[SOLR-10115] Corruption in read-side of SOLR-HDFS stack

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 4.4
• Component/s: Hadoop Integration, hdfs
• Assignee: Yonik Seeley
• Reporter: Yonik Seeley
• Created: 10/Feb/17 02:03
• Updated: 08/Jun/19 15:33

Description

I've been trying to track down some random AIOOB exceptions in Lucene for a customer, and I've managed to reproduce the issue with a unit test of sufficient size in conjunction with highly concurrent read requests.

A typical stack trace looks like:

org.apache.solr.common.SolrException; java.lang.ArrayIndexOutOfBoundsException: 172033655
        at org.apache.lucene.codecs.lucene40.BitVector.get(BitVector.java:149)
        at org.apache.lucene.codecs.lucene41.Lucene41PostingsReader$BlockDocsEnum.nextDoc(Lucene41PostingsReader.java:455)
        at org.apache.lucene.search.MultiTermQueryWrapperFilter.getDocIdSet(MultiTermQueryWrapperFilter.java:111)
        at org.apache.lucene.search.ConstantScoreQuery$ConstantWeight.scorer(ConstantScoreQuery.java:157)

The number of unique stack traces is relatively high, most AIOOB exceptions, but some EOF. Most exceptions occur in the term index, however I believe this may be just an artifact of where highly concurrent access is most likely to occur. The queries that triggered this had many wildcards and other multi-term queries.

Issue Links

• is related to: SOLR-10121 BlockCache corruption with high concurrency
• is related to: SOLR-10116 BlockCache test and documentation improvement
• relates to: SOLR-10141 Caffeine cache causes BlockCache corruption