This topic provides you with instructions on how to set up an active deployment of WSO2 API Manager with multiple datacenters.
Before you begin...
Make sure that you have replicated databases and file systems. For instructions, see Changing the Default API-M Databases.
The following diagram shows the deployment architecture of WSO2 API Manager with multiple datacenters.
A global load balancer is in place to handle API traffic and in this deployment the proposal is to do traffic partitioning based on geography or IP ranges. Based on the LB rules the traffic will simultaneously flow through to both active datacenters, through their datacenter local load balancers. At an event where one datacenter fails, the traffic is routed to the second datacenter where gateway will have data center local high availability.
API creation, throttling policy creation as such management activities are routed towards the designated Active-Master datacenter. Management traffic will only have datacenter local high availability.
Throttling data will be published to the traffic managers of both datacenter. Each datacenter will have a local traffic manager for throttling decision making, however for higher accuracy the gateways will publish events to both traffic managers, one sitting locally and one sitting in the other data center. Throttle out event notification will not occur at once in both datacenter as there is no shared traffic manager topology in place (for efficiency reasons), however the deployment will eventually be consistent as the throttle data is cross published.
Raw data accumulation will only happen to each datacenter and will not be replicated. The summarization data (STATS_DB) in each datacenter will be accepted bi-directionally.
The exception is where API-M alerting usecase will not work in such a deployment is due to file-based indexing storage.
Configure the datacenters
This section explains how to configure the datacenters with separate databases.
The following diagram shows the deployment
Step 1 - Configure PostgreSQL Databases
In this setup, we use shared Event Store DB, Processed Data Store DB, and Stats DB for two Analytics nodes in one data center. The AM_DB, UM_DB, and REG_DB have also been shared between the API-M node and the two Analytics nodes in the data center.
Step 2 - Configure APIM-Analytics 2.1.0 clustered setup
- Configure the two APIM-Analytics nodes clustered setup. Instead of DAS 3.1.0 use API-M Analytics 2.1.0.
When configuring databases, use the PostgreSQL databases configured in the previous step.
If the two API-M Analytics nodes run in the same virtual machine, it is mandatory to have a port offset. Use port offset 1 and 2 for the two Analytics servers.
Step 3 - Configure APIM 2.1.0 with APIM-Analytics 2.1.0 clustered setup
- Configure APIM 2.1.0 and the two APIM-Analytics 2.1.0 nodes. For instructions on how to configure these nodes, see Configuring APIM Analytics.
- When configuring databases, use the same set of databases used in Step 2.
- Open the
<API-M_HOME>/repository/conf/api-manager.xmlfile, after enabling the Analytics.
Add both the Analytics server URLs under the
DASServerURLsection as a comma separated list as shown below.
Apply the solution to add the data center ID
Before you begin...
Make sure that you have configured the databases according to the instructions in the previous section
To ensure that no primary key violation takes place, you have to change the database schema, by adding the data center ID as an extra column for the tables in STATS_DB, and also add it to the primary key combination. This is to make sure that when database syncing happens, both analytics clusters are able to write to their respective databases without conflicts. There is a custom spark User Defined Function (UDF) to read the data center name from a system property and that has been used whenever inserting data to the STATS_DB via the Spark script.
Follow the steps below to apply the changes for each of the datacenter.
- Shut down the APIM 2.1.0 server and the API-M Analytics 2.1.0 servers in the clustered setup.
Add the following parameter to the
<Analytics_Home>/repository/conf/analytics/spark/spark-defaults.conffile, in each Analytics server node.
- Copy and replace the
<Analytics_Home>/repository/conf/template-manager/domain-template/directory in each Analytics server node.
org.wso2.analytics.apim.spark_2.1.0.jaras a patch to each of the APIM Analytics server nodes. This file contains the newly written UDF to get data center ID as system parameter.
- Copy and replace the
<Analytics_Home>/repository/deployment/server/carbonapps/org_wso2_carbon_analytics_apim-1.0.0.carfile with this CApp, for each Analytics server nodes.
Run the following PostgreSQL script against the WSO2AM_STATS_DB.Expand to see the script...
- Restart the APIM 2.1.0 server and the API-M Analytics 2.1.0 servers in the clustered setup.
Synchronize the databases
Why do we need to maintain a same data in the STAT_DB of both data centers?
In the active-active data center architecture, the request may come to one of the datacenters and be fulfilled by that datacenter. The analytics-related details of that request will be stored in the STATS_DB of the same data center. Therefore, when requesting for analytics-related details, both datacenters can provide different details according to their STATS_DBs. To avoid this, we need to maintain same set of data in the STATS_DBs of both the data centers.
You can synchronize databases by sharing the STATS_DB or by using a replication mechanism. Inserting the data center ID to the primary key into all the tables in the STATS_DB and include it in the composite key can be done in two methods.
- Using a bi-directional replication mechanism - This is a master-master node replication, where changes done in one node will be replicated in other nodes.
- Master-slave mechanism - The STATS_DB will be shared among all the nodes. When the master node becomes unavailable, the slave nodes will function as the master node.
Follow the steps below to synchronize the databases using the bi-directional replication(BDR) mechanism.
Note that these instructions are tested with Ubuntu OS and PostgreSQL
Before you begin...
Install and enable the PostgreSQL apt repository for PGDG. This repository is required by the BDR packages.
Create a file with the name pgdg.list in /etc/apt/sources.list.d/ and add the below line
Replace the codename according to your OS. E.g., Ubuntu 14.04 (trusty), 16.04 (xenial), 17.04 (zesty)
Create a 2ndquadrant.list file in the
/etc/apt/sources.list.d/with the repository URL given below. Change codename according to your OS version
Import the repository key from here. Update the package lists and install the packages.
postgresql-9.4packages, if you have them installed already.
BDR requires a patched version of PostgreSQL 9.4 that conflicts with the official packages. If you already have PostgreSQL 9.4 installed either from apt.postgresql.org or your official distribution repository, you will need to make a dump of all your databases, then uninstall the official PostgreSQL 9.4 packages before you install the BDR.
Install the BDR packages. Sample commands are given below.
Make the following changes to the files in the /etc/postgresql/9.4/main/ directory in both nodes.
Restart PostgreSQL in both nodes. Sample commands are given below.
Create the STATS_DB database and users.
Create BDR extension on the STATS_DB in both nodes. Sample commands are given below.
You can check the BDR extension as follows:
Create the first master node.
Do this step only in Node 1.
You can verify this as shown below.
Create the second master node.
Do this step only in Node 2.
You can verify this with the same command given in the previous step.
Restore database data.
Do this step in only one of the two nodes.
You have now successfully set up an active multi data center deployment.