http://git-wip-us.apache.org/repos/asf/hbase-site/blob/6f2e75f2/devapidocs/src-html/org/apache/hadoop/hbase/io/hfile/HFileBlock.Writer.State.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/io/hfile/HFileBlock.Writer.State.html b/devapidocs/src-html/org/apache/hadoop/hbase/io/hfile/HFileBlock.Writer.State.html index a665139..3fedd0b 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/io/hfile/HFileBlock.Writer.State.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/io/hfile/HFileBlock.Writer.State.html @@ -879,1201 +879,1221 @@ <span class="sourceLineNo">871</span> // includes the header size also.<a name="line.871"></a> <span class="sourceLineNo">872</span> private int unencodedDataSizeWritten;<a name="line.872"></a> <span class="sourceLineNo">873</span><a name="line.873"></a> -<span class="sourceLineNo">874</span> /**<a name="line.874"></a> -<span class="sourceLineNo">875</span> * Bytes to be written to the file system, including the header. Compressed<a name="line.875"></a> -<span class="sourceLineNo">876</span> * if compression is turned on. It also includes the checksum data that<a name="line.876"></a> -<span class="sourceLineNo">877</span> * immediately follows the block data. (header + data + checksums)<a name="line.877"></a> -<span class="sourceLineNo">878</span> */<a name="line.878"></a> -<span class="sourceLineNo">879</span> private ByteArrayOutputStream onDiskBlockBytesWithHeader;<a name="line.879"></a> -<span class="sourceLineNo">880</span><a name="line.880"></a> -<span class="sourceLineNo">881</span> /**<a name="line.881"></a> -<span class="sourceLineNo">882</span> * The size of the checksum data on disk. It is used only if data is<a name="line.882"></a> -<span class="sourceLineNo">883</span> * not compressed. If data is compressed, then the checksums are already<a name="line.883"></a> -<span class="sourceLineNo">884</span> * part of onDiskBytesWithHeader. If data is uncompressed, then this<a name="line.884"></a> -<span class="sourceLineNo">885</span> * variable stores the checksum data for this block.<a name="line.885"></a> -<span class="sourceLineNo">886</span> */<a name="line.886"></a> -<span class="sourceLineNo">887</span> private byte[] onDiskChecksum = HConstants.EMPTY_BYTE_ARRAY;<a name="line.887"></a> -<span class="sourceLineNo">888</span><a name="line.888"></a> -<span class="sourceLineNo">889</span> /**<a name="line.889"></a> -<span class="sourceLineNo">890</span> * Current block's start offset in the {@link HFile}. Set in<a name="line.890"></a> -<span class="sourceLineNo">891</span> * {@link #writeHeaderAndData(FSDataOutputStream)}.<a name="line.891"></a> -<span class="sourceLineNo">892</span> */<a name="line.892"></a> -<span class="sourceLineNo">893</span> private long startOffset;<a name="line.893"></a> -<span class="sourceLineNo">894</span><a name="line.894"></a> -<span class="sourceLineNo">895</span> /**<a name="line.895"></a> -<span class="sourceLineNo">896</span> * Offset of previous block by block type. Updated when the next block is<a name="line.896"></a> -<span class="sourceLineNo">897</span> * started.<a name="line.897"></a> -<span class="sourceLineNo">898</span> */<a name="line.898"></a> -<span class="sourceLineNo">899</span> private long[] prevOffsetByType;<a name="line.899"></a> -<span class="sourceLineNo">900</span><a name="line.900"></a> -<span class="sourceLineNo">901</span> /** The offset of the previous block of the same type */<a name="line.901"></a> -<span class="sourceLineNo">902</span> private long prevOffset;<a name="line.902"></a> -<span class="sourceLineNo">903</span> /** Meta data that holds information about the hfileblock**/<a name="line.903"></a> -<span class="sourceLineNo">904</span> private HFileContext fileContext;<a name="line.904"></a> -<span class="sourceLineNo">905</span><a name="line.905"></a> -<span class="sourceLineNo">906</span> /**<a name="line.906"></a> -<span class="sourceLineNo">907</span> * @param dataBlockEncoder data block encoding algorithm to use<a name="line.907"></a> -<span class="sourceLineNo">908</span> */<a name="line.908"></a> -<span class="sourceLineNo">909</span> public Writer(HFileDataBlockEncoder dataBlockEncoder, HFileContext fileContext) {<a name="line.909"></a> -<span class="sourceLineNo">910</span> if (fileContext.getBytesPerChecksum() < HConstants.HFILEBLOCK_HEADER_SIZE) {<a name="line.910"></a> -<span class="sourceLineNo">911</span> throw new RuntimeException("Unsupported value of bytesPerChecksum. " +<a name="line.911"></a> -<span class="sourceLineNo">912</span> " Minimum is " + HConstants.HFILEBLOCK_HEADER_SIZE + " but the configured value is " +<a name="line.912"></a> -<span class="sourceLineNo">913</span> fileContext.getBytesPerChecksum());<a name="line.913"></a> -<span class="sourceLineNo">914</span> }<a name="line.914"></a> -<span class="sourceLineNo">915</span> this.dataBlockEncoder = dataBlockEncoder != null?<a name="line.915"></a> -<span class="sourceLineNo">916</span> dataBlockEncoder: NoOpDataBlockEncoder.INSTANCE;<a name="line.916"></a> -<span class="sourceLineNo">917</span> this.dataBlockEncodingCtx = this.dataBlockEncoder.<a name="line.917"></a> -<span class="sourceLineNo">918</span> newDataBlockEncodingContext(HConstants.HFILEBLOCK_DUMMY_HEADER, fileContext);<a name="line.918"></a> -<span class="sourceLineNo">919</span> // TODO: This should be lazily instantiated since we usually do NOT need this default encoder<a name="line.919"></a> -<span class="sourceLineNo">920</span> this.defaultBlockEncodingCtx = new HFileBlockDefaultEncodingContext(null,<a name="line.920"></a> -<span class="sourceLineNo">921</span> HConstants.HFILEBLOCK_DUMMY_HEADER, fileContext);<a name="line.921"></a> -<span class="sourceLineNo">922</span> // TODO: Set BAOS initial size. Use fileContext.getBlocksize() and add for header/checksum<a name="line.922"></a> -<span class="sourceLineNo">923</span> baosInMemory = new ByteArrayOutputStream();<a name="line.923"></a> -<span class="sourceLineNo">924</span> prevOffsetByType = new long[BlockType.values().length];<a name="line.924"></a> -<span class="sourceLineNo">925</span> for (int i = 0; i < prevOffsetByType.length; ++i) {<a name="line.925"></a> -<span class="sourceLineNo">926</span> prevOffsetByType[i] = UNSET;<a name="line.926"></a> -<span class="sourceLineNo">927</span> }<a name="line.927"></a> -<span class="sourceLineNo">928</span> // TODO: Why fileContext saved away when we have dataBlockEncoder and/or<a name="line.928"></a> -<span class="sourceLineNo">929</span> // defaultDataBlockEncoder?<a name="line.929"></a> -<span class="sourceLineNo">930</span> this.fileContext = fileContext;<a name="line.930"></a> -<span class="sourceLineNo">931</span> }<a name="line.931"></a> -<span class="sourceLineNo">932</span><a name="line.932"></a> -<span class="sourceLineNo">933</span> /**<a name="line.933"></a> -<span class="sourceLineNo">934</span> * Starts writing into the block. The previous block's data is discarded.<a name="line.934"></a> -<span class="sourceLineNo">935</span> *<a name="line.935"></a> -<span class="sourceLineNo">936</span> * @return the stream the user can write their data into<a name="line.936"></a> -<span class="sourceLineNo">937</span> * @throws IOException<a name="line.937"></a> -<span class="sourceLineNo">938</span> */<a name="line.938"></a> -<span class="sourceLineNo">939</span> DataOutputStream startWriting(BlockType newBlockType)<a name="line.939"></a> -<span class="sourceLineNo">940</span> throws IOException {<a name="line.940"></a> -<span class="sourceLineNo">941</span> if (state == State.BLOCK_READY && startOffset != -1) {<a name="line.941"></a> -<span class="sourceLineNo">942</span> // We had a previous block that was written to a stream at a specific<a name="line.942"></a> -<span class="sourceLineNo">943</span> // offset. Save that offset as the last offset of a block of that type.<a name="line.943"></a> -<span class="sourceLineNo">944</span> prevOffsetByType[blockType.getId()] = startOffset;<a name="line.944"></a> -<span class="sourceLineNo">945</span> }<a name="line.945"></a> -<span class="sourceLineNo">946</span><a name="line.946"></a> -<span class="sourceLineNo">947</span> startOffset = -1;<a name="line.947"></a> -<span class="sourceLineNo">948</span> blockType = newBlockType;<a name="line.948"></a> -<span class="sourceLineNo">949</span><a name="line.949"></a> -<span class="sourceLineNo">950</span> baosInMemory.reset();<a name="line.950"></a> -<span class="sourceLineNo">951</span> baosInMemory.write(HConstants.HFILEBLOCK_DUMMY_HEADER);<a name="line.951"></a> -<span class="sourceLineNo">952</span><a name="line.952"></a> -<span class="sourceLineNo">953</span> state = State.WRITING;<a name="line.953"></a> -<span class="sourceLineNo">954</span><a name="line.954"></a> -<span class="sourceLineNo">955</span> // We will compress it later in finishBlock()<a name="line.955"></a> -<span class="sourceLineNo">956</span> userDataStream = new ByteBufferWriterDataOutputStream(baosInMemory);<a name="line.956"></a> -<span class="sourceLineNo">957</span> if (newBlockType == BlockType.DATA) {<a name="line.957"></a> -<span class="sourceLineNo">958</span> this.dataBlockEncoder.startBlockEncoding(dataBlockEncodingCtx, userDataStream);<a name="line.958"></a> -<span class="sourceLineNo">959</span> }<a name="line.959"></a> -<span class="sourceLineNo">960</span> this.unencodedDataSizeWritten = 0;<a name="line.960"></a> -<span class="sourceLineNo">961</span> return userDataStream;<a name="line.961"></a> -<span class="sourceLineNo">962</span> }<a name="line.962"></a> -<span class="sourceLineNo">963</span><a name="line.963"></a> -<span class="sourceLineNo">964</span> /**<a name="line.964"></a> -<span class="sourceLineNo">965</span> * Writes the Cell to this block<a name="line.965"></a> -<span class="sourceLineNo">966</span> * @param cell<a name="line.966"></a> -<span class="sourceLineNo">967</span> * @throws IOException<a name="line.967"></a> -<span class="sourceLineNo">968</span> */<a name="line.968"></a> -<span class="sourceLineNo">969</span> void write(Cell cell) throws IOException{<a name="line.969"></a> -<span class="sourceLineNo">970</span> expectState(State.WRITING);<a name="line.970"></a> -<span class="sourceLineNo">971</span> this.unencodedDataSizeWritten +=<a name="line.971"></a> -<span class="sourceLineNo">972</span> this.dataBlockEncoder.encode(cell, dataBlockEncodingCtx, this.userDataStream);<a name="line.972"></a> -<span class="sourceLineNo">973</span> }<a name="line.973"></a> -<span class="sourceLineNo">974</span><a name="line.974"></a> -<span class="sourceLineNo">975</span> /**<a name="line.975"></a> -<span class="sourceLineNo">976</span> * Returns the stream for the user to write to. The block writer takes care<a name="line.976"></a> -<span class="sourceLineNo">977</span> * of handling compression and buffering for caching on write. Can only be<a name="line.977"></a> -<span class="sourceLineNo">978</span> * called in the "writing" state.<a name="line.978"></a> -<span class="sourceLineNo">979</span> *<a name="line.979"></a> -<span class="sourceLineNo">980</span> * @return the data output stream for the user to write to<a name="line.980"></a> -<span class="sourceLineNo">981</span> */<a name="line.981"></a> -<span class="sourceLineNo">982</span> DataOutputStream getUserDataStream() {<a name="line.982"></a> -<span class="sourceLineNo">983</span> expectState(State.WRITING);<a name="line.983"></a> -<span class="sourceLineNo">984</span> return userDataStream;<a name="line.984"></a> -<span class="sourceLineNo">985</span> }<a name="line.985"></a> -<span class="sourceLineNo">986</span><a name="line.986"></a> -<span class="sourceLineNo">987</span> /**<a name="line.987"></a> -<span class="sourceLineNo">988</span> * Transitions the block writer from the "writing" state to the "block<a name="line.988"></a> -<span class="sourceLineNo">989</span> * ready" state. Does nothing if a block is already finished.<a name="line.989"></a> -<span class="sourceLineNo">990</span> */<a name="line.990"></a> -<span class="sourceLineNo">991</span> void ensureBlockReady() throws IOException {<a name="line.991"></a> -<span class="sourceLineNo">992</span> Preconditions.checkState(state != State.INIT,<a name="line.992"></a> -<span class="sourceLineNo">993</span> "Unexpected state: " + state);<a name="line.993"></a> -<span class="sourceLineNo">994</span><a name="line.994"></a> -<span class="sourceLineNo">995</span> if (state == State.BLOCK_READY) {<a name="line.995"></a> -<span class="sourceLineNo">996</span> return;<a name="line.996"></a> -<span class="sourceLineNo">997</span> }<a name="line.997"></a> -<span class="sourceLineNo">998</span><a name="line.998"></a> -<span class="sourceLineNo">999</span> // This will set state to BLOCK_READY.<a name="line.999"></a> -<span class="sourceLineNo">1000</span> finishBlock();<a name="line.1000"></a> -<span class="sourceLineNo">1001</span> }<a name="line.1001"></a> -<span class="sourceLineNo">1002</span><a name="line.1002"></a> -<span class="sourceLineNo">1003</span> /**<a name="line.1003"></a> -<span class="sourceLineNo">1004</span> * Finish up writing of the block.<a name="line.1004"></a> -<span class="sourceLineNo">1005</span> * Flushes the compressing stream (if using compression), fills out the header,<a name="line.1005"></a> -<span class="sourceLineNo">1006</span> * does any compression/encryption of bytes to flush out to disk, and manages<a name="line.1006"></a> -<span class="sourceLineNo">1007</span> * the cache on write content, if applicable. Sets block write state to "block ready".<a name="line.1007"></a> -<span class="sourceLineNo">1008</span> */<a name="line.1008"></a> -<span class="sourceLineNo">1009</span> private void finishBlock() throws IOException {<a name="line.1009"></a> -<span class="sourceLineNo">1010</span> if (blockType == BlockType.DATA) {<a name="line.1010"></a> -<span class="sourceLineNo">1011</span> this.dataBlockEncoder.endBlockEncoding(dataBlockEncodingCtx, userDataStream,<a name="line.1011"></a> -<span class="sourceLineNo">1012</span> baosInMemory.getBuffer(), blockType);<a name="line.1012"></a> -<span class="sourceLineNo">1013</span> blockType = dataBlockEncodingCtx.getBlockType();<a name="line.1013"></a> -<span class="sourceLineNo">1014</span> }<a name="line.1014"></a> -<span class="sourceLineNo">1015</span> userDataStream.flush();<a name="line.1015"></a> -<span class="sourceLineNo">1016</span> prevOffset = prevOffsetByType[blockType.getId()];<a name="line.1016"></a> -<span class="sourceLineNo">1017</span><a name="line.1017"></a> -<span class="sourceLineNo">1018</span> // We need to set state before we can package the block up for cache-on-write. In a way, the<a name="line.1018"></a> -<span class="sourceLineNo">1019</span> // block is ready, but not yet encoded or compressed.<a name="line.1019"></a> -<span class="sourceLineNo">1020</span> state = State.BLOCK_READY;<a name="line.1020"></a> -<span class="sourceLineNo">1021</span> Bytes compressAndEncryptDat;<a name="line.1021"></a> -<span class="sourceLineNo">1022</span> if (blockType == BlockType.DATA || blockType == BlockType.ENCODED_DATA) {<a name="line.1022"></a> -<span class="sourceLineNo">1023</span> compressAndEncryptDat = dataBlockEncodingCtx.<a name="line.1023"></a> -<span class="sourceLineNo">1024</span> compressAndEncrypt(baosInMemory.getBuffer(), 0, baosInMemory.size());<a name="line.1024"></a> -<span class="sourceLineNo">1025</span> } else {<a name="line.1025"></a> -<span class="sourceLineNo">1026</span> compressAndEncryptDat = defaultBlockEncodingCtx.<a name="line.1026"></a> -<span class="sourceLineNo">1027</span> compressAndEncrypt(baosInMemory.getBuffer(), 0, baosInMemory.size());<a name="line.1027"></a> -<span class="sourceLineNo">1028</span> }<a name="line.1028"></a> -<span class="sourceLineNo">1029</span> if (compressAndEncryptDat == null) {<a name="line.1029"></a> -<span class="sourceLineNo">1030</span> compressAndEncryptDat = new Bytes(baosInMemory.getBuffer(), 0, baosInMemory.size());<a name="line.1030"></a> -<span class="sourceLineNo">1031</span> }<a name="line.1031"></a> -<span class="sourceLineNo">1032</span> if (onDiskBlockBytesWithHeader == null) {<a name="line.1032"></a> -<span class="sourceLineNo">1033</span> onDiskBlockBytesWithHeader = new ByteArrayOutputStream(compressAndEncryptDat.getLength());<a name="line.1033"></a> -<span class="sourceLineNo">1034</span> }<a name="line.1034"></a> -<span class="sourceLineNo">1035</span> onDiskBlockBytesWithHeader.reset();<a name="line.1035"></a> -<span class="sourceLineNo">1036</span> onDiskBlockBytesWithHeader.write(compressAndEncryptDat.get(),<a name="line.1036"></a> -<span class="sourceLineNo">1037</span> compressAndEncryptDat.getOffset(), compressAndEncryptDat.getLength());<a name="line.1037"></a> -<span class="sourceLineNo">1038</span> // Calculate how many bytes we need for checksum on the tail of the block.<a name="line.1038"></a> -<span class="sourceLineNo">1039</span> int numBytes = (int) ChecksumUtil.numBytes(<a name="line.1039"></a> -<span class="sourceLineNo">1040</span> onDiskBlockBytesWithHeader.size(),<a name="line.1040"></a> -<span class="sourceLineNo">1041</span> fileContext.getBytesPerChecksum());<a name="line.1041"></a> -<span class="sourceLineNo">1042</span><a name="line.1042"></a> -<span class="sourceLineNo">1043</span> // Put the header for the on disk bytes; header currently is unfilled-out<a name="line.1043"></a> -<span class="sourceLineNo">1044</span> putHeader(onDiskBlockBytesWithHeader,<a name="line.1044"></a> -<span class="sourceLineNo">1045</span> onDiskBlockBytesWithHeader.size() + numBytes,<a name="line.1045"></a> -<span class="sourceLineNo">1046</span> baosInMemory.size(), onDiskBlockBytesWithHeader.size());<a name="line.1046"></a> -<span class="sourceLineNo">1047</span> if (onDiskChecksum.length != numBytes) {<a name="line.1047"></a> -<span class="sourceLineNo">1048</span> onDiskChecksum = new byte[numBytes];<a name="line.1048"></a> -<span class="sourceLineNo">1049</span> }<a name="line.1049"></a> -<span class="sourceLineNo">1050</span> ChecksumUtil.generateChecksums(<a name="line.1050"></a> -<span class="sourceLineNo">1051</span> onDiskBlockBytesWithHeader.getBuffer(), 0,onDiskBlockBytesWithHeader.size(),<a name="line.1051"></a> -<span class="sourceLineNo">1052</span> onDiskChecksum, 0, fileContext.getChecksumType(), fileContext.getBytesPerChecksum());<a name="line.1052"></a> -<span class="sourceLineNo">1053</span> }<a name="line.1053"></a> -<span class="sourceLineNo">1054</span><a name="line.1054"></a> -<span class="sourceLineNo">1055</span> /**<a name="line.1055"></a> -<span class="sourceLineNo">1056</span> * Put the header into the given byte array at the given offset.<a name="line.1056"></a> -<span class="sourceLineNo">1057</span> * @param onDiskSize size of the block on disk header + data + checksum<a name="line.1057"></a> -<span class="sourceLineNo">1058</span> * @param uncompressedSize size of the block after decompression (but<a name="line.1058"></a> -<span class="sourceLineNo">1059</span> * before optional data block decoding) including header<a name="line.1059"></a> -<span class="sourceLineNo">1060</span> * @param onDiskDataSize size of the block on disk with header<a name="line.1060"></a> -<span class="sourceLineNo">1061</span> * and data but not including the checksums<a name="line.1061"></a> -<span class="sourceLineNo">1062</span> */<a name="line.1062"></a> -<span class="sourceLineNo">1063</span> private void putHeader(byte[] dest, int offset, int onDiskSize,<a name="line.1063"></a> -<span class="sourceLineNo">1064</span> int uncompressedSize, int onDiskDataSize) {<a name="line.1064"></a> -<span class="sourceLineNo">1065</span> offset = blockType.put(dest, offset);<a name="line.1065"></a> -<span class="sourceLineNo">1066</span> offset = Bytes.putInt(dest, offset, onDiskSize - HConstants.HFILEBLOCK_HEADER_SIZE);<a name="line.1066"></a> -<span class="sourceLineNo">1067</span> offset = Bytes.putInt(dest, offset, uncompressedSize - HConstants.HFILEBLOCK_HEADER_SIZE);<a name="line.1067"></a> -<span class="sourceLineNo">1068</span> offset = Bytes.putLong(dest, offset, prevOffset);<a name="line.1068"></a> -<span class="sourceLineNo">1069</span> offset = Bytes.putByte(dest, offset, fileContext.getChecksumType().getCode());<a name="line.1069"></a> -<span class="sourceLineNo">1070</span> offset = Bytes.putInt(dest, offset, fileContext.getBytesPerChecksum());<a name="line.1070"></a> -<span class="sourceLineNo">1071</span> Bytes.putInt(dest, offset, onDiskDataSize);<a name="line.1071"></a> -<span class="sourceLineNo">1072</span> }<a name="line.1072"></a> -<span class="sourceLineNo">1073</span><a name="line.1073"></a> -<span class="sourceLineNo">1074</span> private void putHeader(ByteArrayOutputStream dest, int onDiskSize,<a name="line.1074"></a> -<span class="sourceLineNo">1075</span> int uncompressedSize, int onDiskDataSize) {<a name="line.1075"></a> -<span class="sourceLineNo">1076</span> putHeader(dest.getBuffer(),0, onDiskSize, uncompressedSize, onDiskDataSize);<a name="line.1076"></a> -<span class="sourceLineNo">1077</span> }<a name="line.1077"></a> -<span class="sourceLineNo">1078</span><a name="line.1078"></a> -<span class="sourceLineNo">1079</span> /**<a name="line.1079"></a> -<span class="sourceLineNo">1080</span> * Similar to {@link #writeHeaderAndData(FSDataOutputStream)}, but records<a name="line.1080"></a> -<span class="sourceLineNo">1081</span> * the offset of this block so that it can be referenced in the next block<a name="line.1081"></a> -<span class="sourceLineNo">1082</span> * of the same type.<a name="line.1082"></a> -<span class="sourceLineNo">1083</span> *<a name="line.1083"></a> -<span class="sourceLineNo">1084</span> * @param out<a name="line.1084"></a> -<span class="sourceLineNo">1085</span> * @throws IOException<a name="line.1085"></a> -<span class="sourceLineNo">1086</span> */<a name="line.1086"></a> -<span class="sourceLineNo">1087</span> void writeHeaderAndData(FSDataOutputStream out) throws IOException {<a name="line.1087"></a> -<span class="sourceLineNo">1088</span> long offset = out.getPos();<a name="line.1088"></a> -<span class="sourceLineNo">1089</span> if (startOffset != UNSET && offset != startOffset) {<a name="line.1089"></a> -<span class="sourceLineNo">1090</span> throw new IOException("A " + blockType + " block written to a "<a name="line.1090"></a> -<span class="sourceLineNo">1091</span> + "stream twice, first at offset " + startOffset + ", then at "<a name="line.1091"></a> -<span class="sourceLineNo">1092</span> + offset);<a name="line.1092"></a> -<span class="sourceLineNo">1093</span> }<a name="line.1093"></a> -<span class="sourceLineNo">1094</span> startOffset = offset;<a name="line.1094"></a> -<span class="sourceLineNo">1095</span><a name="line.1095"></a> -<span class="sourceLineNo">1096</span> finishBlockAndWriteHeaderAndData((DataOutputStream) out);<a name="line.1096"></a> -<span class="sourceLineNo">1097</span> }<a name="line.1097"></a> -<span class="sourceLineNo">1098</span><a name="line.1098"></a> -<span class="sourceLineNo">1099</span> /**<a name="line.1099"></a> -<span class="sourceLineNo">1100</span> * Writes the header and the compressed data of this block (or uncompressed<a name="line.1100"></a> -<span class="sourceLineNo">1101</span> * data when not using compression) into the given stream. Can be called in<a name="line.1101"></a> -<span class="sourceLineNo">1102</span> * the "writing" state or in the "block ready" state. If called in the<a name="line.1102"></a> -<span class="sourceLineNo">1103</span> * "writing" state, transitions the writer to the "block ready" state.<a name="line.1103"></a> -<span class="sourceLineNo">1104</span> *<a name="line.1104"></a> -<span class="sourceLineNo">1105</span> * @param out the output stream to write the<a name="line.1105"></a> -<span class="sourceLineNo">1106</span> * @throws IOException<a name="line.1106"></a> -<span class="sourceLineNo">1107</span> */<a name="line.1107"></a> -<span class="sourceLineNo">1108</span> protected void finishBlockAndWriteHeaderAndData(DataOutputStream out)<a name="line.1108"></a> -<span class="sourceLineNo">1109</span> throws IOException {<a name="line.1109"></a> -<span class="sourceLineNo">1110</span> ensureBlockReady();<a name="line.1110"></a> -<span class="sourceLineNo">1111</span> out.write(onDiskBlockBytesWithHeader.getBuffer(), 0, onDiskBlockBytesWithHeader.size());<a name="line.1111"></a> -<span class="sourceLineNo">1112</span> out.write(onDiskChecksum);<a name="line.1112"></a> -<span class="sourceLineNo">1113</span> }<a name="line.1113"></a> -<span class="sourceLineNo">1114</span><a name="line.1114"></a> -<span class="sourceLineNo">1115</span> /**<a name="line.1115"></a> -<span class="sourceLineNo">1116</span> * Returns the header or the compressed data (or uncompressed data when not<a name="line.1116"></a> -<span class="sourceLineNo">1117</span> * using compression) as a byte array. Can be called in the "writing" state<a name="line.1117"></a> -<span class="sourceLineNo">1118</span> * or in the "block ready" state. If called in the "writing" state,<a name="line.1118"></a> -<span class="sourceLineNo">1119</span> * transitions the writer to the "block ready" state. This returns<a name="line.1119"></a> -<span class="sourceLineNo">1120</span> * the header + data + checksums stored on disk.<a name="line.1120"></a> -<span class="sourceLineNo">1121</span> *<a name="line.1121"></a> -<span class="sourceLineNo">1122</span> * @return header and data as they would be stored on disk in a byte array<a name="line.1122"></a> -<span class="sourceLineNo">1123</span> * @throws IOException<a name="line.1123"></a> -<span class="sourceLineNo">1124</span> */<a name="line.1124"></a> -<span class="sourceLineNo">1125</span> byte[] getHeaderAndDataForTest() throws IOException {<a name="line.1125"></a> -<span class="sourceLineNo">1126</span> ensureBlockReady();<a name="line.1126"></a> -<span class="sourceLineNo">1127</span> // This is not very optimal, because we are doing an extra copy.<a name="line.1127"></a> -<span class="sourceLineNo">1128</span> // But this method is used only by unit tests.<a name="line.1128"></a> -<span class="sourceLineNo">1129</span> byte[] output =<a name="line.1129"></a> -<span class="sourceLineNo">1130</span> new byte[onDiskBlockBytesWithHeader.size()<a name="line.1130"></a> -<span class="sourceLineNo">1131</span> + onDiskChecksum.length];<a name="line.1131"></a> -<span class="sourceLineNo">1132</span> System.arraycopy(onDiskBlockBytesWithHeader.getBuffer(), 0, output, 0,<a name="line.1132"></a> -<span class="sourceLineNo">1133</span> onDiskBlockBytesWithHeader.size());<a name="line.1133"></a> -<span class="sourceLineNo">1134</span> System.arraycopy(onDiskChecksum, 0, output,<a name="line.1134"></a> -<span class="sourceLineNo">1135</span> onDiskBlockBytesWithHeader.size(), onDiskChecksum.length);<a name="line.1135"></a> -<span class="sourceLineNo">1136</span> return output;<a name="line.1136"></a> -<span class="sourceLineNo">1137</span> }<a name="line.1137"></a> -<span class="sourceLineNo">1138</span><a name="line.1138"></a> -<span class="sourceLineNo">1139</span> /**<a name="line.1139"></a> -<span class="sourceLineNo">1140</span> * Releases resources used by this writer.<a name="line.1140"></a> -<span class="sourceLineNo">1141</span> */<a name="line.1141"></a> -<span class="sourceLineNo">1142</span> void release() {<a name="line.1142"></a> -<span class="sourceLineNo">1143</span> if (dataBlockEncodingCtx != null) {<a name="line.1143"></a> -<span class="sourceLineNo">1144</span> dataBlockEncodingCtx.close();<a name="line.1144"></a> -<span class="sourceLineNo">1145</span> dataBlockEncodingCtx = null;<a name="line.1145"></a> -<span class="sourceLineNo">1146</span> }<a name="line.1146"></a> -<span class="sourceLineNo">1147</span> if (defaultBlockEncodingCtx != null) {<a name="line.1147"></a> -<span class="sourceLineNo">1148</span> defaultBlockEncodingCtx.close();<a name="line.1148"></a> -<span class="sourceLineNo">1149</span> defaultBlockEncodingCtx = null;<a name="line.1149"></a> -<span class="sourceLineNo">1150</span> }<a name="line.1150"></a> -<span class="sourceLineNo">1151</span> }<a name="line.1151"></a> -<span class="sourceLineNo">1152</span><a name="line.1152"></a> -<span class="sourceLineNo">1153</span> /**<a name="line.1153"></a> -<span class="sourceLineNo">1154</span> * Returns the on-disk size of the data portion of the block. This is the<a name="line.1154"></a> -<span class="sourceLineNo">1155</span> * compressed size if compression is enabled. Can only be called in the<a name="line.1155"></a> -<span class="sourceLineNo">1156</span> * "block ready" state. Header is not compressed, and its size is not<a name="line.1156"></a> -<span class="sourceLineNo">1157</span> * included in the return value.<a name="line.1157"></a> -<span class="sourceLineNo">1158</span> *<a name="line.1158"></a> -<span class="sourceLineNo">1159</span> * @return the on-disk size of the block, not including the header.<a name="line.1159"></a> -<span class="sourceLineNo">1160</span> */<a name="line.1160"></a> -<span class="sourceLineNo">1161</span> int getOnDiskSizeWithoutHeader() {<a name="line.1161"></a> -<span class="sourceLineNo">1162</span> expectState(State.BLOCK_READY);<a name="line.1162"></a> -<span class="sourceLineNo">1163</span> return onDiskBlockBytesWithHeader.size() +<a name="line.1163"></a> -<span class="sourceLineNo">1164</span> onDiskChecksum.length - HConstants.HFILEBLOCK_HEADER_SIZE;<a name="line.1164"></a> -<span class="sourceLineNo">1165</span> }<a name="line.1165"></a> -<span class="sourceLineNo">1166</span><a name="line.1166"></a> -<span class="sourceLineNo">1167</span> /**<a name="line.1167"></a> -<span class="sourceLineNo">1168</span> * Returns the on-disk size of the block. Can only be called in the<a name="line.1168"></a> -<span class="sourceLineNo">1169</span> * "block ready" state.<a name="line.1169"></a> -<span class="sourceLineNo">1170</span> *<a name="line.1170"></a> -<span class="sourceLineNo">1171</span> * @return the on-disk size of the block ready to be written, including the<a name="line.1171"></a> -<span class="sourceLineNo">1172</span> * header size, the data and the checksum data.<a name="line.1172"></a> -<span class="sourceLineNo">1173</span> */<a name="line.1173"></a> -<span class="sourceLineNo">1174</span> int getOnDiskSizeWithHeader() {<a name="line.1174"></a> -<span class="sourceLineNo">1175</span> expectState(State.BLOCK_READY);<a name="line.1175"></a> -<span class="sourceLineNo">1176</span> return onDiskBlockBytesWithHeader.size() + onDiskChecksum.length;<a name="line.1176"></a> -<span class="sourceLineNo">1177</span> }<a name="line.1177"></a> -<span class="sourceLineNo">1178</span><a name="line.1178"></a> -<span class="sourceLineNo">1179</span> /**<a name="line.1179"></a> -<span class="sourceLineNo">1180</span> * The uncompressed size of the block data. Does not include header size.<a name="line.1180"></a> -<span class="sourceLineNo">1181</span> */<a name="line.1181"></a> -<span class="sourceLineNo">1182</span> int getUncompressedSizeWithoutHeader() {<a name="line.1182"></a> -<span class="sourceLineNo">1183</span> expectState(State.BLOCK_READY);<a name="line.1183"></a> -<span class="sourceLineNo">1184</span> return baosInMemory.size() - HConstants.HFILEBLOCK_HEADER_SIZE;<a name="line.1184"></a> -<span class="sourceLineNo">1185</span> }<a name="line.1185"></a> -<span class="sourceLineNo">1186</span><a name="line.1186"></a> -<span class="sourceLineNo">1187</span> /**<a name="line.1187"></a> -<span class="sourceLineNo">1188</span> * The uncompressed size of the block data, including header size.<a name="line.1188"></a> -<span class="sourceLineNo">1189</span> */<a name="line.1189"></a> -<span class="sourceLineNo">1190</span> int getUncompressedSizeWithHeader() {<a name="line.1190"></a> -<span class="sourceLineNo">1191</span> expectState(State.BLOCK_READY);<a name="line.1191"></a> -<span class="sourceLineNo">1192</span> return baosInMemory.size();<a name="line.1192"></a> -<span class="sourceLineNo">1193</span> }<a name="line.1193"></a> -<span class="sourceLineNo">1194</span><a name="line.1194"></a> -<span class="sourceLineNo">1195</span> /** @return true if a block is being written */<a name="line.1195"></a> -<span class="sourceLineNo">1196</span> boolean isWriting() {<a name="line.1196"></a> -<span class="sourceLineNo">1197</span> return state == State.WRITING;<a name="line.1197"></a> -<span class="sourceLineNo">1198</span> }<a name="line.1198"></a> -<span class="sourceLineNo">1199</span><a name="line.1199"></a> -<span class="sourceLineNo">1200</span> /**<a name="line.1200"></a> -<span class="sourceLineNo">1201</span> * Returns the number of bytes written into the current block so far, or<a name="line.1201"></a> -<span class="sourceLineNo">1202</span> * zero if not writing the block at the moment. Note that this will return<a name="line.1202"></a> -<span class="sourceLineNo">1203</span> * zero in the "block ready" state as well.<a name="line.1203"></a> -<span class="sourceLineNo">1204</span> *<a name="line.1204"></a> -<span class="sourceLineNo">1205</span> * @return the number of bytes written<a name="line.1205"></a> -<span class="sourceLineNo">1206</span> */<a name="line.1206"></a> -<span class="sourceLineNo">1207</span> int blockSizeWritten() {<a name="line.1207"></a> -<span class="sourceLineNo">1208</span> if (state != State.WRITING) return 0;<a name="line.1208"></a> -<span class="sourceLineNo">1209</span> return this.unencodedDataSizeWritten;<a name="line.1209"></a> -<span class="sourceLineNo">1210</span> }<a name="line.1210"></a> -<span class="sourceLineNo">1211</span><a name="line.1211"></a> -<span class="sourceLineNo">1212</span> /**<a name="line.1212"></a> -<span class="sourceLineNo">1213</span> * Clones the header followed by the uncompressed data, even if using<a name="line.1213"></a> -<span class="sourceLineNo">1214</span> * compression. This is needed for storing uncompressed blocks in the block<a name="line.1214"></a> -<span class="sourceLineNo">1215</span> * cache. Can be called in the "writing" state or the "block ready" state.<a name="line.1215"></a> -<span class="sourceLineNo">1216</span> * Returns only the header and data, does not include checksum data.<a name="line.1216"></a> -<span class="sourceLineNo">1217</span> *<a name="line.1217"></a> -<span class="sourceLineNo">1218</span> * @return Returns a copy of uncompressed block bytes for caching on write<a name="line.1218"></a> -<span class="sourceLineNo">1219</span> */<a name="line.1219"></a> -<span class="sourceLineNo">1220</span> @VisibleForTesting<a name="line.1220"></a> -<span class="sourceLineNo">1221</span> ByteBuffer cloneUncompressedBufferWithHeader() {<a name="line.1221"></a> -<span class="sourceLineNo">1222</span> expectState(State.BLOCK_READY);<a name="line.1222"></a> -<span class="sourceLineNo">1223</span> byte[] uncompressedBlockBytesWithHeader = baosInMemory.toByteArray();<a name="line.1223"></a> -<span class="sourceLineNo">1224</span> int numBytes = (int) ChecksumUtil.numBytes(<a name="line.1224"></a> -<span class="sourceLineNo">1225</span> onDiskBlockBytesWithHeader.size(),<a name="line.1225"></a> -<span class="sourceLineNo">1226</span> fileContext.getBytesPerChecksum());<a name="line.1226"></a> -<span class="sourceLineNo">1227</span> putHeader(uncompressedBlockBytesWithHeader, 0,<a name="line.1227"></a> -<span class="sourceLineNo">1228</span> onDiskBlockBytesWithHeader.size() + numBytes,<a name="line.1228"></a> -<span class="sourceLineNo">1229</span> baosInMemory.size(), onDiskBlockBytesWithHeader.size());<a name="line.1229"></a> -<span class="sourceLineNo">1230</span> return ByteBuffer.wrap(uncompressedBlockBytesWithHeader);<a name="line.1230"></a> -<span class="sourceLineNo">1231</span> }<a name="line.1231"></a> -<span class="sourceLineNo">1232</span><a name="line.1232"></a> -<span class="sourceLineNo">1233</span> /**<a name="line.1233"></a> -<span class="sourceLineNo">1234</span> * Clones the header followed by the on-disk (compressed/encoded/encrypted) data. This is<a name="line.1234"></a> -<span class="sourceLineNo">1235</span> * needed for storing packed blocks in the block cache. Expects calling semantics identical to<a name="line.1235"></a> -<span class="sourceLineNo">1236</span> * {@link #getUncompressedBufferWithHeader()}. Returns only the header and data,<a name="line.1236"></a> -<span class="sourceLineNo">1237</span> * Does not include checksum data.<a name="line.1237"></a> -<span class="sourceLineNo">1238</span> *<a name="line.1238"></a> -<span class="sourceLineNo">1239</span> * @return Returns a copy of block bytes for caching on write<a name="line.1239"></a> -<span class="sourceLineNo">1240</span> */<a name="line.1240"></a> -<span class="sourceLineNo">1241</span> private ByteBuffer cloneOnDiskBufferWithHeader() {<a name="line.1241"></a> +<span class="sourceLineNo">874</span> // Size of actual data being written. considering the block encoding. This<a name="line.874"></a> +<span class="sourceLineNo">875</span> // includes the header size also.<a name="line.875"></a> +<span class="sourceLineNo">876</span> private int encodedDataSizeWritten;<a name="line.876"></a> +<span class="sourceLineNo">877</span><a name="line.877"></a> +<span class="sourceLineNo">878</span> /**<a name="line.878"></a> +<span class="sourceLineNo">879</span> * Bytes to be written to the file system, including the header. Compressed<a name="line.879"></a> +<span class="sourceLineNo">880</span> * if compression is turned on. It also includes the checksum data that<a name="line.880"></a> +<span class="sourceLineNo">881</span> * immediately follows the block data. (header + data + checksums)<a name="line.881"></a> +<span class="sourceLineNo">882</span> */<a name="line.882"></a> +<span class="sourceLineNo">883</span> private ByteArrayOutputStream onDiskBlockBytesWithHeader;<a name="line.883"></a> +<span class="sourceLineNo">884</span><a name="line.884"></a> +<span class="sourceLineNo">885</span> /**<a name="line.885"></a> +<span class="sourceLineNo">886</span> * The size of the checksum data on disk. It is used only if data is<a name="line.886"></a> +<span class="sourceLineNo">887</span> * not compressed. If data is compressed, then the checksums are already<a name="line.887"></a> +<span class="sourceLineNo">888</span> * part of onDiskBytesWithHeader. If data is uncompressed, then this<a name="line.888"></a> +<span class="sourceLineNo">889</span> * variable stores the checksum data for this block.<a name="line.889"></a> +<span class="sourceLineNo">890</span> */<a name="line.890"></a> +<span class="sourceLineNo">891</span> private byte[] onDiskChecksum = HConstants.EMPTY_BYTE_ARRAY;<a name="line.891"></a> +<span class="sourceLineNo">892</span><a name="line.892"></a> +<span class="sourceLineNo">893</span> /**<a name="line.893"></a> +<span class="sourceLineNo">894</span> * Current block's start offset in the {@link HFile}. Set in<a name="line.894"></a> +<span class="sourceLineNo">895</span> * {@link #writeHeaderAndData(FSDataOutputStream)}.<a name="line.895"></a> +<span class="sourceLineNo">896</span> */<a name="line.896"></a> +<span class="sourceLineNo">897</span> private long startOffset;<a name="line.897"></a> +<span class="sourceLineNo">898</span><a name="line.898"></a> +<span class="sourceLineNo">899</span> /**<a name="line.899"></a> +<span class="sourceLineNo">900</span> * Offset of previous block by block type. Updated when the next block is<a name="line.900"></a> +<span class="sourceLineNo">901</span> * started.<a name="line.901"></a> +<span class="sourceLineNo">902</span> */<a name="line.902"></a> +<span class="sourceLineNo">903</span> private long[] prevOffsetByType;<a name="line.903"></a> +<span class="sourceLineNo">904</span><a name="line.904"></a> +<span class="sourceLineNo">905</span> /** The offset of the previous block of the same type */<a name="line.905"></a> +<span class="sourceLineNo">906</span> private long prevOffset;<a name="line.906"></a> +<span class="sourceLineNo">907</span> /** Meta data that holds information about the hfileblock**/<a name="line.907"></a> +<span class="sourceLineNo">908</span> private HFileContext fileContext;<a name="line.908"></a> +<span class="sourceLineNo">909</span><a name="line.909"></a> +<span class="sourceLineNo">910</span> /**<a name="line.910"></a> +<span class="sourceLineNo">911</span> * @param dataBlockEncoder data block encoding algorithm to use<a name="line.911"></a> +<span class="sourceLineNo">912</span> */<a name="line.912"></a> +<span class="sourceLineNo">913</span> public Writer(HFileDataBlockEncoder dataBlockEncoder, HFileContext fileContext) {<a name="line.913"></a> +<span class="sourceLineNo">914</span> if (fileContext.getBytesPerChecksum() < HConstants.HFILEBLOCK_HEADER_SIZE) {<a name="line.914"></a> +<span class="sourceLineNo">915</span> throw new RuntimeException("Unsupported value of bytesPerChecksum. " +<a name="line.915"></a> +<span class="sourceLineNo">916</span> " Minimum is " + HConstants.HFILEBLOCK_HEADER_SIZE + " but the configured value is " +<a name="line.916"></a> +<span class="sourceLineNo">917</span> fileContext.getBytesPerChecksum());<a name="line.917"></a> +<span class="sourceLineNo">918</span> }<a name="line.918"></a> +<span class="sourceLineNo">919</span> this.dataBlockEncoder = dataBlockEncoder != null?<a name="line.919"></a> +<span class="sourceLineNo">920</span> dataBlockEncoder: NoOpDataBlockEncoder.INSTANCE;<a name="line.920"></a> +<span class="sourceLineNo">921</span> this.dataBlockEncodingCtx = this.dataBlockEncoder.<a name="line.921"></a> +<span class="sourceLineNo">922</span> newDataBlockEncodingContext(HConstants.HFILEBLOCK_DUMMY_HEADER, fileContext);<a name="line.922"></a> +<span class="sourceLineNo">923</span> // TODO: This should be lazily instantiated since we usually do NOT need this default encoder<a name="line.923"></a> +<span class="sourceLineNo">924</span> this.defaultBlockEncodingCtx = new HFileBlockDefaultEncodingContext(null,<a name="line.924"></a> +<span class="sourceLineNo">925</span> HConstants.HFILEBLOCK_DUMMY_HEADER, fileContext);<a name="line.925"></a> +<span class="sourceLineNo">926</span> // TODO: Set BAOS initial size. Use fileContext.getBlocksize() and add for header/checksum<a name="line.926"></a> +<span class="sourceLineNo">927</span> baosInMemory = new ByteArrayOutputStream();<a name="line.927"></a> +<span class="sourceLineNo">928</span> prevOffsetByType = new long[BlockType.values().length];<a name="line.928"></a> +<span class="sourceLineNo">929</span> for (int i = 0; i < prevOffsetByType.length; ++i) {<a name="line.929"></a> +<span class="sourceLineNo">930</span> prevOffsetByType[i] = UNSET;<a name="line.930"></a> +<span class="sourceLineNo">931</span> }<a name="line.931"></a> +<span class="sourceLineNo">932</span> // TODO: Why fileContext saved away when we have dataBlockEncoder and/or<a name="line.932"></a> +<span class="sourceLineNo">933</span> // defaultDataBlockEncoder?<a name="line.933"></a> +<span class="sourceLineNo">934</span> this.fileContext = fileContext;<a name="line.934"></a> +<span class="sourceLineNo">935</span> }<a name="line.935"></a> +<span class="sourceLineNo">936</span><a name="line.936"></a> +<span class="sourceLineNo">937</span> /**<a name="line.937"></a> +<span class="sourceLineNo">938</span> * Starts writing into the block. The previous block's data is discarded.<a name="line.938"></a> +<span class="sourceLineNo">939</span> *<a name="line.939"></a> +<span class="sourceLineNo">940</span> * @return the stream the user can write their data into<a name="line.940"></a> +<span class="sourceLineNo">941</span> * @throws IOException<a name="line.941"></a> +<span class="sourceLineNo">942</span> */<a name="line.942"></a> +<span class="sourceLineNo">943</span> DataOutputStream startWriting(BlockType newBlockType)<a name="line.943"></a> +<span class="sourceLineNo">944</span> throws IOException {<a name="line.944"></a> +<span class="sourceLineNo">945</span> if (state == State.BLOCK_READY && startOffset != -1) {<a name="line.945"></a> +<span class="sourceLineNo">946</span> // We had a previous block that was written to a stream at a specific<a name="line.946"></a> +<span class="sourceLineNo">947</span> // offset. Save that offset as the last offset of a block of that type.<a name="line.947"></a> +<span class="sourceLineNo">948</span> prevOffsetByType[blockType.getId()] = startOffset;<a name="line.948"></a> +<span class="sourceLineNo">949</span> }<a name="line.949"></a> +<span class="sourceLineNo">950</span><a name="line.950"></a> +<span class="sourceLineNo">951</span> startOffset = -1;<a name="line.951"></a> +<span class="sourceLineNo">952</span> blockType = newBlockType;<a name="line.952"></a> +<span class="sourceLineNo">953</span><a name="line.953"></a> +<span class="sourceLineNo">954</span> baosInMemory.reset();<a name="line.954"></a> +<span class="sourceLineNo">955</span> baosInMemory.write(HConstants.HFILEBLOCK_DUMMY_HEADER);<a name="line.955"></a> +<span class="sourceLineNo">956</span><a name="line.956"></a> +<span class="sourceLineNo">957</span> state = State.WRITING;<a name="line.957"></a> +<span class="sourceLineNo">958</span><a name="line.958"></a> +<span class="sourceLineNo">959</span> // We will compress it later in finishBlock()<a name="line.959"></a> +<span class="sourceLineNo">960</span> userDataStream = new ByteBufferWriterDataOutputStream(baosInMemory);<a name="line.960"></a> +<span class="sourceLineNo">961</span> if (newBlockType == BlockType.DATA) {<a name="line.961"></a> +<span class="sourceLineNo">962</span> this.dataBlockEncoder.startBlockEncoding(dataBlockEncodingCtx, userDataStream);<a name="line.962"></a> +<span class="sourceLineNo">963</span> }<a name="line.963"></a> +<span class="sourceLineNo">964</span> this.unencodedDataSizeWritten = 0;<a name="line.964"></a> +<span class="sourceLineNo">965</span> this.encodedDataSizeWritten = 0;<a name="line.965"></a> +<span class="sourceLineNo">966</span> return userDataStream;<a name="line.966"></a> +<span class="sourceLineNo">967</span> }<a name="line.967"></a> +<span class="sourceLineNo">968</span><a name="line.968"></a> +<span class="sourceLineNo">969</span> /**<a name="line.969"></a> +<span class="sourceLineNo">970</span> * Writes the Cell to this block<a name="line.970"></a> +<span class="sourceLineNo">971</span> * @param cell<a name="line.971"></a> +<span class="sourceLineNo">972</span> * @throws IOException<a name="line.972"></a> +<span class="sourceLineNo">973</span> */<a name="line.973"></a> +<span class="sourceLineNo">974</span> void write(Cell cell) throws IOException{<a name="line.974"></a> +<span class="sourceLineNo">975</span> expectState(State.WRITING);<a name="line.975"></a> +<span class="sourceLineNo">976</span> int posBeforeEncode = this.userDataStream.size();<a name="line.976"></a> +<span class="sourceLineNo">977</span> this.unencodedDataSizeWritten +=<a name="line.977"></a> +<span class="sourceLineNo">978</span> this.dataBlockEncoder.encode(cell, dataBlockEncodingCtx, this.userDataStream);<a name="line.978"></a> +<span class="sourceLineNo">979</span> this.encodedDataSizeWritten += this.userDataStream.size() - posBeforeEncode;<a name="line.979"></a> +<span class="sourceLineNo">980</span> }<a name="line.980"></a> +<span class="sourceLineNo">981</span><a name="line.981"></a> +<span class="sourceLineNo">982</span> /**<a name="line.982"></a> +<span class="sourceLineNo">983</span> * Returns the stream for the user to write to. The block writer takes care<a name="line.983"></a> +<span class="sourceLineNo">984</span> * of handling compression and buffering for caching on write. Can only be<a name="line.984"></a> +<span class="sourceLineNo">985</span> * called in the "writing" state.<a name="line.985"></a> +<span class="sourceLineNo">986</span> *<a name="line.986"></a> +<span class="sourceLineNo">987</span> * @return the data output stream for the user to write to<a name="line.987"></a> +<span class="sourceLineNo">988</span> */<a name="line.988"></a> +<span class="sourceLineNo">989</span> DataOutputStream getUserDataStream() {<a name="line.989"></a> +<span class="sourceLineNo">990</span> expectState(State.WRITING);<a name="line.990"></a> +<span class="sourceLineNo">991</span> return userDataStream;<a name="line.991"></a> +<span class="sourceLineNo">992</span> }<a name="line.992"></a> +<span class="sourceLineNo">993</span><a name="line.993"></a> +<span class="sourceLineNo">994</span> /**<a name="line.994"></a> +<span class="sourceLineNo">995</span> * Transitions the block writer from the "writing" state to the "block<a name="line.995"></a> +<span class="sourceLineNo">996</span> * ready" state. Does nothing if a block is already finished.<a name="line.996"></a> +<span class="sourceLineNo">997</span> */<a name="line.997"></a> +<span class="sourceLineNo">998</span> void ensureBlockReady() throws IOException {<a name="line.998"></a> +<span class="sourceLineNo">999</span> Preconditions.checkState(state != State.INIT,<a name="line.999"></a> +<span class="sourceLineNo">1000</span> "Unexpected state: " + state);<a name="line.1000"></a> +<span class="sourceLineNo">1001</span><a name="line.1001"></a> +<span class="sourceLineNo">1002</span> if (state == State.BLOCK_READY) {<a name="line.1002"></a> +<span class="sourceLineNo">1003</span> return;<a name="line.1003"></a> +<span class="sourceLineNo">1004</span> }<a name="line.1004"></a> +<span class="sourceLineNo">1005</span><a name="line.1005"></a> +<span class="sourceLineNo">1006</span> // This will set state to BLOCK_READY.<a name="line.1006"></a> +<span class="sourceLineNo">1007</span> finishBlock();<a name="line.1007"></a> +<span class="sourceLineNo">1008</span> }<a name="line.1008"></a> +<span class="sourceLineNo">1009</span><a name="line.1009"></a> +<span class="sourceLineNo">1010</span> /**<a name="line.1010"></a> +<span class="sourceLineNo">1011</span> * Finish up writing of the block.<a name="line.1011"></a> +<span class="sourceLineNo">1012</span> * Flushes the compressing stream (if using compression), fills out the header,<a name="line.1012"></a> +<span class="sourceLineNo">1013</span> * does any compression/encryption of bytes to flush out to disk, and manages<a name="line.1013"></a> +<span class="sourceLineNo">1014</span> * the cache on write content, if applicable. Sets block write state to "block ready".<a name="line.1014"></a> +<span class="sourceLineNo">1015</span> */<a name="line.1015"></a> +<span class="sourceLineNo">1016</span> private void finishBlock() throws IOException {<a name="line.1016"></a> +<span class="sourceLineNo">1017</span> if (blockType == BlockType.DATA) {<a name="line.1017"></a> +<span class="sourceLineNo">1018</span> this.dataBlockEncoder.endBlockEncoding(dataBlockEncodingCtx, userDataStream,<a name="line.1018"></a> +<span class="sourceLineNo">1019</span> baosInMemory.getBuffer(), blockType);<a name="line.1019"></a> +<span class="sourceLineNo">1020</span> blockType = dataBlockEncodingCtx.getBlockType();<a name="line.1020"></a> +<span class="sourceLineNo">1021</span> }<a name="line.1021"></a> +<span class="sourceLineNo">1022</span> userDataStream.flush();<a name="line.1022"></a> +<span class="sourceLineNo">1023</span> prevOffset = prevOffsetByType[blockType.getId()];<a name="line.1023"></a> +<span class="sourceLineNo">1024</span><a name="line.1024"></a> +<span class="sourceLineNo">1025</span> // We need to set state before we can package the block up for cache-on-write. In a way, the<a name="line.1025"></a> +<span class="sourceLineNo">1026</span> // block is ready, but not yet encoded or compressed.<a name="line.1026"></a> +<span class="sourceLineNo">1027</span> state = State.BLOCK_READY;<a name="line.1027"></a> +<span class="sourceLineNo">1028</span> Bytes compressAndEncryptDat;<a name="line.1028"></a> +<span class="sourceLineNo">1029</span> if (blockType == BlockType.DATA || blockType == BlockType.ENCODED_DATA) {<a name="line.1029"></a> +<span class="sourceLineNo">1030</span> compressAndEncryptDat = dataBlockEncodingCtx.<a name="line.1030"></a> +<span class="sourceLineNo">1031</span> compressAndEncrypt(baosInMemory.getBuffer(), 0, baosInMemory.size());<a name="line.1031"></a> +<span class="sourceLineNo">1032</span> } else {<a name="line.1032"></a> +<span class="sourceLineNo">1033</span> compressAndEncryptDat = defaultBlockEncodingCtx.<a name="line.1033"></a> +<span class="sourceLineNo">1034</span> compressAndEncrypt(baosInMemory.getBuffer(), 0, baosInMemory.size());<a name="line.1034"></a> +<span class="sourceLineNo">1035</span> }<a name="line.1035"></a> +<span class="sourceLineNo">1036</span> if (compressAndEncryptDat == null) {<a name="line.1036"></a> +<span class="sourceLineNo">1037</span> compressAndEncryptDat = new Bytes(baosInMemory.getBuffer(), 0, baosInMemory.size());<a name="line.1037"></a> +<span class="sourceLineNo">1038</span> }<a name="line.1038"></a> +<span class="sourceLineNo">1039</span> if (onDiskBlockBytesWithHeader == null) {<a name="line.1039"></a> +<span class="sourceLineNo">1040</span> onDiskBlockBytesWithHeader = new ByteArrayOutputStream(compressAndEncryptDat.getLength());<a name="line.1040"></a> +<span class="sourceLineNo">1041</span> }<a name="line.1041"></a> +<span class="sourceLineNo">1042</span> onDiskBlockBytesWithHeader.reset();<a name="line.1042"></a> +<span class="sourceLineNo">1043</span> onDiskBlockBytesWithHeader.write(compressAndEncryptDat.get(),<a name="line.1043"></a> +<span class="sourceLineNo">1044</span> compressAndEncryptDat.getOffset(), compressAndEncryptDat.getLength());<a name="line.1044"></a> +<span class="sourceLineNo">1045</span> // Calculate how many bytes we need for checksum on the tail of the block.<a name="line.1045"></a> +<span class="sourceLineNo">1046</span> int numBytes = (int) ChecksumUtil.numBytes(<a name="line.1046"></a> +<span class="sourceLineNo">1047</span> onDiskBlockBytesWithHeader.size(),<a name="line.1047"></a> +<span class="sourceLineNo">1048</span> fileContext.getBytesPerChecksum());<a name="line.1048"></a> +<span class="sourceLineNo">1049</span><a name="line.1049"></a> +<span class="sourceLineNo">1050</span> // Put the header for the on disk bytes; header currently is unfilled-out<a name="line.1050"></a> +<span class="sourceLineNo">1051</span> putHeader(onDiskBlockBytesWithHeader,<a name="line.1051"></a> +<span class="sourceLineNo">1052</span> onDiskBlockBytesWithHeader.size() + numBytes,<a name="line.1052"></a> +<span class="sourceLineNo">1053</span> baosInMemory.size(), onDiskBlockBytesWithHeader.size());<a name="line.1053"></a> +<span class="sourceLineNo">1054</span> if (onDiskChecksum.length != numBytes) {<a name="line.1054"></a> +<span class="sourceLineNo">1055</span> onDiskChecksum = new byte[numBytes];<a name="line.1055"></a> +<span class="sourceLineNo">1056</span> }<a name="line.1056"></a> +<span class="sourceLineNo">1057</span> ChecksumUtil.generateChecksums(<a name="line.1057"></a> +<span class="sourceLineNo">1058</span> onDiskBlockBytesWithHeader.getBuffer(), 0,onDiskBlockBytesWithHeader.size(),<a name="line.1058"></a> +<span class="sourceLineNo">1059</span> onDiskChecksum, 0, fileContext.getChecksumType(), fileContext.getBytesPerChecksum());<a name="line.1059"></a> +<span class="sourceLineNo">1060</span> }<a name="line.1060"></a> +<span class="sourceLineNo">1061</span><a name="line.1061"></a> +<span class="sourceLineNo">1062</span> /**<a name="line.1062"></a> +<span class="sourceLineNo">1063</span> * Put the header into the given byte array at the given offset.<a name="line.1063"></a> +<span class="sourceLineNo">1064</span> * @param onDiskSize size of the block on disk header + data + checksum<a name="line.1064"></a> +<span class="sourceLineNo">1065</span> * @param uncompressedSize size of the block after decompression (but<a name="line.1065"></a> +<span class="sourceLineNo">1066</span> * before optional data block decoding) including header<a name="line.1066"></a> +<span class="sourceLineNo">1067</span> * @param onDiskDataSize size of the block on disk with header<a name="line.1067"></a> +<span class="sourceLineNo">1068</span> * and data but not including the checksums<a name="line.1068"></a> +<span class="sourceLineNo">1069</span> */<a name="line.1069"></a> +<span class="sourceLineNo">1070</span> private void putHeader(byte[] dest, int offset, int onDiskSize,<a name="line.1070"></a> +<span class="sourceLineNo">1071</span> int uncompressedSize, int onDiskDataSize) {<a name="line.1071"></a> +<span class="sourceLineNo">1072</span> offset = blockType.put(dest, offset);<a name="line.1072"></a> +<span class="sourceLineNo">1073</span> offset = Bytes.putInt(dest, offset, onDiskSize - HConstants.HFILEBLOCK_HEADER_SIZE);<a name="line.1073"></a> +<span class="sourceLineNo">1074</span> offset = Bytes.putInt(dest, offset, uncompressedSize - HConstants.HFILEBLOCK_HEADER_SIZE);<a name="line.1074"></a> +<span class="sourceLineNo">1075</span> offset = Bytes.putLong(dest, offset, prevOffset);<a name="line.1075"></a> +<span class="sourceLineNo">1076</span> offset = Bytes.putByte(dest, offset, fileContext.getChecksumType().getCode());<a name="line.1076"></a> +<span class="sourceLineNo">1077</span> offset = Bytes.putInt(dest, offset, fileContext.getBytesPerChecksum());<a name="line.1077"></a> +<span class="sourceLineNo">1078</span> Bytes.putInt(dest, offset, onDiskDataSize);<a name="line.1078"></a> +<span class="sourceLineNo">1079</span> }<a name="line.1079"></a> +<span class="sourceLineNo">1080</span><a name="line.1080"></a> +<span class="sourceLineNo">1081</span> private void putHeader(ByteArrayOutputStream dest, int onDiskSize,<a name="line.1081"></a> +<span class="sourceLineNo">1082</span> int uncompressedSize, int onDiskDataSize) {<a name="line.1082"></a> +<span class="sourceLineNo">1083</span> putHeader(dest.getBuffer(),0, onDiskSize, uncompressedSize, onDiskDataSize);<a name="line.1083"></a> +<span class="sourceLineNo">1084</span> }<a name="line.1084"></a> +<span class="sourceLineNo">1085</span><a name="line.1085"></a> +<span class="sourceLineNo">1086</span> /**<a name="line.1086"></a> +<span class="sourceLineNo">1087</span> * Similar to {@link #writeHeaderAndData(FSDataOutputStream)}, but records<a name="line.1087"></a> +<span class="sourceLineNo">1088</span> * the offset of this block so that it can be referenced in the next block<a name="line.1088"></a> +<span class="sourceLineNo">1089</span> * of the same type.<a name="line.1089"></a> +<span class="sourceLineNo">1090</span> *<a name="line.1090"></a> +<span class="sourceLineNo">1091</span> * @param out<a name="line.1091"></a> +<span class="sourceLineNo">1092</span> * @throws IOException<a name="line.1092"></a> +<span class="sourceLineNo">1093</span> */<a name="line.1093"></a> +<span class="sourceLineNo">1094</span> void writeHeaderAndData(FSDataOutputStream out) throws IOException {<a name="line.1094"></a> +<span class="sourceLineNo">1095</span> long offset = out.getPos();<a name="line.1095"></a> +<span class="sourceLineNo">1096</span> if (startOffset != UNSET && offset != startOffset) {<a name="line.1096"></a> +<span class="sourceLineNo">1097</span> throw new IOException("A " + blockType + " block written to a "<a name="line.1097"></a> +<span class="sourceLineNo">1098</span> + "stream twice, first at offset " + startOffset + ", then at "<a name="line.1098"></a> +<span class="sourceLineNo">1099</span> + offset);<a name="line.1099"></a> +<span class="sourceLineNo">1100</span> }<a name="line.1100"></a> +<span class="sourceLineNo">1101</span> startOffset = offset;<a name="line.1101"></a> +<span class="sourceLineNo">1102</span><a name="line.1102"></a> +<span class="sourceLineNo">1103</span> finishBlockAndWriteHeaderAndData((DataOutputStream) out);<a name="line.1103"></a> +<span class="sourceLineNo">1104</span> }<a name="line.1104"></a> +<span class="sourceLineNo">1105</span><a name="line.1105"></a> +<span class="sourceLineNo">1106</span> /**<a name="line.1106"></a> +<span class="sourceLineNo">1107</span> * Writes the header and the compressed data of this block (or uncompressed<a name="line.1107"></a> +<span class="sourceLineNo">1108</span> * data when not using compression) into the given stream. Can be called in<a name="line.1108"></a> +<span class="sourceLineNo">1109</span> * the "writing" state or in the "block ready" state. If called in the<a name="line.1109"></a> +<span class="sourceLineNo">1110</span> * "writing" state, transitions the writer to the "block ready" state.<a name="line.1110"></a> +<span class="sourceLineNo">1111</span> *<a name="line.1111"></a> +<span class="sourceLineNo">1112</span> * @param out the output stream to write the<a name="line.1112"></a> +<span class="sourceLineNo">1113</span> * @throws IOException<a name="line.1113"></a> +<span class="sourceLineNo">1114</span> */<a name="line.1114"></a> +<span class="sourceLineNo">1115</span> protected void finishBlockAndWriteHeaderAndData(DataOutputStream out)<a name="line.1115"></a> +<span class="sourceLineNo">1116</span> throws IOException {<a name="line.1116"></a> +<span class="sourceLineNo">1117</span> ensureBlockReady();<a name="line.1117"></a> +<span class="sourceLineNo">1118</span> out.write(onDiskBlockBytesWithHeader.getBuffer(), 0, onDiskBlockBytesWithHeader.size());<a name="line.1118"></a> +<span class="sourceLineNo">1119</span> out.write(onDiskChecksum);<a name="line.1119"></a> +<span class="sourceLineNo">1120</span> }<a name="line.1120"></a> +<span class="sourceLineNo">1121</span><a name="line.1121"></a> +<span class="sourceLineNo">1122</span> /**<a name="line.1122"></a> +<span class="sourceLineNo">1123</span> * Returns the header or the compressed data (or uncompressed data when not<a name="line.1123"></a> +<span class="sourceLineNo">1124</span> * using compression) as a byte array. Can be called in the "writing" state<a name="line.1124"></a> +<span class="sourceLineNo">1125</span> * or in the "block ready" state. If called in the "writing" state,<a name="line.1125"></a> +<span class="sourceLineNo">1126</span> * transitions the writer to the "block ready" state. This returns<a name="line.1126"></a> +<span class="sourceLineNo">1127</span> * the header + data + checksums stored on disk.<a name="line.1127"></a> +<span class="sourceLineNo">1128</span> *<a name="line.1128"></a> +<span class="sourceLineNo">1129</span> * @return header and data as they would be stored on disk in a byte array<a name="line.1129"></a> +<span class="sourceLineNo">1130</span> * @throws IOException<a name="line.1130"></a> +<span class="sourceLineNo">1131</span> */<a name="line.1131"></a> +<span class="sourceLineNo">1132</span> byte[] getHeaderAndDataForTest() throws IOException {<a name="line.1132"></a> +<span class="sourceLineNo">1133</span> ensureBlockReady();<a name="line.1133"></a> +<span class="sourceLineNo">1134</span> // This is not very optimal, because we are doing an extra copy.<a name="line.1134"></a> +<span class="sourceLineNo">1135</span> // But this method is used only by unit tests.<a name="line.1135"></a> +<span class="sourceLineNo">1136</span> byte[] output =<a name="line.1136"></a> +<span class="sourceLineNo">1137</span> new byte[onDiskBlockBytesWithHeader.size()<a name="line.1137"></a> +<span class="sourceLineNo">1138</span> + onDiskChecksum.length];<a name="line.1138"></a> +<span class="sourceLineNo">1139</span> System.arraycopy(onDiskBlockBytesWithHeader.getBuffer(), 0, output, 0,<a name="line.1139"></a> +<span class="sourceLineNo">1140</span> onDiskBlockBytesWithHeader.size());<a name="line.1140"></a> +<span class="sourceLineNo">1141</span> System.arraycopy(onDiskChecksum, 0, output,<a name="line.1141"></a> +<span class="sourceLineNo">1142</span> onDiskBlockBytesWithHeader.size(), onDiskChecksum.length);<a name="line.1142"></a> +<span class="sourceLineNo">1143</span> return output;<a name="line.1143"></a> +<span class="sourceLineNo">1144</span> }<a name="line.1144"></a> +<span class="sourceLineNo">1145</span><a name="line.1145"></a> +<span class="sourceLineNo">1146</span> /**<a name="line.1146"></a> +<span class="sourceLineNo">1147</span> * Releases resources used by this writer.<a name="line.1147"></a> +<span class="sourceLineNo">1148</span> */<a name="line.1148"></a> +<span class="sourceLineNo">1149</span> void release() {<a name="line.1149"></a> +<span class="sourceLineNo">1150</span> if (dataBlockEncodingCtx != null) {<a name="line.1150"></a> +<span class="sourceLineNo">1151</span> dataBlockEncodingCtx.close();<a name="line.1151"></a> +<span class="sourceLineNo">1152</span> dataBlockEncodingCtx = null;<a name="line.1152"></a> +<span class="sourceLineNo">1153</span> }<a name="line.1153"></a> +<span class="sourceLineNo">1154</span> if (defaultBlockEncodingCtx != null) {<a name="line.1154"></a> +<span class="sourceLineNo">1155</span> defaultBlockEncodingCtx.close();<a name="line.1155"></a> +<span class="sourceLineNo">1156</span> defaultBlockEncodingCtx = null;<a name="line.1156"></a> +<span class="sourceLineNo">1157</span> }<a name="line.1157"></a> +<span class="sourceLineNo">1158</span> }<a name="line.1158"></a> +<span class="sourceLineNo">1159</span><a name="line.1159"></a> +<span class="sourceLineNo">1160</span> /**<a name="line.1160"></a> +<span class="sourceLineNo">1161</span> * Returns the on-disk size of the data portion of the block. This is the<a name="line.1161"></a> +<span class="sourceLineNo">1162</span> * compressed size if compression is enabled. Can only be called in the<a name="line.1162"></a> +<span class="sourceLineNo">1163</span> * "block ready" state. Header is not compressed, and its size is not<a name="line.1163"></a> +<span class="sourceLineNo">1164</span> * included in the return value.<a name="line.1164"></a> +<span class="sourceLineNo">1165</span> *<a name="line.1165"></a> +<span class="sourceLineNo">1166</span> * @return the on-disk size of the block, not including the header.<a name="line.1166"></a> +<span class="sourceLineNo">1167</span> */<a name="line.1167"></a> +<span class="sourceLineNo">1168</span> int getOnDiskSizeWithoutHeader() {<a name="line.1168"></a> +<span class="sourceLineNo">1169</span> expectState(State.BLOCK_READY);<a name="line.1169"></a> +<span class="sourceLineNo">1170</span> return onDiskBlockBytesWithHeader.size() +<a name="line.1170"></a> +<span class="sourceLineNo">1171</span> onDiskChecksum.length - HConstants.HFILEBLOCK_HEADER_SIZE;<a name="line.1171"></a> +<span class="sourceLineNo">1172</span> }<a name="line.1172"></a> +<span class="sourceLineNo">1173</span><a name="line.1173"></a> +<span class="sourceLineNo">1174</span> /**<a name="line.1174"></a> +<span class="sourceLineNo">1175</span> * Returns the on-disk size of the block. Can only be called in the<a name="line.1175"></a> +<span class="sourceLineNo">1176</span> * "block ready" state.<a name="line.1176"></a> +<span class="sourceLineNo">1177</span> *<a name="line.1177"></a> +<span class="sourceLineNo">1178</span> * @return the on-disk size of the block ready to be written, including the<a name="line.1178"></a> +<span class="sourceLineNo">1179</span> * header size, the data and the checksum data.<a name="line.1179"></a> +<span class="sourceLineNo">1180</span> */<a name="line.1180"></a> +<span class="sourceLineNo">1181</span> int getOnDiskSizeWithHeader() {<a name="line.1181"></a> +<span class="sourceLineNo">1182</span> expectState(State.BLOCK_READY);<a name="line.1182"></a> +<span class="sourceLineNo">1183</span> return onDiskBlockBytesWithHeader.size() + onDiskChecksum.length;<a name="line.1183"></a> +<span class="sourceLineNo">1184</span> }<a name="line.1184"></a> +<span class="sourceLineNo">1185</span><a name="line.1185"></a> +<span class="sourceLineNo">1186</span> /**<a name="line.1186"></a> +<span class="sourceLineNo">1187</span> * The uncompressed size of the block data. Does not include header size.<a name="line.1187"></a> +<span class="sourceLineNo">1188</span> */<a name="line.1188"></a> +<span class="sourceLineNo">1189</span> int getUncompressedSizeWithoutHeader() {<a name="line.1189"></a> +<span class="sourceLineNo">1190</span> expectState(State.BLOCK_READY);<a name="line.1190"></a> +<span class="sourceLineNo">1191</span> return baosInMemory.size() - HConstants.HFILEBLOCK_HEADER_SIZE;<a name="line.1191"></a> +<span class="sourceLineNo">1192</span> }<a name="line.1192"></a> +<span class="sourceLineNo">1193</span><a name="line.1193"></a> +<span class="sourceLineNo">1194</span> /**<a name="line.1194"></a> +<span class="sourceLineNo">1195</span> * The uncompressed size of the block data, including header size.<a name="line.1195"></a> +<span class="sourceLineNo">1196</span> */<a name="line.1196"></a> +<span class="sourceLineNo">1197</span> int getUncompressedSizeWithHeader() {<a name="line.1197"></a> +<span class="sourceLineNo">1198</span> expectState(State.BLOCK_READY);<a name="line.1198"></a> +<span class="sourceLineNo">1199</span> return baosInMemory.size();<a name="line.1199"></a> +<span class="sourceLineNo">1200</span> }<a name="line.1200"></a> +<span class="sourceLineNo">1201</span><a name="line.1201"></a> +<span class="sourceLineNo">1202</span> /** @return true if a block is being written */<a name="line.1202"></a> +<span class="sourceLineNo">1203</span> boolean isWriting() {<a name="line.1203"></a> +<span class="sourceLineNo">1204</span> return state == State.WRITING;<a name="line.1204"></a> +<span class="sourceLineNo">1205</span> }<a name="line.1205"></a> +<span class="sourceLineNo">1206</span><a name="line.1206"></a> +<span class="sourceLineNo">1207</span> /**<a name="line.1207"></a> +<span class="sourceLineNo">1208</span> * Returns the number of bytes written into the current block so far, or<a name="line.1208"></a> +<span class="sourceLineNo">1209</span> * zero if not writing the block at the moment. Note that this will return<a name="line.1209"></a> +<span class="sourceLineNo">1210</span> * zero in the "block ready" state as well.<a name="line.1210"></a> +<span class="sourceLineNo">1211</span> *<a name="line.1211"></a> +<span class="sourceLineNo">1212</span> * @return the number of bytes written<a name="line.1212"></a> +<span class="sourceLineNo">1213</span> */<a name="line.1213"></a> +<span class="sourceLineNo">1214</span> public int encodedBlockSizeWritten() {<a name="line.1214"></a> +<span class="sourceLineNo">1215</span> if (state != State.WRITING)<a name="line.1215"></a> +<span class="sourceLineNo">1216</span> return 0;<a name="line.1216"></a> +<span class="sourceLineNo">1217</span> return this.encodedDataSizeWritten;<a name="line.1217"></a> +<span class="sourceLineNo">1218</span> }<a name="line.1218"></a> +<span class="sourceLineNo">1219</span><a name="line.1219"></a> +<span class="sourceLineNo">1220</span> /**<a name="line.1220"></a> +<span class="sourceLineNo">1221</span> * Returns the number of bytes written into the current block so far, or<a name="line.1221"></a> +<span class="sourceLineNo">1222</span> * zero if not writing the block at the moment. Note that this will return<a name="line.1222"></a> +<span class="sourceLineNo">1223</span> * zero in the "block ready" state as well.<a name="line.1223"></a> +<span class="sourceLineNo">1224</span> *<a name="line.1224"></a> +<span class="sourceLineNo">1225</span> * @return the number of bytes written<a name="line.1225"></a> +<span class="sourceLineNo">1226</span> */<a name="line.1226"></a> +<span class="sourceLineNo">1227</span> int blockSizeWritten() {<a name="line.1227"></a> +<span class="sourceLineNo">1228</span> if (state != State.WRITING) return 0;<a name="line.1228"></a> +<span class="sourceLineNo">1229</span> return this.unencodedDataSizeWritten;<a name="line.1229"></a> +<span class="sourceLineNo">1230</span> }<a name="line.1230"></a> +<span class="sourceLineNo">1231</span><a name="line.1231"></a> +<span class="sourceLineNo">1232</span> /**<a name="line.1232"></a> +<span class="sourceLineNo">1233</span> * Clones the header followed by the uncompressed data, even if using<a name="line.1233"></a> +<span class="sourceLineNo">1234</span> * compression. This is needed for storing uncompressed blocks in the block<a name="line.1234"></a> +<span class="sourceLineNo">1235</span> * cache. Can be called in the "writing" state or the "block ready" state.<a name="line.1235"></a> +<span class="sourceLineNo">1236</span> * Returns only the header and data, does not include checksum data.<a name="line.1236"></a> +<span class="sourceLineNo">1237</span> *<a name="line.1237"></a> +<span class="sourceLineNo">1238</span> * @return Returns a copy of uncompressed block bytes for caching on write<a name="line.1238"></a> +<span class="sourceLineNo">1239</span> */<a name="line.1239"></a> +<span class="sourceLineNo">1240</span> @VisibleForTesting<a name="line.1240"></a> +<span class="sourceLineNo">1241</span> ByteBuffer cloneUncompressedBufferWithHeader() {<a name="line.1241"></a> <span class="sourceLineNo">1242</span> expectState(State.BLOCK_READY);<a name="line.1242"></a> -<span class="sourceLineNo">1243</span> return ByteBuffer.wrap(onDiskBlockBytesWithHeader.toByteArray());<a name="line.1243"></a> -<span class="sourceLineNo">1244</span> }<a name="line.1244"></a> -<span class="sourceLineNo">1245</span><a name="line.1245"></a> -<span class="sourceLineNo">1246</span> private void expectState(State expectedState) {<a name="line.1246"></a> -<span class="sourceLineNo">1247</span> if (state != expectedState) {<a name="line.1247"></a> -<span class="sourceLineNo">1248</span> throw new IllegalStateException("Expected state: " + expectedState +<a name="line.1248"></a> -<span class="sourceLineNo">1249</span> ", actual state: " + state);<a name="line.1249"></a> -<span class="sourceLineNo">1250</span> }<a name="line.1250"></a> +<span class="sourceLineNo">1243</span> byte[] uncompressedBlockBytesWithHeader = baosInMemory.toByteArray();<a name="line.1243"></a> +<span class="sourceLineNo">1244</span> int numBytes = (int) ChecksumUtil.numBytes(<a name="line.1244"></a> +<span class="sourceLineNo">1245</span> onDiskBlockBytesWithHeader.size(),<a name="line.1245"></a> +<span class="sourceLineNo">1246</span> fileContext.getBytesPerChecksum());<a name="line.1246"></a> +<span class="sourceLineNo">1247</span> putHeader(uncompressedBlockBytesWithHeader, 0,<a name="line.1247"></a> +<span class="sourceLineNo">1248</span> onDiskBlockBytesWithHeader.size() + numBytes,<a name="line.1248"></a> +<span class="sourceLineNo">1249</span> baosInMemory.size(), onDiskBlockBytesWithHeader.size());<a name="line.1249"></a> +<span class="sourceLineNo">1250</span> return ByteBuffer.wrap(uncompressedBlockBytesWithHeader);<a name="line.1250"></a> <span class="sourceLineNo">1251</span> }<a name="line.1251"></a> <span class="sourceLineNo">1252</span><a name="line.1252"></a> <span class="sourceLineNo">1253</span> /**<a name="line.1253"></a> -<span class="sourceLineNo">1254</span> * Takes the given {@link BlockWritable} instance, creates a new block of<a name="line.1254"></a> -<span class="sourceLineNo">1255</span> * its appropriate type, writes the writable into this block, and flushes<a name="line.1255"></a> -<span class="sourceLineNo">1256</span> * the block into the output stream. The writer is instructed not to buffer<a name="line.1256"></a> -<span class="sourceLineNo">1257</span> * uncompressed bytes for cache-on-write.<a name="line.1257"></a> +<span class="sourceLineNo">1254</span> * Clones the header followed by the on-disk (compressed/encoded/encrypted) data. This is<a name="line.1254"></a> +<span class="sourceLineNo">1255</span> * needed for storing packed blocks in the block cache. Expects calling semantics identical to<a name="line.1255"></a> +<span class="sourceLineNo">1256</span> * {@link #getUncompressedBufferWithHeader()}. Returns only the header and data,<a name="line.1256"></a> +<span class="sourceLineNo">1257</span> * Does not include checksum data.<a name="line.1257"></a> <span class="sourceLineNo">1258</span> *<a name="line.1258"></a> -<span class="sourceLineNo">1259</span> * @param bw the block-writable object to write as a block<a name="line.1259"></a> -<span class="sourceLineNo">1260</span> * @param out the file system output stream<a name="line.1260"></a> -<span class="sourceLineNo">1261</span> * @throws IOException<a name="line.1261"></a> -<span class="sourceLineNo">1262</span> */<a name="line.1262"></a> -<span class="sourceLineNo">1263</span> void writeBlock(BlockWritable bw, FSDataOutputStream out)<a name="line.1263"></a> -<span class="sourceLineNo">1264</span> throws IOException {<a name="line.1264"></a> -<span class="sourceLineNo">1265</span> bw.writeToBlock(startWriting(bw.getBlockType()));<a name="line.1265"></a> -<span class="sourceLineNo">1266</span> writeHeaderAndData(out);<a name="line.1266"></a> -<span class="sourceLineNo">1267</span> }<a name="line.1267"></a> -<span class="sourceLineNo">1268</span><a name="line.1268"></a> -<span class="sourceLineNo">1269</span> /**<a name="line.1269"></a> -<span class="sourceLineNo">1270</span> * Creates a new HFileBlock. Checksums have already been validated, so<a name="line.1270"></a> -<span class="sourceLineNo">1271</span> * the byte buffer passed into the constructor of this newly created<a name="line.1271"></a> -<span class="sourceLineNo">1272</span> * block does not have checksum data even though the header minor<a name="line.1272"></a> -<span class="sourceLineNo">1273</span> * version is MINOR_VERSION_WITH_CHECKSUM. This is indicated by setting a<a name="line.1273"></a> -<span class="sourceLineNo">1274</span> * 0 value in bytesPerChecksum. This method copies the on-disk or<a name="line.1274"></a> -<span class="sourceLineNo">1275</span> * uncompressed data to build the HFileBlock which is used only<a name="line.1275"></a> -<span class="sourceLineNo">1276</span> * while writing blocks and caching.<a name="line.1276"></a> -<span class="sourceLineNo">1277</span> *<a name="line.1277"></a> -<span class="sourceLineNo">1278</span> * <p>TODO: Should there be an option where a cache can ask that hbase preserve block<a name="line.1278"></a> -<span class="sourceLineNo">1279</span> * checksums for checking after a block comes out of the cache? Otehrwise, cache is responsible<a name="line.1279"></a> -<span class="sourceLineNo">1280</span> * for blocks being wholesome (ECC memory or if file-backed, it does checksumming).<a name="line.1280"></a> -<span class="sourceLineNo">1281</span> */<a name="line.1281"></a> -<span class="sourceLineNo">1282</span> HFileBlock getBlockForCaching(CacheConfig cacheConf) {<a name="line.1282"></a> -<span class="sourceLineNo">1283</span> HFileContext newContext = new HFileContextBuilder()<a name="line.1283"></a> -<span class="sourceLineNo">1284</span> .withBlockSize(fileContext.getBlocksize())<a name="line.1284"></a> -<span class="sourceLineNo">1285</span> .withBytesPerCheckSum(0)<a name="line.1285"></a> -<span class="sourceLineNo">1286</span> .withChecksumType(ChecksumType.NULL) // no checksums in cached data<a name="line.1286"></a> -<span class="sourceLineNo">1287</span> .withCompression(fileContext.getCompression())<a name="line.1287"></a> -<span class="sourceLineNo">1288</span> .withDataBlockEncoding(fileContext.getDataBlockEncoding())<a name="line.1288"></a> -<span class="sourceLineNo">1289</span> .withHBaseCheckSum(fileContext.isUseHBaseChecksum())<a name="line.1289"></a> -<span class="sourceLineNo">1290</span> .withCompressTags(fileContext.isCompressTags())<a name="line.1290"></a> -<span class="sourceLineNo">1291</span> .withIncludesMvcc(fileContext.isIncludesMvcc())<a name="line.1291"></a> -<span class="sourceLineNo">1292</span> .withIncludesTags(fileContext.isIncludesTags())<a name="line.1292"></a> -<span class="sourceLineNo">1293</span> .build();<a name="line.1293"></a> -<span class="sourceLineNo">1294</span> return new HFileBlock(blockType, getOnDiskSizeWithoutHeader(),<a name="line.1294"></a> -<span class="sourceLineNo">1295</span> getUncompressedSizeWithoutHeader(), prevOffset,<a name="line.1295"></a> -<span class="sourceLineNo">1296</span> cacheConf.shouldCacheCompressed(blockType.getCategory())?<a name="line.1296"></a> -<span class="sourceLineNo">1297</span> cloneOnDiskBufferWithHeader() :<a name="line.1297"></a> -<span class="sourceLineNo">1298</span> cloneUncompressedBufferWithHeader(),<a name="line.1298"></a> -<span class="sourceLineNo">1299</span> FILL_HEADER, startOffset, UNSET,<a name="line.1299"></a> -<span class="sourceLineNo">1300</span> onDiskBlockBytesWithHeader.size() + onDiskChecksum.length, newContext);<a name="line.1300"></a> -<span class="sourceLineNo">1301</span> }<a name="line.1301"></a> -<span class="sourceLineNo">1302</span> }<a name="line.1302"></a> -<span class="sourceLineNo">1303</span><a name="line.1303"></a> -<span class="sourceLineNo">1304</span> /** Something that can be written into a block. */<a name="line.1304"></a> -<span class="sourceLineNo">1305</span> interface BlockWritable {<a name="line.1305"></a> -<span class="sourceLineNo">1306</span><a name="line.1306"></a> -<span class="sourceLineNo">1307</span> /** The type of block this data should use. */<a name="line.1307"></a> -<span class="sourceLineNo">1308</span> BlockType getBlockType();<a name="line.1308"></a> -<span class="sourceLineNo">1309</span><a name="line.1309"></a> -<span class="sourceLineNo">1310</span> /**<a name="line.1310"></a> -<span class="sourceLineNo">1311</span> * Writes the block to the provided stream. Must not write any magic<a name="line.1311"></a> -<span class="sourceLineNo">1312</span> * records.<a name="line.1312"></a> -<span class="sourceLineNo">1313</span> *<a name="line.1313"></a> -<span class="sourceLineNo">1314</span> * @param out a stream to write uncompressed data into<a name="line.1314"></a> -<span class="sourceLineNo">1315</span> */<a name="line.1315"></a> -<span class="sourceLineNo">1316</span> void writeToBlock(DataOutput out) throws IOException;<a name="line.1316"></a> -<span class="sourceLineNo">1317</span> }<a name="line.1317"></a> -<span class="sourceLineNo">1318</span><a name="line.1318"></a> -<span class="sourceLineNo">1319</span> // Block readers and writers<a name="line.1319"></a> -<span class="sourceLineNo">1320</span><a name="line.1320"></a> -<span class="sourceLineNo">1321</span> /** An interface allowing to iterate {@link HFileBlock}s. */<a name="line.1321"></a> -<span class="sourceLineNo">1322</span> interface BlockIterator {<a name="line.1322"></a> +<span class="sourceLineNo">1259</span> * @return Returns a copy of block bytes for caching on write<a name="line.1259"></a> +<span class="sourceLineNo">1260</span> */<a name="line.1260"></a> +<span class="sourceLineNo">1261</span> private ByteBuffer cloneOnDiskBufferWithHeader() {<a name="line.1261"></a> +<span class="sourceLineNo">1262</span> expectState(State.BLOCK_READY);<a name="line.1262"></a> +<span class="sourceLineNo">1263</span> return ByteBuffer.wrap(onDiskBlockBytesWithHeader.toByteArray());<a name="line.1263"></a> +<span class="sourceLineNo">1264</span> }<a name="line.1264"></a> +<span class="sourceLineNo">1265</span><a name="line.1265"></a> +<span class="sourceLineNo">1266</span> private void expectState(State expectedState) {<a name="line.1266"></a> +<span class="sourceLineNo">1267</span> if (state != expectedState) {<a name="line.1267"></a> +<span class="sourceLineNo">1268</span> throw new IllegalStateException("Expected state: " + expectedState +<a name="line.1268"></a> +<span class="sourceLineNo">1269</span> ", actual state: " + state);<a name="line.1269"></a> +<span class="sourceLineNo">1270</span> }<a name="line.1270"></a> +<span class="sourceLineNo">1271</span> }<a name="line.1271"></a> +<span class="sourceLineNo">1272</span><a name="line.1272"></a> +<span class="sourceLineNo">1273</span> /**<a name="line.1273"></a> +<span class="sourceLineNo">1274</span> * Takes the given {@link BlockWritable} instance, creates a new block of<a name="line.1274"></a> +<span class="sourceLineNo">1275</span> * its appropriate type, writes the writable into this block, and flushes<a name="line.1275"></a> +<span class="sourceLineNo">1276</span> * the block into the output stream. The writer is instructed not to buffer<a name="line.1276"></a> +<span class="sourceLineNo">1277</span> * uncompressed bytes for cache-on-write.<a name="line.1277"></a> +<span class="sourceLineNo">1278</span> *<a name="line.1278"></a> +<span class="sourceLineNo">1279</span> * @param bw the block-writable object to write as a block<a name="line.1279"></a> +<span class="sourceLineNo">1280</span> * @param out the file system output stream<a name="line.1280"></a> +<span class="sourceLineNo">1281</span> * @throws IOException<a name="line.1281"></a> +<span class="sourceLineNo">1282</span> */<a name="line.1282"></a> +<span class="sourceLineNo">1283</span> void writeBlock(BlockWritable bw, FSDataOutputStream out)<a name="line.1283"></a> +<span class="sourceLineNo">1284</span> throws IOException {<a name="line.1284"></a> +<span class="sourceLineNo">1285</span> bw.writeToBlock(startWriting(bw.getBlockType()));<a name="line.1285"></a> +<span class="sourceLineNo">1286</span> writeHeaderAndData(out);<a name="line.1286"></a
<TRUNCATED>