…directly now, need to set $PIG_HOME
The changes were: - Comment out some @Override annotations - Replace Arrays.copyOf calls with System.arraycopy - Remove FloatHistogram, Histogram, StatsPlugin and StatsServlet. This is not ideal (particularly the last line), but it will have to do for now as Histogram uses TreeMap.floorEntry which does not exist in JDK5. People on JDK6 should use the offiical Avro 1.3.0 jar instead.
* master: Do not throw IOException from encode. Remove unnecessary cast from byte to int. Reformat files in compress.lzf to use the formatting standards used by Voldemort. Remove dependency on h2-lzf.jar as we now include a LZF implementation in the source code. Change LzfCompressionStrategy to use LZFEncoder and LZFDecoder. Move LZF files to the right folder. Refactoring: moved lzf codec stuff in voldemort.store.compress.lzf Added license information (and related) Fixed 2 remaining bugs: now passes large collection of documents without hitch Added a unit test (pointed out one bug that's fixed; also, seems like there is one rarer one to fix) Last part of refactoring. Additional refactoring. Refactoring Added compressor, seems to work for both directions. Note: compression results vary bit from c version due to different hash method, block sizes First changes: added simple command-line tool (LZF.java) to check compress/decompress; created block codec (currently just implements decompress). But decompress code now uses "official" format, and works (as per brief testing, can correctly decompress lzf command line tool compressed files). And block codec is more efficient than streaming, for specific use case of V (as far as I can see). Import from SVN r2110. Conflicts: .classpath
…n in the source code.
* 'master' of github.com:voldemort/voldemort: (158 commits) Updated lib/h2-lzf.jar to be 1.5 compatible. Fixed bad log in bootstrap retry logic. fixed compilation issues after merge. unit test fixes. refactored AdminBasicTest. Strictly checking for size == -1, passing Multi JVM test now: was failing due to a protocol buffers bug. Refactored AdminTest for clarity. Fixed unit tests. Removed AdminSocketRequestHandlerFactory. clean test logs. Created new ServerJVMTestUtils to provide start/stop/manage VoldemortServer as separate JVMs. fixed serverTestUtils to make valid metadata store. Changed AdminServerMultiJVM test to pass classpath as argument. sanitize classpath string in AdminServerMultiJVM tests. Added remote server console output to AdminServiceMultiJVMTest console. Added error logging to server side. Added a MultiJVM AdminTest. Refactoring: using AsyncOperationStatus class, s/request/operation/ in variable and method names. Made AsyncOperationRunner and associated state (e.g. last request id) global via VoldemortServer. small changes in Admin Async request handler. ... Conflicts: .classpath
* 'master' of github.com:voldemort/voldemort: Implementation and tests for Issue 172: Support LZF compression. Extract StreamCompressionStrategy from GzipCompressionStrategy. Incorporated a patch by Benjamin Schmaus<email@example.com>: changes to command line shell client (adding support for reading commands from a text file, changing "locate" to "preflist"). Tweak fix for issue #170, throw the exception instead of returning it and ignore InterruptedException. Introduce BdbIterator and refactor BdbStoreIterator and BdbKeysIterator to use it. Make sure to rethrow exception after adding it to startedStatusQueue. revised test.rb to use localhost Fix and test for Issue 170: Hang if port is being used by another process. Added Ruby client library Implement Issue 168: Introduce StorageEngine.keys(). Conflicts: .classpath
We rely on the implementation from the H2 database. Note that we only include the LZF-related classes (4 classes in org.h2.compress) from SVN revision 2063 of H2.
… makes changes easier and possible in "the field".
…moke test and restructured code a little to maintain a cleaner API.
…izes the entire makeObject() call. This means that only one connection to ANY server can be connecting at a time. This problems shows up quite severely when you have a host that is hard down on the network, and hence the connection will timeout. Exacerbating this it turns out that java.net.Socket does not use the soTimeout when establishing the connection, so we were getting a near infinite timeout which blocked all other connections from being established. Fix is to use soTimeout when establishing a connection, and also upgrade to new version of commons pool which seems to have fixed (or mostly fixed) this problem.
…as a continual stream. The new PB protocol is <REQ_SIZE><REQUEST><RESP_SIZE><RESPONSE>... Seems to resolve difficulties.
…last jetty checked-in Modified RemoteTest to take better style parameters and support reading keys from a text file
Also Regenerated the VProto as the new protobuf expects the generated message class to include a method called toBuilder.
…mark. Avoid recopying byte arrays in hadoop job. Make jvm use -d64 so that we can mmap large chunks in startup scripts.