Repository navigation
Improve iterator performance and add better stream support #269
Description
I've implemented replacements for CursorIterable that appear to be faster. Raising this issue to coincide with the iterator-performance branch.
{{ message }}
Activity
#269 Initial iterator performance enhancements and testing.
stroomdev66 commentedon Nov 4, 2025
See draft PR #270
stroomdev66 commentedon Nov 4, 2025
My initial, and admittedly not very scientific, performance tests show the following improvements:
The X axis shows different key ranges and the Y axis is milliseconds spent executing range scans.
stroomdev66 commentedon Nov 4, 2025
Another set of tests with 100M rows
[-]Improve iterator performance and add better streaming support[/-][+]Improve iterator performance and add better stream support[/+]#269 Initial iterator performance enhancements and testing.
#269 Initial iterator performance enhancements and testing.
#269 Initial iterator performance enhancements and testing.
#269 Initial iterator performance enhancements and testing.
#269 Initial iterator performance enhancements and testing.
#269 Initial iterator performance enhancements and testing.
#269 Initial iterator performance enhancements and testing.
#269 Initial iterator performance enhancements and testing.
cdprete commentedon Feb 20, 2026
Very nice to see the support for
Stream<T>now being offered 👍I will be able with this to drop some code from my side which was doing the same 😁
Add tests for iterator/stream helpers; fix LmdbStream SORTED contract
bernardladenthin commentedon Jul 24, 2026
I spent some time reviewing the
iterator-performancebranch and opened a small follow-up PR stacked on it (targetsiterator-performance, notmaster): #292.It fixes a couple of things in the new iterator/stream code and adds the missing test coverage:
0xFF.incrementLeastSignificantBytekept the trailing0xFF({0x01,0xFF}→{0x02,0xFF}instead of the tight{0x02}), so the reverse-prefix seek over-shot and iteration yielded nothing. Fixed across all four proxies, with a regression test.LmdbStream's Spliterator advertisedSORTEDwith anullcomparator. Now reportsORDERED | NONNULLandgetComparator()throws per the contract; the dead comparator stubs are removed.KeyRange.getType()returnednullfor builder-created ranges (would NPE the legacyCursorIterablepath) — now derives the equivalentKeyRangeType.KeyRangebuilder,Dbi.newIterate/stream/getNameAsString, and the spliterator contract.Full suite is green (excluding the pre-existing flaky
TestLmdbStreamBenchmark). The bigger design calls (API naming, retiringCursorIterable, rebasing, etc.) are intentionally left for you — noted in the PR. Happy to adjust anything.