Mesos Simulator Usage

mesos-simulator-usage.md

Usage

1. Compile this Mesos fork like you normally would.

Notice that we are currently on the custom-exec branch.

2. Set the environment variable MESOS_DEFAULT_EXECUTOR_PATH to $BUILD_DIR/src/mesos-simexecutor

BUILD_DIR is your build directory.

3. Set the environment variable MESOS_SIMULATOR_CONFIG_PATH to point to a config file. The config file must conform to the format described below.

4. Before you launch mesos-local, it's helpful to run the following commands, especially if you want to run a lot of local slaves:

export MESOS_REGISTRY=in_memory
export MESOS_MAX_SLAVE_PING_TIMEOUTS=30000
export MESOS_SLAVE_PING_TIMEOUT=15mins

# Only if you are running mesos-local on a Mac
sudo launchctl limit maxfiles 1000000 1000000
sudo launchctl limit maxproc 2500
ulimit -n 40960

5. Start the newly built mesos-local (under $BUILD_DIR/bin). Use the --num_slaves flag if you want to spawn multiple slaves. Then just start your framework and launch tasks as usual.

Config Format

The config needs to be a valid JSON file. Here is an example:

{
    "task_specs": [
        {
            "pattern": "SOME_PREFIX.*",
            "failure_rate": 0.3,
            "length_mean": 2.9827,
            "length_std": 1.231
        },
        {
            "pattern": ".*",
            "failure_rate": 0.1,
            "length_mean": 10.23052,
            "length_std": 5.25329
        }
    ]
}

pattern is a regex pattern used to match task IDs.

length_mean and length_std are in seconds.

To explain, the first spec says that: all tasks whose IDs start with SOME_PREFIX have a 30% chance of failing. On average, they should run 2.9827 seconds. The standard deviation for their run time is 1.231 seconds. Note that the task lengths follow a normal distribution.

Use .* to match all tasks, as in the second spec.