dht: support auto-NUFA option

Many people have asked for behavior like the old NUFA, which builds and
seems to run but was previously impossible to enable/configure in a
standard way.  This change allows NUFA to be enabled instead of DHT from
the command line, with automatic selection of the local subvolume on each
host.

Change-Id: I0065938db3922361fd450a6c1919a4cbbf6f202e
BUG: 882278
Signed-off-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-on: http://review.gluster.org/4234
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>

4 files changed, 154 insertions, 10 deletions

diff --git a/tests/bugs/bug-882278.t b/tests/bugs/bug-882278.t
new file mode 100755
index 000000000..7933e1863
--- /dev/null
+++ b/tests/bugs/bug-882278.t
@@ -0,0 +1,72 @@
+#!/bin/bash
+
+. $(dirname $0)/../include.rc
+cleanup
+
+# Is there a good reason to require --fqdn elsewhere?  It's worse than useless
+# here.
+H0=$(hostname -s)
+
+function recreate {
+	# The rm is necessary so we don't get fooled by leftovers from old runs.
+	rm -rf $1 && mkdir -p $1
+}
+
+function count_lines {
+	grep "$1" $2/* | wc -l
+}
+
+TEST glusterd
+TEST pidof glusterd
+TEST $CLI volume info;
+
+## Start and create a volume
+TEST recreate ${B0}/${V0}-0
+TEST recreate ${B0}/${V0}-1
+TEST $CLI volume create $V0 $H0:$B0/${V0}-{0,1}
+TEST $CLI volume set $V0 cluster.nufa on
+
+function volinfo_field()
+{
+    local vol=$1;
+    local field=$2;
+
+    $CLI volume info $vol | grep "^$field: " | sed 's/.*: //';
+}
+
+
+## Verify volume is created
+EXPECT "$V0" volinfo_field $V0 'Volume Name';
+EXPECT 'Created' volinfo_field $V0 'Status';
+
+## Start volume and verify
+TEST $CLI volume start $V0;
+EXPECT 'Started' volinfo_field $V0 'Status';
+
+## Mount native
+special_option="--xlator-option ${V0}-dht.local-volume-name=${V0}-client-1"
+TEST glusterfs --volfile-server=$H0 --volfile-id=$V0 $special_option $M0
+
+## Create a bunch of test files.
+for i in $(seq 0 99); do
+	echo hello > $(printf $M0/file%02d $i)
+done
+
+## Make sure the files went to the right place.  There might be link files in
+## the other brick, but they won't have any contents.
+EXPECT "0" count_lines hello ${B0}/${V0}-0
+EXPECT "100" count_lines hello ${B0}/${V0}-1
+
+if [ "$EXIT_EARLY" = "1" ]; then
+	exit 0;
+fi
+
+## Finish up
+TEST umount $M0;
+TEST $CLI volume stop $V0;
+EXPECT 'Stopped' volinfo_field $V0 'Status';
+
+TEST $CLI volume delete $V0;
+TEST ! $CLI volume info $V0;
+
+cleanup;
diff --git a/tests/include.rc b/tests/include.rc
index cfa9390aa..b572fbc63 100644
--- a/tests/include.rc
+++ b/tests/include.rc
@@ -39,6 +39,9 @@ function test_footer()
 	     	echo "ok $t";
 	else
 		echo "not ok $t";
+		if [ "$EXIT_EARLY" = "1" ]; then
+			exit $RET
+		fi
 	fi
 
 	dbg "RESULT $t: $RET";
diff --git a/xlators/cluster/dht/src/nufa.c b/xlators/cluster/dht/src/nufa.c
index a2a97ab3c..701d7ae8d 100644
--- a/xlators/cluster/dht/src/nufa.c
+++ b/xlators/cluster/dht/src/nufa.c
@@ -491,6 +491,26 @@ fini (xlator_t *this)
         return;
 }
 
+gf_boolean_t
+same_first_part (char *str1, char term1, char *str2, char term2)
+{
+        gf_boolean_t    ended1;
+        gf_boolean_t    ended2;
+
+        for (;;) {
+                ended1 = ((*str1 == '\0') || (*str1 == term1));
+                ended2 = ((*str2 == '\0') || (*str2 == term2));
+                if (ended1 && ended2) {
+                        return _gf_true;
+                }
+                if (ended1 || ended2 || (*str1 != *str2)) {
+                        return _gf_false;
+                }
+                ++str1;
+                ++str2;
+        }
+}
+
 int
 init (xlator_t *this)
 {
@@ -504,6 +524,9 @@ init (xlator_t *this)
         char           my_hostname[256];
         double         temp_free_disk = 0;
         uint64_t       size = 0;
+        xlator_t      *local_subvol = NULL;
+        char          *brick_host = NULL;
+        xlator_t      *kid = NULL;
 
         if (!this->children) {
                 gf_log (this->name, GF_LOG_CRITICAL,
@@ -562,23 +585,58 @@ init (xlator_t *this)
                 local_volname = data->data;
         }
 
-        trav = this->children;
-        while (trav) {
+        for (trav = this->children; trav; trav = trav->next) {
                 if (strcmp (trav->xlator->name, local_volname) == 0)
                         break;
-                trav = trav->next;
+                if (local_subvol) {
+                        continue;
+                }
+                kid = trav->xlator;
+                for (;;) {
+                        if (dict_get_str(trav->xlator->options,"remote-host",
+                                         &brick_host) == 0) {
+                                /* Found it. */
+                                break;
+                        }
+                        if (!kid->children) {
+                                /* Nowhere further to look. */
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "could not get remote-host");
+                                goto err;
+                        }
+                        if (kid->children->next) {
+                                /* Multiple choices, can't/shouldn't decide. */
+                                gf_log (this->name, GF_LOG_ERROR,
+                                        "NUFA found fan-out (type %s) volume",
+                                        kid->type);
+                                goto err;
+                        }
+                        /* One-to-one xlators are OK, try the next one. */
+                        kid = kid->children->xlator;
+                }
+                if (same_first_part(my_hostname,'.',brick_host,'.')) {
+                        local_subvol = trav->xlator;
+                }
         }
 
-        if (!trav) {
+        if (trav) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Using specified subvol %s", local_volname);
+                conf->private = trav->xlator;
+        }
+        else if (local_subvol) {
+                gf_log (this->name, GF_LOG_INFO,
+                        "Using first local subvol %s", local_subvol->name);
+                conf->private = local_subvol;
+        }
+        else {
                 gf_log (this->name, GF_LOG_ERROR,
-                        "Could not find subvolume named '%s'. "
-                        "Please define volume with the name as the hostname "
-                        "or override it with 'option local-volume-name'",
-                        local_volname);
+                        "Could not find specified or local subvol");
                 goto err;
+                
         }
+
         /* The volume specified exists */
-        conf->private = trav->xlator;
 
         conf->min_free_disk = 10;
         conf->disk_unit = 'p';
diff --git a/xlators/mgmt/glusterd/src/glusterd-volgen.c b/xlators/mgmt/glusterd/src/glusterd-volgen.c
index da1c2a4ef..40b2e32a2 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volgen.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volgen.c
@@ -108,6 +108,7 @@ static struct volopt_map_entry glusterd_volopt_map[] = {
         {"cluster.rebalance-stats",              "cluster/distribute", NULL, NULL, NO_DOC, 0, 2},
         {"cluster.subvols-per-directory",        "cluster/distribute", "directory-layout-spread", NULL, NO_DOC, 0, 2},
         {"cluster.readdir-optimize",             "cluster/distribute", NULL, NULL, NO_DOC, 0, 2},
+        {"cluster.nufa",                         "cluster/distribute", "!nufa", NULL, NO_DOC, 0, 2},
 
         /* AFR xlator options */
         {"cluster.entry-change-log",             "cluster/replicate",  NULL, NULL, NO_DOC, 0, 1},
@@ -2399,9 +2400,19 @@ volgen_graph_build_dht_cluster (volgen_graph_t *graph,
         int                     ret                      = -1;
         char                    *decommissioned_children = NULL;
         xlator_t                *dht                     = NULL;
+        char                    *optstr                  = NULL;
+        gf_boolean_t             use_nufa                = _gf_false;
 
+        if (dict_get_str(volinfo->dict,"cluster.nufa",&optstr) == 0) {
+                /* Keep static analyzers quiet by "using" the value. */
+                ret = gf_string2boolean(optstr,&use_nufa);
+        }
+                        
         clusters = volgen_graph_build_clusters (graph,  volinfo,
-                                                "cluster/distribute", "%s-dht",
+                                                use_nufa
+                                                        ? "cluster/nufa"
+                                                        : "cluster/distribute",
+                                                "%s-dht",
                                                 child_count, child_count);
         if (clusters < 0)
                 goto out;