Github user cloud-fan commented on a diff in the pull request: https://github.com/apache/spark/pull/18503#discussion_r129474900 --- Diff: sql/core/src/main/scala/org/apache/spark/sql/execution/streaming/state/HDFSBackedStateStoreProvider.scala --- @@ -363,7 +363,11 @@ private[state] class HDFSBackedStateStoreProvider extends StateStoreProvider wit val valueRowBuffer = new Array[Byte](valueSize) ByteStreams.readFully(input, valueRowBuffer, 0, valueSize) val valueRow = new UnsafeRow(valueSchema.fields.length) - valueRow.pointTo(valueRowBuffer, valueSize) + // If valueSize in existing file is not multiple of 8, floor it to multiple of 8. + // This is work around for the following. + // Pre-Spark 2.3 mistakenly append 4 bytes to the value row in + // `FixedLengthRowBasedKeyValueBatch`, which gets persisted into the checkpoint data + valueRow.pointTo(valueRowBuffer, (valueSize / 8) * 8) --- End diff -- yes I'm absolutely sure, as we store aggregate buffers in checkpoint files. BTW, I run [this test](https://github.com/apache/spark/pull/18503/commits/01597010145f7769506cfa51c0450a2db779fc07#diff-b601a70172e88e90cea0020b978d0b5bL482) without the workaround here, and it can still pass. Which means, after this PR, `RowBasedKeyValueBatch` is fixed, and the generated checkpoint files have corrected row length. And I also run the test without the fix for `RowBasedKeyValueBatch`, and it fails. I think this proves that `RowBasedKeyValueBatch` is the reason why we have wrong row length in checkpoit files.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. --- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org