New Cloudbreak release - support for HDP 2.2 Krisztian Horvath 23 December 2014

The last two weeks were pretty busy for us – we have Dockerized the new release of Ambari (1.7.0), integrated Periscope with Cloudbreak and just now we are announcing a new Cloudbreak release which uses Ambari 1.7.0 and has full support for Hortonworks HDP 2.2 and Apache Bigtop stacks. But first – since this has been asked many times – see a short movie about Cloudbreak and Periscope in action.

On-demand Hadoop cluster with autoscaling

Ambari 1.7.0

The Ambari community recently released the 1.7.0 version which comes with lots of new features and bug fixes. We’ve been testing the new version internally for a while now and finally made it to Cloudbreak. Just to highlight the important ones:

  • Ambari Views framework
  • Ambari Administration
    • Management of users/groups
    • Management of view instances
    • Management of cluster permissions
  • Cancel/Abort background operation requests
  • Expose Ambari UI for config versioning, history and rollback
  • Ability to manage -env.sh configuration files
  • Recommendations and validations (via a “Stack Advisor”)
  • Export service configurations via Blueprint
  • Install + Manage Flume
  • HDFS Rebalance
  • ResourceManager HA

These are nice features but for us one of the most important thing is that it allows you to install the latest versions of the Hadoop ecosystem. As usual the Docker image is available for local deployments as well, described here.

Note: There were small changes around the API so if you built an application on top of it check your REST calls. The Ambari Shell and the underlying Groovy rest client have been updated and will go into the Apache repository once it's passed the reviews.

Hadoop 2.6

Since with Ambari 1.7.0 we’re able to install Hadoop 2.6 let’s see what happened in YARN in the last couple of months (it’s stunning):

  • Support for long running services – install Slider with Ambari and scale your Hadoop services!
    • Service Registry for applications
  • Support for rolling upgrades – wow!
    • Work-preserving restarts of ResourceManager
    • Container-preserving restart of NodeManager
  • Supports node labels during scheduling – label based scaling is on the way with Periscope
  • Support for time-based resource reservations in Capacity Scheduler (beta) – more on this awesome feature soon
  • Support running of applications natively in Docker containers (alpha) – Docker in Docker

I’m excited about these great innovations (not, because we’re involved in a few of them), but because people can leverage them by using Cloudbreak.

HDP 2.2 blueprint

I have created a blueprint which is not an official one, but it contains a few from the new services like: SLIDER, KAFKA, FLUME.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
{
  "configurations": [
  {
    "nagios-env": {
      "nagios_contact": "admin@localhost"
    }
    },
    {
      "hive-site": {
        "javax.jdo.option.ConnectionUserName": "hive",
        "javax.jdo.option.ConnectionPassword": "hive"
      }
    }
  ],
  "host_groups": [
    {
      "name": "master_1",
      "components": [
        {
          "name": "NAMENODE"
        },
        {
          "name": "ZOOKEEPER_SERVER"
        },
        {
          "name": "HBASE_MASTER"
        },
        {
          "name": "GANGLIA_SERVER"
        },
        {
          "name": "HDFS_CLIENT"
        },
        {
          "name": "YARN_CLIENT"
        },
        {
          "name": "HCAT"
        },
        {
          "name": "GANGLIA_MONITOR"
        },
        {
          "name": "FALCON_SERVER"
        },
        {
          "name": "FLUME_HANDLER"
        },
        {
          "name": "KAFKA_BROKER"
        }
      ],
      "cardinality": "1"
    },
    {
      "name": "master_2",
      "components": [
        {
          "name": "ZOOKEEPER_CLIENT"
        },
        {
          "name": "HISTORYSERVER"
        },
        {
          "name": "HIVE_SERVER"
        },
        {
          "name": "SECONDARY_NAMENODE"
        },
        {
          "name": "HIVE_METASTORE"
        },
        {
          "name": "HDFS_CLIENT"
        },
        {
          "name": "HIVE_CLIENT"
        },
        {
          "name": "YARN_CLIENT"
        },
        {
          "name": "MYSQL_SERVER"
        },
        {
          "name": "GANGLIA_MONITOR"
        },
        {
          "name": "WEBHCAT_SERVER"
        }
      ],
      "cardinality": "1"
    },
    {
      "name": "master_3",
      "components": [
        {
          "name": "RESOURCEMANAGER"
        },
        {
          "name": "APP_TIMELINE_SERVER"
        },
        {
          "name": "SLIDER"
        },
        {
          "name": "ZOOKEEPER_SERVER"
        },
        {
          "name": "GANGLIA_MONITOR"
        }
      ],
      "cardinality": "1"
    },
    {
      "name": "master_4",
      "components": [
        {
          "name": "OOZIE_SERVER"
        },
        {
          "name": "ZOOKEEPER_SERVER"
        },
        {
          "name": "GANGLIA_MONITOR"
        }
      ],
      "cardinality": "1"
    },
    {
      "name": "slave_1",
      "components": [
        {
          "name": "HBASE_REGIONSERVER"
        },
        {
          "name": "NODEMANAGER"
        },
        {
          "name": "DATANODE"
        },
        {
          "name": "GANGLIA_MONITOR"
        },
        {
          "name": "FALCON_CLIENT"
        },
        {
          "name": "OOZIE_CLIENT"
        }
      ],
      "cardinality": "${slavesCount}"
    },
    {
      "name": "gateway",
      "components": [
        {
          "name": "AMBARI_SERVER"
        },
        {
          "name": "NAGIOS_SERVER"
        },
        {
          "name": "ZOOKEEPER_CLIENT"
        },
        {
          "name": "PIG"
        },
        {
          "name": "OOZIE_CLIENT"
        },
        {
          "name": "HBASE_CLIENT"
        },
        {
          "name": "HCAT"
        },
        {
          "name": "SQOOP"
        },
        {
          "name": "HDFS_CLIENT"
        },
        {
          "name": "HIVE_CLIENT"
        },
        {
          "name": "YARN_CLIENT"
        },
        {
          "name": "MAPREDUCE2_CLIENT"
        },
        {
          "name": "GANGLIA_MONITOR"
        },
        {
          "name": "KNOX_GATEWAY"
        }
      ],
      "cardinality": "1"
    }
    ],
    "Blueprints": {
      "blueprint_name": "hdp-multinode-sequenceiq",
      "stack_name": "HDP",
      "stack_version": "2.2"
    }
}

What’s next?

Do the difficult things while they are easy and do the great things while they are small. A journey of a thousand miles must begin with a single step.

We’ve walked a long journey since we started the company almost a year ago to reach where we are now, but our products are not complete yet. We have big plans with our product stacks. A couple of things from our roadmap:

Cloudbreak

  • Cloudbreak currently supports homogeneous cluster deployments which we’re going to change. The heterogeneous stack structure is more convenient from Hadoop’s perspective. The ability to define different type of cloud instances is a must, giving the users the option to use much more powerful instances for the ResourceManager and NameNodes.
  • Service discovery and decentralization is always a key aspect. At the moment we’re using Serf and dnsmasq, but we’re already started the integration with Consul which generally is a better fit. It provides service registration via DNS, key-value store and decentralization across datacenters.
  • The deployment of Cloudbreak itself is going to change and use Consul with other side projects like Consul templates or Registrator. The deployment is already based on Docker, but will be much more simplified.
  • Custom stack deployments with Ambari will be supported as "recipes".
  • Generating reports of cloud instance usages and cost calculation.
  • Web hooks to subscribe to different cluster events.
  • Shared/company accounts.

Periscope

  • Add more YARN and NameNode related metrics.
  • Node label based scaling.
  • Pluggable metric system for custom metrics.
  • Application movement in Capacity Scheduler queues enforcing SLAs.
  • Time-based resource reservations in Capacity Scheduler for Applications.
  • Integration with ELK.

Happy Holidays

We’re taking a short break of writing new blog posts until next year. You can still reach us on the usual social sites, but you can expect small delays for answering questions. Happy Holidays everyone.

LinkedIn Twitter Facebook

Comments

Recent Posts