GabrielCWT opened a new pull request, #18686: URL: https://github.com/apache/druid/pull/18686
This PR implements an optimisation for the Interval.toString method. Based on the flame graph before optimisation, we can see that the toString method takes up a significant amount of the running time (~16%). After the optimisation, the optimised format method only accounts for ~4% of the running time. We use the algorithm found [here](https://howardhinnant.github.io/date_algorithms.html#civil_from_days) in order to optimise the toString method. The implementation differs slightly as we know Druid's intervals. As such the following tweaks have been made: - We fix era to be 5 since we know the dates will be within [2000-03-01T00:00:00.000Z, 2400-02-29T23:59:59.999Z] - We omit any negative dates checks. Any dates which do not fall within the era or doesn't have UTC timezone will use `Interval.toString()` as the fallback. ``` Benchmark Mode Cnt Score Error Units DefaultQueryMetricsBenchmark.baselineMetrics avgt 40 1.207 ± 0.022 us/op DefaultQueryMetricsBenchmark.withCustomIntervalToStringDimension avgt 40 0.413 ± 0.010 us/op ``` Note: I do agree that the code is difficult to understand and there are magic numbers still existing. I did my best to refactor most of the numbers to variables if they were constants and I've also tried to maintain the same structure as the algorithm provided in order to make it easier to understand. ## Flame Graph before Optimisation <img width="2830" height="1274" alt="image" src="https://github.com/user-attachments/assets/bd61ead5-8798-4624-bcf8-4dfed9e071a4"; /> ## Flame Graph after Optimisation <img width="3000" height="1492" alt="image" src="https://github.com/user-attachments/assets/3470432b-cb4e-48a5-b8dc-af805495f49e"; /> -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
