Apache Hadoop 0.20.2 Release Notes

These release notes cover new developer and user-facing incompatibilities, important issues, features, and major improvements.

Correct synchronization error in IPC where handler thread could hang if request reader got an error.

If an IPC server response buffer has grown to than 1MB, it is replaced by a smaller buffer to free up the Java heap that was used. This will improve the longevity of the name service.

Corrected arithmetic error that made sleep times less than zero.

Allow a general mechanism to disable the cache on a per filesystem basis by using property fs.<schemename>.impl.disable.cache. eg. fs.har.impl.disable.cache in core-default.xml

Bugs fixed for Hadoop archives: character escaping in paths, LineReader and file system caching.

WARNING: No release note provided for this incompatible change.

Correct PendingDeletionBlocks metric to properly decrement counts.

Corrected an error when checking quota policy that resulted in a failure to read the edits log, stopping the primary/secondary name node.

Rename properly considers the case where both source and destination are over quota; operation will fail with error indication.

Memory leak in function hdfsFreeFileInfo in libhdfs. This bug affects fuse-dfs severely.

Modifies shuffle related memory parameters to use ‘long’ from ‘int’ so that sizes greater than maximum integer size are handled correctly

Adds a counter to track the number of records emitted by map writing directly to HDFS i.e map tasks of job with 0 reducers.

Fix streaming job to show proper message if file is is not present, for -file option.

Added support to fallback to new task memory configuration when deprecated memory configuration values are set to disabled.

Use ToolRunner for archives job and return non zero error code on failure.

Removes javac warnings by either resolving them or suppressing them (wherever resolution is not possible)

Resolves the test failure by modifying the test to base it on spill counters rather than on bytes read/written. It also introduces a new configuration parameter “mapred.job.shuffle.input.buffer.percent” to provide finer grained control on the memory limit to be used during shuffle.

Updates of counters for reduce input and output records were added in the new API so they are available for jobs using the new API.