Skip to content

Latest commit

 

History

History
465 lines (284 loc) · 11.2 KB

deployment.md

File metadata and controls

465 lines (284 loc) · 11.2 KB
Raw

A complete MZBench installation consists of two parts:

  • MZBench API server that accepts client requests and provides the dashboard.
  • A set of hosts operating as a worker nodes.

Once launched, the API server is permanent, whereas the worker nodes are dynamically allocated on demand on AWS or another cloud provider.

Here's how you install and configure MZBench for real-life use.

Installation

From RPM and Pip

Available for CentOS 7 and Amazon Linux.

Requirement:

  • Pip Python package manager

Download MZBench RPM from Github releases page

# Install RPM
sudo yum install -y <rpm_file_downloaded_from_github_releases>

# Install Python package
sudo pip install mzbench_api_client

# Start the server
mzbench start_server

MZBench server from RPM has all Github workers bundled. If you need to add a worker to this installation, please use mzbench add_worker <tgz_file> command. Please refer to "How to write your own worker" guide to learn more about generation of this tarball.

From Docker container

Docker is a container platform, more information is available at its website. If you have Docker up and running, use the following command to start MZBench server:

docker run -d -p 4800:80 --name mzbench_server docker.io/ridrisov/mzbench

After that, open http://localhost:4800/ to see the dashboard. Sources for this docker image are available on github.

From sources

Requirements:

  • Erlang R17+
  • C++ compiler (preinstalled on most UNIX-based systems)
  • Python 2.6 or 2.7
  • Pip Python package manager

Install and start the MZBench API server:

# Clone the current MZBench source code
git clone https://github.com/gojekfarm/mzbench.git

# Install Python requirements
sudo pip install -r mzbench/requirements.txt

# Start the server
./mzbench/bin/mzbench start_server

Configuration

!!!note Every time you update the configuration, restart the server.

Format

The MZBench server parameters are defined in the configuration file.

By default, the server tries to load configuration from ~/.config/mzbench/server.config and /etc/mzbench/server.config.

To specify a configuration file from a different location, use the --config param:

$ ./bin/mzbench start_server --config /path/to/server.config

The configuration file is an Erlang list with the mzbench_api tuple. This tuple holds the list of the server parameters:

[
    {mzbench_api, [
        {network_interface, "127.0.0.1"},
        {listen_port, 80}
    ]}
].

Here, two parameters are specified: network_interface and listen_port.

All parameters are optional.

Parameters

cloud_plugins

{cloud_plugins: [
    {<PluginName>, #{module => <ModuleName>,
                    <Option1> => <Value1>,
                    <Option2> => <Value2>,
                    ...
                    }
    },
    ...
    ]
}

List of cloud plugins that can be used to allocate nodes. One plugin can be listed multiple times with different settings and under different names.

<PluginName> is an atom identifying a plugin instance.

<ModuleName> is the name of the plugin module. Each module has its specific <Options>.

There are four built-in plugins:

mzb_api_ec2_plugin : Allocate hosts from the Amazon EC2 cloud.

mzb_staticcloud_plugin : Allocates hosts from a static pool.

mzb_dummycloud_plugin : Dummy provider, treats localhost as unlimited number of hosts, useful for debug. It can be also used as a reference during cloud plugin development.

mzb_multicloud_plugin : Allocate hosts from multiple sources with the given ratio.

network_interface

{network_interface, "<ip address>"}

Specify the IP address for the dashboard to run on. By default it's "127.0.0.1", so the dashboard is unavailable for external connections.

To open the dashboard to the public, set this param to "0.0.0.0".

!!!warning By default MZBench provides no authentication. Opening the dashboard to the public makes your server vulnerable.

To protect your server, please, see [authentication](deployment.md#authentication) and [protocol](deployment.md#protocol).

listen_port

{listen_port, <port>}

Specify the port to access the dashboard.

Default value: 4800.

protocol

By default protocol is set to http, but https is also available. MZBench generates self-signed certificates on first start. If you need to replace them, please use the following configuration parameters:

{protocol, https},
{cacertfile, none},
{certfile, "~/.local/share/mzbench_api/server.crt"},
{keyfile, "~/.local/share/mzbench_api/server.key"},

CA certificate is not required unless you use custom CA.

authentication

API server supports Google and GitHub auth.

  • To create Google credentials open Google API manager page. Click Credentials -> Create credentials -> OAuth Client ID -> Web Application, then specify your server URL. Copy client_id and client_secret to a following structure.
{user_authentication,
         [
          {"google", [{caption, "Google"},
                      {client_id, "..."},
                      {client_secret, "..."},
                      {redirect_url, "http://localhost:4800"}]}
         ]
     }

http://localhost:4800 should be replaced with your server's address.

  • To create GitHub credentials open GitHub developer application page. Click Register New Application. Put your server url to "Homepage URL" and "Authorization callback URL" and click "Register application".
{user_authentication,
         [
          {"github", [{caption, "GitHub"},
                      {client_id, "..."},
                      {client_secret, "..."}]}
         ]
     }

If GitHub Enterprise is used it may be usefull to add the following two parameters:

    {url, "https://<GitHub URL>"},
    {api_url, "https://<GitHub API URL>"},

After successful setup you will be able to authorize yourself at dashboard and create tokens for Command Line Utilities (CLI). To create one hover your name at top-right corner of the dashboard and click "Generate token" link. When token is generated it should be saved in ~/.config/mzbench/token file. If multiple mzbench servers are used token file could be organized the following way:

<Server1> <Token1>
<Server2> <Token2>
<Server3> <Token3>
...

access_lists

It is possible to specify white and black user lists in a following way:

    {admin_list, ["*@mydomain-admins.com", "root@mydomain.com"]},
    {black_list, ["hacker*", "single-hacker@mydomain.com"]},
    {white_list, ["tester*@mydomain.com", "single-tester@mydomain.com"]},

Administrators have privileges to stop any benchmark. Blacklisted users have no access. If white list is used, all users, except this list have no access.

bench_log_file

{bench_log_file, "<filename>"}

The name of the benchmark log files. The files are stored in <bench_data_dir>/<bench_id> where <bench_data_dir> is defined in the bench_data_dir param and <id> is the ID of a particular benchmark.

Default value: "log.txt"

bench_log_compression

{bench_log_compression, (deflate|none)}

Enable or disable log compression with the DEFLATE algorithm.

Default value: deflate

bench_metrics_file

{bench_metrics_file, "<filename>"}

The name of the benchmark metrics data files. The files are stored in <bench_data_dir>/<bench_id> where <bench_data_dir> is defined in the bench_data_dir param and <id> is the ID of a particular benchmark.

Default value: "metrics.txt"

bench_metrics_compression

{bench_metrics_compression, (none|deflate)}

Enable or disable metrics data compression with the DEFLATE algorithm.

Default value: none

node_git

{node_git, "<url>"}

The MZBench git repository used to deploy worker nodes.

By default, the MZBench source code is taken from https://github.com/gojekfarm/mzbench.git.

node_commit

{node_commit, "<string>"}

The git commit SHA or branch name used to deploy worker nodes.

By default, the latest revision is used.

node_rsync

{node_rsync, "<folder>"}

Use local folder when deploying worker nodes. This option has a precedency over node_git.

node_deployment_path

{node_deployment_path, "<path>"}

The path to the MZBench installation on node machines.

Default value: "/.local/share"

worker_deployment_path

{worker_deployment_path, "<path>"}

The the workers installation path on the node machines.

Default value: "~/.local/share/mzbench_workers"

load_workers_subdirs

{load_workers_subdirs, (true|false)}

During worker node startup, it could load all the modules from all subdirs (namely, $$worker_deployment_path$$//ebin/).

Default value: false

plugins_dir

{plugins_dir, "<path>"}

Directory with additional cloud plugins.

Default value: "../../../../plugins"

bench_data_dir

{bench_data_dir, "<path>"}

The location to store the data generated during the benchmark.

Default value: "~/.local/share/mzbench_api/data".

tgz_packages_dir

{tgz_packages_dir, "<path>"}

The location to store prebuilt worker archives.

Default value: "~/.local/cache/mzbench_api/packages".

max_bench_num

{max_bench_num, <integer>}

Maximal number of benchmarks running at the same time.

Default value: 1000.

vm_args

{vm_args, <args>}

Additional arguments for the [Erlang VM](Additional arguments for the Erlang VM.

Default value: [].

ntp_max_timediff_s

{ntp_max_timediff_s, <float>}

Maximum timeout between node creation in seconds.

This check is optional and only prints a warning if fails.

Default value: 0.1.

Dev Parameters

Set these params only if you are an MZBench developer.

bench_read_at_once

{bench_read_at_once, <integer>}

The number of bytes to read from the logs and metrics feed per request.

Default value: 1024

bench_poll_timeout

{bench_poll_timeout, <integer>}

The timeout between requests to logs and metrics feeds in milliseconds.

Default value: 1000

node_log_port

{node_log_port, <integer>}

The TCP port for the logs feed.

Default value: 4801.

node_management_port

{node_management_port, <integer>}

The TCP port used to control the server internally.

Default value: 4802.