[erlang-questions] [ANN] Silly benchmarking
Tue Apr 30 23:26:33 CEST 2013
On 30 Apr 2013, at 20:54, Noah Diewald wrote:
> I've always wished that they had a network of computers as one of the options
> instead of just single computers.
https://github.com/nebularis/systest exists to address the issue of running tests against multiple nodes, although it's still in quite an early stage (no remote nodes via ssh support yet) and already in need of some refactoring. It currently supports running in standalone (or shell) mode and integrating with common_test (via the ct_hooks mechanism) and its goal is to manage test resources (e.g., a node, a cluster of nodes, etc) during a test run. There's nothing to prevent this from being used for benchmarks instead of integration tests.
You can see example (test) usage in https://github.com/rabbitmq/rabbitmq-test/tree/bug25421/multi-node, common_test integration via https://github.com/rabbitmq/rabbitmq-test/tree/bug25421/multi-node/test and SysTest API calls in https://github.com/rabbitmq/rabbitmq-test/blob/bug25421/multi-node/src/rabbit_ha_test_utils.erl
There's plenty of things wrong with the current implementation though - put together in a bit of a rush - if you're interested in the dirty laundry and/or state of usefulness, https://github.com/nebularis/systest/wiki/The-Big-Refactor outlines some of the existing (structural) things that I'm in the process of fixing. Next big ticket feature is an ssh based layer to manage remote nodes during test runs. Nonetheless, we've found and fixed plenty of bugs using it to test clustering/HA scenarios.
Suggestions, comments and/or pull requests are welcome, though do bear in mind I'm going to rip up large swathes of it and rewrite them very soon. :)
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 235 bytes
Desc: Message signed with OpenPGP using GPGMail
More information about the erlang-questions