amalsaidov opened a new pull request, #20246:
URL: https://github.com/apache/pinot/pull/20246
## Problem
`dateBin(binWidth, source, origin)` aligns a timestamp to the bin boundary
at or before it, with bins anchored at `origin`. It computed the bin index with
plain integer division:
```java
long offsetFromOrigin = sourceMillisEpoch - originMillisEpoch;
long binCount = offsetFromOrigin / binWidthMillis;
```
Integer division truncates toward zero. When the source is before the
origin, `offsetFromOrigin` is negative, so the truncation rounds *up* toward
the origin and picks the bin boundary **after** the source. The source then
falls outside the bin it was assigned to.
Example:
```
dateBin('15m', 2023-12-31 23:45:30, 2024-01-01 00:00:00)
returns 2024-01-01 00:00:00 (14m30s after the source)
expected 2023-12-31 23:45:00
```
This shows up with pre-epoch timestamps and with any query whose `origin` is
later than the data being binned. The existing tests only exercised
source-after-origin, so the case was never caught.
## Fix
Use `Math.floorDiv`, which rounds toward negative infinity, so the boundary
is always at or before the source regardless of sign:
```java
long binCount = Math.floorDiv(offsetFromOrigin, binWidthMillis);
```
For non-negative offsets the result is unchanged, so this is backward
compatible for the already-covered cases.
## Tests
Added assertions to `DateTimeFunctionsTest#testDateBin` for sources before
the origin, including a source that lands exactly on a bin boundary. Verified
they fail on the old code (`expected [2023-12-31 23:45:00.0] but found
[2024-01-01 00:00:00.0]`) and pass with the fix.
```
./mvnw -pl pinot-core -am -Dtest=DateTimeFunctionsTest#testDateBin test
```
--
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]
