Ignite Thin Client in Kubernetes - ignite

I'm trying to set up a distributed cache using Ignite and my java app through a thin client in a Kubernetes environment.
In my Kubernetes cluster, I have 2 pods with the java app and 2 pods of ignite. For the java pods to communicate with ignite pods, I have configured a thin client to connect with the ignite kubernetes service. With this configuration, I was expecting that the load balancing was on the kubernetes side. Here's what I have done in java code:
ClientConfiguration cfg = new ClientConfiguration()
.setAddresses("ignite-service.default.svc.cluster.local:10800")
.setUserName("user")
.setUserPassword("password");
IgniteClient igniteClient = Ignition.startClient(cfg);
While storing and getting objects from ignite, I deleted one of the ignite pods and, after a while, I was getting errors saying that "Ignite cluster is unavailable":
org.apache.ignite.client.ClientConnectionException: Ignite cluster is unavailable
With this behavior, I assume that the method setAddresses in ClientConfiguration class stores one of the IPs of the pods and channels all communication to that pod.
Is this what's happening in this method?
Ignite version 2.7
Kubernetes version 1.12.3

You need to pass several IP addresses to enable the failover (aka. automatic reconnect) on the thin client end. Find more details here.

Although you might have resolved the issue since the question was posted a long time back, but still putting an answer here for others.
With the Apache Ignite version(2.7+), you can modify your deployment to use Kubernetes IP Finder. With this Kubernetes will take care of discovering and connecting all server and client nodes.
TcpDiscoveryKubernetesIpFinder module will help you achieve this.
This is the discovery SPI that needs to be added to your configuration (Replace with appropriate Namespace and Service Name)
<property name="discoverySpi">
<bean class="org.apache.ignite.spi.discovery.tcp.TcpDiscoverySpi">
<property name="ipFinder">
<bean class="org.apache.ignite.spi.discovery.tcp.ipfinder.kubernetes.TcpDiscoveryKubernetesIpFinder">
<constructor-arg>
<bean class="org.apache.ignite.kubernetes.configuration.KubernetesConnectionConfiguration">
<property name="namespace" value="default" />
<property name="serviceName" value="ignite" />
</bean>
</constructor-arg>
</bean>
</property>
</bean>
</property>
Official documentation can be found here - https://ignite.apache.org/docs/latest/installation/kubernetes/amazon-eks-deployment

Related

Ignite cluster communication with multiple nics

Ignite uses one network card during a re-balance. It should use multiple.
Our cluster using more than 1gbps bandwidth during re-balance, so we tried network bonding but ARP cache needs to be refresh. Instead we want to use separate network devices on a virtual machine. But ignite uses one of them per re-balancing. Virtual machines are centos7. ignite is 2.7.0-1
<property name="discoverySpi">
<bean class="org.apache.ignite.spi.discovery.tcp.TcpDiscoverySpi">
<property name="ipFinder">
<bean class="org.apache.ignite.spi.discovery.tcp.ipfinder.vm.TcpDiscoveryVmIpFinder">
<property name="addresses">
<list>
<value>ip1:47500..47509</value>
<value>ip2:47500..47509</value>
<value>ip3:47500..47509</value>
</list>
</property>
</bean>
</property>
</bean>
</property>
We expect ignite to re-balance trough ip1, ip2, ip3 at the same time.
UPDATE
We've made a bonded virtual network device with combining multiple devices, unfortunately it has required some down time. Problem solved.
Two options here:
Can you create a virtual NIC which will join two physical ones together? I think this should be doable.
Failing that, you can have two nodes per VM, one with localAddress nic1 and other with localAddress nic2. Please note that you should define localAddress on TcpCommunicationSpi since that's where traffic is. Share RAM fairly between those two nodes.
Maybe you could also have a custom TcpCommunicationSpi which will use two NICs, but I'm not sure if traffic will be distributed fairly even then.

Apache Ignite SQLClient Connection from outside cluster

Apache Ignite is running in the 5 node hadoop cluster. Ignite Visor top command shows all the recognized nodes accurately. Outside the cluster, only one node is exposed as an edge node, using external ip. I am unable to connect to the Apache Ignite Cluster from outside the cluster using the exposed ip of the edge node.
Working within cluster : jdbc:ignite:thin://127.0.0.1/
Working within cluster : jdbc:ignite:thin://internal-ip.labs.net/
Not Working Outside cluster : jdbc:ignite:thin://external-ip.labs.net/
Please advise if any additional configuration is needed in the edge node to make the jdbc url work using the external ip address also. I am trying to do this in order to connect to the ignite cluster from outside using a sql client so that I can run all the sqls.
My Current Config
<bean id="grid.cfg" class="org.apache.ignite.configuration.IgniteConfiguration">
<property name="discoverySpi">
<bean class="org.apache.ignite.spi.discovery.tcp.TcpDiscoverySpi">
<property name="ipFinder">
<bean class="org.apache.ignite.spi.discovery.tcp.ipfinder.sharedfs.TcpDiscoverySharedFsIpFinder">
<property name="path" value="/storage/softwares/ignite/addresses"/>
</bean>
</property>
</bean>
</property>
</bean>
Apache Ignite JDBC driver operates over port 10800 by default. You need to forward it from external IP to your Ignite node to be able to connect to the cluster using JDBC.

Apache Ignite - Node running on remote machine not discovered

Apache Ignite Version is: 2.1.0
I am using TcpDiscoveryVmIpFinder to configure the nodes in an Apache Ignite cluster to setup a compute grid. Below is my configuration which is nothing but the example-default.xml, edited for the IP addresses:
<property name="discoverySpi">
<bean class="org.apache.ignite.spi.discovery.tcp.TcpDiscoverySpi">
<property name="ipFinder">
<!--
Ignite provides several options for automatic discovery that can be used
instead os static IP based discovery. For information on all options refer
to our documentation: http://apacheignite.readme.io/docs/cluster-config
-->
<!-- Uncomment static IP finder to enable static-based discovery of initial nodes. -->
<!--<bean class="org.apache.ignite.spi.discovery.tcp.ipfinder.vm.TcpDiscoveryVmIpFinder">-->
<!-- <bean class="org.apache.ignite.spi.discovery.tcp.ipfinder.multicast.TcpDiscoveryMulticastIpFinder"> -->
<bean class="org.apache.ignite.spi.discovery.tcp.ipfinder.vm.TcpDiscoveryVmIpFinder">
<property name="addresses">
<list>
<!-- In distributed environment, replace with actual host IP address. -->
<value>xxx.40.16.yyy:47500..47509</value>
<value>xx.40.16.zzz:47500..47509</value>
</list>
</property>
</bean>
</property>
</bean>
</property>
If I start multiple nodes on individual machine, the nodes on respective machines discover each other and form a cluster. But, the nodes on the remote machines do not discover each other.
Any advise will be helpful...
First of all, make sure that you really use this config file and not a default config. With default configuration, nodes can find each other only on the same machine.
Once you've checked it, you also need to test that it's possible to connect from host 106.40.16.64 to 106.40.16.121(and vice versa) via 47500..47509 ports. It's possible that there is a firewall blocked connections or these ports is simply closed.
For example, it's possible to check it with netcat, run this from 106.40.16.64 host:
nc -z 106.40.16.121 47500

SharedRDD code for ignite works on setup of single server but fails with exception when additional server added

I have 2 server nodes running collocated with spark worker. I am using shared ignite RDD to save my dataframe. My code works fine when I work with only one server node stared, if I start both server nodes code fails with
Grid is in invalid state to perform this operation. It either not started yet or has already being or have stopped [gridName=null, state=STOPPING]
DiscoverySpi is configured as below
<property name="discoverySpi">
<bean class="org.apache.ignite.spi.discovery.tcp.TcpDiscoverySpi">
<property name="ipFinder">
<!--
Ignite provides several options for automatic discovery that can be used
instead os static IP based discovery. For information on all options refer
to our documentation: http://apacheignite.readme.io/docs/cluster-config
-->
<!-- Uncomment static IP finder to enable static-based discovery of initial nodes. -->
<bean class="org.apache.ignite.spi.discovery.tcp.ipfinder.vm.TcpDiscoveryVmIpFinder">
<!--<bean class="org.apache.ignite.spi.discovery.tcp.ipfinder.multicast.TcpDiscoveryMulticastIpFinder">-->
<property name="shared" value="true"/>
<property name="addresses">
<list>
<!-- In distributed environment, replace with actual host IP address. -->
<value>v-in-spark-01:47500..47509</value>
<value>v-in-spark-02:47500..47509</value>
</list>
</property>
</bean>
</property>
</bean>
</property>
I know this exception generally means ignite instanace either not started or stopped and operation tried with same, but I don't think this is the case for reasons that with single server node it works fine and also I am not explicitly closing ignite instance in my program.
Also in my code flow I do perform operations in transaction which works, so it is like
create cache1 : works fine
Create cache2 : works fine
put value in cache1 ; works fine
igniteRDD.saveValues on cache2 : This step failes with above mentioned exception.
USE this link for complete error trace
caused by part is pasted below here also
Caused by: java.lang.IllegalStateException: Grid is in invalid state to perform this operation. It either not started yet or has already being or have stopped [gridName=null, state=STOPPING]
at org.apache.ignite.internal.GridKernalGatewayImpl.illegalState(GridKernalGatewayImpl.java:190)
at org.apache.ignite.internal.GridKernalGatewayImpl.readLock(GridKernalGatewayImpl.java:90)
at org.apache.ignite.internal.IgniteKernal.guard(IgniteKernal.java:3151)
at org.apache.ignite.internal.IgniteKernal.getOrCreateCache(IgniteKernal.java:2739)
at org.apache.ignite.spark.impl.IgniteAbstractRDD.ensureCache(IgniteAbstractRDD.scala:39)
at org.apache.ignite.spark.IgniteRDD$$anonfun$saveValues$1.apply(IgniteRDD.scala:164)
at org.apache.ignite.spark.IgniteRDD$$anonfun$saveValues$1.apply(IgniteRDD.scala:161)
at org.apache.spark.rdd.RDD$$anonfun$foreachPartition$1$$anonfun$apply$28.apply(RDD.scala:883)
at org.apache.spark.rdd.RDD$$anonfun$foreachPartition$1$$anonfun$apply$28.apply(RDD.scala:883)
at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1897)
at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1897)
at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:70)
at org.apache.spark.scheduler.Task.run(Task.scala:85)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:274)
... 3 more</pre>
It looks like the node embedded in the executor process is stopped for some reason while you are still trying to run the job. To my knowledge the only way for this to happen is to stop the executor process. Can this be the case? Is there anything in the log except the trace?

How to monitor PooledConnectionFactory (via JMX?)

I have a client app that is consuming from a queue in an activemq cluster. The app is running in tomcat 7 and uses camel (v2.10.3) and spring 3.1.2. I use a PooledConnectionFactory to connect.
Everything works for a while (sometimes days), but then all of the connections go away in the pool (the activemq broker web console shows no consumers. I figured it was the idletimeout issue, but adding the suggested config didn't help. I also upgraded to activemq-pool-5.10.0.jar, but also no luck.
SO, I'm trying to find out what is going on and was hoping to use JMX, but I can not find any related mbeans (via jconsole) that the pool registers. Is there a way to monitor/control the pool via JMX (or another/better way)?
My config fyi:
<bean id="jmsConnectionFactory" class="org.apache.activemq.ActiveMWSslConnectionFactory">
<property name="brokerURL" value="failover://ssl://...."/>
</bean>
<bean id="pooledConnectionFactory" class="org.apache.activemq.pool.PooledConnectionFactory" init-method="start" destroy-method="stop">
<property name="connectionFactory" ref="jmsConnectionFactory"/>
<property name="idleTimeout" value="0"/>
</bean>
As simple as it sounds, I don't see any other option other than to turn on TRACE level logging for that class. Check out the logs of this question.