false; + } +if (s->s_ttl &&time_after (jiffies, s->s_ttl)) { +if (s->s_state ==CEPH_MDS_SESSION_OPEN ) { + s->s_state =CEPH_MDS_SESSION_HUNG ; +pr_info ("mds%d hung\n" , s->s_mds); + } + } +if (s->s_state ==CEPH_MDS_SESSION_NEW || + s->s_state ==CEPH_MDS_SESSION_RESTARTING || + s->s_state ==CEPH_MDS_SESSION_REJECTED ) +/* this mds is failed or recovering, just wait */ +return false; + +return true; } +/* +* delayed work -- periodically trim expired leases, renew caps with mds +*/ static void delayed_work (struct work_struct *work) {int i;
@@ -4116,23 +4267,8 @@ static void delayed_work(struct work_struct *work)struct ceph_mds_session *s =__ceph_lookup_mds_session (mdsc, i);if (!s)continue ; -if (s->s_state ==CEPH_MDS_SESSION_CLOSING ) { -dout ("resending session close request for mds%d\n" , - s->s_mds); -request_close_session (mdsc, s); -ceph_put_mds_session (s); -continue ; - } -if (s->s_ttl &&time_after (jiffies, s->s_ttl)) { -if (s->s_state ==CEPH_MDS_SESSION_OPEN ) { - s->s_state =CEPH_MDS_SESSION_HUNG ; -pr_info ("mds%d hung\n" , s->s_mds); - } - } -if (s->s_state ==CEPH_MDS_SESSION_NEW || - s->s_state ==CEPH_MDS_SESSION_RESTARTING || - s->s_state ==CEPH_MDS_SESSION_REJECTED ) { -/* this mds is failed or recovering, just wait */ + +if (!check_session_state (mdsc, s)) {ceph_put_mds_session (s);continue ; }
@@ -4164,8 +4300,53 @@ static void delayed_work(struct work_struct *work)schedule_delayed (mdsc); } -static int ceph_mdsc_metric_init (struct ceph_client_metric *metric) +static void metric_delayed_work (struct work_struct *work) +{ +struct ceph_mds_client *mdsc = + container_of(work, struct ceph_mds_client, metric_delayed_work.work); +struct ceph_mds_session *s; +u64 nr_caps =0 ; +bool ret; +int i; + +if (!metric_send_interval) + goto idle; + +dout ("mdsc metric_delayed_work\n" ); + +mutex_lock (&mdsc->mutex); +for (i =0 ; i < mdsc->max_sessions; i++) { + s =__ceph_lookup_mds_session (mdsc, i); +if (!s) +continue ; + nr_caps += s->s_nr_caps; +ceph_put_mds_session (s); + } + +for (i =0 ; i < mdsc->max_sessions; i++) { + s =__ceph_lookup_mds_session (mdsc, i); +if (!s) +continue ; +if (!check_session_state (mdsc, s)) { +ceph_put_mds_session (s); +continue ; + } + +/* Only send the metric once in any available session */ + ret =ceph_mdsc_send_metrics (mdsc, s, nr_caps); +ceph_put_mds_session (s); +if (ret) +break ; + } +mutex_unlock (&mdsc->mutex); + +idle: +metric_schedule_delayed (mdsc); +} + +static int ceph_mdsc_metric_init (struct ceph_mds_client *mdsc) { +struct ceph_client_metric *metric = &mdsc->metric;int ret;if (!metric)
@@ -4289,7 +4470,8 @@ int ceph_mdsc_init(struct ceph_fs_client *fsc)init_waitqueue_head (&mdsc->cap_flushing_wq);INIT_WORK (&mdsc->cap_reclaim_work, ceph_cap_reclaim_work);atomic_set (&mdsc->cap_reclaim_pending,0 ); - err =ceph_mdsc_metric_init (&mdsc->metric); +INIT_DELAYED_WORK (&mdsc->metric_delayed_work, metric_delayed_work); + err =ceph_mdsc_metric_init (mdsc);if (err) goto err_mdsmap;
@@ -4511,6 +4693,7 @@ void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)cancel_work_sync (&mdsc->cap_reclaim_work);cancel_delayed_work_sync (&mdsc->delayed_work);/* cancel timer */ +cancel_delayed_work_sync (&mdsc->metric_delayed_work);/* cancel timer */ dout ("stopped\n" ); }
@@ -4553,6 +4736,7 @@ static void ceph_mdsc_stop(struct ceph_mds_client *mdsc) {dout ("stop\n" );cancel_delayed_work_sync (&mdsc->delayed_work);/* cancel timer */ +cancel_delayed_work_sync (&mdsc->metric_delayed_work);/* cancel timer */ if (mdsc->mdsmap)ceph_mdsmap_destroy (mdsc->mdsmap);kfree (mdsc->sessions);
@@ -4719,6 +4903,7 @@ void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)mutex_unlock (&mdsc->mutex);schedule_delayed (mdsc); +metric_schedule_delayed (mdsc);return ; bad_unlock:
fs/ceph/mds_client.h +2 -0
@@ -448,7 +448,9 @@ struct ceph_mds_client {struct list_head dentry_leases;/* fifo list */ struct list_head dentry_dir_leases;/* lru list */ +/* metrics */ struct ceph_client_metric metric; +struct delayed_work metric_delayed_work;/* delayed work */ spinlock_t snapid_map_lock;struct rb_root snapid_map_tree;
fs/ceph/metric.h +76 -0
82 lines not on this page · raw message
fs/ceph/super.c +4 -0
@@ -1315,6 +1315,10 @@ bool enable_async_dirops;module_param (enable_async_dirops, bool,0644 );MODULE_PARM_DESC (enable_async_dirops,"Asynchronous directory operations enabled" ); +unsigned int metric_send_interval; +module_param (metric_send_interval, uint,0644 ); +MODULE_PARM_DESC (metric_send_interval,"Interval (in seconds) of sending perf metric to ceph cluster (default: 0)" ); +module_init (init_ceph);module_exit (exit_ceph);
fs/ceph/super.h +1 -0
@@ -73,6 +73,7 @@#define CEPH_CAPS_WANTED_DELAY_MAX_DEFAULT 60/* cap release delay */ extern bool enable_async_dirops; +extern unsigned int metric_send_interval;struct ceph_mount_options {unsigned int flags;
include/linux/ceph/ceph_fs.h +1 -0
@@ -130,6 +130,7 @@ struct ceph_dir_layout {#define CEPH_MSG_CLIENT_REQUEST 24#define CEPH_MSG_CLIENT_REQUEST_FORWARD 25#define CEPH_MSG_CLIENT_REPLY 26 +#define CEPH_MSG_CLIENT_METRICS 29#define CEPH_MSG_CLIENT_CAPS 0x310#define CEPH_MSG_CLIENT_LEASE 0x311#define CEPH_MSG_CLIENT_SNAP 0x312
Signature · 1 line
2.21.0
Ilya Dryomov
maintainerRe: [PATCH v6 6/9] ceph: periodically send perf metrics to ceph
Quoted 264 lines · On Mon, Feb 10, 2020 at 6:34 AM <[email protected]> wrote:
From: Xiubo Li <[email protected]>
Add
metric_send_intervalmodule parameter support, the default valume is 0, means disabled. If none zero it will enable the transmission of the metrics to the ceph cluster periodically permetric_send_intervalseconds.This will send the caps, dentry lease and read/write/metadata perf metrics to any available MDS only once per
metric_send_intervalseconds.URL: https://tracker.ceph.com/issues/43215
Signed-off-by Xiubo Li <[email protected]>--- fs/ceph/mds_client.c | 235 +++++++++++++++++++++++++++++++---- fs/ceph/mds_client.h | 2 + fs/ceph/metric.h | 76 +++++++++++ fs/ceph/super.c | 4 + fs/ceph/super.h | 1 + include/linux/ceph/ceph_fs.h | 1 + 6 files changed, 294 insertions(+), 25 deletions(-)fs/ceph/mds_client.c +191
-24@@ -4085,16 +4085,167 @@ static void maybe_recover_session(struct ceph_mds_client *mdsc)ceph_force_reconnect (fsc->sb); } -/* -* delayed work -- periodically trim expired leases, renew caps with mds -*/ +static bool ceph_mdsc_send_metrics (struct ceph_mds_client *mdsc, +struct ceph_mds_session *s, +u64 nr_caps) +{ +struct ceph_metric_head *head; +struct ceph_metric_cap *cap; +struct ceph_metric_dentry_lease *lease; +struct ceph_metric_read_latency *read; +struct ceph_metric_write_latency *write; +struct ceph_metric_metadata_latency *meta; +struct ceph_msg *msg; +struct timespec64 ts; +s64 sum, total; +s32 items =0 ; +s32 len; + +if (!mdsc || !s) +return false; + + len =sizeof (*head) +sizeof (*cap) +sizeof (*lease) +sizeof (*read) + +sizeof (*write) +sizeof (*meta); + + msg =ceph_msg_new (CEPH_MSG_CLIENT_METRICS , len,GFP_NOFS , true); +if (!msg) { +pr_err ("send metrics to mds%d, failed to allocate message\n" , + s->s_mds); +return false; + } + + head = msg->front.iov_base; + +/* encode the cap metric */ + cap = (struct ceph_metric_cap *)(head +1 ); + cap->type =cpu_to_le32 (CLIENT_METRIC_TYPE_CAP_INFO ); + cap->ver =1 ; + cap->compat =1 ; + cap->data_len =cpu_to_le32 (sizeof (*cap) -10 ); + cap->hit =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.i_caps_hit)); + cap->mis =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.i_caps_mis)); + cap->total =cpu_to_le64 (nr_caps); + items++; + +dout ("cap metric hit %lld, mis %lld, total caps %lld" , +le64_to_cpu (cap->hit),le64_to_cpu (cap->mis), +le64_to_cpu (cap->total)); + +/* encode the read latency metric */ + read = (struct ceph_metric_read_latency *)(cap +1 ); + read->type =cpu_to_le32 (CLIENT_METRIC_TYPE_READ_LATENCY ); + read->ver =1 ; + read->compat =1 ; + read->data_len =cpu_to_le32 (sizeof (*read) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_reads), + sum =percpu_counter_sum (&mdsc->metric.read_latency_sum); +jiffies_to_timespec64 (sum, &ts); + read->sec =cpu_to_le32 (ts.tv_sec); + read->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("read latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the write latency metric */ + write = (struct ceph_metric_write_latency *)(read +1 ); + write->type =cpu_to_le32 (CLIENT_METRIC_TYPE_WRITE_LATENCY ); + write->ver =1 ; + write->compat =1 ; + write->data_len =cpu_to_le32 (sizeof (*write) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_writes), + sum =percpu_counter_sum (&mdsc->metric.write_latency_sum); +jiffies_to_timespec64 (sum, &ts); + write->sec =cpu_to_le32 (ts.tv_sec); + write->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("write latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the metadata latency metric */ + meta = (struct ceph_metric_metadata_latency *)(write +1 ); + meta->type =cpu_to_le32 (CLIENT_METRIC_TYPE_METADATA_LATENCY ); + meta->ver =1 ; + meta->compat =1 ; + meta->data_len =cpu_to_le32 (sizeof (*meta) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_metadatas), + sum =percpu_counter_sum (&mdsc->metric.metadata_latency_sum); +jiffies_to_timespec64 (sum, &ts); + meta->sec =cpu_to_le32 (ts.tv_sec); + meta->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("metadata latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the dentry lease metric */ + lease = (struct ceph_metric_dentry_lease *)(meta +1 ); + lease->type =cpu_to_le32 (CLIENT_METRIC_TYPE_DENTRY_LEASE ); + lease->ver =1 ; + lease->compat =1 ; + lease->data_len =cpu_to_le32 (sizeof (*lease) -10 ); + lease->hit =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.d_lease_hit)); + lease->mis =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.d_lease_mis)); + lease->total =cpu_to_le64 (atomic64_read (&mdsc->metric.total_dentries)); + items++; +dout ("dentry lease metric hit %lld, mis %lld, total dentries %lld" , +le64_to_cpu (lease->hit),le64_to_cpu (lease->mis), +le64_to_cpu (lease->total)); + +put_unaligned_le32 (items, &head->num); + msg->front.iov_len =cpu_to_le32 (len); + msg->hdr.version =cpu_to_le16 (1 ); + msg->hdr.compat_version =cpu_to_le16 (1 ); + msg->hdr.front_len =cpu_to_le32 (msg->front.iov_len); +dout ("send metrics to mds%d %p\n" , s->s_mds, msg); +ceph_con_send (&s->s_con, msg); + +return true; +} + +#define CEPH_WORK_DELAY_DEF 5 +static void __schedule_delayed (struct delayed_work *work,int delay) +{ +unsigned int hz =round_jiffies_relative (HZ * delay); + +schedule_delayed_work (work, hz); +} +static void schedule_delayed (struct ceph_mds_client *mdsc) { -int delay =5 ; -unsigned hz =round_jiffies_relative (HZ * delay); -schedule_delayed_work (&mdsc->delayed_work, hz); +__schedule_delayed (&mdsc->delayed_work,CEPH_WORK_DELAY_DEF ); +} + +static void metric_schedule_delayed (struct ceph_mds_client *mdsc) +{ +/* delay CEPH_WORK_DELAY_DEF seconds when idle */ +int delay = metric_send_interval ? :CEPH_WORK_DELAY_DEF ; + +__schedule_delayed (&mdsc->metric_delayed_work, delay); +} + +static bool check_session_state (struct ceph_mds_client *mdsc, +struct ceph_mds_session *s) +{ +if (s->s_state ==CEPH_MDS_SESSION_CLOSING ) { +dout ("resending session close request for mds%d\n" , + s->s_mds); +request_close_session (mdsc, s); +return false; + } +if (s->s_ttl &&time_after (jiffies, s->s_ttl)) { +if (s->s_state ==CEPH_MDS_SESSION_OPEN ) { + s->s_state =CEPH_MDS_SESSION_HUNG ; +pr_info ("mds%d hung\n" , s->s_mds); + } + } +if (s->s_state ==CEPH_MDS_SESSION_NEW || + s->s_state ==CEPH_MDS_SESSION_RESTARTING || + s->s_state ==CEPH_MDS_SESSION_REJECTED ) +/* this mds is failed or recovering, just wait */ +return false; + +return true; } +/* +* delayed work -- periodically trim expired leases, renew caps with mds +*/ static void delayed_work (struct work_struct *work) {int i;@@ -4116,23 +4267,8 @@ static void delayed_work(struct work_struct *work)struct ceph_mds_session *s =__ceph_lookup_mds_session (mdsc, i);if (!s)continue ; -if (s->s_state ==CEPH_MDS_SESSION_CLOSING ) { -dout ("resending session close request for mds%d\n" , - s->s_mds); -request_close_session (mdsc, s); -ceph_put_mds_session (s); -continue ; - } -if (s->s_ttl &&time_after (jiffies, s->s_ttl)) { -if (s->s_state ==CEPH_MDS_SESSION_OPEN ) { - s->s_state =CEPH_MDS_SESSION_HUNG ; -pr_info ("mds%d hung\n" , s->s_mds); - } - } -if (s->s_state ==CEPH_MDS_SESSION_NEW || - s->s_state ==CEPH_MDS_SESSION_RESTARTING || - s->s_state ==CEPH_MDS_SESSION_REJECTED ) { -/* this mds is failed or recovering, just wait */ + +if (!check_session_state (mdsc, s)) {ceph_put_mds_session (s);continue ; }@@ -4164,8 +4300,53 @@ static void delayed_work(struct work_struct *work)schedule_delayed (mdsc); } -static int ceph_mdsc_metric_init (struct ceph_client_metric *metric) +static void metric_delayed_work (struct work_struct *work) +{ +struct ceph_mds_client *mdsc = + container_of(work, struct ceph_mds_client, metric_delayed_work.work); +struct ceph_mds_session *s; +u64 nr_caps =0 ; +bool ret; +int i; + +if (!metric_send_interval) + goto idle; + +dout ("mdsc metric_delayed_work\n" ); + +mutex_lock (&mdsc->mutex); +for (i =0 ; i < mdsc->max_sessions; i++) { + s =__ceph_lookup_mds_session (mdsc, i); +if (!s) +continue ; + nr_caps += s->s_nr_caps; +ceph_put_mds_session (s); + } + +for (i =0 ; i < mdsc->max_sessions; i++) { + s =__ceph_lookup_mds_session (mdsc, i); +if (!s) +continue ; +if (!check_session_state (mdsc, s)) { +ceph_put_mds_session (s); +continue ; + } +
fs/ceph/mds_client.c static void delayed_work(struct work_struct *work) · line 4344
Quoted 10 lines
fs/ceph/mds_client.c +10
-0+/* Only send the metric once in any available session */ + ret =ceph_mdsc_send_metrics (mdsc, s, nr_caps); +ceph_put_mds_session (s); +if (ret) +break ; + } +mutex_unlock (&mdsc->mutex); + +idle: +metric_schedule_delayed (mdsc);
Looks like this will schedule metric_delayed_work() every 5 seconds even if metric_send_interval = 0 (i.e. sending is disabled). What is the reason for that?
Thanks,
Ilya
Re: [PATCH v6 6/9] ceph: periodically send perf metrics to ceph
Quoted 269 lines · Ilya Dryomov
Quoted 268 lines · On Mon, Feb 10, 2020 at 6:34 AM <[email protected]> wrote:
From: Xiubo Li <[email protected]>
Add
metric_send_intervalmodule parameter support, the default valume is 0, means disabled. If none zero it will enable the transmission of the metrics to the ceph cluster periodically permetric_send_intervalseconds.This will send the caps, dentry lease and read/write/metadata perf metrics to any available MDS only once per
metric_send_intervalseconds.URL: https://tracker.ceph.com/issues/43215
Signed-off-by Xiubo Li <[email protected]>--- fs/ceph/mds_client.c | 235 +++++++++++++++++++++++++++++++---- fs/ceph/mds_client.h | 2 + fs/ceph/metric.h | 76 +++++++++++ fs/ceph/super.c | 4 + fs/ceph/super.h | 1 + include/linux/ceph/ceph_fs.h | 1 + 6 files changed, 294 insertions(+), 25 deletions(-)fs/ceph/mds_client.c +196
-24@@ -4085,16 +4085,167 @@ static void maybe_recover_session(struct ceph_mds_client *mdsc)ceph_force_reconnect (fsc->sb); } -/* -* delayed work -- periodically trim expired leases, renew caps with mds -*/ +static bool ceph_mdsc_send_metrics (struct ceph_mds_client *mdsc, +struct ceph_mds_session *s, +u64 nr_caps) +{ +struct ceph_metric_head *head; +struct ceph_metric_cap *cap; +struct ceph_metric_dentry_lease *lease; +struct ceph_metric_read_latency *read; +struct ceph_metric_write_latency *write; +struct ceph_metric_metadata_latency *meta; +struct ceph_msg *msg; +struct timespec64 ts; +s64 sum, total; +s32 items =0 ; +s32 len; + +if (!mdsc || !s) +return false; + + len =sizeof (*head) +sizeof (*cap) +sizeof (*lease) +sizeof (*read) + +sizeof (*write) +sizeof (*meta); + + msg =ceph_msg_new (CEPH_MSG_CLIENT_METRICS , len,GFP_NOFS , true); +if (!msg) { +pr_err ("send metrics to mds%d, failed to allocate message\n" , + s->s_mds); +return false; + } + + head = msg->front.iov_base; + +/* encode the cap metric */ + cap = (struct ceph_metric_cap *)(head +1 ); + cap->type =cpu_to_le32 (CLIENT_METRIC_TYPE_CAP_INFO ); + cap->ver =1 ; + cap->compat =1 ; + cap->data_len =cpu_to_le32 (sizeof (*cap) -10 ); + cap->hit =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.i_caps_hit)); + cap->mis =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.i_caps_mis)); + cap->total =cpu_to_le64 (nr_caps); + items++; + +dout ("cap metric hit %lld, mis %lld, total caps %lld" , +le64_to_cpu (cap->hit),le64_to_cpu (cap->mis), +le64_to_cpu (cap->total)); + +/* encode the read latency metric */ + read = (struct ceph_metric_read_latency *)(cap +1 ); + read->type =cpu_to_le32 (CLIENT_METRIC_TYPE_READ_LATENCY ); + read->ver =1 ; + read->compat =1 ; + read->data_len =cpu_to_le32 (sizeof (*read) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_reads), + sum =percpu_counter_sum (&mdsc->metric.read_latency_sum); +jiffies_to_timespec64 (sum, &ts); + read->sec =cpu_to_le32 (ts.tv_sec); + read->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("read latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the write latency metric */ + write = (struct ceph_metric_write_latency *)(read +1 ); + write->type =cpu_to_le32 (CLIENT_METRIC_TYPE_WRITE_LATENCY ); + write->ver =1 ; + write->compat =1 ; + write->data_len =cpu_to_le32 (sizeof (*write) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_writes), + sum =percpu_counter_sum (&mdsc->metric.write_latency_sum); +jiffies_to_timespec64 (sum, &ts); + write->sec =cpu_to_le32 (ts.tv_sec); + write->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("write latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the metadata latency metric */ + meta = (struct ceph_metric_metadata_latency *)(write +1 ); + meta->type =cpu_to_le32 (CLIENT_METRIC_TYPE_METADATA_LATENCY ); + meta->ver =1 ; + meta->compat =1 ; + meta->data_len =cpu_to_le32 (sizeof (*meta) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_metadatas), + sum =percpu_counter_sum (&mdsc->metric.metadata_latency_sum); +jiffies_to_timespec64 (sum, &ts); + meta->sec =cpu_to_le32 (ts.tv_sec); + meta->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("metadata latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the dentry lease metric */ + lease = (struct ceph_metric_dentry_lease *)(meta +1 ); + lease->type =cpu_to_le32 (CLIENT_METRIC_TYPE_DENTRY_LEASE ); + lease->ver =1 ; + lease->compat =1 ; + lease->data_len =cpu_to_le32 (sizeof (*lease) -10 ); + lease->hit =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.d_lease_hit)); + lease->mis =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.d_lease_mis)); + lease->total =cpu_to_le64 (atomic64_read (&mdsc->metric.total_dentries)); + items++; +dout ("dentry lease metric hit %lld, mis %lld, total dentries %lld" , +le64_to_cpu (lease->hit),le64_to_cpu (lease->mis), +le64_to_cpu (lease->total)); + +put_unaligned_le32 (items, &head->num); + msg->front.iov_len =cpu_to_le32 (len); + msg->hdr.version =cpu_to_le16 (1 ); + msg->hdr.compat_version =cpu_to_le16 (1 ); + msg->hdr.front_len =cpu_to_le32 (msg->front.iov_len); +dout ("send metrics to mds%d %p\n" , s->s_mds, msg); +ceph_con_send (&s->s_con, msg); + +return true; +} + +#define CEPH_WORK_DELAY_DEF 5 +static void __schedule_delayed (struct delayed_work *work,int delay) +{ +unsigned int hz =round_jiffies_relative (HZ * delay); + +schedule_delayed_work (work, hz); +} +static void schedule_delayed (struct ceph_mds_client *mdsc) { -int delay =5 ; -unsigned hz =round_jiffies_relative (HZ * delay); -schedule_delayed_work (&mdsc->delayed_work, hz); +__schedule_delayed (&mdsc->delayed_work,CEPH_WORK_DELAY_DEF ); +} + +static void metric_schedule_delayed (struct ceph_mds_client *mdsc) +{ +/* delay CEPH_WORK_DELAY_DEF seconds when idle */ +int delay = metric_send_interval ? :CEPH_WORK_DELAY_DEF ; + +__schedule_delayed (&mdsc->metric_delayed_work, delay); +} + +static bool check_session_state (struct ceph_mds_client *mdsc, +struct ceph_mds_session *s) +{ +if (s->s_state ==CEPH_MDS_SESSION_CLOSING ) { +dout ("resending session close request for mds%d\n" , + s->s_mds); +request_close_session (mdsc, s); +return false; + } +if (s->s_ttl &&time_after (jiffies, s->s_ttl)) { +if (s->s_state ==CEPH_MDS_SESSION_OPEN ) { + s->s_state =CEPH_MDS_SESSION_HUNG ; +pr_info ("mds%d hung\n" , s->s_mds); + } + } +if (s->s_state ==CEPH_MDS_SESSION_NEW || + s->s_state ==CEPH_MDS_SESSION_RESTARTING || + s->s_state ==CEPH_MDS_SESSION_REJECTED ) +/* this mds is failed or recovering, just wait */ +return false; + +return true; } +/* +* delayed work -- periodically trim expired leases, renew caps with mds +*/ static void delayed_work (struct work_struct *work) {int i;@@ -4116,23 +4267,8 @@ static void delayed_work(struct work_struct *work)struct ceph_mds_session *s =__ceph_lookup_mds_session (mdsc, i);if (!s)continue ; -if (s->s_state ==CEPH_MDS_SESSION_CLOSING ) { -dout ("resending session close request for mds%d\n" , - s->s_mds); -request_close_session (mdsc, s); -ceph_put_mds_session (s); -continue ; - } -if (s->s_ttl &&time_after (jiffies, s->s_ttl)) { -if (s->s_state ==CEPH_MDS_SESSION_OPEN ) { - s->s_state =CEPH_MDS_SESSION_HUNG ; -pr_info ("mds%d hung\n" , s->s_mds); - } - } -if (s->s_state ==CEPH_MDS_SESSION_NEW || - s->s_state ==CEPH_MDS_SESSION_RESTARTING || - s->s_state ==CEPH_MDS_SESSION_REJECTED ) { -/* this mds is failed or recovering, just wait */ + +if (!check_session_state (mdsc, s)) {ceph_put_mds_session (s);continue ; }@@ -4164,8 +4300,53 @@ static void delayed_work(struct work_struct *work)schedule_delayed (mdsc); } -static int ceph_mdsc_metric_init (struct ceph_client_metric *metric) +static void metric_delayed_work (struct work_struct *work) +{ +struct ceph_mds_client *mdsc = + container_of(work, struct ceph_mds_client, metric_delayed_work.work); +struct ceph_mds_session *s; +u64 nr_caps =0 ; +bool ret; +int i; + +if (!metric_send_interval) + goto idle; + +dout ("mdsc metric_delayed_work\n" ); + +mutex_lock (&mdsc->mutex); +for (i =0 ; i < mdsc->max_sessions; i++) { + s =__ceph_lookup_mds_session (mdsc, i); +if (!s) +continue ; + nr_caps += s->s_nr_caps; +ceph_put_mds_session (s); + } + +for (i =0 ; i < mdsc->max_sessions; i++) { + s =__ceph_lookup_mds_session (mdsc, i); +if (!s) +continue ; +if (!check_session_state (mdsc, s)) { +ceph_put_mds_session (s); +continue ; + } + +/* Only send the metric once in any available session */ + ret =ceph_mdsc_send_metrics (mdsc, s, nr_caps); +ceph_put_mds_session (s); +if (ret) +break ;
Quoted 8 lines
fs/ceph/mds_client.c static void delayed_work(struct work_struct *work) · line 4344
Quoted 5 lines
fs/ceph/mds_client.c +5
-0+ } +mutex_unlock (&mdsc->mutex); + +idle: +metric_schedule_delayed (mdsc);Looks like this will schedule
metric_delayed_work()every 5 seconds even ifmetric_send_interval= 0 (i.e. sending is disabled). What is the reason for that?
Hi Ilya,
Before I folded the metric_delayed_work() into delayed_work(). But for the this version since the interval is settable, so it hard to calculate the next schedule delay for that.
When it is idle just looping every 5 seconds, I thought though this is not a very graceful approach it won't introduce too much overload. If we do not like this, let's switch it to a completion.
Thanks,
Quoted 4 lines
Thanks,
Ilya
Ilya Dryomov
maintainerRe: [PATCH v6 6/9] ceph: periodically send perf metrics to ceph
Quoted 282 lines · Xiubo Li
Quoted 269 lines · Ilya Dryomov
Quoted 268 lines · On Mon, Feb 10, 2020 at 6:34 AM <[email protected]> wrote:
From: Xiubo Li <[email protected]>
Add
metric_send_intervalmodule parameter support, the default valume is 0, means disabled. If none zero it will enable the transmission of the metrics to the ceph cluster periodically permetric_send_intervalseconds.This will send the caps, dentry lease and read/write/metadata perf metrics to any available MDS only once per
metric_send_intervalseconds.URL: https://tracker.ceph.com/issues/43215
Signed-off-by Xiubo Li <[email protected]>--- fs/ceph/mds_client.c | 235 +++++++++++++++++++++++++++++++---- fs/ceph/mds_client.h | 2 + fs/ceph/metric.h | 76 +++++++++++ fs/ceph/super.c | 4 + fs/ceph/super.h | 1 + include/linux/ceph/ceph_fs.h | 1 + 6 files changed, 294 insertions(+), 25 deletions(-)fs/ceph/mds_client.c +196
-24@@ -4085,16 +4085,167 @@ static void maybe_recover_session(struct ceph_mds_client *mdsc)ceph_force_reconnect (fsc->sb); } -/* -* delayed work -- periodically trim expired leases, renew caps with mds -*/ +static bool ceph_mdsc_send_metrics (struct ceph_mds_client *mdsc, +struct ceph_mds_session *s, +u64 nr_caps) +{ +struct ceph_metric_head *head; +struct ceph_metric_cap *cap; +struct ceph_metric_dentry_lease *lease; +struct ceph_metric_read_latency *read; +struct ceph_metric_write_latency *write; +struct ceph_metric_metadata_latency *meta; +struct ceph_msg *msg; +struct timespec64 ts; +s64 sum, total; +s32 items =0 ; +s32 len; + +if (!mdsc || !s) +return false; + + len =sizeof (*head) +sizeof (*cap) +sizeof (*lease) +sizeof (*read) + +sizeof (*write) +sizeof (*meta); + + msg =ceph_msg_new (CEPH_MSG_CLIENT_METRICS , len,GFP_NOFS , true); +if (!msg) { +pr_err ("send metrics to mds%d, failed to allocate message\n" , + s->s_mds); +return false; + } + + head = msg->front.iov_base; + +/* encode the cap metric */ + cap = (struct ceph_metric_cap *)(head +1 ); + cap->type =cpu_to_le32 (CLIENT_METRIC_TYPE_CAP_INFO ); + cap->ver =1 ; + cap->compat =1 ; + cap->data_len =cpu_to_le32 (sizeof (*cap) -10 ); + cap->hit =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.i_caps_hit)); + cap->mis =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.i_caps_mis)); + cap->total =cpu_to_le64 (nr_caps); + items++; + +dout ("cap metric hit %lld, mis %lld, total caps %lld" , +le64_to_cpu (cap->hit),le64_to_cpu (cap->mis), +le64_to_cpu (cap->total)); + +/* encode the read latency metric */ + read = (struct ceph_metric_read_latency *)(cap +1 ); + read->type =cpu_to_le32 (CLIENT_METRIC_TYPE_READ_LATENCY ); + read->ver =1 ; + read->compat =1 ; + read->data_len =cpu_to_le32 (sizeof (*read) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_reads), + sum =percpu_counter_sum (&mdsc->metric.read_latency_sum); +jiffies_to_timespec64 (sum, &ts); + read->sec =cpu_to_le32 (ts.tv_sec); + read->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("read latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the write latency metric */ + write = (struct ceph_metric_write_latency *)(read +1 ); + write->type =cpu_to_le32 (CLIENT_METRIC_TYPE_WRITE_LATENCY ); + write->ver =1 ; + write->compat =1 ; + write->data_len =cpu_to_le32 (sizeof (*write) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_writes), + sum =percpu_counter_sum (&mdsc->metric.write_latency_sum); +jiffies_to_timespec64 (sum, &ts); + write->sec =cpu_to_le32 (ts.tv_sec); + write->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("write latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the metadata latency metric */ + meta = (struct ceph_metric_metadata_latency *)(write +1 ); + meta->type =cpu_to_le32 (CLIENT_METRIC_TYPE_METADATA_LATENCY ); + meta->ver =1 ; + meta->compat =1 ; + meta->data_len =cpu_to_le32 (sizeof (*meta) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_metadatas), + sum =percpu_counter_sum (&mdsc->metric.metadata_latency_sum); +jiffies_to_timespec64 (sum, &ts); + meta->sec =cpu_to_le32 (ts.tv_sec); + meta->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("metadata latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the dentry lease metric */ + lease = (struct ceph_metric_dentry_lease *)(meta +1 ); + lease->type =cpu_to_le32 (CLIENT_METRIC_TYPE_DENTRY_LEASE ); + lease->ver =1 ; + lease->compat =1 ; + lease->data_len =cpu_to_le32 (sizeof (*lease) -10 ); + lease->hit =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.d_lease_hit)); + lease->mis =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.d_lease_mis)); + lease->total =cpu_to_le64 (atomic64_read (&mdsc->metric.total_dentries)); + items++; +dout ("dentry lease metric hit %lld, mis %lld, total dentries %lld" , +le64_to_cpu (lease->hit),le64_to_cpu (lease->mis), +le64_to_cpu (lease->total)); + +put_unaligned_le32 (items, &head->num); + msg->front.iov_len =cpu_to_le32 (len); + msg->hdr.version =cpu_to_le16 (1 ); + msg->hdr.compat_version =cpu_to_le16 (1 ); + msg->hdr.front_len =cpu_to_le32 (msg->front.iov_len); +dout ("send metrics to mds%d %p\n" , s->s_mds, msg); +ceph_con_send (&s->s_con, msg); + +return true; +} + +#define CEPH_WORK_DELAY_DEF 5 +static void __schedule_delayed (struct delayed_work *work,int delay) +{ +unsigned int hz =round_jiffies_relative (HZ * delay); + +schedule_delayed_work (work, hz); +} +static void schedule_delayed (struct ceph_mds_client *mdsc) { -int delay =5 ; -unsigned hz =round_jiffies_relative (HZ * delay); -schedule_delayed_work (&mdsc->delayed_work, hz); +__schedule_delayed (&mdsc->delayed_work,CEPH_WORK_DELAY_DEF ); +} + +static void metric_schedule_delayed (struct ceph_mds_client *mdsc) +{ +/* delay CEPH_WORK_DELAY_DEF seconds when idle */ +int delay = metric_send_interval ? :CEPH_WORK_DELAY_DEF ; + +__schedule_delayed (&mdsc->metric_delayed_work, delay); +} + +static bool check_session_state (struct ceph_mds_client *mdsc, +struct ceph_mds_session *s) +{ +if (s->s_state ==CEPH_MDS_SESSION_CLOSING ) { +dout ("resending session close request for mds%d\n" , + s->s_mds); +request_close_session (mdsc, s); +return false; + } +if (s->s_ttl &&time_after (jiffies, s->s_ttl)) { +if (s->s_state ==CEPH_MDS_SESSION_OPEN ) { + s->s_state =CEPH_MDS_SESSION_HUNG ; +pr_info ("mds%d hung\n" , s->s_mds); + } + } +if (s->s_state ==CEPH_MDS_SESSION_NEW || + s->s_state ==CEPH_MDS_SESSION_RESTARTING || + s->s_state ==CEPH_MDS_SESSION_REJECTED ) +/* this mds is failed or recovering, just wait */ +return false; + +return true; } +/* +* delayed work -- periodically trim expired leases, renew caps with mds +*/ static void delayed_work (struct work_struct *work) {int i;@@ -4116,23 +4267,8 @@ static void delayed_work(struct work_struct *work)struct ceph_mds_session *s =__ceph_lookup_mds_session (mdsc, i);if (!s)continue ; -if (s->s_state ==CEPH_MDS_SESSION_CLOSING ) { -dout ("resending session close request for mds%d\n" , - s->s_mds); -request_close_session (mdsc, s); -ceph_put_mds_session (s); -continue ; - } -if (s->s_ttl &&time_after (jiffies, s->s_ttl)) { -if (s->s_state ==CEPH_MDS_SESSION_OPEN ) { - s->s_state =CEPH_MDS_SESSION_HUNG ; -pr_info ("mds%d hung\n" , s->s_mds); - } - } -if (s->s_state ==CEPH_MDS_SESSION_NEW || - s->s_state ==CEPH_MDS_SESSION_RESTARTING || - s->s_state ==CEPH_MDS_SESSION_REJECTED ) { -/* this mds is failed or recovering, just wait */ + +if (!check_session_state (mdsc, s)) {ceph_put_mds_session (s);continue ; }@@ -4164,8 +4300,53 @@ static void delayed_work(struct work_struct *work)schedule_delayed (mdsc); } -static int ceph_mdsc_metric_init (struct ceph_client_metric *metric) +static void metric_delayed_work (struct work_struct *work) +{ +struct ceph_mds_client *mdsc = + container_of(work, struct ceph_mds_client, metric_delayed_work.work); +struct ceph_mds_session *s; +u64 nr_caps =0 ; +bool ret; +int i; + +if (!metric_send_interval) + goto idle; + +dout ("mdsc metric_delayed_work\n" ); + +mutex_lock (&mdsc->mutex); +for (i =0 ; i < mdsc->max_sessions; i++) { + s =__ceph_lookup_mds_session (mdsc, i); +if (!s) +continue ; + nr_caps += s->s_nr_caps; +ceph_put_mds_session (s); + } + +for (i =0 ; i < mdsc->max_sessions; i++) { + s =__ceph_lookup_mds_session (mdsc, i); +if (!s) +continue ; +if (!check_session_state (mdsc, s)) { +ceph_put_mds_session (s); +continue ; + } + +/* Only send the metric once in any available session */ + ret =ceph_mdsc_send_metrics (mdsc, s, nr_caps); +ceph_put_mds_session (s); +if (ret) +break ;Quoted 8 lines
fs/ceph/mds_client.c static void delayed_work(struct work_struct *work) · line 4344
Quoted 5 lines
fs/ceph/mds_client.c +5
-0+ } +mutex_unlock (&mdsc->mutex); + +idle: +metric_schedule_delayed (mdsc);Looks like this will schedule
metric_delayed_work()every 5 seconds even ifmetric_send_interval= 0 (i.e. sending is disabled). What is the reason for that?Hi Ilya,
Quoted 7 lines
Before I folded the
metric_delayed_work()intodelayed_work(). But for the this version since the interval is settable, so it hard to calculate the next schedule delay for that.When it is idle just looping every 5 seconds, I thought though this is not a very graceful approach it won't introduce too much overload. If we do not like this, let's switch it to a completion.
Take a look at module_param_cb macro. I think you can provide a setter and schedule the first work / modify the delay from there.
That said, I'm not sure making the interval configurable is a good idea. I'm not saying you need to change anything -- just that if it was me, I would send these metrics once per tick (i.e. delayed_work) with an on/off switch and no other tunables.
Thanks,
Ilya
Re: [PATCH v6 6/9] ceph: periodically send perf metrics to ceph
Quoted 285 lines · Ilya Dryomov
Quoted 284 lines · Xiubo Li
Quoted 269 lines · Ilya Dryomov
Quoted 268 lines · On Mon, Feb 10, 2020 at 6:34 AM <[email protected]> wrote:
From: Xiubo Li <[email protected]>
Add
metric_send_intervalmodule parameter support, the default valume is 0, means disabled. If none zero it will enable the transmission of the metrics to the ceph cluster periodically permetric_send_intervalseconds.This will send the caps, dentry lease and read/write/metadata perf metrics to any available MDS only once per
metric_send_intervalseconds.URL: https://tracker.ceph.com/issues/43215
Signed-off-by Xiubo Li <[email protected]>--- fs/ceph/mds_client.c | 235 +++++++++++++++++++++++++++++++---- fs/ceph/mds_client.h | 2 + fs/ceph/metric.h | 76 +++++++++++ fs/ceph/super.c | 4 + fs/ceph/super.h | 1 + include/linux/ceph/ceph_fs.h | 1 + 6 files changed, 294 insertions(+), 25 deletions(-)fs/ceph/mds_client.c +196
-24@@ -4085,16 +4085,167 @@ static void maybe_recover_session(struct ceph_mds_client *mdsc)ceph_force_reconnect (fsc->sb); } -/* -* delayed work -- periodically trim expired leases, renew caps with mds -*/ +static bool ceph_mdsc_send_metrics (struct ceph_mds_client *mdsc, +struct ceph_mds_session *s, +u64 nr_caps) +{ +struct ceph_metric_head *head; +struct ceph_metric_cap *cap; +struct ceph_metric_dentry_lease *lease; +struct ceph_metric_read_latency *read; +struct ceph_metric_write_latency *write; +struct ceph_metric_metadata_latency *meta; +struct ceph_msg *msg; +struct timespec64 ts; +s64 sum, total; +s32 items =0 ; +s32 len; + +if (!mdsc || !s) +return false; + + len =sizeof (*head) +sizeof (*cap) +sizeof (*lease) +sizeof (*read) + +sizeof (*write) +sizeof (*meta); + + msg =ceph_msg_new (CEPH_MSG_CLIENT_METRICS , len,GFP_NOFS , true); +if (!msg) { +pr_err ("send metrics to mds%d, failed to allocate message\n" , + s->s_mds); +return false; + } + + head = msg->front.iov_base; + +/* encode the cap metric */ + cap = (struct ceph_metric_cap *)(head +1 ); + cap->type =cpu_to_le32 (CLIENT_METRIC_TYPE_CAP_INFO ); + cap->ver =1 ; + cap->compat =1 ; + cap->data_len =cpu_to_le32 (sizeof (*cap) -10 ); + cap->hit =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.i_caps_hit)); + cap->mis =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.i_caps_mis)); + cap->total =cpu_to_le64 (nr_caps); + items++; + +dout ("cap metric hit %lld, mis %lld, total caps %lld" , +le64_to_cpu (cap->hit),le64_to_cpu (cap->mis), +le64_to_cpu (cap->total)); + +/* encode the read latency metric */ + read = (struct ceph_metric_read_latency *)(cap +1 ); + read->type =cpu_to_le32 (CLIENT_METRIC_TYPE_READ_LATENCY ); + read->ver =1 ; + read->compat =1 ; + read->data_len =cpu_to_le32 (sizeof (*read) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_reads), + sum =percpu_counter_sum (&mdsc->metric.read_latency_sum); +jiffies_to_timespec64 (sum, &ts); + read->sec =cpu_to_le32 (ts.tv_sec); + read->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("read latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the write latency metric */ + write = (struct ceph_metric_write_latency *)(read +1 ); + write->type =cpu_to_le32 (CLIENT_METRIC_TYPE_WRITE_LATENCY ); + write->ver =1 ; + write->compat =1 ; + write->data_len =cpu_to_le32 (sizeof (*write) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_writes), + sum =percpu_counter_sum (&mdsc->metric.write_latency_sum); +jiffies_to_timespec64 (sum, &ts); + write->sec =cpu_to_le32 (ts.tv_sec); + write->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("write latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the metadata latency metric */ + meta = (struct ceph_metric_metadata_latency *)(write +1 ); + meta->type =cpu_to_le32 (CLIENT_METRIC_TYPE_METADATA_LATENCY ); + meta->ver =1 ; + meta->compat =1 ; + meta->data_len =cpu_to_le32 (sizeof (*meta) -10 ); + total =percpu_counter_sum (&mdsc->metric.total_metadatas), + sum =percpu_counter_sum (&mdsc->metric.metadata_latency_sum); +jiffies_to_timespec64 (sum, &ts); + meta->sec =cpu_to_le32 (ts.tv_sec); + meta->nsec =cpu_to_le32 (ts.tv_nsec); + items++; +dout ("metadata latency metric total %lld, sum lat %lld" , total, sum); + +/* encode the dentry lease metric */ + lease = (struct ceph_metric_dentry_lease *)(meta +1 ); + lease->type =cpu_to_le32 (CLIENT_METRIC_TYPE_DENTRY_LEASE ); + lease->ver =1 ; + lease->compat =1 ; + lease->data_len =cpu_to_le32 (sizeof (*lease) -10 ); + lease->hit =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.d_lease_hit)); + lease->mis =cpu_to_le64 (percpu_counter_sum (&mdsc->metric.d_lease_mis)); + lease->total =cpu_to_le64 (atomic64_read (&mdsc->metric.total_dentries)); + items++; +dout ("dentry lease metric hit %lld, mis %lld, total dentries %lld" , +le64_to_cpu (lease->hit),le64_to_cpu (lease->mis), +le64_to_cpu (lease->total)); + +put_unaligned_le32 (items, &head->num); + msg->front.iov_len =cpu_to_le32 (len); + msg->hdr.version =cpu_to_le16 (1 ); + msg->hdr.compat_version =cpu_to_le16 (1 ); + msg->hdr.front_len =cpu_to_le32 (msg->front.iov_len); +dout ("send metrics to mds%d %p\n" , s->s_mds, msg); +ceph_con_send (&s->s_con, msg); + +return true; +} + +#define CEPH_WORK_DELAY_DEF 5 +static void __schedule_delayed (struct delayed_work *work,int delay) +{ +unsigned int hz =round_jiffies_relative (HZ * delay); + +schedule_delayed_work (work, hz); +} +static void schedule_delayed (struct ceph_mds_client *mdsc) { -int delay =5 ; -unsigned hz =round_jiffies_relative (HZ * delay); -schedule_delayed_work (&mdsc->delayed_work, hz); +__schedule_delayed (&mdsc->delayed_work,CEPH_WORK_DELAY_DEF ); +} + +static void metric_schedule_delayed (struct ceph_mds_client *mdsc) +{ +/* delay CEPH_WORK_DELAY_DEF seconds when idle */ +int delay = metric_send_interval ? :CEPH_WORK_DELAY_DEF ; + +__schedule_delayed (&mdsc->metric_delayed_work, delay); +} + +static bool check_session_state (struct ceph_mds_client *mdsc, +struct ceph_mds_session *s) +{ +if (s->s_state ==CEPH_MDS_SESSION_CLOSING ) { +dout ("resending session close request for mds%d\n" , + s->s_mds); +request_close_session (mdsc, s); +return false; + } +if (s->s_ttl &&time_after (jiffies, s->s_ttl)) { +if (s->s_state ==CEPH_MDS_SESSION_OPEN ) { + s->s_state =CEPH_MDS_SESSION_HUNG ; +pr_info ("mds%d hung\n" , s->s_mds); + } + } +if (s->s_state ==CEPH_MDS_SESSION_NEW || + s->s_state ==CEPH_MDS_SESSION_RESTARTING || + s->s_state ==CEPH_MDS_SESSION_REJECTED ) +/* this mds is failed or recovering, just wait */ +return false; + +return true; } +/* +* delayed work -- periodically trim expired leases, renew caps with mds +*/ static void delayed_work (struct work_struct *work) {int i;@@ -4116,23 +4267,8 @@ static void delayed_work(struct work_struct *work)struct ceph_mds_session *s =__ceph_lookup_mds_session (mdsc, i);if (!s)continue ; -if (s->s_state ==CEPH_MDS_SESSION_CLOSING ) { -dout ("resending session close request for mds%d\n" , - s->s_mds); -request_close_session (mdsc, s); -ceph_put_mds_session (s); -continue ; - } -if (s->s_ttl &&time_after (jiffies, s->s_ttl)) { -if (s->s_state ==CEPH_MDS_SESSION_OPEN ) { - s->s_state =CEPH_MDS_SESSION_HUNG ; -pr_info ("mds%d hung\n" , s->s_mds); - } - } -if (s->s_state ==CEPH_MDS_SESSION_NEW || - s->s_state ==CEPH_MDS_SESSION_RESTARTING || - s->s_state ==CEPH_MDS_SESSION_REJECTED ) { -/* this mds is failed or recovering, just wait */ + +if (!check_session_state (mdsc, s)) {ceph_put_mds_session (s);continue ; }@@ -4164,8 +4300,53 @@ static void delayed_work(struct work_struct *work)schedule_delayed (mdsc); } -static int ceph_mdsc_metric_init (struct ceph_client_metric *metric) +static void metric_delayed_work (struct work_struct *work) +{ +struct ceph_mds_client *mdsc = + container_of(work, struct ceph_mds_client, metric_delayed_work.work); +struct ceph_mds_session *s; +u64 nr_caps =0 ; +bool ret; +int i; + +if (!metric_send_interval) + goto idle; + +dout ("mdsc metric_delayed_work\n" ); + +mutex_lock (&mdsc->mutex); +for (i =0 ; i < mdsc->max_sessions; i++) { + s =__ceph_lookup_mds_session (mdsc, i); +if (!s) +continue ; + nr_caps += s->s_nr_caps; +ceph_put_mds_session (s); + } + +for (i =0 ; i < mdsc->max_sessions; i++) { + s =__ceph_lookup_mds_session (mdsc, i); +if (!s) +continue ; +if (!check_session_state (mdsc, s)) { +ceph_put_mds_session (s); +continue ; + } + +/* Only send the metric once in any available session */ + ret =ceph_mdsc_send_metrics (mdsc, s, nr_caps); +ceph_put_mds_session (s); +if (ret) +break ;Quoted 8 lines
fs/ceph/mds_client.c static void delayed_work(struct work_struct *work) · line 4344
Quoted 5 lines
fs/ceph/mds_client.c +5
-0+ } +mutex_unlock (&mdsc->mutex); + +idle: +metric_schedule_delayed (mdsc);Looks like this will schedule
metric_delayed_work()every 5 seconds even ifmetric_send_interval= 0 (i.e. sending is disabled). What is the reason for that?Hi Ilya,
Before I folded the
metric_delayed_work()intodelayed_work(). But for the this version since the interval is settable, so it hard to calculate the next schedule delay for that.
Quoted 5 lines
Quoted 3 lines
When it is idle just looping every 5 seconds, I thought though this is not a very graceful approach it won't introduce too much overload. If we do not like this, let's switch it to a completion.
Take a look at
module_param_cbmacro. I think you can provide a setter and schedule the first work / modify the delay from there.
Hi Ilya,
Yeah, this is what I was trying to switch to.
Quoted 4 lines
That said, I'm not sure making the interval configurable is a good idea. I'm not saying you need to change anything -- just that if it was me, I would send these metrics once per tick (i.e.
delayed_work) with an on/off switch and no other tunables.
Currently I still couldn't be sure whether per second will introduce any potential overload in some cases. But for now I couldn't foresee it will.
Thanks,
Xiubo
Quoted 5 lines
Thanks,
Ilya
PATCH v6 7/9 ceph: add CEPH_DEFINE_RW_FUNC helper support
In reply to [email protected]From: Xiubo Li <[email protected]>
This will support the string store.
--- include/linux/ceph/debugfs.h | 14 ++++++++++++++ 1 file changed, 14 insertions(+)
include/linux/ceph/debugfs.h +14 -0
@@ -4,6 +4,20 @@#include <linux/ceph/types.h> +#define CEPH_DEFINE_RW_FUNC (name) \ +static int name##_open(struct inode *inode, struct file *file) \ +{ \ + return single_open(file, name##_show, inode->i_private); \ +} \ + \ +static const struct file_operations name##_fops = { \ + .open = name##_open, \ + .read = seq_read, \ + .write = name##_store, \ + .llseek = seq_lseek, \ + .release = single_release, \ +} +/* debugfs.c */ extern void ceph_debugfs_init (void );extern void ceph_debugfs_cleanup (void );
Signature · 1 line
2.21.0
PATCH v6 8/9 ceph: add reset metrics support
In reply to [email protected]From: Xiubo Li <[email protected]>
Sometimes we need to discard the old perf metrics and start to get new ones. And this will reset the most metric counters, except the total numbers for caps and dentries.
URL: https://tracker.ceph.com/issues/43215
--- fs/ceph/debugfs.c | 38 +++++++++++++++++++++++++++++++++++++- 1 file changed, 37 insertions(+), 1 deletion(-)
fs/ceph/debugfs.c +37 -1
@@ -179,6 +179,43 @@ static int metric_show(struct seq_file *s, void *p)return 0 ; } +static ssize_t metric_store (struct file *file,const char __user *user_buf, +size_t count,loff_t *ppos) +{ +struct seq_file *s = file->private_data; +struct ceph_fs_client *fsc = s->private; +struct ceph_mds_client *mdsc = fsc->mdsc; +struct ceph_client_metric *metric = &mdsc->metric; +char buf[8 ]; + +if (copy_from_user (buf, user_buf,8 )) +return -EFAULT ; + +if (strncmp (buf,"reset" ,strlen ("reset" ))) { +pr_err ("Invalid set value '%s', only 'reset' is valid\n" , buf); +return -EINVAL ; + } + +percpu_counter_set (&metric->d_lease_hit,0 ); +percpu_counter_set (&metric->d_lease_mis,0 ); + +percpu_counter_set (&metric->i_caps_hit,0 ); +percpu_counter_set (&metric->i_caps_mis,0 ); + +percpu_counter_set (&metric->read_latency_sum,0 ); +percpu_counter_set (&metric->total_reads,0 ); + +percpu_counter_set (&metric->write_latency_sum,0 ); +percpu_counter_set (&metric->total_writes,0 ); + +percpu_counter_set (&metric->metadata_latency_sum,0 ); +percpu_counter_set (&metric->total_metadatas,0 ); + +return count; +} + +CEPH_DEFINE_RW_FUNC (metric); +static int caps_show_cb (struct inode *inode,struct ceph_cap *cap,void *p) {struct seq_file *s = p;
@@ -277,7 +314,6 @@ DEFINE_SHOW_ATTRIBUTE(mdsmap);DEFINE_SHOW_ATTRIBUTE (mdsc);DEFINE_SHOW_ATTRIBUTE (caps);DEFINE_SHOW_ATTRIBUTE (mds_sessions); -DEFINE_SHOW_ATTRIBUTE (metric);/*
Signature · 1 line
2.21.0
Ilya Dryomov
maintainerRe: [PATCH v6 8/9] ceph: add reset metrics support
Quoted 27 lines · On Mon, Feb 10, 2020 at 6:34 AM <[email protected]> wrote:
From: Xiubo Li <[email protected]>
Sometimes we need to discard the old perf metrics and start to get new ones. And this will reset the most metric counters, except the total numbers for caps and dentries.
URL: https://tracker.ceph.com/issues/43215
Signed-off-by Xiubo Li <[email protected]>--- fs/ceph/debugfs.c | 38 +++++++++++++++++++++++++++++++++++++- 1 file changed, 37 insertions(+), 1 deletion(-)fs/ceph/debugfs.c +6
-0@@ -179,6 +179,43 @@ static int metric_show(struct seq_file *s, void *p)return 0 ; } +static ssize_t metric_store (struct file *file,const char __user *user_buf, +size_t count,loff_t *ppos) +{ +struct seq_file *s = file->private_data; +struct ceph_fs_client *fsc = s->private; +struct ceph_mds_client *mdsc = fsc->mdsc;
fs/ceph/debugfs.c static int metric_show(struct seq_file *s, void *p) · line 197
Quoted 10 lines
fs/ceph/debugfs.c +10
-0+struct ceph_client_metric *metric = &mdsc->metric; +char buf[8 ]; + +if (copy_from_user (buf, user_buf,8 )) +return -EFAULT ; + +if (strncmp (buf,"reset" ,strlen ("reset" ))) { +pr_err ("Invalid set value '%s', only 'reset' is valid\n" , buf); +return -EINVAL ; + }
Hi Xiubo,
Why strncmp? How does this handle inputs like "resetfoobar"?
Quoted 10 lines
fs/ceph/debugfs.c +10
-0+ +percpu_counter_set (&metric->d_lease_hit,0 ); +percpu_counter_set (&metric->d_lease_mis,0 ); + +percpu_counter_set (&metric->i_caps_hit,0 ); +percpu_counter_set (&metric->i_caps_mis,0 ); + +percpu_counter_set (&metric->read_latency_sum,0 ); +percpu_counter_set (&metric->total_reads,0 ); +
fs/ceph/debugfs.c static int metric_show(struct seq_file *s, void *p) · line 217
Quoted 10 lines
fs/ceph/debugfs.c +10
-0+percpu_counter_set (&metric->write_latency_sum,0 ); +percpu_counter_set (&metric->total_writes,0 ); + +percpu_counter_set (&metric->metadata_latency_sum,0 ); +percpu_counter_set (&metric->total_metadatas,0 ); + +return count; +} + +CEPH_DEFINE_RW_FUNC (metric);
More broadly, how are these metrics going to be used? I suspect the MDSes will gradually start relying on the them in the future and probably make decisions based off of them? If that is the case, did you think about clients being able to mess with that by zeroing these counters on a regular basis?
It looks like all of this is still in flight on the userspace side, but I don't see anything similar in https://github.com/ceph/ceph/pull/32120.
Is there a different PR or is this kernel-only for some reason?
Thanks,
Ilya
PATCH v6 9/9 ceph: send client provided metric flags in client metadata
In reply to [email protected]From: Xiubo Li <[email protected]>
Will send the metric flags to MDS, currently it supports the cap, dentry lease, read latency, write latency and metadata latency.
URL: https://tracker.ceph.com/issues/43435
--- fs/ceph/mds_client.c | 47 ++++++++++++++++++++++++++++++++++++++++++-- fs/ceph/metric.h | 14 +++++++++++++ 2 files changed, 59 insertions(+), 2 deletions(-)
fs/ceph/mds_client.c +45 -2
@@ -1082,6 +1082,41 @@ static void encode_supported_features(void **p, void *end) } } +static const unsigned char metric_bits[] =CEPHFS_METRIC_SPEC_CLIENT_SUPPORTED ; +#define METRIC_BYTES (cnt) (DIV_ROUND_UP((size_t)metric_bits[cnt - 1] + 1, 64) * 8) +static void encode_metric_spec (void **p,void *end) +{ +static const size_t count =ARRAY_SIZE (metric_bits); + +/* header */ +BUG_ON (*p +2 > end); +ceph_encode_8 (p,1 );/* version */ +ceph_encode_8 (p,1 );/* compat */ + +if (count >0 ) { +size_t i; +size_t size =METRIC_BYTES (count); + +BUG_ON (*p +4 +4 + size > end); + +/* metric spec info length */ +ceph_encode_32 (p,4 + size); + +/* metric spec */ +ceph_encode_32 (p, size); +memset (*p,0 , size); +for (i =0 ; i < count; i++) + ((unsigned char *)(*p))[i /8 ] |=BIT (metric_bits[i] %8 ); + *p += size; + }else { +BUG_ON (*p +4 +4 > end); +/* metric spec info length */ +ceph_encode_32 (p,4 ); +/* metric spec */ +ceph_encode_32 (p,0 ); + } +} +/* * session message, specialization for CEPH_SESSION_REQUEST_OPEN * to include additional client metadata fields.
@@ -1121,6 +1156,13 @@ static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u6 size =FEATURE_BYTES (count); extra_bytes +=4 + size; +/* metric spec */ + size =0 ; + count =ARRAY_SIZE (metric_bits); +if (count >0 ) + size =METRIC_BYTES (count); + extra_bytes +=2 +4 +4 + size; +/* Allocate the message */ msg =ceph_msg_new (CEPH_MSG_CLIENT_SESSION ,sizeof (*h) + extra_bytes,GFP_NOFS , false);
@@ -1139,9 +1181,9 @@ static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u6* Serialize client metadata into waiting buffer space, using * the format that userspace expects for map<string, string> * -* ClientSession messages with metadata are v3 +* ClientSession messages with metadata are v4 */ - msg->hdr.version =cpu_to_le16 (3 ); + msg->hdr.version =cpu_to_le16 (4 ); msg->hdr.compat_version =cpu_to_le16 (1 );/* The write pointer, following the session_head structure */
@@ -1164,6 +1206,7 @@ static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u6 }encode_supported_features (&p, end); +encode_metric_spec (&p, end); msg->front.iov_len = p - msg->front.iov_base; msg->hdr.front_len =cpu_to_le32 (msg->front.iov_len);
fs/ceph/metric.h +14 -0
@@ -14,6 +14,20 @@ enum ceph_metric_type { CLIENT_METRIC_TYPE_MAX = CLIENT_METRIC_TYPE_DENTRY_LEASE, }; +/* +* This will always have the highest metric bit value +* as the last element of the array. +*/ +#define CEPHFS_METRIC_SPEC_CLIENT_SUPPORTED { \ + CLIENT_METRIC_TYPE_CAP_INFO, \ + CLIENT_METRIC_TYPE_READ_LATENCY, \ + CLIENT_METRIC_TYPE_WRITE_LATENCY, \ + CLIENT_METRIC_TYPE_METADATA_LATENCY, \ + CLIENT_METRIC_TYPE_DENTRY_LEASE, \ + \ + CLIENT_METRIC_TYPE_MAX, \ +} +/* metric caps header */ struct ceph_metric_cap {__le32 type;/* ceph metric type */
Signature · 1 line
2.21.0
Quoted 7 lines · On 2020/2/10 13:33, [email protected] wrote:
Xiubo Li (9):
ceph: add global dentry lease metric support
ceph: add caps perf metric for each session
ceph: add global read latency metric support
ceph: add global write latency metric support
ceph: add global metadata perf metric support
Hi Jeff, Ilya
Currently the corresponding PR in the ceph is still not merged, so the following 4 patches we could ignore for now. And I will address the new comments and post them after that PR get merged.
The above 5 ones are only kclient concerned, if the above is okay could we split this series and test/merge them ?
Thanks
BRs
Xiubo
Quoted 25 lines
ceph: periodically send perf metrics to ceph
ceph: addCEPH_DEFINE_RW_FUNChelper support
ceph: add reset metrics support
ceph: send client provided metric flags in client metadatafs/ceph/acl.c | 2 + fs/ceph/addr.c | 13 ++ fs/ceph/caps.c | 29 +++ fs/ceph/debugfs.c | 107 ++++++++- fs/ceph/dir.c | 25 ++- fs/ceph/file.c | 22 ++ fs/ceph/mds_client.c | 381 +++++++++++++++++++++++++++++--- fs/ceph/mds_client.h | 6 + fs/ceph/metric.h | 155 +++++++++++++ fs/ceph/quota.c | 9 +- fs/ceph/super.c | 4 + fs/ceph/super.h | 11 + fs/ceph/xattr.c | 17 +- include/linux/ceph/ceph_fs.h | 1 + include/linux/ceph/debugfs.h | 14 ++ include/linux/ceph/osd_client.h | 1 + net/ceph/osd_client.c | 2 + 17 files changed, 759 insertions(+), 40 deletions(-)create mode 100644
fs/ceph/metric.h