cluster/ec: Implement gfid-hash read-policy

Add a policy in ec to performs reads from same bricks as long as they
are good. Based on the gfid of the file/directory it determines the
bricks to be considered for reading.

Change-Id: Ic97b5c54c086a28b5e07a330a4fd448551b49376
BUG: 1261260
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/12133
Tested-by: NetBSD Build System <jenkins@build.gluster.org>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Xavier Hernandez <xhernandez@datalab.es>

6 files changed, 137 insertions, 11 deletions

diff --git a/libglusterfs/src/globals.h b/libglusterfs/src/globals.h
index 6934aec5ed1..88e5f77721b 100644
--- a/libglusterfs/src/globals.h
+++ b/libglusterfs/src/globals.h
@@ -38,7 +38,7 @@
  */
 #define GD_OP_VERSION_MIN  1 /* MIN is the fresh start op-version, mostly
                                 should not change */
-#define GD_OP_VERSION_MAX  GD_OP_VERSION_3_7_5 /* MAX VERSION is the maximum
+#define GD_OP_VERSION_MAX  GD_OP_VERSION_3_7_6 /* MAX VERSION is the maximum
                                                   count in VME table, should
                                                   keep changing with
                                                   introduction of newer
@@ -58,6 +58,8 @@
 
 #define GD_OP_VERSION_3_7_5    30705 /* Op-version for GlusterFS 3.7.5 */
 
+#define GD_OP_VERSION_3_7_6    30706 /* Op-version for GlusterFS 3.7.6 */
+
 #define GD_OP_VER_PERSISTENT_AFR_XATTRS GD_OP_VERSION_3_6_0
 
 #include "xlator.h"
diff --git a/tests/basic/ec/ec-read-policy.t b/tests/basic/ec/ec-read-policy.t
new file mode 100644
index 00000000000..891508063e6
--- /dev/null
+++ b/tests/basic/ec/ec-read-policy.t
@@ -0,0 +1,53 @@
+#!/bin/bash
+
+. $(dirname $0)/../../include.rc
+. $(dirname $0)/../../volume.rc
+
+cleanup
+
+TEST glusterd
+TEST pidof glusterd
+TEST $CLI volume create $V0 disperse 6 redundancy 2 $H0:$B0/${V0}{0..5}
+TEST $CLI volume set $V0 performance.quick-read off
+TEST $CLI volume set $V0 performance.io-cache off
+TEST $CLI volume set $V0 performance.write-behind off
+TEST $CLI volume set $V0 performance.stat-prefetch off
+TEST $CLI volume set $V0 performance.read-ahead off
+TEST $CLI volume heal $V0 disable
+TEST $CLI volume start $V0
+
+#Disable all caching
+TEST glusterfs --direct-io-mode=yes --entry-timeout=0 --attribute-timeout=0 -s $H0 --volfile-id $V0 $M0
+EXPECT_WITHIN $CHILD_UP_TIMEOUT "6" ec_child_up_count $V0 0
+#TEST volume operations work fine
+EXPECT "round-robin" mount_get_option_value $M0 $V0-disperse-0 read-policy
+TEST $CLI volume set $V0 disperse.read-policy gfid-hash
+EXPECT_WITHIN $CONFIG_UPDATE_TIMEOUT "gfid-hash" mount_get_option_value $M0 $V0-disperse-0 read-policy
+TEST $CLI volume reset $V0 disperse.read-policy
+EXPECT_WITHIN $CONFIG_UPDATE_TIMEOUT "round-robin" mount_get_option_value $M0 $V0-disperse-0 read-policy
+
+#TEST if the option gives the intended behavior. The way we perform this test
+#is by performing reads from the mount and write to /dev/null. If the
+#read-policy is round-robin, then all bricks should have read-fop where as
+#with gfid-hash number of bricks with reads should be equal to (num-bricks - redundancy)
+#count
+
+TEST $CLI volume profile $V0 start
+TEST dd if=/dev/zero of=$M0/1 bs=1M count=4
+#Perform reads now from file on the mount, this only tests dispatch_min
+TEST dd if=$M0/1 of=/dev/null bs=1M count=4
+#TEST that reads are executed on all bricks
+rr_reads=$($CLI volume profile $V0 info cumulative| grep READ | wc -l)
+EXPECT "^6$" echo $rr_reads
+TEST $CLI volume profile $V0 info clear
+
+TEST $CLI volume set $V0 disperse.read-policy gfid-hash
+EXPECT_WITHIN $CONFIG_UPDATE_TIMEOUT "gfid-hash" mount_get_option_value $M0 $V0-disperse-0 read-policy
+
+#Perform reads now from file on the mount, this only tests dispatch_min
+TEST dd if=$M0/1 of=/dev/null bs=1M count=4
+#TEST that reads are executed on all bricks
+gh_reads=$($CLI volume profile $V0 info cumulative| grep READ |  wc -l)
+EXPECT "^4$" echo $gh_reads
+
+cleanup;
diff --git a/xlators/cluster/ec/src/ec-common.c b/xlators/cluster/ec/src/ec-common.c
index d0c9f97ab28..39a529d3a0b 100644
--- a/xlators/cluster/ec/src/ec-common.c
+++ b/xlators/cluster/ec/src/ec-common.c
@@ -9,6 +9,7 @@
 */
 
 #include "byte-order.h"
+#include "hashfn.h"
 
 #include "ec-mem-types.h"
 #include "ec-data.h"
@@ -20,6 +21,25 @@
 #include "ec.h"
 #include "ec-messages.h"
 
+uint32_t
+ec_select_first_by_read_policy (ec_t *ec, ec_fop_data_t *fop)
+{
+        if (ec->read_policy == EC_ROUND_ROBIN) {
+                return ec->idx;
+        } else if (ec->read_policy == EC_GFID_HASH) {
+                if (fop->use_fd) {
+                        return SuperFastHash((char *)fop->fd->inode->gfid,
+                                   sizeof(fop->fd->inode->gfid)) % ec->nodes;
+                } else {
+                        if (gf_uuid_is_null (fop->loc[0].gfid))
+                                loc_gfid (&fop->loc[0], fop->loc[0].gfid);
+                        return SuperFastHash((char *)fop->loc[0].gfid,
+                                   sizeof(fop->loc[0].gfid)) % ec->nodes;
+                }
+        }
+        return 0;
+}
+
 int32_t ec_child_valid(ec_t * ec, ec_fop_data_t * fop, int32_t idx)
 {
     return (idx < ec->nodes) && (((fop->remaining >> idx) & 1) == 1);
@@ -415,12 +435,13 @@ int32_t ec_child_select(ec_fop_data_t * fop)
             fop->minimum = 1;
     }
 
-    first = ec->idx;
-    if (++first >= ec->nodes)
-    {
-        first = 0;
+    if (ec->read_policy == EC_ROUND_ROBIN) {
+            first = ec->idx;
+            if (++first >= ec->nodes) {
+                first = 0;
+            }
+            ec->idx = first;
     }
-    ec->idx = first;
 
     /*Unconditionally wind on healing subvolumes*/
     fop->mask |= fop->healing;
@@ -518,14 +539,12 @@ void ec_dispatch_start(ec_fop_data_t * fop)
 
 void ec_dispatch_one(ec_fop_data_t * fop)
 {
-    ec_t * ec = fop->xl->private;
-
     ec_dispatch_start(fop);
 
     if (ec_child_select(fop))
     {
         fop->expected = 1;
-        fop->first = ec->idx;
+        fop->first = ec_select_first_by_read_policy (fop->xl->private, fop);
 
         ec_dispatch_next(fop, fop->first);
     }
@@ -589,7 +608,7 @@ void ec_dispatch_min(ec_fop_data_t * fop)
     if (ec_child_select(fop))
     {
         fop->expected = count = ec->fragments;
-        fop->first = ec->idx;
+        fop->first = ec_select_first_by_read_policy (fop->xl->private, fop);
         idx = fop->first - 1;
         mask = 0;
         while (count-- > 0)
diff --git a/xlators/cluster/ec/src/ec.c b/xlators/cluster/ec/src/ec.c
index 11c71743458..06f814f9f5c 100644
--- a/xlators/cluster/ec/src/ec.c
+++ b/xlators/cluster/ec/src/ec.c
@@ -21,6 +21,11 @@
 #include "ec-messages.h"
 #include "ec-heald.h"
 
+static char *ec_read_policies[EC_READ_POLICY_MAX + 1] = {
+        [EC_ROUND_ROBIN] = "round-robin",
+        [EC_GFID_HASH] = "gfid-hash",
+        [EC_READ_POLICY_MAX] = NULL
+};
 #define EC_MAX_FRAGMENTS EC_METHOD_MAX_FRAGMENTS
 /* The maximum number of nodes is derived from the maximum allowed fragments
  * using the rule that redundancy cannot be equal or greater than the number
@@ -231,10 +236,24 @@ ec_configure_background_heal_opts (ec_t *ec, int background_heals,
         ec->background_heals = background_heals;
 }
 
+int
+ec_assign_read_policy (ec_t *ec, char *read_policy)
+{
+        int read_policy_idx = -1;
+
+        read_policy_idx = gf_get_index_by_elem (ec_read_policies, read_policy);
+        if (read_policy_idx < 0 || read_policy_idx >= EC_READ_POLICY_MAX)
+                return -1;
+
+        ec->read_policy = read_policy_idx;
+        return 0;
+}
+
 int32_t
 reconfigure (xlator_t *this, dict_t *options)
 {
         ec_t     *ec              = this->private;
+        char     *read_policy     = NULL;
         uint32_t heal_wait_qlen   = 0;
         uint32_t background_heals = 0;
 
@@ -250,6 +269,10 @@ reconfigure (xlator_t *this, dict_t *options)
                           int32, failed);
         ec_configure_background_heal_opts (ec, background_heals,
                                            heal_wait_qlen);
+        GF_OPTION_RECONF ("read-policy", read_policy, options, str, failed);
+        if (ec_assign_read_policy (ec, read_policy))
+                goto failed;
+
         return 0;
 failed:
         return -1;
@@ -514,7 +537,8 @@ notify (xlator_t *this, int32_t event, void *data, ...)
 int32_t
 init (xlator_t *this)
 {
-    ec_t *ec = NULL;
+    ec_t *ec          = NULL;
+    char *read_policy = NULL;
 
     if (this->parents == NULL)
     {
@@ -576,6 +600,9 @@ init (xlator_t *this)
     GF_OPTION_INIT ("heal-wait-qlength", ec->heal_wait_qlen, uint32, failed);
     ec_configure_background_heal_opts (ec, ec->background_heals,
                                        ec->heal_wait_qlen);
+    GF_OPTION_INIT ("read-policy", read_policy, str, failed);
+    if (ec_assign_read_policy (ec, read_policy))
+            goto failed;
 
     if (ec->shd.iamshd)
             ec_selfheal_daemon_init (this);
@@ -1191,6 +1218,7 @@ int32_t ec_dump_private(xlator_t *this)
     gf_proc_dump_write("heal-wait-qlength", "%d", ec->heal_wait_qlen);
     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]);
 
     return 0;
 }
@@ -1298,5 +1326,14 @@ struct volume_options options[] =
       .description = "time interval for checking the need to self-heal "
                      "in self-heal-daemon"
     },
+    { .key = {"read-policy" },
+      .type = GF_OPTION_TYPE_STR,
+      .value = {"round-robin", "gfid-hash"},
+      .default_value = "round-robin",
+      .description = "inode-read fops happen only on 'k' number of bricks in"
+              " n=k+m disperse subvolume. 'round-robin' selects the read"
+              " subvolume using round-robin algo. 'gfid-hash' selects read"
+              " subvolume based on hash of the gfid of that file/directory.",
+    },
     { }
 };
diff --git a/xlators/cluster/ec/src/ec.h b/xlators/cluster/ec/src/ec.h
index f335fd52afc..4ee7983b289 100644
--- a/xlators/cluster/ec/src/ec.h
+++ b/xlators/cluster/ec/src/ec.h
@@ -25,6 +25,12 @@
 
 #define EC_VERSION_SIZE 2
 
+typedef enum {
+        EC_ROUND_ROBIN,
+        EC_GFID_HASH,
+        EC_READ_POLICY_MAX
+} ec_read_policy_t;
+
 struct _ec
 {
     xlator_t *        xl;
@@ -58,6 +64,7 @@ struct _ec
     ec_self_heald_t   shd;
     char              vol_uuid[UUID_SIZE + 1];
     dict_t           *leaf_to_subvolid;
+    ec_read_policy_t  read_policy;
 };
 
 void ec_pending_fops_completed(ec_t *ec);
diff --git a/xlators/mgmt/glusterd/src/glusterd-volume-set.c b/xlators/mgmt/glusterd/src/glusterd-volume-set.c
index e93a22eafdd..c62f2d79c1f 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volume-set.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volume-set.c
@@ -2082,10 +2082,12 @@ struct volopt_map_entry glusterd_volopt_map[] = {
         { .key         = "disperse.background-heals",
           .voltype     = "cluster/disperse",
           .op_version  = GD_OP_VERSION_3_7_3,
+          .flags       = OPT_FLAG_CLIENT_OPT
         },
         { .key         = "disperse.heal-wait-qlength",
           .voltype     = "cluster/disperse",
           .op_version  = GD_OP_VERSION_3_7_3,
+          .flags       = OPT_FLAG_CLIENT_OPT
         },
         { .key        = "cluster.heal-timeout",
           .voltype    = "cluster/disperse",
@@ -2098,6 +2100,12 @@ struct volopt_map_entry glusterd_volopt_map[] = {
           .voltype     = "cluster/distribute",
           .option      = "use-readdirp",
           .op_version  = GD_OP_VERSION_3_7_5,
+          .flags       = OPT_FLAG_CLIENT_OPT
+        },
+        { .key         = "disperse.read-policy",
+          .voltype     = "cluster/disperse",
+          .op_version  = GD_OP_VERSION_3_7_6,
+          .flags       = OPT_FLAG_CLIENT_OPT
         },
         { .key         = NULL
         }