Performance harness

Performance harness

What is the harness?

It's a tool that allows running test scenarios against an rchain network.

So, what is the harness?

A grpc client extracted from the rchain project and bridged to gatling (gatling.io).

Where can it be found?

https://github.com/lukasz-golebiewski-org/rchain-perf-harness in the folder templater

It should become part of rchain soon.

How can I use it?

Prerequisite: an rchain network running (e.g. one standalone node).

Three approaches to running the harness:

  1. run it as a test

  2. run it as a standalone application

  3. bypass gatling and run it directly

Ad. 1)

coop.rchain.perf.DeployProposeSimulation

This simulation is configured in templater/runner/test/reources/appliaction.conf

rnodes=["localhost:40401", "127.0.0.1:40401"]

sbt runner/gatling:test

Ad. 2)

coop.rchain.perf.ContinuousRunner

This is a typical Main style app. behaves similar to the above simulation but can be run instead of tested.

The params are as follows:

#!/bin/bash set -axe java -Dhosts="$1" -Dpath=$2 -Dsessions=$3 -Dratio=$4 -Dloops=$5 -jar ./runner.jar

 

Ad. 3)

coop.rchain.perf.Runner

object Propose { def propose(session: Session)( client: ClientWithDetails): (Session, DeployServiceResponse) = { (session.remove("client"), client.client.createBlock(Empty())) } } object Deploy { def deploy(session: Session)( client: ClientWithDetails): (Session, DeployServiceResponse) = { val (_, contract): (String, String) = session("contract") .as[(String, String)] val d = DeployData() .withTimestamp(System.currentTimeMillis()) .withTerm(contract) .withFrom("0x1") .withPhloLimit(0) .withPhloPrice(0) .withNonce(0) (session.set("client", client), client.client.doDeploy(d)) } }

You can use these in any way.

 

 

But I don't have a network!

There  a myriad ways of setting up a network:

 

 

Problems

Dupe Contract

We encountered an issue with the contract:

contract @"dupe"(@depth) = { if (depth <= 0) { Nil } else { @"dupe"!(depth-1) | @"dupe"!(depth-1) | @"dupe"!(depth-1) | @"dupe"!(depth-1) | @"dupe"!(depth-1) | @"dupe"!(depth-1) | @"dupe"!(depth-1) | @"dupe"!(depth-1) | @"dupe"!(depth-1) | @"dupe"!(depth-1) } } | @"dupe"!(5)

The scenario was as follows:

10 deploys, 1 propose, 10 threads running "at once", 100 loops. A network of 5 rnodes.

The problem observed was Kamon Http client throwing exceptions. This was most likely due to node running out of memory and becoming unstable.

A good starting point would be to reproduce, make sure it's the heap running out and getting heap dumps.