siddharthteotia commented on code in PR #8398:
URL: https://github.com/apache/pinot/pull/8398#discussion_r842448677
##########
pinot-segment-local/src/main/java/org/apache/pinot/segment/local/segment/creator/impl/SegmentColumnarIndexCreator.java:
##########
@@ -260,6 +260,26 @@ private boolean
createDictionaryForColumn(ColumnIndexCreationInfo info, SegmentG
.containsKey(column)) {
return false;
}
+
+ // Do not create dictionary if size with dictionary is going to be larger
than size without dictionary
+ // This is done to reduce the cost of dictionary for high cardinality
columns
+ // Off by default and needs optimizeDictionaryEnabled to be set to true
+ if (config.isOptimizeDictionaryEnabled() && spec.getFieldType() ==
FieldType.METRIC
+ && spec.isSingleValueField() && spec.getDataType().isFixedWidth()) {
+ long dictionarySize = info.getDistinctValueCount() *
spec.getDataType().size();
+ long forwardIndexSize =
+ ((long) info.getTotalNumberOfEntries() *
PinotDataBitSet.getNumBitsPerValue(info.getDistinctValueCount() - 1)
+ + Byte.SIZE - 1) / Byte.SIZE;
+
+ double indexWithDictSize = dictionarySize + forwardIndexSize;
+ double indexWithoutDictSize = info.getTotalNumberOfEntries() *
spec.getDataType().size();
+
+ double storageSaved = (indexWithDictSize - indexWithoutDictSize) /
indexWithDictSize;
Review Comment:
We can instead do a check based on ratio ?
`indexWithDictSize / indexWithoutDictSize`
Also, if we want to use percentage, then the reverse calculation will be
more intuitive imo as we want to compute the storage saved by creating
dictionary over no-dictionary
`double storageSaved = (indexWithoutDictSize - indexWithDictSize) /
indexWithoutDictSize;`
e.g -- long column with 1million rows and 16 cardinality
(8 * 1024 * 1024) - (128 + 500000) / (8 * 1024 * 1024) = 0.94 or 94%
--
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]
