zfs-patches/0046-Change-checksum-IO-delay-ratelimit-values.patch

   1 From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
   2 From: Tony Hutter <hutter2@llnl.gov>
   3 Date: Sun, 4 Mar 2018 17:34:51 -0800
   4 Subject: [PATCH] Change checksum & IO delay ratelimit values
   5 MIME-Version: 1.0
   6 Content-Type: text/plain; charset=UTF-8
   7 Content-Transfer-Encoding: 8bit
   8
   9 Change checksum & IO delay ratelimit thresholds from 5/sec to 20/sec.
  10 This allows zed to actually trigger if a bunch of these events arrive in
  11 a short period of time (zed has a threshold of 10 events in 10 sec).
  12 Previously, if you had, say, 100 checksum errors in 1 sec, it would get
  13 ratelimited to 5/sec which wouldn't trigger zed to fault the drive.
  14
  15 Also, convert the checksum and IO delay thresholds to module params for
  16 easy testing.
  17
  18 Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
  19 Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
  20 Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
  21 Signed-off-by: Tony Hutter <hutter2@llnl.gov>
  22 Closes #7252
  23 (cherry picked from commit 6dc40e2ada2d0d008bd314ff3525f2b0acc2bb01)
  24 Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
  25 ---
  26  include/sys/vdev_impl.h          |  2 --
  27  include/sys/zfs_ratelimit.h      | 12 +++++++++---
  28  module/zcommon/zfs_comutil.c     |  4 ++--
  29  module/zfs/vdev.c                | 23 +++++++++++++++++++++--
  30  man/man5/zfs-module-parameters.5 | 39 +++++++++++++++++++++++++++++++++++++++
  31  5 files changed, 71 insertions(+), 9 deletions(-)
  32
  33 diff --git a/include/sys/vdev_impl.h b/include/sys/vdev_impl.h
  34 index 13c495822..4f9f1a903 100644
  35 --- a/include/sys/vdev_impl.h
  36 +++ b/include/sys/vdev_impl.h
  37 @@ -255,8 +255,6 @@ struct vdev {
  38          * We rate limit ZIO delay and ZIO checksum events, since they
  39          * can flood ZED with tons of events when a drive is acting up.
  40          */
  41 -#define        DELAYS_PER_SECOND 5
  42 -#define        CHECKSUMS_PER_SECOND 5
  43         zfs_ratelimit_t vdev_delay_rl;
  44         zfs_ratelimit_t vdev_checksum_rl;
  45  };
  46 diff --git a/include/sys/zfs_ratelimit.h b/include/sys/zfs_ratelimit.h
  47 index f36e07841..012825fad 100644
  48 --- a/include/sys/zfs_ratelimit.h
  49 +++ b/include/sys/zfs_ratelimit.h
  50 @@ -25,13 +25,19 @@
  51  typedef struct {
  52         hrtime_t start;
  53         unsigned int count;
  54 -       unsigned int burst;             /* Number to allow per interval */
  55 -       unsigned int interval;          /* Interval length in seconds */
  56 +
  57 +       /*
  58 +        * Pointer to number of events per interval.  We do this to
  59 +        * allow the burst to be a (changeable) module parameter.
  60 +        */
  61 +       unsigned int *burst;
  62 +
  63 +       unsigned int interval;  /* Interval length in seconds */
  64         kmutex_t lock;
  65  } zfs_ratelimit_t;
  66
  67  int zfs_ratelimit(zfs_ratelimit_t *rl);
  68 -void zfs_ratelimit_init(zfs_ratelimit_t *rl, unsigned int burst,
  69 +void zfs_ratelimit_init(zfs_ratelimit_t *rl, unsigned int *burst,
  70      unsigned int interval);
  71  void zfs_ratelimit_fini(zfs_ratelimit_t *rl);
  72
  73 diff --git a/module/zcommon/zfs_comutil.c b/module/zcommon/zfs_comutil.c
  74 index 52cb7e365..44cdc8523 100644
  75 --- a/module/zcommon/zfs_comutil.c
  76 +++ b/module/zcommon/zfs_comutil.c
  77 @@ -215,7 +215,7 @@ const char *zfs_history_event_names[ZFS_NUM_LEGACY_HISTORY_EVENTS] = {
  78   * interval:   Interval time in seconds
  79   */
  80  void
  81 -zfs_ratelimit_init(zfs_ratelimit_t *rl, unsigned int burst,
  82 +zfs_ratelimit_init(zfs_ratelimit_t *rl, unsigned int *burst,
  83      unsigned int interval)
  84  {
  85         rl->count = 0;
  86 @@ -270,7 +270,7 @@ zfs_ratelimit(zfs_ratelimit_t *rl)
  87                 rl->start = now;
  88                 rl->count = 0;
  89         } else {
  90 -               if (rl->count >= rl->burst) {
  91 +               if (rl->count >= *rl->burst) {
  92                         rc = 0; /* We're ratelimiting */
  93                 }
  94         }
  95 diff --git a/module/zfs/vdev.c b/module/zfs/vdev.c
  96 index df07d893d..0786fbb83 100644
  97 --- a/module/zfs/vdev.c
  98 +++ b/module/zfs/vdev.c
  99 @@ -56,6 +56,16 @@
 100   */
 101  int metaslabs_per_vdev = 200;
 102
 103 +/*
 104 + * Rate limit delay events to this many IO delays per second.
 105 + */
 106 +unsigned int zfs_delays_per_second = 20;
 107 +
 108 +/*
 109 + * Rate limit checksum events after this many checksum errors per second.
 110 + */
 111 +unsigned int zfs_checksums_per_second = 20;
 112 +
 113  /*
 114   * Virtual device management.
 115   */
 116 @@ -357,8 +367,8 @@ vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
 117          * and checksum events so that we don't overwhelm ZED with thousands
 118          * of events when a disk is acting up.
 119          */
 120 -       zfs_ratelimit_init(&vd->vdev_delay_rl, DELAYS_PER_SECOND, 1);
 121 -       zfs_ratelimit_init(&vd->vdev_checksum_rl, CHECKSUMS_PER_SECOND, 1);
 122 +       zfs_ratelimit_init(&vd->vdev_delay_rl, &zfs_delays_per_second, 1);
 123 +       zfs_ratelimit_init(&vd->vdev_checksum_rl, &zfs_checksums_per_second, 1);
 124
 125         list_link_init(&vd->vdev_config_dirty_node);
 126         list_link_init(&vd->vdev_state_dirty_node);
 127 @@ -3776,5 +3786,14 @@ module_param(metaslabs_per_vdev, int, 0644);
 128  MODULE_PARM_DESC(metaslabs_per_vdev,
 129         "Divide added vdev into approximately (but no more than) this number "
 130         "of metaslabs");
 131 +
 132 +module_param(zfs_delays_per_second, uint, 0644);
 133 +MODULE_PARM_DESC(zfs_delays_per_second, "Rate limit delay events to this many "
 134 +       "IO delays per second");
 135 +
 136 +module_param(zfs_checksums_per_second, uint, 0644);
 137 +       MODULE_PARM_DESC(zfs_checksums_per_second, "Rate limit checksum events "
 138 +       "to this many checksum errors per second (do not set below zed"
 139 +       "threshold).");
 140  /* END CSTYLED */
 141  #endif
 142 diff --git a/man/man5/zfs-module-parameters.5 b/man/man5/zfs-module-parameters.5
 143 index d4daffde6..8d5ac2576 100644
 144 --- a/man/man5/zfs-module-parameters.5
 145 +++ b/man/man5/zfs-module-parameters.5
 146 @@ -739,6 +739,34 @@ Disable pool import at module load by ignoring the cache file (typically \fB/etc
 147  Use \fB1\fR for yes (default) and \fB0\fR for no.
 148  .RE
 149
 150 +.sp
 151 +.ne 2
 152 +.na
 153 +\fBzfs_checksums_per_second\fR (int)
 154 +.ad
 155 +.RS 12n
 156 +Rate limit checksum events to this many per second.  Note that this should
 157 +not be set below the zed thresholds (currently 10 checksums over 10 sec)
 158 +or else zed may not trigger any action.
 159 +.sp
 160 +Default value: 20
 161 +.RE
 162 +
 163 +.sp
 164 +.ne 2
 165 +.na
 166 +\fBzfs_commit_timeout_pct\fR (int)
 167 +.ad
 168 +.RS 12n
 169 +This controls the amount of time that a ZIL block (lwb) will remain "open"
 170 +when it isn't "full", and it has a thread waiting for it to be committed to
 171 +stable storage.  The timeout is scaled based on a percentage of the last lwb
 172 +latency to avoid significantly impacting the latency of each individual
 173 +transaction record (itx).
 174 +.sp
 175 +Default value: \fB5\fR%.
 176 +.RE
 177 +
 178  .sp
 179  .ne 2
 180  .na
 181 @@ -866,6 +894,17 @@ Note: \fBzfs_delay_scale\fR * \fBzfs_dirty_data_max\fR must be < 2^64.
 182  Default value: \fB500,000\fR.
 183  .RE
 184
 185 +.sp
 186 +.ne 2
 187 +.na
 188 +\fBzfs_delays_per_second\fR (int)
 189 +.ad
 190 +.RS 12n
 191 +Rate limit IO delay events to this many per second.
 192 +.sp
 193 +Default value: 20
 194 +.RE
 195 +
 196  .sp
 197  .ne 2
 198  .na
 199 --
 200 2.14.2
 201