Registry DNS Server
The document describes the internals of Registry DNS server. It is based on the YARN service registry which is backed by a zookeeper cluster.
Introduction
The Registry DNS Server provides a standard DNS interface to the information posted into the YARN Registry by deployed applications. The DNS service serves the following functions:
- Exposing existing service-discovery information via DNS - Information provided in the current YARN service registry’s records will be converted into DNS entries, thus allowing users to discover information about YARN applications using standard DNS client mechanisms (e.g. a DNS SRV Record specifying the hostname and port number for services).
- Enabling Container to IP mappings - Enables discovery of the IPs of containers via standard DNS lookups. Given the availability of the records via DNS, container name-based communication will be facilitated (e.g. curl http://solr-0.solr-service.devuser.yarncluster:8983/solr/admin/collections?action=LIST).
Service Properties
The existing YARN Service Registry is leveraged as the source of information for the DNS Service.
The following core functions are supported by the DNS-Server:
Functional properties
- Supports creation of DNS records for end-points of the deployed YARN applications
- Record names remain unchanged during restart of containers and/or applications
- Supports reverse lookups (name based on IP). Note, this works only for Docker containers because other containers share the IP of the host
- Supports security using the standards defined by The Domain Name System Security Extensions (DNSSEC)
- Highly available
- Scalable - The service provides the responsiveness (e.g. low-latency) required to respond to DNS queries (timeouts yield attempts to invoke other configured name servers).
Deployment properties
- Supports integration with existing DNS assets (e.g. a corporate DNS server) by acting as a DNS server for a Hadoop cluster zone/domain. The server is not intended to act as a primary DNS server and does not forward requests to other servers. Rather, a primary DNS server can be configured to forward a zone to the registry DNS server.
- The DNS Server exposes a port that can receive both TCP and UDP requests per DNS standards. The default port for DNS protocols is not in the restricted range (5353). However, existing DNS assets may only allow zone forwarding to non-custom ports. To support this, the registry DNS server can be started in privileged mode.
DNS Record Name Structure
The DNS names of generated records are composed from the following elements (labels). Note that these elements must be compatible with DNS conventions (see “Preferred Name Syntax” in RFC 1035):
- domain - the name of the cluster DNS domain. This name is provided as a configuration property. In addition, it is this name that is configured at a parent DNS server as the zone name for the defined registry DNS zone (the zone for which the parent DNS server will forward requests to registry DNS). E.g. yarncluster.com
- username - the name of the application deployer. This name is the simple short-name (for e.g. the primary component of the Kerberos principal) associated with the user launching the application. As the username is one of the elements of DNS names, it is expected that this also conforms to DNS name conventions (RFC 1035 linked above), so it is converted to a valid DNS hostname entries using the punycode convention used for internationalized DNS.
- application name - the name of the deployed YARN application. This name is inferred from the YARN registry path to the application’s node. Application name, rather than application id, was chosen as a way of making it easy for users to refer to human-readable DNS names. This obviously mandates certain uniqueness properties on application names.
- container id - the YARN assigned ID to a container (e.g. container_e3741_1454001598828_01_000004)
- component name - the name assigned to the deployed component (for e.g. a master component). A component is a distributed element of an application or service that is launched in a YARN container (e.g. an HBase master). One can imagine multiple components within an application. A component name is not yet a first class concept in YARN, but is a very useful one that we are introducing here for the sake of registry DNS entries. Many frameworks like MapReduce, Slider already have component names (though, as mentioned, they are not yet supported in YARN in a first class fashion).
- api - the api designation for the exposed endpoint
Notes about DNS Names
- In most instances, the DNS names can be easily distinguished by the number of elements/labels that compose the name. The cluster’s domain name is always the last element. After that element is parsed out, reading from right to left, the first element maps to the application user and so on. Wherever it is not easily distinguishable, naming conventions are used to disambiguate the name using a prefix such as “container” or suffix such as “api”. For example, an endpoint published as a management endpoint will be referenced with the name management-api.griduser.yarncluster.com.
- Unique application name (per user) is not currently supported/guaranteed by YARN, but it is supported by the YARN service framework. The registry DNS service currently leverages the last element of the ZK path entry for the application as an application name. These application names have to be unique for a given user.
DNS Server Functionality
The primary functions of the DNS service are illustrated in the following diagram:
DNS record creation
The following figure illustrates at slightly greater detail the DNS record creation and registration sequence (NOTE: service record updates would follow a similar sequence of steps, distinguished only by the different event type):
DNS record removal
Similarly, record removal follows a similar sequence
(NOTE: The DNS Zone requires a record as an argument for the deletion method, thus requiring similar parsing logic to identify the specific records that should be removed).
DNS Service initialization
- The DNS service initializes both UDP and TCP listeners on a configured port. If a port in the restricted range is desired (such as the standard DNS port 53), the DNS service can be launched using jsvc as described in the section on starting the DNS server.
- Subsequently, the DNS service listens for inbound DNS requests. Those requests are standard DNS requests from users or other DNS servers (for example, DNS servers that have the RegistryDNS service configured as a forwarder).
Start the DNS Server
By default, the DNS server runs on non-privileged port 5353. Start the server with:
yarn --daemon start registrydns
If the DNS server is configured to use the standard privileged port 53, the environment variables YARN_REGISTRYDNS_SECURE_USER and YARN_REGISTRYDNS_SECURE_EXTRA_OPTS must be uncommented in the yarn-env.sh file. The DNS server should then be launched as root and jsvc will be used to reduce the privileges of the daemon after the port has been bound.
Configuration
The Registry DNS server reads its configuration properties from the yarn-site.xml file. The following are the DNS associated configuration properties:
| Name |
Description |
| hadoop.registry.zk.quorum |
A comma separated list of hostname:port pairs defining the zookeeper quorum for the YARN registry. |
| hadoop.registry.dns.enabled |
The DNS functionality is enabled for the cluster. Default is false. |
| hadoop.registry.dns.domain-name |
The domain name for Hadoop cluster associated records. |
| hadoop.registry.dns.bind-address |
Address associated with the network interface to which the DNS listener should bind. |
| hadoop.registry.dns.bind-port |
The port number for the DNS listener. The default port is 5353. |
| hadoop.registry.dns.dnssec.enabled |
Indicates whether the DNSSEC support is enabled. Default is false. |
| hadoop.registry.dns.public-key |
The base64 representation of the server’s public key. Leveraged for creating the DNSKEY Record provided for DNSSEC client requests. |
| hadoop.registry.dns.private-key-file |
The path to the standard DNSSEC private key file. Must only be readable by the DNS launching identity. See dnssec-keygen documentation. |
| hadoop.registry.dns-ttl |
The default TTL value to associate with DNS records. The default value is set to 1 (a value of 0 has undefined behavior). A typical value should be approximate to the time it takes YARN to restart a failed container. |
| hadoop.registry.dns.zone-subnet |
An indicator of the IP range associated with the cluster containers. The setting is utilized for the generation of the reverse zone name. |
| hadoop.registry.dns.zone-mask |
The network mask associated with the zone IP range. If specified, it is utilized to ascertain the IP range possible and come up with an appropriate reverse zone name. |
| hadoop.registry.dns.zones-dir |
A directory containing zone configuration files to read during zone initialization. This directory can contain zone master files named zone-name.zone. See here for zone master file documentation. |
Sample configurations
<property>
<description>The domain name for Hadoop cluster associated records.</description>
<name>hadoop.registry.dns.domain-name</name>
<value>ycluster</value>
</property>
<property>
<description>The port number for the DNS listener. The default port is 5353.
If the standard privileged port 53 is used, make sure start the DNS with jsvc support.</description>
<name>hadoop.registry.dns.bind-port</name>
<value>5353</value>
</property>
<property>
<description>The DNS functionality is enabled for the cluster. Default is false.</description>
<name>hadoop.registry.dns.enabled</name>
<value>true</value>
</property>
<property>
<description>Address associated with the network interface to which the DNS listener should bind.</description>
<name>hadoop.registry.dns.bind-address</name>
<value>localhost</value>
</property>
<property>
<description>A comma separated list of hostname:port pairs defining the zookeeper quorum for the YARN registry</description>
<name>hadoop.registry.zk.quorum</name>
<value>localhost:2181</value>
</property>
