You are viewing a plain text version of this content. The canonical link for it is here.
Posted to common-commits@hadoop.apache.org by bi...@apache.org on 2017/09/28 00:38:28 UTC
hadoop git commit: YARN-7191. Improve yarn-service documentation.
Contributed by Jian He
Repository: hadoop
Updated Branches:
refs/heads/yarn-native-services 3f7a50d8d -> 281c1d1e8
YARN-7191. Improve yarn-service documentation. Contributed by Jian He
Project: http://git-wip-us.apache.org/repos/asf/hadoop/repo
Commit: http://git-wip-us.apache.org/repos/asf/hadoop/commit/281c1d1e
Tree: http://git-wip-us.apache.org/repos/asf/hadoop/tree/281c1d1e
Diff: http://git-wip-us.apache.org/repos/asf/hadoop/diff/281c1d1e
Branch: refs/heads/yarn-native-services
Commit: 281c1d1e87aeea227a7174a72d9957d8af6faf07
Parents: 3f7a50d
Author: Billie Rinaldi <bi...@apache.org>
Authored: Wed Sep 27 15:08:33 2017 -0700
Committer: Billie Rinaldi <bi...@apache.org>
Committed: Wed Sep 27 15:08:33 2017 -0700
----------------------------------------------------------------------
.../src/site/markdown/yarn-service/Concepts.md | 47 +---
.../src/site/markdown/yarn-service/Overview.md | 3 +-
.../site/markdown/yarn-service/QuickStart.md | 34 +--
.../site/markdown/yarn-service/RegistryDNS.md | 166 +++++++++++++
.../markdown/yarn-service/ServiceDiscovery.md | 235 ++++++++-----------
5 files changed, 286 insertions(+), 199 deletions(-)
----------------------------------------------------------------------
http://git-wip-us.apache.org/repos/asf/hadoop/blob/281c1d1e/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/Concepts.md
----------------------------------------------------------------------
diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/Concepts.md b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/Concepts.md
index 7b62c36..e567d03 100644
--- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/Concepts.md
+++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/Concepts.md
@@ -22,6 +22,8 @@ It also does all the heavy lifting work such as resolving the service definition
failed containers, monitoring components' healthiness and readiness, ensuring dependency start order across components, flexing up/down components,
upgrading components etc. The end goal of the framework is to make sure the service is up and running as the state that user desired.
+In addition, it leverages a lot of features in YARN core to accomplish scheduling constraints, such as
+affinity and anti-affinity scheduling, log aggregation for services, automatically restart a container if it fails, and do in-place upgrade of a container.
### A Restful API-Server for deploying/managing services on YARN
A restful API server is developed to allow users to deploy/manage their services on YARN via a simple JSON spec. This avoids users
@@ -34,44 +36,11 @@ support HA, distribute the load etc.
### Service Discovery
A DNS server is implemented to enable discovering services on YARN via the standard mechanism: DNS lookup.
-The DNS server essentially exposes the information in YARN service registry by translating them into DNS records such as A record and SRV record.
-Clients can discover the IPs of containers via standard DNS lookup.
-The previous read mechanisms of YARN Service Registry were limited to a registry specific (java) API and a REST interface and are difficult
-to wireup existing clients and services. The DNS based service discovery eliminates this gap. Please refer to this [DNS doc](ServiceDiscovery.md)
-for more details.
-
-### Scheduling
-
-A host of scheduling features are being developed to support long running services.
-
-* Affinity and anti-affinity scheduling across containers ([YARN-6592](https://issues.apache.org/jira/browse/YARN-6592)).
-* Container resizing ([YARN-1197](https://issues.apache.org/jira/browse/YARN-1197))
-* Special handling of container preemption/reservation for services
-
-### Container auto-restarts
-
-[YARN-3998](https://issues.apache.org/jira/browse/YARN-3998) implements a retry-policy to let NM re-launch a service container when it fails.
-The service REST API provides users a way to enable NodeManager to automatically restart the container if it fails.
-The advantage is that it avoids the entire cycle of releasing the failed containers, re-asking new containers, re-do resource localizations and so on, which
-greatly minimizes container downtime.
+The framework posts container information such as hostname and ip into the [YARN service registry](../registry/index.md). And the DNS server essentially exposes the
+information in YARN service registry by translating them into DNS records such as A record and SRV record.
+Clients can then discover the IPs of containers via standard DNS lookup.
-### Container in-place upgrade
-
-[YARN-4726](https://issues.apache.org/jira/browse/YARN-4726) aims to support upgrading containers in-place, that is, without losing the container allocations.
-It opens up a few APIs in NodeManager to allow ApplicationMasters to upgrade their containers via a simple API call.
-Under the hood, NodeManager does below steps:
-* Downloading the new resources such as jars, docker container images, new configurations.
-* Stop the old container.
-* Start the new container with the newly downloaded resources.
-
-At the time of writing this document, core changes are done but the feature is not usable end-to-end.
-
-### Resource Profiles
-
-In [YARN-3926](https://issues.apache.org/jira/browse/YARN-3926), YARN introduces Resource Profiles which extends the YARN resource model for easier
-resource-type management and profiles.
-It primarily solves two problems:
-* Make it easy to support new resource types such as network bandwith([YARN-2140](https://issues.apache.org/jira/browse/YARN-2140)), disks([YARN-2139](https://issues.apache.org/jira/browse/YARN-2139)).
- Under the hood, it unifies the scheduler codebase to essentially parameterize the resource types.
-* User can specify the container resource requirement by a profile name, rather than fiddling with varying resource-requirements for each resource type.
+The previous read mechanisms of YARN Service Registry were limited to a registry specific (java) API and a REST interface and are difficult
+to wireup existing clients and services. The DNS based service discovery eliminates this gap. Please refer to this [Service Discovery doc](ServiceDiscovery.md)
+for more details.
\ No newline at end of file
http://git-wip-us.apache.org/repos/asf/hadoop/blob/281c1d1e/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/Overview.md
----------------------------------------------------------------------
diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/Overview.md b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/Overview.md
index 407fbc0..58daee5 100644
--- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/Overview.md
+++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/Overview.md
@@ -52,7 +52,8 @@ The benefits of combining these workloads are two-fold:
* [Concepts](Concepts.md): Describes the internals of the framework and some features in YARN core to support running services on YARN.
* [Service REST API](YarnServiceAPI.md): The API doc for deploying/managing services on YARN.
-* [Service Discovery](ServiceDiscovery.md): Deep dives into the YARN DNS internals.
+* [Service Discovery](ServiceDiscovery.md): Descirbes the service discovery mechanism on YARN.
+* [Registry DNS](RegistryDNS.md): Deep dives into the Registry DNS internals.
* [Examples](Examples.md): List some example service definitions (`Yarnfile`).
http://git-wip-us.apache.org/repos/asf/hadoop/blob/281c1d1e/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/QuickStart.md
----------------------------------------------------------------------
diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/QuickStart.md b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/QuickStart.md
index ab415de..15df0cd 100644
--- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/QuickStart.md
+++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/QuickStart.md
@@ -194,32 +194,10 @@ If you are building from source code, make sure you use `-Pyarn-ui` in the `mvn`
</property>
```
-## Service Discovery with YARN DNS
-YARN Service framework comes with a DNS server (backed by YARN Service Registry) which enables DNS based discovery of services deployed on YARN.
-That is, user can simply access their services in a well-defined naming format as below:
+# Try with Docker
+The above example is only for a non-docker container based service. YARN Service Framework also provides first-class support for managing docker based services.
+Most of the steps for managing docker based services are the same except that in docker the `Artifact` type for a component is `DOCKER` and the Artifact `id` is the name of the docker image.
+For details in how to setup docker on YARN, please check [Docker on YARN](../DockerContainers.md).
-```
-${COMPONENT_INSTANCE_NAME}.${SERVICE_NAME}.${USER}.${DOMAIN}
-```
-For example, in a cluster whose domain name is `yarncluster` (as defined by the `hadoop.registry.dns.domain-name` in `yarn-site.xml`), a service named `hbase` deployed by user `dev`
-with two components `hbasemaster` and `regionserver` can be accessed as below:
-
-This URL points to the usual hbase master UI
-```
-http://hbasemaster-0.hbase.dev.yarncluster:16010/master-status
-```
-
-
-Note that YARN service framework assigns COMPONENT_INSTANCE_NAME for each container in a sequence of monotonically increasing integers. For example, `hbasemaster-0` gets
-assigned `0` since it is the first and only instance for the `hbasemaster` component. In case of `regionserver` component, it can have multiple containers
- and so be named as such: `regionserver-0`, `regionserver-1`, `regionserver-2` ... etc
-
-`Disclaimer`: The DNS implementation is still experimental. It should not be used as a fully-functional corporate DNS.
-
-### Start the DNS server
-By default, the DNS runs on non-privileged port `5353`.
-If it is configured to use the standard privileged port `53`, the DNS server needs to be run as root:
-```
-sudo su - -c "yarn org.apache.hadoop.registry.server.dns.RegistryDNSServer > /${HADOOP_LOG_FOLDER}/registryDNS.log 2>&1 &" root
-```
-Please refer to [YARN DNS doc](ServicesDiscovery.md) for the full list of configurations.
\ No newline at end of file
+With docker support, it also opens up a set of new possibilities to implement features such as discovering service containers on YARN with DNS.
+Check [ServiceDiscovery](ServiceDiscovery.md) for more details.
http://git-wip-us.apache.org/repos/asf/hadoop/blob/281c1d1e/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/RegistryDNS.md
----------------------------------------------------------------------
diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/RegistryDNS.md b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/RegistryDNS.md
new file mode 100644
index 0000000..ef395fc
--- /dev/null
+++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/RegistryDNS.md
@@ -0,0 +1,166 @@
+<!---
+ Licensed under the Apache License, Version 2.0 (the "License");
+ you may not use this file except in compliance with the License.
+ You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+ Unless required by applicable law or agreed to in writing, software
+ distributed under the License is distributed on an "AS IS" BASIS,
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ See the License for the specific language governing permissions and
+ limitations under the License. See accompanying LICENSE file.
+-->
+
+# Registry DNS Server
+
+<!-- MACRO{toc|fromDepth=0|toDepth=3} -->
+
+## 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:
+
+1. **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).
+2. **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
+
+1. Supports creation of DNS records for end-points of the deployed YARN applications
+2. Record names remain unchanged during restart of containers and/or applications
+3. Supports reverse lookups (name based on IP). Note, this works only for
+Docker containers because other containers share the IP of the host
+4. Supports security using the standards defined by The Domain Name System Security
+Extensions (DNSSEC)
+5. Highly available
+6. 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
+
+1. 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.
+2. 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](https://www.ietf.org/rfc/rfc1035.txt)):
+
+* **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 frameworks such as Apache Slider. 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.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](https://ftp.isc.org/isc/bind/cur/9.9/doc/arm/man.dnssec-keygen.html) 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](http://www.zytrax.com/books/dns/ch6/mydomain.html) for zone master file documentation.|
\ No newline at end of file
http://git-wip-us.apache.org/repos/asf/hadoop/blob/281c1d1e/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/ServiceDiscovery.md
----------------------------------------------------------------------
diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/ServiceDiscovery.md b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/ServiceDiscovery.md
index 6318a07..a5dd0d2 100644
--- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/ServiceDiscovery.md
+++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-site/src/site/markdown/yarn-service/ServiceDiscovery.md
@@ -12,139 +12,112 @@
limitations under the License. See accompanying LICENSE file.
-->
-# YARN DNS Server
-
-<!-- MACRO{toc|fromDepth=0|toDepth=3} -->
-
-## Introduction
-
-The YARN 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:
-
-1. **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 (for e.g. a DNS SRV Record specifying the hostname and port
-number for services).
-2. **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://myContainer.myDomain.com/endpoint’).
-
-## 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
-
-1. Supports creation of DNS records for end-points of the deployed YARN applications
-2. Record names remain unchanged during restart of containers and/or applications
-3. Supports reverse lookups (name based on IP). Note, this works only for Docker containers.
-4. Supports security using the standards defined by The Domain Name System Security
-Extensions (DNSSEC)
-5. Highly available
-6. 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
-
-1. 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.
-2. The DNS Server exposes a port that can receive both TCP and UDP requests per
-DNS standards. The default port for DNS protocols is in a restricted, administrative port
-range (5353), so the port is configurable for deployments in which the service may
-not be managed via an administrative account.
-
-## 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 yDNS zone (the zone for which the parent DNS
-server will forward requests to yDNS). 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 confirms DNS name conventions (RFC 1035 linked above), so special translation is performed for names with special characters like hyphens and spaces.
-* **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 thn 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 yDNS
-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 frameworks such as Apache Slider. The yDNS 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. As
-noted above, the default port of 5353 is in a restricted range that is only accessible to an
-account with administrative privileges.
-* 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
-YARN DNS service configured as a forwarder).
+# Service Discovery
+
+This document describes the mechanism of service discovery on YARN and the
+steps for enabling it.
+
+## Overview
+A [DNS server](RegistryDNS.md) is implemented to enable discovering services on YARN via
+the standard mechanism: DNS lookup.
+
+The framework ApplicationMaster posts the container information such as hostname and IP address into
+the YARN service registry. The DNS server exposes the information in YARN service registry by translating them into DNS
+records such as A record and SRV record. Clients can then discover the IPs of containers via standard DNS lookup.
+
+For non-docker containers (containers with null `Artifact` or with `Artifact` type set to `TARBALL`), since all containers on the same host share the same ip address,
+the DNS supports forward DNS lookup, but not support reverse DNS lookup.
+With docker, it supports both forward and reverse lookup, since each container
+can be configured to have its own unique IP. In addition, the DNS also supports configuring static zone files for both foward and reverse lookup.
+
+## Docker Container IP Management in Cluster
+To support the use-case of per container per IP, containers must be launched with `bridge` network. However, with `bridge` network, containers
+running on one node are not routable from other nodes by default. This is not an issue if you are only doing single node testing, however, for
+a multi-node environment, containers must be made routable from other nodes.
+
+There are several approaches to solve this depending on the platforms like GCE or AWS. Please refer to specific platform documentations for how to enable this.
+For on-prem cluster, one way to solve this issue is, on each node, configure the docker daemon to use a custom bridge say `br0` which is routable from all nodes.
+Also, assign an exclusive, contiguous range of IP addresses expressed in CIDR form e.g `172.21.195.240/26 (64 IPs)` to each docker
+daemon using the `fixed-cidr` option like below in the docker `daemon.json`:
+```
+"bridge": "br0"
+"fixed-cidr": "172.21.195.240/26"
+```
+Check how to [customize docker bridge network](https://docs.docker.com/engine/userguide/networking/default_network/custom-docker0/) for details.
+
+
+## Naming Convention with Registry DNS
+With the DNS support, user can simply access their services in a well-defined naming format as below:
+
+```
+${COMPONENT_INSTANCE_NAME}.${SERVICE_NAME}.${USER}.${DOMAIN}
+```
+For example, in a cluster whose domain name is `yarncluster` (as defined by the `hadoop.registry.dns.domain-name` in `yarn-site.xml`), a service named `hbase` deployed by user `devuser`
+with two components `hbasemaster` and `regionserver` can be accessed as below:
+
+This URL points to the usual hbase master UI
+```
+http://hbasemaster-0.hbase.devuser.yarncluster:16010/master-status
+```
+
+
+Note that YARN service framework assigns `COMPONENT_INSTANCE_NAME` for each container in a sequence of monotonically increasing integers. For example, `hbasemaster-0` gets
+assigned `0` since it is the first and only instance for the `hbasemaster` component. In case of `regionserver` component, it can have multiple containers
+ and so be named as such: `regionserver-0`, `regionserver-1`, `regionserver-2` ... etc
+
+`Disclaimer`: The DNS implementation is still experimental. It should not be used as a fully-functional DNS.
+
+
+## Configure Registry DNS
+
+Below is the set of configurations in `yarn-site.xml` required for enabling Registry DNS. A full list of properties can be found in the Configuration
+section of [Registry DNS](RegistryDNS.md).
+
+```
+ <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>53</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>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.</description>
+ <name>hadoop.registry.dns.zone-mask</name>
+ <value>255.255.255.0</value>
+ </property>
+
+ <property>
+ <description>An indicator of the IP range associated with the cluster containers. The setting is utilized for the
+ generation of the reverse zone name.</description>
+ <name>hadoop.registry.dns.zone-subnet</name>
+ <value>172.17.0</value>
+ </property>
+
+```
## Start the DNS Server
-By default, the DNS runs on non-privileged port `5353`.
-If it is configured to use the standard privileged port `53`, the DNS server needs to be run as root:
+By default, the DNS server runs on non-privileged port `5353`. Start the server
+with:
```
-sudo su - -c "yarn org.apache.hadoop.registry.server.dns.RegistryDNSServer > /${HADOOP_LOG_FOLDER}/registryDNS.log 2>&1 &" root
+yarn --daemon start registrydns
```
-## Configuration
-The YARN DNS server reads its configuration properties from the yarn-site.xml file. The following are the DNS associated configuration properties:
-
-| Name | Description |
-| ------------ | ------------- |
-| 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. However, since that port falls in a administrator-only range, typical deployments may need to specify an alternate port. |
-| 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](https://ftp.isc.org/isc/bind/cur/9.9/doc/arm/man.dnssec-keygen.html) 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](http://www.zytrax.com/books/dns/ch6/mydomain.html) for zone master file documentation.|
+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.
---------------------------------------------------------------------
To unsubscribe, e-mail: common-commits-unsubscribe@hadoop.apache.org
For additional commands, e-mail: common-commits-help@hadoop.apache.org