I made a simple presentation to explain the basic concepts of docker.
We had a scenario where we were using the JarLauncher of springboot but had to add additional jars to classpath.
i.e. we would launch the springboot app using the following command
java -jar fat_app.jar
Since springboot when using JarLaunhcer , ignores -classpath or -cp argument of java, our attempt to add a jar via -cp argument fails.
Solution as of now is to modify the pom.xml and change to Properties launcher i.e. change pom.xml as follows
<build> <plugins> <plugin> <groupId>org.springframework.boot</groupId> <artifactId>spring-boot-maven-plugin</artifactId> <configuration> <!-- added --> <layout>ZIP</layout> <!-- to use PropertiesLaunchar --> </configuration> </plugin> </plugins> </build>
But we DID NOT want to do the above.
We wanted to add a jar, without changing the pom.xml
i.e. continue using the JAR Launcher.
When we use the jar launcher i.e. run via the following command
java -jar fat_app.jar # The main class in JAR manifest # is set to 'org.springframework.boot.loader.JarLauncher'
The solution is to go the old java way and provide the main class yourself:
i.e. the solution is
java \ -cp fat_app.jar \ -Dloader.path=<path_to_your_additional_jars> \ org.springframework.boot.loader.PropertiesLauncher # It should work now, even though the main class in manifest # of fat_app.jar is set to '
' # and your additional classes will be picked up by springboot
#Thanks to my ex colleague @Dapeng for the idea
So recently while being oncall, i noticed that the load average was very high on my aws ec2 instance.
The first thing i did, was run TOP command: which showed me the following:
Ntop – taking 105 % cpu and 500MB !!!!
Now thats Fishy !!
I googled and found out these two links which suggested that NTOP is broken:
So i proceeded to stop ntop, but stop failed, so i just killed it !
kill -9 <ntop_pid>
wallah the load averages returned to normal 🙂
- Ntop running for a while.
- load average slowly increasing
- increase in network traffic
Recently the mysql community got an awesome monitoring solution for mysql
You can actually monitor Amazon RDS instance with the same steps mentioned in the above post but with a few changes:
The monitoring framework consists of 4 components:
- Prometheus server on port 9090
- Grafana server on port 3000
- MySQL exporter process which connects to the MySQL Server
- Node Exporter process which gets system metrics of the box hosting MySQL
- Create an RDS instance.
- create an amazon ec2-instance. (this will host all the 4 components)
Component 3 & 4: – Node Exporter & MySQL exporter process :
Amazon _DOES_NOT_ allow us to install anything on the RDS box.
So, I am sorry we will not be able get the System metrics of RDS – please rely on cloudwatch / Rds console for load averages, cpu usage , io etc etc.
So Follow the steps as mentioned in the nice post BUT make the following changes,
- Install the Node exporter & MySQL exporter processes on the ec-instance ,
- So the ‘/opt/prometheus/prometheus.yml’ file will look like
- i.e. you are now monitoring system metrics of the ec2-instance not RDS box !
cat << EOF > /opt/prometheus/prometheus.yml global: scrape_interval: 5s evaluation_interval: 5s scrape_configs: - job_name: linux target_groups: - targets: ['localhost:9100'] labels: alias: db1 - job_name: mysql target_groups: - targets: ['localhost:9104'] labels: alias: db1 EOF
But we need to tell the MySQL exporter to pull from RDS endpoint, so the
my.cnf file for MySQL exporter should be as follows:
[root@centos7 prometheus_exporters]# cat << EOF > .my.cnf [client] user=prom password=abc123 host=amazon-rds-instance.amazonaws.com EOF
Component 1 & 2: – Grafana & Prometheus:
Just Follow the steps as mentioned in the nice post .
And walllaaaah…. you should be able graph Amazon RDS metrics 🙂
a Possible solution to handle deletion of shards in a stream with apache storm.
I have been working a lot with Transactional topologies of Apache Storm these days.
In the course of my work, i have come up with questions like
- Why is the topology stuck ?
- Which part of the topology is the batch stuck at ?
So I came up with an idea based on a punch clock.
To Find out …
- When did the Person enter / exit the office ?
- Who is still in office ?
Apply Punch clock to Storm:
In the emitBatch method of Partitioned Transactional Spout:
punchCardId = "SPOUT__"+ InetAddress.getLocalHost().getHostAddress()+Thread.currentThread().getId()+"__"+System.currentTimeMillis();
PunchClock.getInstance().punchIn(punchCardId); // Punch In collector.emit(tuples); // Emit tuple(s) PunchClock.getInstance().punchOut(punchCardId); // Punch Out
Prepare method of Transactional Bolt:
punchCardId ="Bolt__"+Thread.currentThread().getId()+"__"+System.currentTimeMillis(); //Create Punch Card for txn
Execute method of Transactional Bolt:
PunchClock.getInstance().punchIn(punchCardId); // Punch In
In the finishBatch method of Transactional Bolt:
PunchClock.getInstance().punchOut(punchCardId); // Punch Out
PS: if there are no punch cards available anywhere & topology is stuck, then the problem is probably not your bolts/spout.
(1) logging while entering and exiting
(2) using http://riemann.io/ -> suggested by my friend Angad @Inmobi
Hope you like the idea & hope its useful to you.
Thank you for reading & any feedback is welcome.