gsyncd: distribute the crawling load

* also consume changelog for change detection.
* Status fixes
* Use new libgfchangelog done API
* process (and sync) one changelog at a time

Change-Id: I24891615bb762e0741b1819ddfdef8802326cb16
BUG: 847839
Original Author: Csaba Henk <csaba@redhat.com>
Original Author: Aravinda VK <avishwan@redhat.com>
Original Author: Venky Shankar <vshankar@redhat.com>
Original Author: Amar Tumballi <amarts@redhat.com>
Original Author: Avra Sengupta <asengupt@redhat.com>
Signed-off-by: Avra Sengupta <asengupt@redhat.com>
Reviewed-on: http://review.gluster.org/5131
Reviewed-by: Vijay Bellur <vbellur@redhat.com>
Tested-by: Vijay Bellur <vbellur@redhat.com>

5 files changed, 207 insertions, 2 deletions

diff --git a/geo-replication/src/Makefile.am b/geo-replication/src/Makefile.am
index 9e410cda633..6feeda8e68c 100644
--- a/geo-replication/src/Makefile.am
+++ b/geo-replication/src/Makefile.am
@@ -1,5 +1,12 @@
+
 gsyncddir = $(libexecdir)/glusterfs
 
+gsyncd_SCRIPTS = gverify.sh peer_add_secret_pub peer_gsec_create
+
+# peer_gsec_create is not added to EXTRA_DIST as it's derived
+# from a .in file
+EXTRA_DIST = gverify.sh peer_add_secret_pub
+
 gsyncd_PROGRAMS = gsyncd
 
 gsyncd_SOURCES = gsyncd.c procdiggy.c
diff --git a/geo-replication/src/gsyncd.c b/geo-replication/src/gsyncd.c
index 9c4a5bdffb3..68446d9ad34 100644
--- a/geo-replication/src/gsyncd.c
+++ b/geo-replication/src/gsyncd.c
@@ -285,6 +285,46 @@ invoke_rsync (int argc, char **argv)
         return 1;
 }
 
+static int
+invoke_gluster (int argc, char **argv)
+{
+        int i = 0;
+        int j = 0;
+        int optsover = 0;
+        char *ov = NULL;
+
+        for (i = 1; i < argc; i++) {
+                ov = strtail (argv[i], "--");
+                if (ov && !optsover) {
+                        if (*ov == '\0')
+                                optsover = 1;
+                        continue;
+                }
+                switch (++j) {
+                case 1:
+                        if (strcmp (argv[i], "volume") != 0)
+                                goto error;
+                        break;
+                case 2:
+                        if (strcmp (argv[i], "info") != 0)
+                                goto error;
+                        break;
+                case 3:
+                        break;
+                default:
+                        goto error;
+                }
+        }
+
+        argv[0] = "gluster";
+        execvp (SBIN_DIR"/gluster", argv);
+        fprintf (stderr, "exec of gluster failed\n");
+        return 127;
+
+ error:
+        fprintf (stderr, "disallowed gluster invocation\n");
+        return 1;
+}
 
 struct invocable {
         char *name;
@@ -292,8 +332,9 @@ struct invocable {
 };
 
 struct invocable invocables[] = {
-        { "rsync",  invoke_rsync  },
-        { "gsyncd", invoke_gsyncd },
+        { "rsync",   invoke_rsync  },
+        { "gsyncd",  invoke_gsyncd },
+        { "gluster", invoke_gluster },
         { NULL, NULL}
 };
 
diff --git a/geo-replication/src/gverify.sh b/geo-replication/src/gverify.sh
new file mode 100755
index 00000000000..186af53a407
--- /dev/null
+++ b/geo-replication/src/gverify.sh
@@ -0,0 +1,142 @@
+#!/bin/bash
+
+# Script to verify the Master and Slave Gluster compatibility.
+# To use ./gverify <master volume> <slave host> <slave volume>
+# Returns 0 if master and slave compatible.
+
+BUFFER_SIZE=1000;
+slave_log_file=`gluster --print-logdir`/geo-replication-slaves/slave.log
+
+function SSHM()
+{
+    ssh -q \
+	-oPasswordAuthentication=no \
+	-oStrictHostKeyChecking=no \
+	-oControlMaster=yes \
+	"$@";
+}
+
+function cmd_master()
+{
+    VOL=$1;
+    local cmd_line;
+    cmd_line=$(cat <<EOF
+function do_verify() {
+v=\$1;
+d=\$(mktemp -d 2>/dev/null);
+glusterfs -s localhost --xlator-option="*dht.lookup-unhashed=off" --volfile-id \$v -l $slave_log_file \$d;
+i=\$(stat -c "%i" \$d);
+if [[ "\$i" -ne "1" ]]; then
+echo 0:0;
+exit 1;
+fi;
+cd \$d;
+available_size=\$(df \$d | tail -1 | awk "{print \\\$2}");
+umount -l \$d;
+rmdir \$d;
+ver=\$(gluster --version | head -1 | cut -f2 -d " ");
+echo \$available_size:\$ver;
+};
+cd /tmp;
+[ x$VOL != x ] && do_verify $VOL;
+EOF
+);
+
+echo $cmd_line;
+}
+
+function cmd_slave()
+{
+    VOL=$1;
+    local cmd_line;
+    cmd_line=$(cat <<EOF
+function do_verify() {
+v=\$1;
+d=\$(mktemp -d 2>/dev/null);
+glusterfs -s localhost --xlator-option="*dht.lookup-unhashed=off" --volfile-id \$v -l $slave_log_file \$d;
+i=\$(stat -c "%i" \$d);
+if [[ "\$i" -ne "1" ]]; then
+echo 0:0;
+exit 1;
+fi;
+cd \$d;
+available_size=\$(df \$d | tail -1 | awk "{print \\\$4}");
+umount -l \$d;
+rmdir \$d;
+ver=\$(gluster --version | head -1 | cut -f2 -d " ");
+echo \$available_size:\$ver;
+};
+cd /tmp;
+[ x$VOL != x ] && do_verify $VOL;
+EOF
+);
+
+echo $cmd_line;
+}
+
+function master_stats()
+{
+    MASTERVOL=$1;
+    local cmd_line;
+    cmd_line=$(cmd_master $MASTERVOL);
+    bash -c "$cmd_line";
+}
+
+
+function slave_stats()
+{
+    SLAVEHOST=$1;
+    SLAVEVOL=$2;
+    local cmd_line;
+    cmd_line=$(cmd_slave $SLAVEVOL);
+    SSHM $SLAVEHOST bash -c "'$cmd_line'";
+}
+
+
+function main()
+{
+    ERRORS=0;
+    master_data=$(master_stats $1);
+    slave_data=$(slave_stats $2 $3);
+    master_size=$(echo $master_data | cut -f1 -d':');
+    slave_size=$(echo $slave_data | cut -f1 -d':');
+    master_version=$(echo $master_data | cut -f2 -d':');
+    slave_version=$(echo $slave_data | cut -f2 -d':');
+    log_file=$4
+
+    if [[ "x$master_size" = "x" || "x$master_version" = "x" || "$master_size" -eq "0" ]]; then
+	echo "Unable to fetch master volume details." > $log_file;
+	exit 1;
+    fi;
+
+    if [[ "x$slave_size" = "x" || "x$slave_version" = "x" || "$slave_size" -eq "0" ]]; then
+        ping -w 5 $2;
+        if [ $? -ne 0 ]; then
+            echo "$2 not reachable." > $log_file
+            exit 1;
+        fi;
+	echo "Unable to fetch slave volume details." > $log_file;
+	exit 1;
+    fi;
+
+    if [ $slave_size -ge $(($master_size - $BUFFER_SIZE )) ]; then
+	echo "Total size of master is lesser than available size of slave." > $log_file;
+    else
+	echo "Total size of master is greater than available size of slave." > $log_file;
+	ERRORS=$(($ERRORS + 1));
+	exit $ERRORS;
+    fi;
+
+    if [[ $master_version < $slave_version || $master_version == $slave_version ]]; then
+	echo "Gluster version of master and slave matches." > $log_file;
+    else
+	echo "Gluster version mismatch between master and slave." > $log_file;
+	ERRORS=$(($ERRORS + 1));
+	exit $ERRORS;
+    fi;
+
+    exit $ERRORS;
+}
+
+
+main "$@";
diff --git a/geo-replication/src/peer_add_secret_pub b/geo-replication/src/peer_add_secret_pub
new file mode 100644
index 00000000000..1ce040d4419
--- /dev/null
+++ b/geo-replication/src/peer_add_secret_pub
@@ -0,0 +1,3 @@
+#!/bin/bash
+
+cat $1 >> ~/.ssh/authorized_keys
diff --git a/geo-replication/src/peer_gsec_create.in b/geo-replication/src/peer_gsec_create.in
new file mode 100755
index 00000000000..ef630bd4417
--- /dev/null
+++ b/geo-replication/src/peer_gsec_create.in
@@ -0,0 +1,12 @@
+#!/bin/bash
+
+prefix=@prefix@
+exec_prefix=@exec_prefix@
+
+if [ ! -f "$GLUSTERD_WORKING_DIR"/geo-replication/secret.pem.pub ]; then
+    \rm -rf "$GLUSTERD_WORKING_DIR"/geo-replication/secret.pem*
+    ssh-keygen -N '' -f "$GLUSTERD_WORKING_DIR"/geo-replication/secret.pem > /dev/null
+fi
+
+output=`echo command=\"@libexecdir@/glusterfs/gsyncd\" " "``cat "$GLUSTERD_WORKING_DIR"/geo-replication/secret.pem.pub`
+echo $output