[HDDS-7303] EC: ECBlockReconstructedStripeInputStream should set initialized false on re-init

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.3.0
• Component/s: EC Client
• Labels: pull-request-available
• Assignee: Stephen O'Donnell
• Reporter: Stephen O'Donnell
• Created: 10/Oct/22 12:02
• Updated: 15/Oct/22 07:06
• Resolved: 15/Oct/22 07:06

Description

In ECBlockReconstructedStripeInputStream, when an exception occurs reading a block, the code calls the `init()` method to setup the missing indexes and buffers.

If an InsufficientLocations exception is thrown part way through that method, the class ends up partly re-initialized. If something then ignores / handles the InsufficientLocations and tries to call read again, it can cause strange results. In one case, we get an illegalArgumentException, which I think is related to the above:

Caused by: java.lang.ArrayIndexOutOfBoundsException: 1
	at org.apache.hadoop.ozone.client.io.ECBlockReconstructedStripeInputStream.assignBuffers(ECBlockReconstructedStripeInputStream.java:289)
	at org.apache.hadoop.ozone.client.io.ECBlockReconstructedStripeInputStream.read(ECBlockReconstructedStripeInputStream.java:360)
	at org.apache.hadoop.ozone.client.io.ECBlockReconstructedStripeInputStream.readStripe(ECBlockReconstructedStripeInputStream.java:345)
	at org.apache.hadoop.ozone.client.io.ECBlockReconstructedInputStream.readStripe(ECBlockReconstructedInputStream.java:214)
	at org.apache.hadoop.ozone.client.io.ECBlockReconstructedInputStream.readAndSeekStripe(ECBlockReconstructedInputStream.java:198)
	at org.apache.hadoop.ozone.client.io.ECBlockReconstructedInputStream.seek(ECBlockReconstructedInputStream.java:192)
	at org.apache.hadoop.ozone.client.io.ECBlockInputStreamProxy.seek(ECBlockInputStreamProxy.java:224)
	at org.apache.hadoop.ozone.client.io.KeyInputStream.seek(KeyInputStream.java:340)
	at org.apache.hadoop.fs.ozone.OzoneFSInputStream.seek(OzoneFSInputStream.java:78)
	at org.apache.hadoop.fs.FSInputStream.read(FSInputStream.java:85)
	at org.apache.hadoop.fs.FSInputStream.readFully(FSInputStream.java:124)
	at org.apache.hadoop.fs.FSDataInputStream.readFully(FSDataInputStream.java:116)
	at org.apache.orc.impl.RecordReaderUtils$DefaultDataReader.readStripeFooter(RecordReaderUtils.java:273)
	at org.apache.orc.impl.RecordReaderImpl.readStripeFooter(RecordReaderImpl.java:308)
	at org.apache.orc.impl.RecordReaderImpl.beginReadStripe(RecordReaderImpl.java:1131)
	at org.apache.orc.impl.RecordReaderImpl.readStripe(RecordReaderImpl.java:1093)
	at org.apache.orc.impl.RecordReaderImpl.advanceStripe(RecordReaderImpl.java:1261)
	at org.apache.orc.impl.RecordReaderImpl.advanceToNextRow(RecordReaderImpl.java:1296)
	at org.apache.orc.impl.RecordReaderImpl.nextBatch(RecordReaderImpl.java:1332)
	at org.apache.hadoop.hive.ql.io.orc.RecordReaderImpl.nextBatch(RecordReaderImpl.java:157)
	at org.apache.hadoop.hive.ql.io.orc.VectorizedOrcAcidRowBatchReader$1.next(VectorizedOrcAcidRowBatchReader.java:175)
	at org.apache.hadoop.hive.ql.io.orc.VectorizedOrcAcidRowBatchReader$1.next(VectorizedOrcAcidRowBatchReader.java:171)
	at org.apache.hadoop.hive.ql.io.orc.VectorizedOrcAcidRowBatchReader.next(VectorizedOrcAcidRowBatchReader.java:871)
	... 26 more

We should simply set initialized to false at the beginning of init and set it to try at the end when the full init method has completed.

Issue Links

• is duplicated by: HDDS-7278 [ozone-LR] ArrayIndexOutOfBoundsException in EC Buckets with fault injections
• links to: GitHub Pull Request #3816