Halo Replication feature for AFR translator

Summary:
Halo Geo-replication is a feature which allows Gluster or NFS clients to write
locally to their region (as defined by a latency "halo" or threshold if you
like), and have their writes asynchronously propagate from their origin to the
rest of the cluster.  Clients can also write synchronously to the cluster
simply by specifying a halo-latency which is very large (e.g. 10seconds) which
will include all bricks.

In other words, it allows clients to decide at mount time if they desire
synchronous or asynchronous IO into a cluster and the cluster can support both
of these modes to any number of clients simultaneously.

There are a few new volume options due to this feature:
  halo-shd-latency:  The threshold below which self-heal daemons will
  consider children (bricks) connected.

  halo-nfsd-latency: The threshold below which NFS daemons will consider
  children (bricks) connected.

  halo-latency: The threshold below which all other clients will
  consider children (bricks) connected.

  halo-min-replicas: The minimum number of replicas which are to
  be enforced regardless of latency specified in the above 3 options.
  If the number of children falls below this threshold the next
  best (chosen by latency) shall be swapped in.

New FUSE mount options:
  halo-latency & halo-min-replicas: As descripted above.

This feature combined with multi-threaded SHD support (D1271745) results in
some pretty cool geo-replication possibilities.

Operational Notes:
- Global consistency is gaurenteed for synchronous clients, this is provided by
  the existing entry-locking mechanism.
- Asynchronous clients on the other hand and merely consistent to their region.
  Writes & deletes will be protected via entry-locks as usual preventing
  concurrent writes into files which are undergoing replication.  Read operations
  on the other hand should never block.
- Writes are allowed from _any_ region and propagated from the origin to all
  other regions.  The take away from this is care should be taken to ensure
  multiple writers do not write the same files resulting in a gfid split-brain
  which will require resolution via split-brain policies (majority, mtime &
  size).  Recommended method for preventing this is using the nfs-auth feature to
  define which region for each share has RW permissions, tiers not in the origin
  region should have RO perms.

TODO:
- Synchronous clients (including the SHD) should choose clients from their own
  region as preferred sources for reads.  Most of the plumbing is in place for
  this via the child_latency array.
- Better GFID split brain handling & better dent type split brain handling
  (i.e. create a trash can and move the offending files into it).
- Tagging in addition to latency as a means of defining which children you wish
  to synchronously write to

Test Plan:
- The usual suspects, clang, gcc w/ address sanitizer & valgrind
- Prove tests

Reviewers: jackl, dph, cjh, meyering

Reviewed By: meyering

Subscribers: ethanr

Differential Revision: https://phabricator.fb.com/D1272053

Tasks: 4117827

Change-Id: I694a9ab429722da538da171ec528406e77b5e6d1
BUG: 1428061
Signed-off-by: Kevin Vigor <kvigor@fb.com>
Reviewed-on: http://review.gluster.org/16099
Reviewed-on: https://review.gluster.org/16177
Tested-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>
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>

5 files changed, 422 insertions, 58 deletions

diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index ac834e90f4b..17943d7baae 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -2531,7 +2531,6 @@ unwind:
         return 0;
 }
 
-
 int
 afr_lookup_entry_heal (call_frame_t *frame, xlator_t *this)
 {
@@ -3227,7 +3226,7 @@ afr_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         }
         UNLOCK (&frame->lock);
 
-	call_count = afr_frame_return (frame);
+        call_count = afr_frame_return (frame);
 
 	if (call_count == 0)
 		AFR_STACK_UNWIND (flush, frame, local->op_ret,
@@ -4655,20 +4654,292 @@ __get_heard_from_all_status (xlator_t *this)
         return heard_from_all;
 }
 
+static int
+find_best_down_child (xlator_t *this)
+{
+        afr_private_t   *priv               = NULL;
+        int             i                   = -1;
+        int32_t         best_child          = -1;
+        int64_t         best_latency        = INT64_MAX;
+
+        priv = this->private;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (priv->child_up[i] &&
+                    priv->child_latency[i] >= 0 &&
+                    priv->child_latency[i] < best_latency) {
+                        best_child = i;
+                        best_latency = priv->child_latency[i];
+                }
+        }
+        if (best_child >= 0) {
+                gf_msg_debug (this->name, 0, "Found best down child (%d) "
+                              "@ %ld ms latency", best_child, best_latency);
+        }
+        return best_child;
+}
+
+int
+find_worst_up_child (xlator_t *this)
+{
+        afr_private_t   *priv               = NULL;
+        int             i                   = -1;
+        int32_t         worst_child         = -1;
+        int64_t         worst_latency       = INT64_MIN;
+
+        priv = this->private;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (priv->child_up[i] &&
+                    priv->child_latency[i] >= 0 &&
+                    priv->child_latency[i] > worst_latency) {
+                        worst_child = i;
+                        worst_latency = priv->child_latency[i];
+                }
+        }
+        if (worst_child >= 0) {
+                gf_msg_debug (this->name, 0, "Found worst up child (%d)"
+                              " @ %ld ms latency", worst_child, worst_latency);
+        }
+        return worst_child;
+}
+
+void
+__afr_handle_ping_event (xlator_t *this, xlator_t *child_xlator,
+                const int idx, int64_t halo_max_latency_msec, int32_t *event,
+                int64_t child_latency_msec)
+{
+        afr_private_t   *priv               = NULL;
+        int             up_children         = 0;
+
+        priv = this->private;
+
+        priv->child_latency[idx] = child_latency_msec;
+        gf_msg_debug (child_xlator->name, 0, "Client ping @ %ld ms",
+                      child_latency_msec);
+
+        up_children = __afr_get_up_children_count (priv);
+
+        if (child_latency_msec > halo_max_latency_msec &&
+            priv->child_up[idx] == 1 &&
+            up_children > priv->halo_min_replicas) {
+                if ((up_children - 1) <
+                    priv->halo_min_replicas) {
+                        gf_log (child_xlator->name, GF_LOG_INFO,
+                               "Overriding halo threshold, "
+                               "min replicas: %d",
+                               priv->halo_min_replicas);
+                } else {
+                        gf_log (child_xlator->name, GF_LOG_INFO,
+                                "Child latency (%ld ms) "
+                                "exceeds halo threshold (%ld), "
+                                "marking child down.",
+                                child_latency_msec,
+                                halo_max_latency_msec);
+                        *event = GF_EVENT_CHILD_DOWN;
+                }
+        } else if (child_latency_msec < halo_max_latency_msec &&
+                   priv->child_up[idx] == 0) {
+                if (up_children < priv->halo_max_replicas) {
+                        gf_log (child_xlator->name, GF_LOG_INFO,
+                                "Child latency (%ld ms) "
+                                "below halo threshold (%ld), "
+                                "marking child up.",
+                                child_latency_msec,
+                                halo_max_latency_msec);
+                        *event = GF_EVENT_CHILD_UP;
+                } else {
+                        gf_log (child_xlator->name, GF_LOG_INFO,
+                            "Not marking child %d up, "
+                            "max replicas (%d) reached.", idx,
+                            priv->halo_max_replicas);
+                }
+        }
+}
+
+void
+__afr_handle_child_up_event (xlator_t *this, xlator_t *child_xlator,
+                const int idx, int64_t halo_max_latency_msec,
+                int32_t *event, int32_t *call_psh, int32_t *up_child)
+{
+        afr_private_t   *priv               = NULL;
+        int             up_children         = 0;
+        int             worst_up_child      = -1;
+
+        priv = this->private;
+
+        /*
+         * This only really counts if the child was never up
+         * (value = -1) or had been down (value = 0).  See
+         * comment at GF_EVENT_CHILD_DOWN for a more detailed
+         * explanation.
+         */
+        if (priv->child_up[idx] != 1) {
+                priv->event_generation++;
+        }
+        priv->child_up[idx] = 1;
+
+        *call_psh = 1;
+        *up_child = idx;
+        up_children = __afr_get_up_children_count (priv);
+
+        /*
+         * Handle the edge case where we exceed
+         * halo_min_replicas and we've got a child which is
+         * marked up as it was helping to satisfy the
+         * halo_min_replicas even though it's latency exceeds
+         * halo_max_latency_msec.
+         */
+        if (up_children > priv->halo_min_replicas) {
+                worst_up_child = find_worst_up_child (this);
+                if (worst_up_child >= 0 &&
+                    priv->child_latency[worst_up_child] >
+                    halo_max_latency_msec) {
+                        gf_msg_debug (this->name, 0, "Marking child %d down, "
+                                "doesn't meet halo threshold (%ld), and > "
+                                "halo_min_replicas (%d)",
+                                worst_up_child, halo_max_latency_msec,
+                                priv->halo_min_replicas);
+                        priv->child_up[worst_up_child] = 0;
+                        up_children--;
+                }
+        }
+        if (up_children > priv->halo_max_replicas &&
+            !priv->shd.iamshd) {
+                worst_up_child = find_worst_up_child (this);
+                if (worst_up_child < 0) {
+                        worst_up_child = idx;
+                }
+                priv->child_up[worst_up_child] = 0;
+                up_children--;
+                gf_msg_debug (this->name, 0, "Marking child %d down, "
+                        "up_children (%d) > halo_max_replicas (%d)",
+                        worst_up_child, up_children, priv->halo_max_replicas);
+        }
+
+        if (up_children == 1) {
+                gf_msg (this->name, GF_LOG_INFO, 0, AFR_MSG_SUBVOL_UP,
+                        "Subvolume '%s' came back up; "
+                        "going online.",
+                        child_xlator->name);
+        } else {
+                *event = GF_EVENT_SOME_DESCENDENT_UP;
+        }
+
+        priv->last_event[idx] = *event;
+}
+
+void
+__afr_handle_child_down_event (xlator_t *this, xlator_t *child_xlator,
+                int idx, int64_t child_latency_msec, int32_t *event,
+                int32_t *call_psh, int32_t *up_child)
+{
+        afr_private_t   *priv               = NULL;
+        int             i                   = 0;
+        int             up_children         = 0;
+        int             down_children       = 0;
+        int             best_down_child     = -1;
+
+        priv = this->private;
+
+        /*
+         * If a brick is down when we start, we'll get a
+         * CHILD_DOWN to indicate its initial state.  There
+         * was never a CHILD_UP in this case, so if we
+         * increment "down_count" the difference between than
+         * and "up_count" will no longer be the number of
+         * children that are currently up.  This has serious
+         * implications e.g. for quorum enforcement, so we
+         * don't increment these values unless the event
+         * represents an actual state transition between "up"
+         * (value = 1) and anything else.
+         */
+        if (priv->child_up[idx] == 1) {
+                priv->event_generation++;
+        }
+
+        /*
+         * If this is an _actual_ CHILD_DOWN event, we
+         * want to set the child_latency to < 0 to indicate
+         * the child is really disconnected.
+         */
+        if (child_latency_msec < 0) {
+                priv->child_latency[idx] = child_latency_msec;
+        }
+        priv->child_up[idx] = 0;
+
+        up_children = __afr_get_up_children_count (priv);
+        /*
+         * Handle the edge case where we need to find the
+         * next best child (to mark up) as marking this child
+         * down would cause us to fall below halo_min_replicas.
+         * We will also force the SHD to heal this child _now_
+         * as we want it to be up to date if we are going to
+         * begin using it synchronously.
+         */
+        if (up_children < priv->halo_min_replicas) {
+                best_down_child = find_best_down_child (this);
+                if (best_down_child >= 0) {
+                        gf_msg_debug (this->name, 0,
+                                "Swapping out child %d for "
+                                "child %d to satisfy halo_min_replicas (%d).",
+                                idx, best_down_child, priv->halo_min_replicas);
+                        priv->child_up[best_down_child] = 1;
+                        *call_psh = 1;
+                        *up_child = best_down_child;
+                }
+        }
+
+        for (i = 0; i < priv->child_count; i++)
+                if (priv->child_up[i] == 0)
+                        down_children++;
+        if (down_children == priv->child_count) {
+                gf_msg (this->name, GF_LOG_ERROR, 0, AFR_MSG_SUBVOLS_DOWN,
+                        "All subvolumes are down. Going "
+                        "offline until atleast one of them "
+                        "comes back up.");
+        } else {
+                *event = GF_EVENT_SOME_DESCENDENT_DOWN;
+        }
+        priv->last_event[idx] = *event;
+}
+
+static int64_t
+afr_get_halo_latency (xlator_t *this)
+{
+        afr_private_t *priv           = NULL;
+        int64_t halo_max_latency_msec = 0;
+
+        priv = this->private;
+
+        if (priv->shd.iamshd) {
+                halo_max_latency_msec = priv->shd.halo_max_latency_msec;
+        } else if (priv->nfsd.iamnfsd) {
+                halo_max_latency_msec =
+                        priv->nfsd.halo_max_latency_msec;
+        } else {
+                halo_max_latency_msec = priv->halo_max_latency_msec;
+        }
+        gf_msg_debug (this->name, 0, "Using halo latency %ld",
+                halo_max_latency_msec);
+        return halo_max_latency_msec;
+}
+
+
 int32_t
 afr_notify (xlator_t *this, int32_t event,
             void *data, void *data2)
 {
         afr_private_t   *priv               = NULL;
+        xlator_t        *child_xlator       = NULL;
         int             i                   = -1;
-        int             up_children         = 0;
-        int             down_children       = 0;
         int             propagate           = 0;
         int             had_heard_from_all  = 0;
         int             have_heard_from_all = 0;
         int             idx                 = -1;
         int             ret                 = -1;
         int             call_psh            = 0;
+        int             up_child            = -1;
         dict_t          *input              = NULL;
         dict_t          *output             = NULL;
         gf_boolean_t    had_quorum          = _gf_false;
@@ -4677,6 +4948,10 @@ afr_notify (xlator_t *this, int32_t event,
         struct gf_upcall_cache_invalidation *up_ci = NULL;
         inode_table_t  *itable              = NULL;
         inode_t        *inode               = NULL;
+        int64_t         halo_max_latency_msec = 0;
+        int64_t         child_latency_msec   = -1;
+
+        child_xlator = (xlator_t *)data;
 
         priv = this->private;
 
@@ -4701,7 +4976,7 @@ afr_notify (xlator_t *this, int32_t event,
          * subsequent revalidate lookup happens on all the dht's subvolumes
          * which triggers afr self-heals if any.
          */
-        idx = find_child_index (this, data);
+        idx = find_child_index (this, child_xlator);
         if (idx < 0) {
                 gf_msg (this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_CHILD_UP,
                         "Received child_up from invalid subvolume");
@@ -4710,6 +4985,30 @@ afr_notify (xlator_t *this, int32_t event,
 
         had_quorum = priv->quorum_count && afr_has_quorum (priv->child_up,
                                                            this);
+        if (priv->halo_enabled) {
+                halo_max_latency_msec = afr_get_halo_latency (this);
+
+                if (event == GF_EVENT_CHILD_PING) {
+                        /* Calculates the child latency and sets event
+                         */
+                        child_latency_msec = (int64_t)(uintptr_t)data2;
+                        LOCK (&priv->lock);
+                        {
+                                __afr_handle_ping_event (this, child_xlator,
+                                        idx, halo_max_latency_msec, &event,
+                                        child_latency_msec);
+                        }
+                        UNLOCK (&priv->lock);
+                }
+        }
+
+        if (event == GF_EVENT_CHILD_PING) {
+                /* This is the only xlator that handles PING, no reason to
+                 * propagate.
+                 */
+                goto out;
+            }
+
         if (event == GF_EVENT_TRANSLATOR_OP) {
                 LOCK (&priv->lock);
                 {
@@ -4736,57 +5035,15 @@ afr_notify (xlator_t *this, int32_t event,
                         propagate = 1;
                         break;
                 case GF_EVENT_CHILD_UP:
-                        /*
-                         * This only really counts if the child was never up
-                         * (value = -1) or had been down (value = 0).  See
-                         * comment at GF_EVENT_CHILD_DOWN for a more detailed
-                         * explanation.
-                         */
-                        if (priv->child_up[idx] != 1) {
-                                priv->event_generation++;
-                        }
-                        priv->child_up[idx] = 1;
-
-                        call_psh = 1;
-                        up_children = __afr_get_up_children_count (priv);
-                        if (up_children == 1) {
-                                gf_msg (this->name, GF_LOG_INFO, 0,
-                                        AFR_MSG_SUBVOL_UP,
-                                        "Subvolume '%s' came back up; "
-                                     "going online.", ((xlator_t *)data)->name);
-                                gf_event (EVENT_AFR_SUBVOL_UP,
-                                          "subvol=%s", this->name);
-
-                        } else {
-                                event = GF_EVENT_SOME_DESCENDENT_UP;
-                        }
-
-                        priv->last_event[idx] = event;
-
+                        __afr_handle_child_up_event (this, child_xlator,
+                                idx, halo_max_latency_msec, &event, &call_psh,
+                                &up_child);
                         break;
 
                 case GF_EVENT_CHILD_DOWN:
-                        if (priv->child_up[idx] == 1) {
-                                priv->event_generation++;
-                        }
-                        priv->child_up[idx] = 0;
-
-                        for (i = 0; i < priv->child_count; i++)
-                                if (priv->child_up[i] == 0)
-                                        down_children++;
-                        if (down_children == priv->child_count) {
-                                gf_msg (this->name, GF_LOG_ERROR, 0,
-                                        AFR_MSG_SUBVOLS_DOWN,
-                                       "All subvolumes are down. Going offline "
-                                    "until atleast one of them comes back up.");
-                                gf_event (EVENT_AFR_SUBVOLS_DOWN,
-                                          "subvol=%s", this->name);
-                        } else {
-                                event = GF_EVENT_SOME_DESCENDENT_DOWN;
-                        }
-
-                        priv->last_event[idx] = event;
-
+                        __afr_handle_child_down_event (this, child_xlator, idx,
+                                child_latency_msec, &event, &call_psh,
+                                &up_child);
                         break;
 
                 case GF_EVENT_CHILD_CONNECTING:
@@ -4839,7 +5096,6 @@ afr_notify (xlator_t *this, int32_t event,
                            had come up, propagate CHILD_UP, but only this time
                         */
                         event = GF_EVENT_CHILD_DOWN;
-                        up_children = __afr_get_up_children_count (priv);
                         for (i = 0; i < priv->child_count; i++) {
                                 if (priv->last_event[i] == GF_EVENT_CHILD_UP) {
                                         event = GF_EVENT_CHILD_UP;
diff --git a/xlators/cluster/afr/src/afr-mem-types.h b/xlators/cluster/afr/src/afr-mem-types.h
index 7f7962013d7..c7d6261b110 100644
--- a/xlators/cluster/afr/src/afr-mem-types.h
+++ b/xlators/cluster/afr/src/afr-mem-types.h
@@ -46,7 +46,8 @@ enum gf_afr_mem_types_ {
 	gf_afr_mt_spbc_timeout_t,
         gf_afr_mt_spb_status_t,
         gf_afr_mt_empty_brick_t,
-        gf_afr_mt_end
+        gf_afr_mt_child_latency_t,
+    gf_afr_mt_end
 };
 #endif
 
diff --git a/xlators/cluster/afr/src/afr-self-heald.h b/xlators/cluster/afr/src/afr-self-heald.h
index c6ac5ebfd1b..4ac1d32f58a 100644
--- a/xlators/cluster/afr/src/afr-self-heald.h
+++ b/xlators/cluster/afr/src/afr-self-heald.h
@@ -58,6 +58,7 @@ typedef struct {
         eh_t                    **statistics;
         uint32_t                max_threads;
         uint32_t                wait_qlength;
+        uint32_t                halo_max_latency_msec;
 } afr_self_heald_t;
 
 
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index ceaa034dbbb..17b34822c17 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -184,6 +184,27 @@ reconfigure (xlator_t *this, dict_t *options)
         GF_OPTION_RECONF ("data-self-heal-algorithm",
                           priv->data_self_heal_algorithm, options, str, out);
 
+        GF_OPTION_RECONF ("halo-enabled",
+                          priv->halo_enabled, options, bool,
+                          out);
+
+        GF_OPTION_RECONF ("halo-shd-max-latency",
+                          priv->shd.halo_max_latency_msec, options, uint32,
+                          out);
+
+        GF_OPTION_RECONF ("halo-nfsd-max-latency",
+                          priv->nfsd.halo_max_latency_msec, options, uint32,
+                          out);
+
+        GF_OPTION_RECONF ("halo-max-latency", priv->halo_max_latency_msec,
+                          options, uint32, out);
+
+        GF_OPTION_RECONF ("halo-max-replicas", priv->halo_max_replicas, options,
+                              uint32, out);
+
+        GF_OPTION_RECONF ("halo-min-replicas", priv->halo_min_replicas, options,
+                              uint32, out);
+
         GF_OPTION_RECONF ("read-subvolume", read_subvol, options, xlator, out);
 
         GF_OPTION_RECONF ("read-hash-mode", priv->hash_mode,
@@ -473,6 +494,24 @@ init (xlator_t *this)
 
         GF_OPTION_INIT ("entry-self-heal", priv->entry_self_heal, bool, out);
 
+        GF_OPTION_INIT ("halo-shd-max-latency", priv->shd.halo_max_latency_msec,
+                        uint32, out);
+
+        GF_OPTION_INIT ("halo-max-latency", priv->halo_max_latency_msec,
+                        uint32, out);
+        GF_OPTION_INIT ("halo-max-replicas", priv->halo_max_replicas, uint32,
+                        out);
+        GF_OPTION_INIT ("halo-min-replicas", priv->halo_min_replicas, uint32,
+                        out);
+
+        GF_OPTION_INIT ("halo-enabled",
+                        priv->halo_enabled, bool, out);
+
+        GF_OPTION_INIT ("halo-nfsd-max-latency",
+                        priv->nfsd.halo_max_latency_msec, uint32, out);
+
+        GF_OPTION_INIT ("iam-nfs-daemon", priv->nfsd.iamnfsd, bool, out);
+
         GF_OPTION_INIT ("data-change-log", priv->data_change_log, bool, out);
 
         GF_OPTION_INIT ("metadata-change-log", priv->metadata_change_log, bool,
@@ -528,7 +567,12 @@ init (xlator_t *this)
 
         priv->child_up = GF_CALLOC (sizeof (unsigned char), child_count,
                                     gf_afr_mt_char);
-        if (!priv->child_up) {
+
+        priv->child_latency = GF_CALLOC (sizeof (*priv->child_latency),
+                                         child_count,
+                                         gf_afr_mt_child_latency_t);
+
+        if (!priv->child_up || !priv->child_latency) {
                 ret = -ENOMEM;
                 goto out;
         }
@@ -736,7 +780,50 @@ struct volume_options options[] = {
                          "jobs that can perform parallel heals in the "
                          "background."
         },
-        { .key  = {"heal-wait-queue-length"},
+        { .key   = {"halo-shd-max-latency"},
+          .type  = GF_OPTION_TYPE_INT,
+          .min   = 1,
+          .max   = 99999,
+          .default_value = "99999",
+           .description = "Maximum latency for shd halo replication in msec."
+        },
+        { .key   = {"halo-enabled"},
+          .type  = GF_OPTION_TYPE_BOOL,
+          .default_value = "False",
+           .description = "Enable Halo (geo) replication mode."
+        },
+        { .key   = {"halo-nfsd-max-latency"},
+          .type  = GF_OPTION_TYPE_INT,
+          .min   = 1,
+          .max   = 99999,
+          .default_value = "5",
+           .description = "Maximum latency for nfsd halo replication in msec."
+        },
+        { .key   = {"halo-max-latency"},
+          .type  = GF_OPTION_TYPE_INT,
+          .min   = 1,
+          .max   = 99999,
+          .default_value = "5",
+           .description = "Maximum latency for halo replication in msec."
+        },
+        { .key   = {"halo-max-replicas"},
+          .type  = GF_OPTION_TYPE_INT,
+          .min   = 1,
+          .max   = 99999,
+          .default_value = "99999",
+           .description = "The maximum number of halo replicas; replicas"
+                          " beyond this value will be written asynchronously"
+                          "via the SHD."
+        },
+        { .key   = {"halo-min-replicas"},
+          .type  = GF_OPTION_TYPE_INT,
+          .min   = 1,
+          .max   = 99999,
+          .default_value = "2",
+           .description = "The minimmum number of halo replicas, before adding "
+                          "out of region replicas."
+         },
+         { .key  = {"heal-wait-queue-length"},
           .type = GF_OPTION_TYPE_INT,
           .min  = 0,
           .max  = 10000, /*Around 100MB with sizeof(afr_local_t)= 10496 bytes*/
@@ -876,6 +963,13 @@ struct volume_options options[] = {
                          "translator is running as part of self-heal-daemon "
                          "or not."
         },
+        { .key = {"iam-nfs-daemon"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "off",
+          .description = "This option differentiates if the replicate "
+                         "translator is running as part of an NFS daemon "
+                         "or not."
+        },
         { .key = {"quorum-type"},
           .type = GF_OPTION_TYPE_STR,
           .value = { "none", "auto", "fixed"},
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 0535e7c7271..3be15175dc7 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -74,6 +74,11 @@ typedef enum {
         AFR_FAV_CHILD_POLICY_MAX,
 } afr_favorite_child_policy;
 
+struct afr_nfsd {
+        gf_boolean_t     iamnfsd;
+        uint32_t         halo_max_latency_msec;
+};
+
 typedef struct _afr_private {
         gf_lock_t lock;               /* to guard access to child_count, etc */
         unsigned int child_count;     /* total number of children   */
@@ -85,6 +90,7 @@ typedef struct _afr_private {
         inode_t *root_inode;
 
         unsigned char *child_up;
+        int64_t *child_latency;
         unsigned char *local;
 
         char **pending_key;
@@ -155,8 +161,14 @@ typedef struct _afr_private {
         gf_boolean_t           ensure_durability;
         char                   *sh_domain;
 	char                   *afr_dirty;
+        gf_boolean_t           halo_enabled;
+
+        uint32_t               halo_max_latency_msec;
+        uint32_t               halo_max_replicas;
+        uint32_t               halo_min_replicas;
 
-	afr_self_heald_t       shd;
+        afr_self_heald_t       shd;
+        struct afr_nfsd        nfsd;
 
         gf_boolean_t           consistent_metadata;
         uint64_t               spb_choice_timeout;