LU-4259 osd-zfs: Add brw_stats collection and display

[fs/lustre-release.git] / lustre / osd-zfs / osd_lproc.c

diff --git a/lustre/osd-zfs/osd_lproc.c b/lustre/osd-zfs/osd_lproc.c
index 622bc5b..8146c16 100644 (file)
--- a/lustre/osd-zfs/osd_lproc.c
+++ b/lustre/osd-zfs/osd_lproc.c
@@ -50,6 +50,110 @@
 
 #ifdef LPROCFS
 
+#define pct(a, b) (b ? a * 100 / b : 0)
+
+static void display_brw_stats(struct seq_file *seq, char *name, char *units,
+                             struct obd_histogram *read,
+                             struct obd_histogram *write, int scale)
+{
+       unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
+       int i;
+
+       seq_printf(seq, "\n%26s read      |     write\n", " ");
+       seq_printf(seq, "%-22s %-5s %% cum %% |  %-11s %% cum %%\n",
+                  name, units, units);
+
+       read_tot = lprocfs_oh_sum(read);
+       write_tot = lprocfs_oh_sum(write);
+       for (i = 0; i < OBD_HIST_MAX; i++) {
+               r = read->oh_buckets[i];
+               w = write->oh_buckets[i];
+               read_cum += r;
+               write_cum += w;
+               if (read_cum == 0 && write_cum == 0)
+                       continue;
+
+               if (!scale)
+                       seq_printf(seq, "%u", i);
+               else if (i < 10)
+                       seq_printf(seq, "%u", scale << i);
+               else if (i < 20)
+                       seq_printf(seq, "%uK", scale << (i-10));
+               else
+                       seq_printf(seq, "%uM", scale << (i-20));
+
+               seq_printf(seq, ":\t\t%10lu %3lu %3lu   | %4lu %3lu %3lu\n",
+                          r, pct(r, read_tot), pct(read_cum, read_tot),
+                          w, pct(w, write_tot), pct(write_cum, write_tot));
+
+               if (read_cum == read_tot && write_cum == write_tot)
+                       break;
+       }
+}
+
+static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
+{
+       struct timeval now;
+
+       /* this sampling races with updates */
+       do_gettimeofday(&now);
+       seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
+                  now.tv_sec, now.tv_usec);
+
+       display_brw_stats(seq, "pages per bulk r/w", "rpcs",
+                         &brw_stats->hist[BRW_R_PAGES],
+                         &brw_stats->hist[BRW_W_PAGES], 1);
+       display_brw_stats(seq, "discontiguous pages", "rpcs",
+                         &brw_stats->hist[BRW_R_DISCONT_PAGES],
+                         &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
+#if 0
+       display_brw_stats(seq, "discontiguous blocks", "rpcs",
+                         &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
+                         &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
+
+       display_brw_stats(seq, "disk fragmented I/Os", "ios",
+                         &brw_stats->hist[BRW_R_DIO_FRAGS],
+                         &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
+#endif
+       display_brw_stats(seq, "disk I/Os in flight", "ios",
+                         &brw_stats->hist[BRW_R_RPC_HIST],
+                         &brw_stats->hist[BRW_W_RPC_HIST], 0);
+
+       display_brw_stats(seq, "I/O time (1/1000s)", "ios",
+                         &brw_stats->hist[BRW_R_IO_TIME],
+                         &brw_stats->hist[BRW_W_IO_TIME], 1000 / HZ);
+
+       display_brw_stats(seq, "disk I/O size", "ios",
+                         &brw_stats->hist[BRW_R_DISK_IOSIZE],
+                         &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
+}
+
+#undef pct
+
+static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
+{
+       struct osd_device *osd = seq->private;
+
+       brw_stats_show(seq, &osd->od_brw_stats);
+
+       return 0;
+}
+
+static ssize_t osd_brw_stats_seq_write(struct file *file, const char *buf,
+                                      size_t len, loff_t *off)
+{
+       struct seq_file *seq = file->private_data;
+       struct osd_device *osd = seq->private;
+       int i;
+
+       for (i = 0; i < BRW_LAST; i++)
+               lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
+
+       return len;
+}
+
+LPROC_SEQ_FOPS(osd_brw_stats);
+
 static int osd_stats_init(struct osd_device *osd)
 {
        int result;
@@ -97,6 +201,8 @@ static int osd_stats_init(struct osd_device *osd)
                                LPROCFS_CNTR_AVGMINMAX,
                                "thandle_closing", "usec");
 #endif
+               result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
+                                           0644, &osd_brw_stats_fops, osd);
        } else {
                result = -ENOMEM;
        }