cluster/ec: Implement self-heal-window_size option

Fix implements the heal window size option for
EC. This option control the maximum size of
read/write operation carried out in self-heal
process.

BUG: 1441491
Change-Id: I6c0ef65c9ca18b0828f91b319d4f52ac5b77d0d8
Signed-off-by: Sunil Kumar Acharya <sheggodu@redhat.com>
Reviewed-on: https://review.gluster.org/17098
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Smoke: Gluster Build System <jenkins@build.gluster.org>
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
CentOS-regression: Gluster Build System <jenkins@build.gluster.org>

4 files changed, 21 insertions, 3 deletions

diff --git a/xlators/cluster/ec/src/ec-heal.c b/xlators/cluster/ec/src/ec-heal.c
index e4f9c0e3539..afe7833f385 100644
--- a/xlators/cluster/ec/src/ec-heal.c
+++ b/xlators/cluster/ec/src/ec-heal.c
@@ -1897,7 +1897,6 @@ ec_rebuild_data (call_frame_t *frame, ec_t *ec, fd_t *fd, uint64_t size,
         ec_heal_t        *heal = NULL;
         int              ret = 0;
         syncbarrier_t    barrier;
-        struct iobuf_pool *pool = NULL;
 
         if (syncbarrier_init (&barrier))
                 return -ENOMEM;
@@ -1907,9 +1906,8 @@ ec_rebuild_data (call_frame_t *frame, ec_t *ec, fd_t *fd, uint64_t size,
         heal->xl = ec->xl;
         heal->data = &barrier;
         syncbarrier_init (heal->data);
-        pool = ec->xl->ctx->iobuf_pool;
         heal->total_size = size;
-        heal->size = iobpool_default_pagesize (pool);
+        heal->size = (128 * GF_UNIT_KB * (ec->self_heal_window_size));
         /* We need to adjust the size to a multiple of the stripe size of the
          * volume. Otherwise writes would need to fill gaps (head and/or tail)
          * with existent data from the bad bricks. This could be garbage on a
diff --git a/xlators/cluster/ec/src/ec-types.h b/xlators/cluster/ec/src/ec-types.h
index f184f459c2e..3e93a1a32cc 100644
--- a/xlators/cluster/ec/src/ec-types.h
+++ b/xlators/cluster/ec/src/ec-types.h
@@ -565,6 +565,7 @@ struct _ec {
     gf_boolean_t       optimistic_changelog;
     uint32_t           background_heals;
     uint32_t           heal_wait_qlen;
+    uint32_t           self_heal_window_size; /* max size of read/writes */
     struct list_head   pending_fops;
     struct list_head   heal_waiting;
     struct list_head   healing;
diff --git a/xlators/cluster/ec/src/ec.c b/xlators/cluster/ec/src/ec.c
index d61ff1bf36a..e628183190e 100644
--- a/xlators/cluster/ec/src/ec.c
+++ b/xlators/cluster/ec/src/ec.c
@@ -280,6 +280,8 @@ reconfigure (xlator_t *this, dict_t *options)
                           uint32, failed);
         GF_OPTION_RECONF ("heal-wait-qlength", heal_wait_qlen, options,
                           uint32, failed);
+        GF_OPTION_RECONF ("self-heal-window-size", ec->self_heal_window_size,
+                          options, uint32, failed);
         GF_OPTION_RECONF ("heal-timeout", ec->shd.timeout, options,
                           int32, failed);
         ec_configure_background_heal_opts (ec, background_heals,
@@ -663,6 +665,8 @@ init (xlator_t *this)
     GF_OPTION_INIT ("eager-lock", ec->eager_lock, bool, failed);
     GF_OPTION_INIT ("background-heals", ec->background_heals, uint32, failed);
     GF_OPTION_INIT ("heal-wait-qlength", ec->heal_wait_qlen, uint32, failed);
+    GF_OPTION_INIT ("self-heal-window-size", ec->self_heal_window_size, uint32,
+                    failed);
     ec_configure_background_heal_opts (ec, ec->background_heals,
                                        ec->heal_wait_qlen);
     GF_OPTION_INIT ("read-policy", read_policy, str, failed);
@@ -1303,6 +1307,8 @@ int32_t ec_dump_private(xlator_t *this)
                        ec_bin(tmp, sizeof(tmp), ec->xl_up, ec->nodes));
     gf_proc_dump_write("background-heals", "%d", ec->background_heals);
     gf_proc_dump_write("heal-wait-qlength", "%d", ec->heal_wait_qlen);
+    gf_proc_dump_write("self-heal-window-size", "%"PRIu32,
+                       ec->self_heal_window_size);
     gf_proc_dump_write("healers", "%d", ec->healers);
     gf_proc_dump_write("heal-waiters", "%d", ec->heal_waiters);
     gf_proc_dump_write("read-policy", "%s", ec_read_policies[ec->read_policy]);
@@ -1466,6 +1472,14 @@ struct volume_options options[] =
         .description = "force the cpu extensions to be used to accelerate the "
                        "galois field computations."
     },
+    { .key  = {"self-heal-window-size"},
+        .type = GF_OPTION_TYPE_INT,
+        .min  = 1,
+        .max  = 1024,
+        .default_value = "1",
+        .description = "Maximum number blocks(128KB) per file for which "
+                       "self-heal process would be applied simultaneously."
+    },
     {   .key = {"optimistic-change-log"},
         .type = GF_OPTION_TYPE_BOOL,
         .default_value = "on",
diff --git a/xlators/mgmt/glusterd/src/glusterd-volume-set.c b/xlators/mgmt/glusterd/src/glusterd-volume-set.c
index d8005fd8d96..08557d1bd86 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volume-set.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volume-set.c
@@ -3212,6 +3212,11 @@ struct volopt_map_entry glusterd_volopt_map[] = {
           .op_version  = GD_OP_VERSION_3_9_0,
           .flags       = OPT_FLAG_CLIENT_OPT
         },
+        { .key         = "disperse.self-heal-window-size",
+          .voltype     = "cluster/disperse",
+          .op_version  = GD_OP_VERSION_3_11_0,
+          .flags       = OPT_FLAG_CLIENT_OPT
+        },
         { .key        = "cluster.use-compound-fops",
           .voltype    = "cluster/replicate",
           .value      = "off",