sunchao commented on code in PR #3726:
URL: https://github.com/apache/parquet-java/pull/3726#discussion_r3887225406
##########
parquet-hadoop/src/main/java/org/apache/parquet/hadoop/ParquetFileReader.java:
##########
@@ -1357,32 +1334,105 @@ private boolean
arePartsValidForVectoredIo(List<ConsecutivePartList> allParts) {
* If directly implemented by a Filesystem then it is likely to be a more
efficient
* operation such as a scatter-gather read (native IO) or set of parallel
* GET requests against an object store.
+ * The allocation limit applies to filesystem buffers; decoders can still
require a
+ * contiguous buffer for an individual logical value larger than that limit.
* @param allParts all parts to be read.
* @param builder used to build chunk list to read the pages for the
different columns.
- * @throws IOException any IOE.
- * @throws IllegalArgumentException arguments are invalid.
- * @throws UnsupportedOperationException if the filesystem does not support
vectored IO.
+ * @throws IOException if submitting or consuming the vectored reads fails.
+ * @throws IllegalArgumentException if range preparation fails before any
reads are submitted.
*/
private void readVectored(List<ConsecutivePartList> allParts,
ChunkListBuilder builder) throws IOException {
-
+ final int maximumAllocation = options.getMaxAllocationSize();
+ Preconditions.checkArgument(maximumAllocation > 0, "Invalid maximum
allocation size %s", maximumAllocation);
+ if (vectoredReadFileLength < 0) {
+ vectoredReadFileLength = file.getLength();
+ }
+ final long fileLength = vectoredReadFileLength;
List<ParquetFileRange> ranges = new ArrayList<>(allParts.size());
+ List<Integer> partRangeCounts = new ArrayList<>(allParts.size());
long totalSize = 0;
for (ConsecutivePartList consecutiveChunks : allParts) {
final long len = consecutiveChunks.length;
- Preconditions.checkArgument(
- len < Integer.MAX_VALUE,
- "Invalid length %s for vectored read operation. It must be less than
max integer value.",
- len);
- ranges.add(new ParquetFileRange(consecutiveChunks.offset, (int) len));
+ final long start = consecutiveChunks.offset;
+ if (start < 0 || len < 0 || start > fileLength || len > fileLength -
start) {
+ throw new IOException(String.format(
+ "Invalid vectored read range (offset %d, length %d) for file
length %d",
+ start, len, fileLength));
+ }
+ final int firstRange = ranges.size();
+ long remaining = len;
+ long offset = start;
+ do {
+ int rangeLength = (int) Math.min(remaining, maximumAllocation);
+ ranges.add(new ParquetFileRange(offset, rangeLength));
+ offset += rangeLength;
+ remaining -= rangeLength;
+ } while (remaining > 0);
+ partRangeCounts.add(ranges.size() - firstRange);
totalSize += len;
}
LOG.debug("Reading {} bytes of data with vectored IO in {} ranges",
totalSize, ranges.size());
- // Request a vectored read;
- f.readVectored(ranges, options.getAllocator());
- int k = 0;
- for (ConsecutivePartList consecutivePart : allParts) {
- ParquetFileRange currRange = ranges.get(k++);
- consecutivePart.readFromVectoredRange(currRange, builder);
+ final long readStart = System.nanoTime();
+ try {
+ // Even a synchronous rejection can follow partial submission. The
Hadoop bridge
+ // publishes futures only after submission returns, so missing futures
do not prove
+ // that no reads started. Once this call is entered, normal-read
fallback is unsafe.
+ f.readVectored(ranges, options.getAllocator());
+ int firstRange = 0;
+ for (int partIndex = 0; partIndex < allParts.size(); partIndex++) {
+ int endRange = firstRange + partRangeCounts.get(partIndex);
+
allParts.get(partIndex).readFromVectoredRanges(ranges.subList(firstRange,
endRange), builder);
+ firstRange = endRange;
+ }
+ } catch (IllegalArgumentException | UnsupportedOperationException e) {
+ // Consumption may also have populated the builder. Do not replay those
chunks.
+ IOException failure =
+ new IOException("Vectored read failed after asynchronous reads may
have been submitted", e);
+ awaitRemainingVectoredReads(ranges, readStart, failure);
+ throw failure;
+ } catch (IOException | RuntimeException e) {
+ awaitRemainingVectoredReads(ranges, readStart, e);
+ throw e;
+ }
+ }
+
+ /**
+ * Wait for submitted reads with published futures to finish before their
stream can be
+ * closed. Cancelling result futures does not stop all Hadoop backends from
continuing IO.
+ */
+ private void awaitRemainingVectoredReads(List<ParquetFileRange> ranges, long
readStart, Throwable failure) {
+ if (Thread.currentThread().isInterrupted()
+ || failure instanceof InterruptedIOException && failure.getCause()
instanceof InterruptedException) {
Review Comment:
Adjusted in
[f82eaa42](https://github.com/apache/parquet-java/pull/3726/commits/f82eaa428528849cf019c99410fe879c4d641875).
An interrupted wait still propagates promptly as `InterruptedIOException`; it
does not block draining siblings. The failed reader is invalidated and its
original stream is handed to deferred cleanup, which cannot overtake a
submission call still running after cancellation.
Cleanup may close a stream with asynchronous reads still pending. Close or
cancellation is not treated as proof that I/O stopped: buffer release requires
every range to have a published, completed, non-cancelled future. Unknown
completion retains the batch's allocations.
Added tests for a submission that ignores interruption and for a pending
read that completes after stream closure, checking that its buffer is not
released early.
--
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]
