hammer

Overview ¶

Hammer is a utility for load-testing HTTP servers. In command-line form you can think of it as an alternative to apachebench or siege, sending requests at a controlled rate to a given URL or list of URLs and reporting on throughput, latency, and the status codes of the responses. It is also structured as a library that can be used to perform more advanced tasks, such as using the response to one request to create a later request, and gathering statistics for each type of request separately.

Example ¶

hammer --threads 10 -r 20 -t 60 http://www.example.com/

Will send 20 requests per second to http://www.example.com/ for 60 seconds, making a maximum of 10 concurrent requests.

Command Line Options ¶

--url, -u: URLs to request

Specify one or more URLs for hammer to send requests to. Any options provided on the commandline without a flag will also be interpreted as URLs, so the use of --url or -u is optional.

--threads, -T: Number of request goroutines

This value controls the number of goroutines that will be used to send requests to servers, which limits the maximum number of parallel requests that can be in flight at one time. Too low a value can result in fewer requests being sent than you requested; too high can result in resource exhaustion on the target server or the hammering machine itself.

--time, --run-for, -t: Runtime (seconds)

Hammer will run until this timer expires, or until it is terminated by a signal.

--rate, -r: Request rate (requests/second)

Hammer will attempt to send this many requests per second in total. If you specify multiple URLs, the requests will be divided up evenly among all of them.

--backlog, -b: Backlog size (requests)

The backlog is a queue of requests between the request rate throttler and the requesting threads, ready to be sent immediately. If the requesting threads are unable to send requests at the desired rate, the backlog will fill up. Later, if possible, requests will be sent as fast as possible until the backlog is empty in an attempt to "catch up". The backlog can help smooth over network or server hiccups, but too large a backlog can result in large request spikes.

--no-read-body, --read-body:

By default, hammer will read the entire HTTP response body and record the time that it took to receive the response. With --no-read-body, hammer will only read until the end of the headers, and no body timings will be available. In the case of large or slow response bodies, you may be able to generate many more requests by ignoring the bodies.

--log-errors, --no-log-errors:

If --log-errors is enabled, hammer will write a file to disk containing the response headers and body whenever a 4xx or 5xx HTTP response is received from the server. This may help to diagnose transient problems that appear under load.

--report: Report filename

If this option is set, it specifies a file to write statistics to every second during the run, in addition to writing them to the console at the end of the run. If the filename contains a '*' character it will be replaced with the value of the --name option.

--stats: Stats filename

If this option is set, it specifies a file to append abbreviated statistics (single-line, tab-separated) to at the end of the run.

--name: Stats name

Name to use for --report and --stats output.

--header, -h:

Use --header to add headers to outgoing requests. Use as --header 'Name: value'. Can be specified multiple times.