Java Bugs in various JVMs affecting Lucene / Solr

Sometimes Lucene runs amok of bugs in JVM implementations from different vendors. In certain cases we whittle it down to a small test case, open an issue with the vendor, and hopefully it gets fixed. In other cases we know the bug is in the JVM but we haven't narrowed it enough to open a bug with the vendor. Sometimes we can work out a simple workaround in Lucene.

We try to open a Lucene mirror bug to provide details on how Lucene is affected, iterate on a compact test case, etc.

Oracle Java / Sun Java / OpenJDK Bugs

If you are affected by one of these issues that Oracle's Java has yet to accept or resolve, or simply have some spare votes, please consider adding your vote to the bug (on Oracle's bug page):

Do not, under any circumstances, run Lucene with the G1 garbage collector. Lucene's test suite fails with the G1 garbage collector on a regular basis, including bugs that cause index corruption. There is no person on this planet that seems to understand such bugs (see https://bugs.openjdk.java.net/browse/JDK-8038348, open for over a year), so don't count on the situation changing soon. This information is not out of date, and don't think that the next oracle java release will fix the situation.

Oracle bug

Lucene mirror issue

Impact to Lucene

State, Priority

Workaround

6265734

LUCENE-573

NIOFSDirectory has very poor performance on Windows

Accepted, Low

Use FSDirectory or MMapDirectory on Windows

6478546

LUCENE-1566

You hit a false OutOfMemoryException when loading norms in an index with many docs

Accepted, Low

Locally patch Lucene to load large contiguous byte sequences in chunks

6707044

LUCENE-1282

Index corruption

Fixed as of 1.6.0_10, High

Lucene code base has a workaround in it

4724038

LUCENE-1658 LUCENE-1669

MMapDirectory won't close files until GC (higher transient disk usage than other FSDir); MMapDirectory returns 0 bytes when used for read/write access in remote (SMB/CIFS) mount

Cause Known, Low

Use a different Directory implementation

(No Oracle bug yet)

LUCENE-1342

SEGV during indexing, with Java 1.6 64 bit

(No compact test case yet)

None known

(No Oracle bug yet)

Discussed here and here

On 64 bit JREs, reading from files may hang (??)

(Still characterizing)

None known

<ac:structured-macro ac:name="unmigrated-wiki-markup" ac:schema-version="1" ac:macro-id="87221b11-cc1e-468d-b506-07ae14c7a6f8"><ac:plain-text-body><![CDATA[

[6588260

https://bugs.openjdk.java.net/browse/JDK-6588260]

[LUCENE-2449

https://issues.apache.org/jira/browse/LUCENE-2449]

No impact as of 3.1/4.0 – we now avoid calling new String(int[], int, int)

Fix Delivered, Bug

 

]]></ac:plain-text-body></ac:structured-macro>

6240963

LUCENE-2685

Fix Delivered, High

xml-query-parser's XSLT transforms fail under some locales

Upgrade to Java 6, use a different JRE vendor, plug in an alternate, more up-to-date XSL engine like Apache XALAN, or set system property telling TransformerFactory to use the non byte-code generating bundled XSL engine in Java 5

5091921

LUCENE-2975

readVInt() returns wrong results

not sure

Lucene code base has a workaround in it

7044738

LUCENE-3346

readVInt() returns wrong results

8-Fix Available, Medium

use Java 6 or Java 7u1

7070134

LUCENE-3335

Porter Stemmer crashes JRE

8-Fix Available, Low

-XX:-UseLoopPredicate or Java 7u1

7104012

LUCENE-3301

BreakIterators (e.g. in analyzers) crash on certain inputs

Accepted, Low

Lucene code base has a workaround in it

6822370

LUCENE-3235

Concurrent Classes suffer from a race in LockSupport park() arising from weak memory models

Fixed, High

Use -XX:+UseMembar if you are running on a JVM < 1.6u18 see this writeup for details

8024830

LUCENE-5212

SEGV or serious index corrumption

Fixed, Critical

Use -XX:-UseSuperWord if you are running on 7u40 <= JVM < 7u55

Oracle JRockit Bugs

Oracle JRockit is no longer available for Java 7+, but it is still used in legacy environments (e.g. Oracle/BEA WebLogic application server). Oracle seems to no longer accept bugs / only accepts bugs from customers with a paid support contract, so there is no chance for the Lucene Contributors to report bugs upstream.

As stated by Oracle, it is recommended to upgrade to Oracle Java 7 or OpenJDK 7 (both update 1 or later). JRockit key features were merged into OpenJDK / Oracle Java.

Lucene issue

Impact to Lucene

Severity

Workaround

(No issue yet)

"If" statements are not being executed

Serious (index corruption)

Use -XnoOpt JVM parameter

IBM J9 Bugs

IBM J9 is available for Java 5, 6, and 7. Unfortunately, there is no public bugtracker available, so the bugs listed below cannot be reported. IBM only has a bug tracker for paying customers with a support contract.

Although some Lucene committers are IBM employees, its impossible to report bugs involving the whole Lucene community. Because of this the status of all bugs listed below is unknown, we have no information about possible fixes.

If you are using SUSE Enterprise Linux which has IBM J9 as their default Java implementation, it is strongly recommend to use Oracle Java 7 or OpenJDK 7 (minimum update 1).

The following bugs are known, but since a while all nightly tests using the IBM J9 were disabled on Lucene's Jenkins servers, because a lot of test fail randomly.

Lucene issue

Impact to Lucene

Severity

Workaround

(No issue yet)

FST.pack() produces corrupt index (Lucene 4) because a loop is miscompiled

Serious (index corruption)

Use -Xjit:exclude={org/apache/lucene/util/fst/FST.pack(IIF)Lorg/apache/lucene/util/fst/FST;} JVM parameter to disable optimizations in FST.pack()

LUCENE-4987

a synchronized method is completely optimized away

tests only

the test framework was updated to version 2.0.10 which contains a workaround

  • No labels