zfs-patches/0007-vdev_mirror-load-balancing-fixes.patch

   1 From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
   2 From: Gvozden Neskovic <neskovic@gmail.com>
   3 Date: Fri, 4 Aug 2017 11:29:56 +0200
   4 Subject: [PATCH] vdev_mirror: load balancing fixes
   5 MIME-Version: 1.0
   6 Content-Type: text/plain; charset=UTF-8
   7 Content-Transfer-Encoding: 8bit
   8
   9 vdev_queue:
  10 - Track the last position of each vdev, including the io size,
  11   in order to detect linear access of the following zio.
  12 - Remove duplicate `vq_lastoffset`
  13
  14 vdev_mirror:
  15 - Correctly calculate the zio offset (signedness issue)
  16 - Deprecate `vdev_queue_register_lastoffset()`
  17 - Add `VDEV_LABEL_START_SIZE` to zio offset of leaf vdevs
  18
  19 Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
  20 Signed-off-by: Gvozden Neskovic <neskovic@gmail.com>
  21 Closes #6461
  22 (cherry picked from commit 06acbbc429bfe7197e5fc3a49acfeef5c37b64c8)
  23 Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
  24 ---
  25  include/sys/vdev.h       |  3 +--
  26  include/sys/vdev_impl.h  |  1 -
  27  module/zfs/vdev_mirror.c | 36 ++++++++++++++++--------------------
  28  module/zfs/vdev_queue.c  | 21 +++++++--------------
  29  4 files changed, 24 insertions(+), 37 deletions(-)
  30
  31 diff --git a/include/sys/vdev.h b/include/sys/vdev.h
  32 index 7157ef43f..473d2691c 100644
  33 --- a/include/sys/vdev.h
  34 +++ b/include/sys/vdev.h
  35 @@ -125,8 +125,7 @@ extern zio_t *vdev_queue_io(zio_t *zio);
  36  extern void vdev_queue_io_done(zio_t *zio);
  37
  38  extern int vdev_queue_length(vdev_t *vd);
  39 -extern uint64_t vdev_queue_lastoffset(vdev_t *vd);
  40 -extern void vdev_queue_register_lastoffset(vdev_t *vd, zio_t *zio);
  41 +extern uint64_t vdev_queue_last_offset(vdev_t *vd);
  42
  43  extern void vdev_config_dirty(vdev_t *vd);
  44  extern void vdev_config_clean(vdev_t *vd);
  45 diff --git a/include/sys/vdev_impl.h b/include/sys/vdev_impl.h
  46 index 7c5e54b08..4c2e3cd2e 100644
  47 --- a/include/sys/vdev_impl.h
  48 +++ b/include/sys/vdev_impl.h
  49 @@ -127,7 +127,6 @@ struct vdev_queue {
  50         hrtime_t        vq_io_delta_ts;
  51         zio_t           vq_io_search; /* used as local for stack reduction */
  52         kmutex_t        vq_lock;
  53 -       uint64_t        vq_lastoffset;
  54  };
  55
  56  /*
  57 diff --git a/module/zfs/vdev_mirror.c b/module/zfs/vdev_mirror.c
  58 index 0439e4b46..d230b4db4 100644
  59 --- a/module/zfs/vdev_mirror.c
  60 +++ b/module/zfs/vdev_mirror.c
  61 @@ -116,7 +116,8 @@ static const zio_vsd_ops_t vdev_mirror_vsd_ops = {
  62  static int
  63  vdev_mirror_load(mirror_map_t *mm, vdev_t *vd, uint64_t zio_offset)
  64  {
  65 -       uint64_t lastoffset;
  66 +       uint64_t last_offset;
  67 +       int64_t offset_diff;
  68         int load;
  69
  70         /* All DVAs have equal weight at the root. */
  71 @@ -129,13 +130,17 @@ vdev_mirror_load(mirror_map_t *mm, vdev_t *vd, uint64_t zio_offset)
  72          * worse overall when resilvering with compared to without.
  73          */
  74
  75 +       /* Fix zio_offset for leaf vdevs */
  76 +       if (vd->vdev_ops->vdev_op_leaf)
  77 +               zio_offset += VDEV_LABEL_START_SIZE;
  78 +
  79         /* Standard load based on pending queue length. */
  80         load = vdev_queue_length(vd);
  81 -       lastoffset = vdev_queue_lastoffset(vd);
  82 +       last_offset = vdev_queue_last_offset(vd);
  83
  84         if (vd->vdev_nonrot) {
  85                 /* Non-rotating media. */
  86 -               if (lastoffset == zio_offset)
  87 +               if (last_offset == zio_offset)
  88                         return (load + zfs_vdev_mirror_non_rotating_inc);
  89
  90                 /*
  91 @@ -148,16 +153,16 @@ vdev_mirror_load(mirror_map_t *mm, vdev_t *vd, uint64_t zio_offset)
  92         }
  93
  94         /* Rotating media I/O's which directly follow the last I/O. */
  95 -       if (lastoffset == zio_offset)
  96 +       if (last_offset == zio_offset)
  97                 return (load + zfs_vdev_mirror_rotating_inc);
  98
  99         /*
 100          * Apply half the seek increment to I/O's within seek offset
 101 -        * of the last I/O queued to this vdev as they should incur less
 102 +        * of the last I/O issued to this vdev as they should incur less
 103          * of a seek increment.
 104          */
 105 -       if (ABS(lastoffset - zio_offset) <
 106 -           zfs_vdev_mirror_rotating_seek_offset)
 107 +       offset_diff = (int64_t)(last_offset - zio_offset);
 108 +       if (ABS(offset_diff) < zfs_vdev_mirror_rotating_seek_offset)
 109                 return (load + (zfs_vdev_mirror_rotating_seek_inc / 2));
 110
 111         /* Apply the full seek increment to all other I/O's. */
 112 @@ -382,29 +387,20 @@ vdev_mirror_child_select(zio_t *zio)
 113                 mm->mm_preferred_cnt++;
 114         }
 115
 116 -       if (mm->mm_preferred_cnt == 1) {
 117 -               vdev_queue_register_lastoffset(
 118 -                   mm->mm_child[mm->mm_preferred[0]].mc_vd, zio);
 119 +       if (mm->mm_preferred_cnt == 1)
 120                 return (mm->mm_preferred[0]);
 121 -       }
 122
 123 -       if (mm->mm_preferred_cnt > 1) {
 124 -               int c = vdev_mirror_preferred_child_randomize(zio);
 125
 126 -               vdev_queue_register_lastoffset(mm->mm_child[c].mc_vd, zio);
 127 -               return (c);
 128 -       }
 129 +       if (mm->mm_preferred_cnt > 1)
 130 +               return (vdev_mirror_preferred_child_randomize(zio));
 131
 132         /*
 133          * Every device is either missing or has this txg in its DTL.
 134          * Look for any child we haven't already tried before giving up.
 135          */
 136         for (c = 0; c < mm->mm_children; c++) {
 137 -               if (!mm->mm_child[c].mc_tried) {
 138 -                       vdev_queue_register_lastoffset(mm->mm_child[c].mc_vd,
 139 -                           zio);
 140 +               if (!mm->mm_child[c].mc_tried)
 141                         return (c);
 142 -               }
 143         }
 144
 145         /*
 146 diff --git a/module/zfs/vdev_queue.c b/module/zfs/vdev_queue.c
 147 index 6b3e87291..40cba340a 100644
 148 --- a/module/zfs/vdev_queue.c
 149 +++ b/module/zfs/vdev_queue.c
 150 @@ -393,7 +393,7 @@ vdev_queue_init(vdev_t *vd)
 151                     sizeof (zio_t), offsetof(struct zio, io_queue_node));
 152         }
 153
 154 -       vq->vq_lastoffset = 0;
 155 +       vq->vq_last_offset = 0;
 156  }
 157
 158  void
 159 @@ -699,9 +699,8 @@ again:
 160          */
 161         tree = vdev_queue_class_tree(vq, p);
 162         vq->vq_io_search.io_timestamp = 0;
 163 -       vq->vq_io_search.io_offset = vq->vq_last_offset + 1;
 164 -       VERIFY3P(avl_find(tree, &vq->vq_io_search,
 165 -           &idx), ==, NULL);
 166 +       vq->vq_io_search.io_offset = vq->vq_last_offset - 1;
 167 +       VERIFY3P(avl_find(tree, &vq->vq_io_search, &idx), ==, NULL);
 168         zio = avl_nearest(tree, idx, AVL_AFTER);
 169         if (zio == NULL)
 170                 zio = avl_first(tree);
 171 @@ -728,7 +727,7 @@ again:
 172         }
 173
 174         vdev_queue_pending_add(vq, zio);
 175 -       vq->vq_last_offset = zio->io_offset;
 176 +       vq->vq_last_offset = zio->io_offset + zio->io_size;
 177
 178         return (zio);
 179  }
 180 @@ -806,7 +805,7 @@ vdev_queue_io_done(zio_t *zio)
 181  }
 182
 183  /*
 184 - * As these three methods are only used for load calculations we're not
 185 + * As these two methods are only used for load calculations we're not
 186   * concerned if we get an incorrect value on 32bit platforms due to lack of
 187   * vq_lock mutex use here, instead we prefer to keep it lock free for
 188   * performance.
 189 @@ -818,15 +817,9 @@ vdev_queue_length(vdev_t *vd)
 190  }
 191
 192  uint64_t
 193 -vdev_queue_lastoffset(vdev_t *vd)
 194 +vdev_queue_last_offset(vdev_t *vd)
 195  {
 196 -       return (vd->vdev_queue.vq_lastoffset);
 197 -}
 198 -
 199 -void
 200 -vdev_queue_register_lastoffset(vdev_t *vd, zio_t *zio)
 201 -{
 202 -       vd->vdev_queue.vq_lastoffset = zio->io_offset + zio->io_size;
 203 +       return (vd->vdev_queue.vq_last_offset);
 204  }
 205
 206  #if defined(_KERNEL) && defined(HAVE_SPL)
 207 --
 208 2.14.2
 209