diff --git a/doc/source/userdoc/cdh_plugin.rst b/doc/source/userdoc/cdh_plugin.rst
index 1c9b129d..57d33b0f 100644
--- a/doc/source/userdoc/cdh_plugin.rst
+++ b/doc/source/userdoc/cdh_plugin.rst
@@ -55,6 +55,7 @@ cloudera plugin versions:
   + Cluster must contain exactly one manager.
   + Cluster must contain exactly one namenode.
   + Cluster must contain exactly one secondarynamenode.
+  + Cluster must contain at least ``dfs_replication`` datanodes.
   + Cluster can contain at most one resourcemanager and this process is also
     required by nodemanager.
   + Cluster can contain at most one jobhistory and this process is also
diff --git a/sahara/plugins/cdh/plugin_utils.py b/sahara/plugins/cdh/plugin_utils.py
index 1f0565b2..17bcc985 100644
--- a/sahara/plugins/cdh/plugin_utils.py
+++ b/sahara/plugins/cdh/plugin_utils.py
@@ -22,6 +22,7 @@ from oslo_log import log as logging
 
 from sahara.conductor import resource as res
 from sahara import context
+from sahara import exceptions as exc
 from sahara.i18n import _
 from sahara.plugins.cdh import commands as cmd
 from sahara.plugins import utils as u
@@ -323,3 +324,15 @@ class AbstractPluginUtils(object):
                 cmd.write_centos_repository(r, cdh5_repo_content, 'cdh')
                 cmd.write_centos_repository(r, cm5_repo_content, 'cm')
                 cmd.update_repository(r)
+
+    def _get_config_value(self, service, name, configs, cluster=None):
+        conf = cluster.cluster_configs
+        if cluster:
+            if service in conf and name in conf[service]:
+                return conf[service][name]
+        for config in configs:
+            if config.applicable_target == service and config.name == name:
+                return config.default_value
+        raise exc.InvalidDataException(
+            _("Unable to find config: {applicable_target: %(target)s, name: "
+              "%(name)s").format(target=service, name=name))
diff --git a/sahara/plugins/cdh/v5/plugin_utils.py b/sahara/plugins/cdh/v5/plugin_utils.py
index 24304dfb..1a53417a 100644
--- a/sahara/plugins/cdh/v5/plugin_utils.py
+++ b/sahara/plugins/cdh/v5/plugin_utils.py
@@ -57,3 +57,7 @@ class PluginUtilsV5(pu.AbstractPluginUtils):
     def start_cloudera_manager(self, cluster):
         self._start_cloudera_manager(
             cluster, c_helper.AWAIT_MANAGER_STARTING_TIMEOUT)
+
+    def get_config_value(self, service, name, cluster=None):
+        configs = c_helper.get_plugin_configs()
+        return self._get_config_value(service, name, configs, cluster)
diff --git a/sahara/plugins/cdh/v5/validation.py b/sahara/plugins/cdh/v5/validation.py
index be76aae0..b80d677a 100644
--- a/sahara/plugins/cdh/v5/validation.py
+++ b/sahara/plugins/cdh/v5/validation.py
@@ -36,6 +36,12 @@ def validate_cluster_creating(cluster):
     if snn_count != 1:
         raise ex.InvalidComponentCountException('HDFS_SECONDARYNAMENODE', 1,
                                                 snn_count)
+    dn_count = _get_inst_count(cluster, 'HDFS_DATANODE')
+    replicas = PU.get_config_value('HDFS', 'dfs_replication', cluster)
+    if dn_count < replicas:
+        raise ex.InvalidComponentCountException(
+            'HDFS_DATANODE', replicas, dn_count,
+            _('Number of datanodes must be not less than dfs_replication.'))
 
     rm_count = _get_inst_count(cluster, 'YARN_RESOURCEMANAGER')
     if rm_count > 1:
@@ -58,7 +64,6 @@ def validate_cluster_creating(cluster):
                 'YARN_RESOURCEMANAGER', required_by='YARN_NODEMANAGER')
 
     oo_count = _get_inst_count(cluster, 'OOZIE_SERVER')
-    dn_count = _get_inst_count(cluster, 'HDFS_DATANODE')
     if oo_count > 1:
         raise ex.InvalidComponentCountException('OOZIE_SERVER', _('0 or 1'),
                                                 oo_count)
@@ -156,7 +161,8 @@ def validate_existing_ng_scaling(cluster, existing):
     dn_to_delete = 0
     for ng in cluster.node_groups:
         if ng.id in existing:
-            if ng.count > existing[ng.id] and "datanode" in ng.node_processes:
+            if (ng.count > existing[ng.id] and
+                    "HDFS_DATANODE" in ng.node_processes):
                 dn_to_delete += ng.count - existing[ng.id]
 
             if not set(ng.node_processes).issubset(scalable_processes):
@@ -165,6 +171,13 @@ def validate_existing_ng_scaling(cluster, existing):
                 raise ex.NodeGroupCannotBeScaled(
                     ng.name, msg % {'processes': ' '.join(ng.node_processes)})
 
+    dn_count = _get_inst_count(cluster, 'HDFS_DATANODE') - dn_to_delete
+    replicas = PU.get_config_value('HDFS', 'dfs_replication', cluster)
+    if dn_count < replicas:
+        raise ex.ClusterCannotBeScaled(
+            cluster,
+            _('Number of datanodes must be not less than dfs_replication.'))
+
 
 def _get_scalable_processes():
     return ['HDFS_DATANODE', 'YARN_NODEMANAGER']
diff --git a/sahara/plugins/cdh/v5_3_0/plugin_utils.py b/sahara/plugins/cdh/v5_3_0/plugin_utils.py
index 032cfdf1..03ffa98e 100644
--- a/sahara/plugins/cdh/v5_3_0/plugin_utils.py
+++ b/sahara/plugins/cdh/v5_3_0/plugin_utils.py
@@ -127,3 +127,7 @@ class PluginUtilsV530(pu.AbstractPluginUtils):
     def start_cloudera_manager(self, cluster):
         self._start_cloudera_manager(
             cluster, c_helper.AWAIT_MANAGER_STARTING_TIMEOUT)
+
+    def get_config_value(self, service, name, cluster=None):
+        configs = c_helper.get_plugin_configs()
+        return self._get_config_value(service, name, configs, cluster)
diff --git a/sahara/plugins/cdh/v5_3_0/validation.py b/sahara/plugins/cdh/v5_3_0/validation.py
index aabc39a8..54d7bc55 100644
--- a/sahara/plugins/cdh/v5_3_0/validation.py
+++ b/sahara/plugins/cdh/v5_3_0/validation.py
@@ -37,6 +37,13 @@ def validate_cluster_creating(cluster):
         raise ex.InvalidComponentCountException('HDFS_SECONDARYNAMENODE', 1,
                                                 snn_count)
 
+    dn_count = _get_inst_count(cluster, 'HDFS_DATANODE')
+    replicas = PU.get_config_value('HDFS', 'dfs_replication', cluster)
+    if dn_count < replicas:
+        raise ex.InvalidComponentCountException(
+            'HDFS_DATANODE', replicas, dn_count,
+            _('Number of datanodes must be not less than dfs_replication.'))
+
     rm_count = _get_inst_count(cluster, 'YARN_RESOURCEMANAGER')
     if rm_count > 1:
         raise ex.InvalidComponentCountException('YARN_RESOURCEMANAGER',
@@ -58,7 +65,6 @@ def validate_cluster_creating(cluster):
                 'YARN_RESOURCEMANAGER', required_by='YARN_NODEMANAGER')
 
     oo_count = _get_inst_count(cluster, 'OOZIE_SERVER')
-    dn_count = _get_inst_count(cluster, 'HDFS_DATANODE')
     if oo_count > 1:
         raise ex.InvalidComponentCountException('OOZIE_SERVER', _('0 or 1'),
                                                 oo_count)
@@ -236,7 +242,8 @@ def validate_existing_ng_scaling(cluster, existing):
     dn_to_delete = 0
     for ng in cluster.node_groups:
         if ng.id in existing:
-            if ng.count > existing[ng.id] and "datanode" in ng.node_processes:
+            if (ng.count > existing[ng.id] and
+                    'HDFS_DATANODE' in ng.node_processes):
                 dn_to_delete += ng.count - existing[ng.id]
 
             if not set(ng.node_processes).issubset(scalable_processes):
@@ -245,6 +252,13 @@ def validate_existing_ng_scaling(cluster, existing):
                 raise ex.NodeGroupCannotBeScaled(
                     ng.name, msg % {'processes': ' '.join(ng.node_processes)})
 
+    dn_count = _get_inst_count(cluster, 'HDFS_DATANODE') - dn_to_delete
+    replicas = PU.get_config_value('HDFS', 'dfs_replication', cluster)
+    if dn_count < replicas:
+        raise ex.ClusterCannotBeScaled(
+            cluster,
+            _('Number of datanodes must be not less than dfs_replication.'))
+
 
 def _get_scalable_processes():
     return ['HDFS_DATANODE', 'YARN_NODEMANAGER']