Optimise seshat:counters/3 #16
Conversation
If someone uses this function, the list of counters to return is likely very short, quite possibly it has 1 element. Rather than building a map of all counters and then dropping the irrelevant ones, only add the relevant counters in the first place. Testing with RabbitMQ an rabbit_quorum_queue:open_files function, which only asks for 1 metric, the avery execution went down form 7ms to 1ms with this change.
|
@mkuratczyk WDYT about going further and adding a separate function head for a single element list? We have been doing similar things on some RabbitMQ hot code paths (routing) since… 2008 or so? Maybe that would not be meaningful enough but also should be easy to try, e.g. build_counters_map(CRef, FieldsSpec, [Name]) ->
%% construct a single element map using `Name` |
|
Makes sense. Let me test that |
given we still have to fold over the full list of fieldsspec I don't think this will yield meaningful value. If we really wanted to optimise we could make |
Previous commit sped up single-item invocation from 7ms to 1ms. With this change, it goes down to 0.1ms.
|
Turns out it does make a significant difference - initially from 7ms to 1ms. With the second commit from 1ms to 0.1ms. :) Nice one @michaelklishin ! |
|
|
If someone uses this function, the list of counters to return is likely very short, quite possibly it has 1 element. Rather than building a map of all counters and then dropping the irrelevant ones, only add the relevant counters in the first place.
Testing with RabbitMQ an rabbit_quorum_queue:open_files function, which only asks for 1 metric, the avery execution went down form 7ms to 1ms with this change.