Thread (30 messages) 30 messages, 7 authors, 2021-03-03

Re: [PATCH v2 2/5] ext4: add mballoc stats proc file

From: harshad shirwadkar <hidden>
Date: 2021-02-12 17:16:51

Thanks Artem for pointing that out. The reason I added
s_bal_groups_considered inside ext4_mb_good_group() was that, in the
next patches in the series, the new allocator structures call
ext4_mb_good_group() before to check if the group looked up using cr0
lists or cr 1 tree is good or not. Having this counter incremented in
ext4_mb_good_group() helps us measure the overhead of the new
structures too. I'd like to keep that. But you're right, even without
those optimizations, ext4_mb_good_group() gets called twice. Which
means, with these optimizations it will get called 3 times. So,
perhaps we can increment s_bal_groups_considered in
ext4_mb_good_group_nolock() as you mentioned, and then we can also
increment it in the following patches just before calling
ext4_mb_good_group(). I'll note that for the next version, thanks!

- Harshad

On Fri, Feb 12, 2021 at 8:52 AM Благодаренко Артём
[off-list ref] wrote:
Hello Harshad,

Thanks for this great work. I am still testing these patches. Right now I have noticed this flow -
s_bal_groups_considered is counted twice. It is placed to ext4_mb_good_group, that is called twice,
one time from  ext4_mb_good_group_nolock and again from  ext4_mb_good_group_nolock.
I’d rather put s_bal_groups_considered into the ext4_mb_good_group_nolock().

Here is simple execution (on non-fragmented partition) with old counter location

[root@CO82 ~]# dd if=/dev/zero of=/mnt/file4 bs=2M count=2 conv=fsync
2+0 records in
2+0 records out
4194304 bytes (4.2 MB, 4.0 MiB) copied, 0.0137407 s, 305 MB/s
[root@CO82 ~]# cat /proc/fs/ext4/sda2/mb_stats
mballoc:
    reqs: 2
    success: 1
    groups_scanned: 2
    groups_considered: 4
    extents_scanned: 2
        goal_hits: 0
        2^n_hits: 2
        breaks: 0
        lost: 0
    useless_c0_loops: 0
    useless_c1_loops: 0
    useless_c2_loops: 0
    useless_c3_loops: 0
    buddies_generated: 60/969
    buddies_time_used: 9178136
    preallocated: 5612
    discarded: 0
[root@CO82 ~]#

And location that suggested
[root@CO82 ~]# dd if=/dev/zero of=/mnt/file3 bs=2M count=2 conv=fsync
2+0 records in
2+0 records out
4194304 bytes (4.2 MB, 4.0 MiB) copied, 0.0155074 s, 270 MB/s
[root@CO82 ~]# cat /proc/fs/ext4/sda2/mb_stats
mballoc:
    reqs: 3
    success: 2
    groups_scanned: 2
    groups_considered: 2
    extents_scanned: 2
        goal_hits: 1
        2^n_hits: 2
        breaks: 0
        lost: 0
    useless_c0_loops: 0
    useless_c1_loops: 0
    useless_c2_loops: 0
    useless_c3_loops: 0
    buddies_generated: 60/969
    buddies_time_used: 2870959
    preallocated: 5626
    discarded: 0
[root@CO82 ~]#

Second looks more rational.

Best regards,
Artem Blagodarenko.
quoted
On 9 Feb 2021, at 23:28, Harshad Shirwadkar [off-list ref] wrote:

Add new stats for measuring the performance of mballoc. This patch is
forked from Artem Blagodarenko's work that can be found here:

https://github.com/lustre/lustre-release/blob/master/ldiskfs/kernel_patches/patches/rhel8/ext4-simple-blockalloc.patch

Signed-off-by: Harshad Shirwadkar <redacted>
---
fs/ext4/ext4.h    |  4 ++++
fs/ext4/mballoc.c | 51 ++++++++++++++++++++++++++++++++++++++++++++++-
fs/ext4/mballoc.h |  1 +
fs/ext4/sysfs.c   |  2 ++
4 files changed, 57 insertions(+), 1 deletion(-)
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 6dd127942208..317b43420ecf 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -1549,6 +1549,8 @@ struct ext4_sb_info {
      atomic_t s_bal_success; /* we found long enough chunks */
      atomic_t s_bal_allocated;       /* in blocks */
      atomic_t s_bal_ex_scanned;      /* total extents scanned */
+     atomic_t s_bal_groups_considered;       /* number of groups considered */
+     atomic_t s_bal_groups_scanned;  /* number of groups scanned */
      atomic_t s_bal_goals;   /* goal hits */
      atomic_t s_bal_breaks;  /* too long searches */
      atomic_t s_bal_2orders; /* 2^order hits */
@@ -1558,6 +1560,7 @@ struct ext4_sb_info {
      atomic_t s_mb_preallocated;
      atomic_t s_mb_discarded;
      atomic_t s_lock_busy;
+     atomic64_t s_bal_cX_failed[4];          /* cX loop didn't find blocks */

      /* locality groups */
      struct ext4_locality_group __percpu *s_locality_groups;
@@ -2808,6 +2811,7 @@ int __init ext4_fc_init_dentry_cache(void);
extern const struct seq_operations ext4_mb_seq_groups_ops;
extern long ext4_mb_stats;
extern long ext4_mb_max_to_scan;
+extern int ext4_seq_mb_stats_show(struct seq_file *seq, void *offset);
extern int ext4_mb_init(struct super_block *);
extern int ext4_mb_release(struct super_block *);
extern ext4_fsblk_t ext4_mb_new_blocks(handle_t *,
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index 07b78a3cc421..fffd0770e930 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -2083,6 +2083,7 @@ static bool ext4_mb_good_group(struct ext4_allocation_context *ac,

      BUG_ON(cr < 0 || cr >= 4);

+     ac->ac_groups_considered++;
      if (unlikely(EXT4_MB_GRP_BBITMAP_CORRUPT(grp)))
              return false;
@@ -2420,6 +2421,9 @@ ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
                      if (ac->ac_status != AC_STATUS_CONTINUE)
                              break;
              }
+             /* Processed all groups and haven't found blocks */
+             if (sbi->s_mb_stats && i == ngroups)
+                     atomic64_inc(&sbi->s_bal_cX_failed[cr]);
      }

      if (ac->ac_b_ex.fe_len > 0 && ac->ac_status != AC_STATUS_FOUND &&
@@ -2548,6 +2552,48 @@ const struct seq_operations ext4_mb_seq_groups_ops = {
      .show   = ext4_mb_seq_groups_show,
};

+int ext4_seq_mb_stats_show(struct seq_file *seq, void *offset)
+{
+     struct super_block *sb = (struct super_block *)seq->private;
+     struct ext4_sb_info *sbi = EXT4_SB(sb);
+
+     seq_puts(seq, "mballoc:\n");
+     if (!sbi->s_mb_stats) {
+             seq_puts(seq, "\tmb stats collection turned off.\n");
+             seq_puts(seq, "\tTo enable, please write \"1\" to sysfs file mb_stats.\n");
+             return 0;
+     }
+     seq_printf(seq, "\treqs: %u\n", atomic_read(&sbi->s_bal_reqs));
+     seq_printf(seq, "\tsuccess: %u\n", atomic_read(&sbi->s_bal_success));
+
+     seq_printf(seq, "\tgroups_scanned: %u\n",  atomic_read(&sbi->s_bal_groups_scanned));
+     seq_printf(seq, "\tgroups_considered: %u\n",  atomic_read(&sbi->s_bal_groups_considered));
+     seq_printf(seq, "\textents_scanned: %u\n", atomic_read(&sbi->s_bal_ex_scanned));
+     seq_printf(seq, "\t\tgoal_hits: %u\n", atomic_read(&sbi->s_bal_goals));
+     seq_printf(seq, "\t\t2^n_hits: %u\n", atomic_read(&sbi->s_bal_2orders));
+     seq_printf(seq, "\t\tbreaks: %u\n", atomic_read(&sbi->s_bal_breaks));
+     seq_printf(seq, "\t\tlost: %u\n", atomic_read(&sbi->s_mb_lost_chunks));
+
+     seq_printf(seq, "\tuseless_c0_loops: %llu\n",
+                (unsigned long long)atomic64_read(&sbi->s_bal_cX_failed[0]));
+     seq_printf(seq, "\tuseless_c1_loops: %llu\n",
+                (unsigned long long)atomic64_read(&sbi->s_bal_cX_failed[1]));
+     seq_printf(seq, "\tuseless_c2_loops: %llu\n",
+                (unsigned long long)atomic64_read(&sbi->s_bal_cX_failed[2]));
+     seq_printf(seq, "\tuseless_c3_loops: %llu\n",
+                (unsigned long long)atomic64_read(&sbi->s_bal_cX_failed[3]));
+     seq_printf(seq, "\tbuddies_generated: %u/%u\n",
+                atomic_read(&sbi->s_mb_buddies_generated),
+                ext4_get_groups_count(sb));
+     seq_printf(seq, "\tbuddies_time_used: %llu\n",
+                atomic64_read(&sbi->s_mb_generation_time));
+     seq_printf(seq, "\tpreallocated: %u\n",
+                atomic_read(&sbi->s_mb_preallocated));
+     seq_printf(seq, "\tdiscarded: %u\n",
+                atomic_read(&sbi->s_mb_discarded));
+     return 0;
+}
+
static struct kmem_cache *get_groupinfo_cache(int blocksize_bits)
{
      int cache_index = blocksize_bits - EXT4_MIN_BLOCK_LOG_SIZE;
@@ -2968,9 +3014,10 @@ int ext4_mb_release(struct super_block *sb)
                              atomic_read(&sbi->s_bal_reqs),
                              atomic_read(&sbi->s_bal_success));
              ext4_msg(sb, KERN_INFO,
-                   "mballoc: %u extents scanned, %u goal hits, "
+                   "mballoc: %u extents scanned, %u groups scanned, %u goal hits, "
                              "%u 2^N hits, %u breaks, %u lost",
                              atomic_read(&sbi->s_bal_ex_scanned),
+                             atomic_read(&sbi->s_bal_groups_scanned),
                              atomic_read(&sbi->s_bal_goals),
                              atomic_read(&sbi->s_bal_2orders),
                              atomic_read(&sbi->s_bal_breaks),
@@ -3579,6 +3626,8 @@ static void ext4_mb_collect_stats(struct ext4_allocation_context *ac)
              if (ac->ac_b_ex.fe_len >= ac->ac_o_ex.fe_len)
                      atomic_inc(&sbi->s_bal_success);
              atomic_add(ac->ac_found, &sbi->s_bal_ex_scanned);
+             atomic_add(ac->ac_groups_scanned, &sbi->s_bal_groups_scanned);
+             atomic_add(ac->ac_groups_considered, &sbi->s_bal_groups_considered);
              if (ac->ac_g_ex.fe_start == ac->ac_b_ex.fe_start &&
                              ac->ac_g_ex.fe_group == ac->ac_b_ex.fe_group)
                      atomic_inc(&sbi->s_bal_goals);
diff --git a/fs/ext4/mballoc.h b/fs/ext4/mballoc.h
index e75b4749aa1c..7597330dbdf8 100644
--- a/fs/ext4/mballoc.h
+++ b/fs/ext4/mballoc.h
@@ -161,6 +161,7 @@ struct ext4_allocation_context {
      /* copy of the best found extent taken before preallocation efforts */
      struct ext4_free_extent ac_f_ex;

+     __u32 ac_groups_considered;
      __u16 ac_groups_scanned;
      __u16 ac_found;
      __u16 ac_tail;
diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
index 4e27fe6ed3ae..752d1c261e2a 100644
--- a/fs/ext4/sysfs.c
+++ b/fs/ext4/sysfs.c
@@ -527,6 +527,8 @@ int ext4_register_sysfs(struct super_block *sb)
                                      ext4_fc_info_show, sb);
              proc_create_seq_data("mb_groups", S_IRUGO, sbi->s_proc,
                              &ext4_mb_seq_groups_ops, sb);
+             proc_create_single_data("mb_stats", 0444, sbi->s_proc,
+                             ext4_seq_mb_stats_show, sb);
      }
      return 0;
}
--
2.30.0.478.g8a0d178c01-goog
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help