HyperLogLogPlusPlus fails on serializeToByteArray

[shnapz]

In Scio we have the following error thrown on version 0.1.0. Unfortunately it is unclear how to reproduce, and I can't provide input data. But stack trace suggests that there is something wrong with management of array slices in com.google.zetasketch.internal.GrowingByteSlice:

16:04:53.990  [info]   java.lang.IllegalArgumentException: 19 > 5
16:04:53.990  [info]   at Due to Exception while trying to `encode` an instance of com.spotify.data.dm.sources.internal.AudienceEngagementMetrics: Can't encode field listeners value com.spotify.scio.extra.hll.zetasketch.ZetaSketchHll@151901cb.(:0)
16:04:53.990  [info]   at java.base/java.util.Arrays.copyOfRange(Arrays.java:4029)
16:04:53.990  [info]   at com.google.zetasketch.internal.GrowingByteSlice.maybeExtendLimit(GrowingByteSlice.java:219)
16:04:53.990  [info]   at com.google.zetasketch.internal.GrowingByteSlice.putNextVarInt(GrowingByteSlice.java:191)
16:04:53.990  [info]   at com.google.zetasketch.internal.GrowingByteSlice.putNextVarInt(GrowingByteSlice.java:30)
16:04:53.990  [info]   at com.google.zetasketch.internal.DifferenceEncoder.putInt(DifferenceEncoder.java:54)
16:04:53.990  [info]   at com.google.zetasketch.internal.hllplus.SparseRepresentation.set(SparseRepresentation.java:424)
16:04:53.990  [info]   at com.google.zetasketch.internal.hllplus.SparseRepresentation.flushBuffer(SparseRepresentation.java:348)
16:04:53.990  [info]   at com.google.zetasketch.internal.hllplus.SparseRepresentation.compact(SparseRepresentation.java:243)
16:04:53.990  [info]   at com.google.zetasketch.HyperLogLogPlusPlus.serializeToByteArray(HyperLogLogPlusPlus.java:298)
16:04:53.990  [info]   at com.spotify.scio.extra.hll.zetasketch.ZetaSketchHll$$anonfun$coder$2.apply(ZetaSketchHLL.scala:121)
16:04:53.990  [info]   at com.spotify.scio.extra.hll.zetasketch.ZetaSketchHll$$anonfun$coder$2.apply(ZetaSketchHLL.scala:121)

So it fails on array = Arrays.copyOfRange(array, arrayOffset(), Math.max(growthCapacity, limit)); and there might be something wrong with Math.max(growthCapacity, limit), apparently it is out of boundaries

[zfraa]

Thank you for the report, we'll have a look.

[AndersonReyes]

to add more context

looks like it only happens when running tests via sbt and the tests only add one to 3 values to the sketch. Production code running with no issues (noticeable failures anyways)

[AndersonReyes]

digging some more and getting random errors i see this as well

[info]   java.util.NoSuchElementException:
[info]   at Due to Exception while trying to `encode` an instance of com.spotify.data.dm.sources.internal.AudienceEngagementMetrics: Can't encode field listeners value com.spotify.scio.extra.hll.zetasketch.ZetaSketchHll@3ae0e410.(:0)
[info]   at com.google.zetasketch.internal.DifferenceDecoder.nextInt(DifferenceDecoder.java:49)
[info]   at com.google.zetasketch.internal.MergedIntIterator.advanceA(MergedIntIterator.java:51)
[info]   at com.google.zetasketch.internal.MergedIntIterator.<init>(MergedIntIterator.java:42)
[info]   at com.google.zetasketch.internal.hllplus.SparseRepresentation.dedupedIterator(SparseRepresentation.java:362)
[info]   at com.google.zetasketch.internal.hllplus.SparseRepresentation.flushBuffer(SparseRepresentation.java:348)
[info]   at com.google.zetasketch.internal.hllplus.SparseRepresentation.compact(SparseRepresentation.java:243)
[info]   at com.google.zetasketch.HyperLogLogPlusPlus.serializeToByteArray(HyperLogLogPlusPlus.java:298)
[info]   at com.spotify.scio.extra.hll.zetasketch.ZetaSketchHll$$anonfun$coder$2.apply(ZetaSketchHLL.scala:121)
[info]   at com.spotify.scio.extra.hll.zetasketch.ZetaSketchHll$$anonfun$coder$2.apply(ZetaSketchHLL.scala:121)
[info]   at com.spotify.scio.coders.TransformCoder.encode(Coder.scala:374)
[info]   at com.spotify.scio.coders.RecordCoder$$anonfun$encode$2.apply$mcV$sp(Coder.scala:263)
[info]   at com.spotify.scio.coders.RecordCoder$$anonfun$encode$2.apply(Coder.scala:263)
[info]   at com.spotify.scio.coders.RecordCoder$$anonfun$encode$2.apply(Coder.scala:263)
[info]   at com.spotify.scio.coders.RecordCoder.onErrorMsg(Coder.scala:249)
[info]   at com.spotify.scio.coders.RecordCoder.encode(Coder.scala:263)
[info]   at org.apache.beam.sdk.coders.Coder.encode(Coder.java:136)

as well which means somehow for some reason some runs show up with