Benchmark Apache Storm cluster under failure conditions based on jfr2102/storm-cluster with help of Pumba.
- Clone this repository:
git clone git@github.com:jfr2102/Thesis.git- Run setup script in driver directory:
./setup.sh- Configure host adresses of used virtual machines in machine_config.sh as key value pairs.
- Configure ssh access between hosts (add ssh public key of driver machine to authorized_hosts on required machines)
- Configure prom-metrics-fetcher as decribed.
- Recommended: run in "background":
nohup ./buildpipeline delay 100 &- e.g: run sets of experiments by groups of 4
nohup ./run_4_times.sh batch_identifier &- ./buildpipeline delay delay_time_ms: Pumba delay on one randomly chosen supervisor container
- ./buldpipeline loss loss_percentage: Pumba package loss on one randomly chosen supervisor container
- vary data generation rate by editing the ./testdriver call (see kafka-generator)
./testdriver.sh -benchmark 1250000 10 10000 50- vary buildpipline.sh sleep times according to number of messages, event-rate or other needs