An agent is the workhorse of the Tensei (転成) system. It uses an actor system to do the actual work of reading, parsing, transforming and writing the data.
It communicates with the tensei-server which is responsible for starting and stopping and agent.
The main website for Tensei-Data is located at: https://www.wegtam.com/products/tensei-data
The Tensei-Data system is build upon three components:
- Tensei-Server
- Tensei-Frontend
- At least one Tensei-Agent
To be able to run Tensei-Data you have to start at least one of each components.
For development purposes it is feasible to simply start each one from the sbt prompt via the run task.
To be able to provision the system components a packaging configuration for the sbt native packager plugin is included. The recommended way is to create debian packages via the debian:packageBin sbt task. Resulting debian packages can be installed on a debian or ubuntu system. Before the package is build the test suite will be executed.
% sbt clean debian:packageBin
We recommend to use the gdebi tool on ubuntu because it will automatically fetch required dependencies.
The packages include system startup scripts that will launch them upon system boot.
There are tests (sbt test) and integration tests (sbt it:test) that can
be executed by the appropriate sbt tasks. The integration tests are tagged
according to their requirements. Please see com.wegtam.scalatest.tags
package.
To run only tests with a specific test use the -n flag:
> it:testOnly -- -n com.wegtam.scalatest.tags.DbTestH2
To run all tests except the ones with specific tags use the -l flag:
> it:testOnly -- -l com.wegtam.scalatest.tags.DbTestFirebird
We use a JAR file that is provided by our own repository. To be able to use
the hyperic sigar extension properly the system property -Djava.library.path
must be set to the folder were the system library of sigar is installed.
This can simply be done by settings the SBT_OPTS environment variable:
% SBT_OPTS="-Djava.library.path=/usr/local/share/java/classes" sbt run
The command above should work on a FreeBSD system that has the sigar port installed.
To profile the agent several JVM options should be included. These can
be specified in the SBT_OPTS environment variable:
% SBT_OPTS="-XX:+PrintGCApplicationStoppedTime" sbt run
The sub module benchmarks includes benchmarks that can be started
using the sbt console.
To compile and run the JMH benchmarks just issue the following commands:
> benchmarks/jmh:clean
> benchmarks/jmh:compile
> benchmarks/jmh:run -i 10 -wi 4 -f3 -t1
To compile and run the Jamm memory benchmarks issue the following commands:
> benchmarks/clean
> benchmarks/compile
> benchmarks/run
Choose a benchmark to run and remember that you might need to increase
the memory settings for the JVM (see javaOptions in run in build.sbt).
The best way to use jHiccup is
probably to use the javaagent parameter:
% SBT_OPTS="-javaagent:/path/to/jHiccup.jar" sbt run
The usage of jHiccup should always be combined with GC analysis.
% SBT_OPTS="-javaagent:/path/to/jHiccup.jar -XX:+PrintGCApplicationStoppedTime" sbt run
For memory (heap) analysis you should specify memory settings and instruct the jvm to write a heap dump if an out of memory error occurs.
% SBT_OPTS="-server -Xms1g -Xmx1g -XX:MaxMetaspaceSize=1g -XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=/tmp/jvm-dumps" sbt run
Alternatively you can create heap dumps via the jmap utility. Replace PID
with the process id of the JVM process.
% jmap -dump:format=b,file=FILENAME.hprof PID
The resulting hprof files can be analysed using the Eclipse Memory Analyzer. Beware that eclipse will need huge ammounts of memory!