Repository: incubator-atlas Updated Branches: refs/heads/master a90800197 -> b8f4ffb68
ATLAS-603 Document High Availability of Atlas (yhemanth via sumasai) Project: http://git-wip-us.apache.org/repos/asf/incubator-atlas/repo Commit: http://git-wip-us.apache.org/repos/asf/incubator-atlas/commit/b8f4ffb6 Tree: http://git-wip-us.apache.org/repos/asf/incubator-atlas/tree/b8f4ffb6 Diff: http://git-wip-us.apache.org/repos/asf/incubator-atlas/diff/b8f4ffb6 Branch: refs/heads/master Commit: b8f4ffb68cf6adab4015d545715d1370a1ee57b9 Parents: a908001 Author: Suma Shivaprasad <[email protected]> Authored: Fri Apr 8 10:35:49 2016 -0700 Committer: Suma Shivaprasad <[email protected]> Committed: Fri Apr 8 10:35:49 2016 -0700 ---------------------------------------------------------------------- docs/src/site/twiki/Configuration.twiki | 41 +++++++ docs/src/site/twiki/HighAvailability.twiki | 157 +++++++++++++++++++++--- release-log.txt | 1 + 3 files changed, 182 insertions(+), 17 deletions(-) ---------------------------------------------------------------------- http://git-wip-us.apache.org/repos/asf/incubator-atlas/blob/b8f4ffb6/docs/src/site/twiki/Configuration.twiki ---------------------------------------------------------------------- diff --git a/docs/src/site/twiki/Configuration.twiki b/docs/src/site/twiki/Configuration.twiki index 460f2aa..023f5a0 100644 --- a/docs/src/site/twiki/Configuration.twiki +++ b/docs/src/site/twiki/Configuration.twiki @@ -163,3 +163,44 @@ The following property is used to toggle the SSL feature. atlas.enableTLS=false </verbatim> +---++ High Availability Properties + +The following properties describe High Availability related configuration options: + +<verbatim> +# Set the following property to true, to enable High Availability. Default = false. +atlas.server.ha.enabled=true + +# Define a unique set of strings to identify each instance that should run an Atlas Web Service instance as a comma separated list. +atlas.server.ids=id1,id2 +# For each string defined above, define the host and port on which Atlas server binds to. +atlas.server.address.id1=host1.company.com:21000 +atlas.server.address.id2=host2.company.com:31000 + +# Specify Zookeeper properties needed for HA. +# Specify the list of services running Zookeeper servers as a comma separated list. +atlas.server.ha.zookeeper.connect=zk1.company.com:2181,zk2.company.com:2181,zk3.company.com:2181 +# Specify how many times should connection try to be established with a Zookeeper cluster, in case of any connection issues. +atlas.server.ha.zookeeper.num.retries=3 +# Specify how much time should the server wait before attempting connections to Zookeeper, in case of any connection issues. +atlas.server.ha.zookeeper.retry.sleeptime.ms=1000 +# Specify how long a session to Zookeeper should last without inactiviy to be deemed as unreachable. +atlas.server.ha.zookeeper.session.timeout.ms=20000 + +# Specify the scheme and the identity to be used for setting up ACLs on nodes created in Zookeeper for HA. +# The format of these options is <scheme>:<identity>. For more information refer to http://zookeeper.apache.org/doc/r3.2.2/zookeeperProgrammers.html#sc_ZooKeeperAccessControl. +# The 'acl' option allows to specify a scheme, identity pair to setup an ACL for. +atlas.server.ha.zookeeper.acl=auth:sasl:[email protected] +# The 'auth' option specifies the authentication that should be used for connecting to Zookeeper. +atlas.server.ha.zookeeper.auth=sasl:[email protected] + +# Since Zookeeper is a shared service that is typically used by many components, +# it is preferable for each component to set its znodes under a namespace. +# Specify the namespace under which the znodes should be written. Default = /apache_atlas +atlas.server.ha.zookeeper.zkroot=/apache_atlas + +# Specify number of times a client should retry with an instance before selecting another active instance, or failing an operation. +atlas.client.ha.retries=4 +# Specify interval between retries for a client. +atlas.client.ha.sleep.interval.ms=5000 +</verbatim> http://git-wip-us.apache.org/repos/asf/incubator-atlas/blob/b8f4ffb6/docs/src/site/twiki/HighAvailability.twiki ---------------------------------------------------------------------- diff --git a/docs/src/site/twiki/HighAvailability.twiki b/docs/src/site/twiki/HighAvailability.twiki index 2a87067..1e52c85 100644 --- a/docs/src/site/twiki/HighAvailability.twiki +++ b/docs/src/site/twiki/HighAvailability.twiki @@ -3,9 +3,9 @@ ---++ Introduction Apache Atlas uses and interacts with a variety of systems to provide metadata management and data lineage to data -administrators. By choosing and configuring these dependencies appropriately, it is possible to achieve a good degree +administrators. By choosing and configuring these dependencies appropriately, it is possible to achieve a high degree of service availability with Atlas. This document describes the state of high availability support in Atlas, -including its capabilities and current limitations, and also the configuration required for achieving a this level of +including its capabilities and current limitations, and also the configuration required for achieving this level of high availability. [[Architecture][The architecture page]] in the wiki gives an overview of the various components that make up Atlas. @@ -14,22 +14,146 @@ review before proceeding to read this page. ---++ Atlas Web Service -Currently, the Atlas Web service has a limitation that it can only have one active instance at a time. Therefore, in -case of errors to the host running the service, a new Atlas web service instance should be brought up and pointed to -from the clients. In future versions of the system, we plan to provide full High Availability of the service, thereby -enabling hot failover. To minimize service loss, we recommend the following: +Currently, the Atlas Web Service has a limitation that it can only have one active instance at a time. In earlier +releases of Atlas, a backup instance could be provisioned and kept available. However, a manual failover was +required to make this backup instance active. - * An extra physical host with the Atlas system software and configuration is available to be brought up on demand. - * It would be convenient to have the web service fronted by a proxy solution like [[https://cbonte.github.io/haproxy-dconv/configuration-1.5.html#5.2][HAProxy]] which can be used to provide both the monitoring and transparent switching of the backend instance clients talk to. - * An example HAProxy configuration of this form will allow a transparent failover to a backup server: +From this release, Atlas will support multiple instances of the Atlas Web service in an active/passive configuration +with automated failover. This means that users can deploy and start multiple instances of the Atlas Web Service on +different physical hosts at the same time. One of these instances will be automatically selected as an 'active' +instance to service user requests. The others will automatically be deemed 'passive'. If the 'active' instance +becomes unavailable either because it is deliberately stopped, or due to unexpected failures, one of the other +instances will automatically be elected as an 'active' instance and start to service user requests. + +An 'active' instance is the only instance that can respond to user requests correctly. It can create, delete, modify +or respond to queries on metadata objects. A 'passive' instance will accept user requests, but will redirect them +using HTTP redirect to the currently known 'active' instance. Specifically, a passive instance will not itself +respond to any queries on metadata objects. However, all instances (both active and passive), will respond to admin +requests that return information about that instance. + +When configured in a High Availability mode, users can get the following operational benefits: + + * *Uninterrupted service during maintenance intervals*: If an active instance of the Atlas Web Service needs to be brought down for maintenance, another instance would automatically become active and can service requests. + * *Uninterrupted service in event of unexpected failures*: If an active instance of the Atlas Web Service fails due to software or hardware errors, another instance would automatically become active and can service requests. + +In the following sub-sections, we describe the steps required to setup High Availability for the Atlas Web Service. +We also describe how the deployment and client can be designed to take advantage of this capability. +Finally, we describe a few details of the underlying implementation. + +---+++ Setting up the High Availability feature in Atlas + +The following pre-requisites must be met for setting up the High Availability feature. + + * Ensure that you install Apache Zookeeper on a cluster of machines (a minimum of 3 servers is recommended for production). + * Select 2 or more physical machines to run the Atlas Web Service instances on. These machines define what we refer to as a 'server ensemble' for Atlas. + +To setup High Availability in Atlas, a few configuration options must be defined in the =atlas-application.properties= +file. While the complete list of configuration items are defined in the [[Configuration][Configuration Page]], this +section lists a few of the main options. + + * High Availability is an optional feature in Atlas. Hence, it must be enabled by setting the configuration option =atlas.server.ha.enabled= to true. + * Next, define a list of identifiers, one for each physical machine you have selected for the Atlas Web Service instance. These identifiers can be simple strings like =id1=, =id2= etc. They should be unique and should not contain a comma. + * Define a comma separated list of these identifiers as the value of the option =atlas.server.ids=. + * For each physical machine, list the IP Address/hostname and port as the value of the configuration =atlas.server.address.id=, where =id= refers to the identifier string for this physical machine. + * For e.g., if you have selected 2 machines with hostnames =host1.company.com= and =host2.company.com=, you can define the configuration options as below: + <verbatim> + atlas.server.ids=id1,id2 + atlas.server.address.id1=host1.company.com:21000 + atlas.server.address.id2=host2.company.com:21000 + </verbatim> + * Define the Zookeeper quorum which will be used by the Atlas High Availability feature. <verbatim> - listen atlas - bind <proxy hostname>:<proxy port> - balance roundrobin - server inst1 <atlas server hostname>:<port> check - server inst2 <atlas backup server hostname>:<port> check backup + atlas.server.ha.zookeeper.connect=zk1.company.com:2181,zk2.company.com:2181,zk3.company.com:2181 </verbatim> - * The stores that hold Atlas data can be configured to be highly available as described below. + * You can review other configuration options that are defined for the High Availability feature, and set them up as desired in the =atlas-application.properties= file. + * For production environments, the components that Atlas depends on must also be set up in High Availability mode. This is described in detail in the following sections. Follow those instructions to setup and configure them. + * Install the Atlas software on the selected physical machines. + * Copy the =atlas-application.properties= file created using the steps above to the configuration directory of all the machines. + * Start the dependent components. + * Start each instance of the Atlas Web Service. + +To verify that High Availability is working, run the following script on each of the instances where Atlas Web Service +is installed. +<verbatim> +$ATLAS_HOME/bin/atlas_admin.py -status +</verbatim> +This script can print one of the values below as response: + + * *ACTIVE*: This instance is active and can respond to user requests. + * *PASSIVE*: This instance is PASSIVE. It will redirect any user requests it receives to the current active instance. + * *BECOMING_ACTIVE*: This would be printed if the server is transitioning to become an ACTIVE instance. The server cannot service any metadata user requests in this state. + * *BECOMING_PASSIVE*: This would be printed if the server is transitioning to become a PASSIVE instance. The server cannot service any metadata user requests in this state. + +Under normal operating circumstances, only one of these instances should print the value *ACTIVE* as response to +the script, and the others would print *PASSIVE*. + +---+++ Configuring clients to use the High Availability feature + +The Atlas Web Service can be accessed in two ways: + + * *Using the Atlas Web UI*: This is a browser based client that can be used to query the metadata stored in Atlas. + * *Using the Atlas REST API*: As Atlas exposes a RESTful API, one can use any standard REST client including libraries in other applications. In fact, Atlas ships with a client called !AtlasClient that can be used as an example to build REST client access. + +In order to take advantage of the High Availability feature in the clients, there are two options possible. + +---++++ Using an intermediate proxy + +The simplest solution to enable highly available access to Atlas is to install and configure some intermediate proxy +that has a capability to transparently switch services based on status. One such proxy solution is [[http://www.haproxy.org/][HAProxy]]. + +Here is an example HAProxy configuration that can be used. Note this is provided for illustration only, and not as a +recommended production configuration. For that, please refer to the HAProxy documentation for appropriate instructions. + +<verbatim> +frontend atlas_fe + bind *:41000 + default_backend atlas_be + +backend atlas_be + mode http + option httpchk get /api/atlas/admin/status + http-check expect string ACTIVE + balance roundrobin + server host1_21000 host1:21000 check + server host2_21000 host2:21000 check backup + +listen atlas + bind localhost:42000 +</verbatim> + +The above configuration binds HAProxy to listen on port 41000 for incoming client connections. It then routes +the connections to either of the hosts host1 or host2 depending on a HTTP status check. The status check is +done using a HTTP GET on the REST URL =/api/atlas/admin/status=, and is deemed successful only if the HTTP response +contains the string ACTIVE. + +---++++ Using automatic detection of active instance + +If one does not want to setup and manage a separate proxy, then the other option to use the High Availability +feature is to build a client application that is capable of detecting status and retrying operations. In such a +setting, the client application can be launched with the URLs of all Atlas Web Service instances that form the +ensemble. The client should then call the REST URL =/api/atlas/admin/status= on each of these to determine which is +the active instance. The response from the Active instance would be of the form ={Status:ACTIVE}=. Also, when the +client faces any exceptions in the course of an operation, it should again determine which of the remaining URLs +is active and retry the operation. + +The !AtlasClient class that ships with Atlas can be used as an example client library that implements the logic +for working with an ensemble and selecting the right Active server instance. + +Utilities in Atlas, like =quick_start.py= and =import-hive.sh= can be configured to run with multiple server +URLs. When launched in this mode, the !AtlasClient automatically selects and works with the current active instance. +If a proxy is set up in between, then its address can be used when running quick_start.py or import-hive.sh. + +---+++ Implementation Details of Atlas High Availability + +The Atlas High Availability work is tracked under the master JIRA [[https://issues.apache.org/jira/browse/ATLAS-510][ATLAS-510]]. +The JIRAs filed under it have detailed information about how the High Availability feature has been implemented. +At a high level the following points can be called out: + + * The automatic selection of an Active instance, as well as automatic failover to a new Active instance happen through a leader election algorithm. + * For leader election, we use the [[http://curator.apache.org/curator-recipes/leader-latch.html][Leader Latch Recipe]] of [[http://curator.apache.org][Apache Curator]]. + * The Active instance is the only one which initializes, modifies or reads state in the backend stores to keep them consistent. + * Also, when an instance is elected as Active, it refreshes any cached information from the backend stores to get up to date. + * A servlet filter ensures that only the active instance services user requests. If a passive instance receives these requests, it automatically redirects them to the current active instance. ---++ Metadata Store @@ -84,5 +208,4 @@ to configure Atlas to use Kafka in HA mode, do the following: ---++ Known Issues - * [[https://issues.apache.org/jira/browse/ATLAS-338][ATLAS-338]]: ATLAS-338: Metadata events generated from a Hive CLI (as opposed to Beeline or any client going !HiveServer2) would be lost if Atlas server is down. - * If the HBase region servers hosting the Atlas âtitanâ HTable are down, Atlas would not be able to store or retrieve metadata from HBase until they are brought back online. + * If the HBase region servers hosting the Atlas âtitanâ HTable are down, Atlas would not be able to store or retrieve metadata from HBase until they are brought back online. \ No newline at end of file http://git-wip-us.apache.org/repos/asf/incubator-atlas/blob/b8f4ffb6/release-log.txt ---------------------------------------------------------------------- diff --git a/release-log.txt b/release-log.txt index aff850f..a2fd4ea 100644 --- a/release-log.txt +++ b/release-log.txt @@ -13,6 +13,7 @@ ATLAS-409 Atlas will not import avro tables with schema read from a file (dosset ATLAS-379 Create sqoop and falcon metadata addons (venkatnrangan,bvellanki,sowmyaramesh via shwethags) ALL CHANGES: +ATLAS-603 Document High Availability of Atlas (yhemanth via sumasai) ATLAS-498 Support Embedded HBase (tbeerbower via sumasai) ATLAS-527 Support lineage for load table, import, export (sumasai via shwethags) ATLAS-572 Handle secure instance of Zookeeper for leader election.(yhemanth via sumasai)
