Skip to content

Netty 4 IndexOutOfBoundsException due to writer index exceeding buffer capacity #113

@benalexau

Description

@benalexau

Netty 4.1.30 and above (plus earlier versions intermittently) fail the following test:

[INFO] Running org.lmdbjava.CursorParamTest
[ERROR] Tests run: 5, Failures: 0, Errors: 1, Skipped: 0, Time elapsed: 0.78 s <<< FAILURE! - in org.lmdbjava.CursorParamTest
[ERROR] execute[4: buffer adapter: org.lmdbjava.CursorParamTest$NettyBufferRunner@68b7bdcb](org.lmdbjava.CursorParamTest)  Time elapsed: 0.015 s  <<< ERROR!
java.lang.IndexOutOfBoundsException: readerIndex: 0, writerIndex: 8 (expected: 0 <= readerIndex <= writerIndex <= capacity(4))
	at io.netty.buffer.AbstractByteBuf.checkIndexBounds(AbstractByteBuf.java:112)
	at io.netty.buffer.AbstractByteBuf.readerIndex(AbstractByteBuf.java:121)
	at org.lmdbjava.ByteBufProxy.out(ByteBufProxy.java:144)
	at org.lmdbjava.ByteBufProxy.out(ByteBufProxy.java:39)
	at org.lmdbjava.KeyVal.valOut(KeyVal.java:133)
	at org.lmdbjava.Cursor.get(Cursor.java:153)
	at org.lmdbjava.CursorParamTest$AbstractBufferRunner.execute(CursorParamTest.java:127)
	at org.lmdbjava.CursorParamTest.execute(CursorParamTest.java:90)

Two issues were identified in ButeBufProxy:

  1. The unsafe setting of the LENGTH_OFFSET was setting the value as a long, whereas the actual field in the Java PooledByteBuf type is an int.
  2. The order of updating the readerIndex before the writerIndex was causing the bounds check on the readerIndex call to fail as the writerIndex had not yet been corrected.

Correction of the above causes tests to pass consistently with Netty 4.1.32 (currently the latest version).

cc #90 #96 #111

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions