Summary
After HDX-3697 adds log-spaced buckets, explore adaptive bucket density that allocates more buckets near the mode and fewer in the sparse tail (similar to exponential histogram approaches used in OpenTelemetry and modern observability systems).
Currently the heatmap uses 80 fixed buckets. With adaptive sizing, bucket boundaries would follow a relative-error guarantee — e.g., 2% relative error at any point in the distribution. This maximizes visual resolution where data actually lives.
This is a deeper change to the bucketing query and may require a different ClickHouse approach (e.g., pre-computed exponential histogram boundaries rather than widthBucket()).
Dependencies
Summary
After HDX-3697 adds log-spaced buckets, explore adaptive bucket density that allocates more buckets near the mode and fewer in the sparse tail (similar to exponential histogram approaches used in OpenTelemetry and modern observability systems).
Currently the heatmap uses 80 fixed buckets. With adaptive sizing, bucket boundaries would follow a relative-error guarantee — e.g., 2% relative error at any point in the distribution. This maximizes visual resolution where data actually lives.
This is a deeper change to the bucketing query and may require a different ClickHouse approach (e.g., pre-computed exponential histogram boundaries rather than
widthBucket()).Dependencies