multi-dbench

Benchmark Kubernetes persistent disk volumes with fio

Usage

Build the docker image using the Dockerfile supplied and upload to your local repo. This will pick up the latest version of FIO.
Edit dbench-statefulset.yaml, updating
(a) the container/repo location, as built above (under containers/image)
(b) to keep things simple, the example uses the default storage class; either define a storageclass as required or set a default sc to match your env
(c) finally update the number of replicas required
Deploy Dbench using: kubectl apply -f dbench-statefulset.yaml
Once deployed, each POD will run a daemon of fio, thus creating a set of worker nodes.
Run a FIO client to run jobs on the worker nodes, see notes below
Once the tests are finished, clean up using: kubectl delete -f dbench-statefulset.yaml and delete the volumes k get pvc | awk '/dbench/{print $1}' | xargs -I {} kubectl delete pvc {}

You can either login to one of the PODs (recommended) or create VM to test with.
If running via a POD, skip to step 5

For the VM, follow steps 1-4 below
NOTE: pods must have routable networking if using a VM

cd fio
./configure
make
sudo make install

Define an FIO file (see example.fio)
Obtain the IP addresses of the pods, e.g.:
kubectl get pods -o wide | awk '/multi-dbench/{print $6}'
Log into one of the PODs or the FIO VM
Use VI to create two new text files, one containing the FIO parameters (xxx.fio) and the other a list of IP addresses (fio_workers)
Run FIO with the IP list and parameter file name, e.g.: fio --client=fio_workers example.fio

If the Persistent Volume Claim is stuck on Pending, check the default storageclass (or add the storageclass to the manifest)
It can take some time for a Persistent Volume to be Bound
When cloning ensure that the docker-entrypoint.sh has EXECUTE permissions