[
https://issues.apache.org/jira/browse/FLINK-10356?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16659363#comment-16659363
]
ASF GitHub Bot commented on FLINK-10356:
----------------------------------------
NicoK commented on a change in pull request #6705: [FLINK-10356][network] add
sanity checks to SpillingAdaptiveSpanningRecordDeserializer
URL: https://github.com/apache/flink/pull/6705#discussion_r226979613
##########
File path:
flink-runtime/src/main/java/org/apache/flink/runtime/io/network/api/serialization/SpillingAdaptiveSpanningRecordDeserializer.java
##########
@@ -549,21 +584,53 @@ private void addNextChunkFromMemorySegment(MemorySegment
segment, int offset, in
}
else {
spillingChannel.close();
+ spillingChannel = null;
- BufferedInputStream inStream = new
BufferedInputStream(new FileInputStream(spillFile), 2 * 1024 * 1024);
+ BufferedInputStream inStream =
+ new BufferedInputStream(
+ new
FileInputStream(checkNotNull(spillFile)),
+ 2 * 1024 * 1024);
this.spillFileReader = new
DataInputViewStreamWrapper(inStream);
}
}
}
- private void
moveRemainderToNonSpanningDeserializer(NonSpanningWrapper deserializer) {
+ private void
moveRemainderToNonSpanningDeserializer(NonSpanningWrapper deserializer) throws
IOException {
+ Optional<String> deserializationError =
getDeserializationError(0);
+ if (deserializationError.isPresent()) {
+ throw new
IOException(deserializationError.get());
+ }
+
deserializer.clear();
if (leftOverData != null) {
deserializer.initializeFromMemorySegment(leftOverData, leftOverStart,
leftOverLimit);
}
}
+ private Optional<String> getDeserializationError(int
addToReadBytes) {
Review comment:
It took me a bit, but then I realised why: If this code throws an
`EOFException`, `remainingSpanningBytes` will report `0` and not fail (and fall
back to the thrown `EOFException` instead of our more detailed custom one):
```
try {
target.read(this.spanningWrapper.getInputView());
} catch (EOFException e) {
```
-> by simulating that we read too many bytes (which is actually true), we
get our exception containing the line `-1 remaining unread byte`. It is a bit
hacky though. Let me try to refactor and add comments nonetheless.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
> Add sanity checks to SpillingAdaptiveSpanningRecordDeserializer
> ---------------------------------------------------------------
>
> Key: FLINK-10356
> URL: https://issues.apache.org/jira/browse/FLINK-10356
> Project: Flink
> Issue Type: Improvement
> Components: Network
> Affects Versions: 1.5.0, 1.5.1, 1.5.2, 1.5.3, 1.5.4, 1.6.0, 1.6.1, 1.7.0
> Reporter: Nico Kruber
> Assignee: Nico Kruber
> Priority: Major
> Labels: pull-request-available
>
> {{SpillingAdaptiveSpanningRecordDeserializer}} doesn't have any consistency
> checks for usage calls or serializers behaving properly, e.g. to read only as
> many bytes as available/promised for that record. At least these checks
> should be added:
> # Check that buffers have not been read from yet before adding them (this is
> an invariant {{SpillingAdaptiveSpanningRecordDeserializer}} works with and
> from what I can see, it is followed now.
> # Check that after deserialization, we actually consumed {{recordLength}}
> bytes
> ** If not, in the spanning deserializer, we currently simply skip the
> remaining bytes.
> ** But in the non-spanning deserializer, we currently continue from the
> wrong offset.
> # Protect against {{setNextBuffer}} being called before draining all
> available records
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
