[Bug] Limiting memory does not degrade query latency #17070
Description
Search before asking
- I searched in the issues and found nothing similar.
Version
1.1.3 and 1.3.4 (tested with standalone Docker image)
Describe the bug and provide the minimal reproduce step
While testing IoTDB under limited memory configurations, using docker for that effect, it was found that the database's performance won't drop until the threshold of 4GB of memory is hit, despite it always using the maximum amount of memory allocated for the container.
It was expected that progressively constraining the database's available memory would lead to gradual performance degradation, but instead IoTDB maintains virtually identical query latency across the tested memory limits, despite always consuming the maximum amount of memory available to the container.
Minimal reprodution steps:
- Launch IoTDB in standalone mode with constrained memory using Docker Compose
- Preload data before executing queries
- Execue a query workload, using the same for every memory configuration
The queries follow the following templates:
-- Aggregation
select agg_func(field) from path where time >= start and time <= end
-- Downsampling
select field from path group by ([start, end), step)
-- Outlier-filter
select field from path where time >= start and time <= end and field [>,>=,<,<=] threshold
What did you expect to see?
Either performance degradation as memory limits get increasingly smaller, or the container not using all the memory available to it, if it is able to maintain performance with less memory usage.
What did you see instead?
Performance didn't degrade until the 4GB memory limit, remaining similar regardless of the limit used, despite the database always using the maximum memory allocated to it.
Anything else?
The tests were run using MulletBench, as well as plot generation.
Are you willing to submit a PR?
- I'm willing to submit a PR!