http://git-wip-us.apache.org/repos/asf/hbase-site/blob/7cf6034b/devapidocs/src-html/org/apache/hadoop/hbase/io/hfile/HFileBlock.BlockDeserializer.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/io/hfile/HFileBlock.BlockDeserializer.html b/devapidocs/src-html/org/apache/hadoop/hbase/io/hfile/HFileBlock.BlockDeserializer.html new file mode 100644 index 0000000..3d1edb3 --- /dev/null +++ b/devapidocs/src-html/org/apache/hadoop/hbase/io/hfile/HFileBlock.BlockDeserializer.html @@ -0,0 +1,2186 @@ +<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd";> +<html lang="en"> +<head> +<title>Source code</title> +<link rel="stylesheet" type="text/css" href="../../../../../../../stylesheet.css" title="Style"> +</head> +<body> +<div class="sourceContainer"> +<pre><span class="sourceLineNo">001</span>/*<a name="line.1"></a> +<span class="sourceLineNo">002</span> * Licensed to the Apache Software Foundation (ASF) under one<a name="line.2"></a> +<span class="sourceLineNo">003</span> * or more contributor license agreements. See the NOTICE file<a name="line.3"></a> +<span class="sourceLineNo">004</span> * distributed with this work for additional information<a name="line.4"></a> +<span class="sourceLineNo">005</span> * regarding copyright ownership. The ASF licenses this file<a name="line.5"></a> +<span class="sourceLineNo">006</span> * to you under the Apache License, Version 2.0 (the<a name="line.6"></a> +<span class="sourceLineNo">007</span> * "License"); you may not use this file except in compliance<a name="line.7"></a> +<span class="sourceLineNo">008</span> * with the License. You may obtain a copy of the License at<a name="line.8"></a> +<span class="sourceLineNo">009</span> *<a name="line.9"></a> +<span class="sourceLineNo">010</span> * http://www.apache.org/licenses/LICENSE-2.0<a name="line.10"></a> +<span class="sourceLineNo">011</span> *<a name="line.11"></a> +<span class="sourceLineNo">012</span> * Unless required by applicable law or agreed to in writing, software<a name="line.12"></a> +<span class="sourceLineNo">013</span> * distributed under the License is distributed on an "AS IS" BASIS,<a name="line.13"></a> +<span class="sourceLineNo">014</span> * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.<a name="line.14"></a> +<span class="sourceLineNo">015</span> * See the License for the specific language governing permissions and<a name="line.15"></a> +<span class="sourceLineNo">016</span> * limitations under the License.<a name="line.16"></a> +<span class="sourceLineNo">017</span> */<a name="line.17"></a> +<span class="sourceLineNo">018</span>package org.apache.hadoop.hbase.io.hfile;<a name="line.18"></a> +<span class="sourceLineNo">019</span><a name="line.19"></a> +<span class="sourceLineNo">020</span>import java.io.DataInputStream;<a name="line.20"></a> +<span class="sourceLineNo">021</span>import java.io.DataOutput;<a name="line.21"></a> +<span class="sourceLineNo">022</span>import java.io.DataOutputStream;<a name="line.22"></a> +<span class="sourceLineNo">023</span>import java.io.IOException;<a name="line.23"></a> +<span class="sourceLineNo">024</span>import java.io.InputStream;<a name="line.24"></a> +<span class="sourceLineNo">025</span>import java.nio.ByteBuffer;<a name="line.25"></a> +<span class="sourceLineNo">026</span>import java.util.concurrent.atomic.AtomicReference;<a name="line.26"></a> +<span class="sourceLineNo">027</span>import java.util.concurrent.locks.Lock;<a name="line.27"></a> +<span class="sourceLineNo">028</span>import java.util.concurrent.locks.ReentrantLock;<a name="line.28"></a> +<span class="sourceLineNo">029</span><a name="line.29"></a> +<span class="sourceLineNo">030</span>import org.apache.hadoop.fs.FSDataInputStream;<a name="line.30"></a> +<span class="sourceLineNo">031</span>import org.apache.hadoop.fs.FSDataOutputStream;<a name="line.31"></a> +<span class="sourceLineNo">032</span>import org.apache.hadoop.fs.Path;<a name="line.32"></a> +<span class="sourceLineNo">033</span>import org.apache.hadoop.hbase.Cell;<a name="line.33"></a> +<span class="sourceLineNo">034</span>import org.apache.hadoop.hbase.HConstants;<a name="line.34"></a> +<span class="sourceLineNo">035</span>import org.apache.yetus.audience.InterfaceAudience;<a name="line.35"></a> +<span class="sourceLineNo">036</span>import org.slf4j.Logger;<a name="line.36"></a> +<span class="sourceLineNo">037</span>import org.slf4j.LoggerFactory;<a name="line.37"></a> +<span class="sourceLineNo">038</span>import org.apache.hadoop.hbase.fs.HFileSystem;<a name="line.38"></a> +<span class="sourceLineNo">039</span>import org.apache.hadoop.hbase.io.ByteArrayOutputStream;<a name="line.39"></a> +<span class="sourceLineNo">040</span>import org.apache.hadoop.hbase.io.ByteBuffInputStream;<a name="line.40"></a> +<span class="sourceLineNo">041</span>import org.apache.hadoop.hbase.io.ByteBufferWriterDataOutputStream;<a name="line.41"></a> +<span class="sourceLineNo">042</span>import org.apache.hadoop.hbase.io.FSDataInputStreamWrapper;<a name="line.42"></a> +<span class="sourceLineNo">043</span>import org.apache.hadoop.hbase.io.encoding.DataBlockEncoding;<a name="line.43"></a> +<span class="sourceLineNo">044</span>import org.apache.hadoop.hbase.io.encoding.HFileBlockDecodingContext;<a name="line.44"></a> +<span class="sourceLineNo">045</span>import org.apache.hadoop.hbase.io.encoding.HFileBlockDefaultDecodingContext;<a name="line.45"></a> +<span class="sourceLineNo">046</span>import org.apache.hadoop.hbase.io.encoding.HFileBlockDefaultEncodingContext;<a name="line.46"></a> +<span class="sourceLineNo">047</span>import org.apache.hadoop.hbase.io.encoding.HFileBlockEncodingContext;<a name="line.47"></a> +<span class="sourceLineNo">048</span>import org.apache.hadoop.hbase.nio.ByteBuff;<a name="line.48"></a> +<span class="sourceLineNo">049</span>import org.apache.hadoop.hbase.nio.MultiByteBuff;<a name="line.49"></a> +<span class="sourceLineNo">050</span>import org.apache.hadoop.hbase.nio.SingleByteBuff;<a name="line.50"></a> +<span class="sourceLineNo">051</span>import org.apache.hadoop.hbase.util.Bytes;<a name="line.51"></a> +<span class="sourceLineNo">052</span>import org.apache.hadoop.hbase.util.ChecksumType;<a name="line.52"></a> +<span class="sourceLineNo">053</span>import org.apache.hadoop.hbase.util.ClassSize;<a name="line.53"></a> +<span class="sourceLineNo">054</span>import org.apache.hadoop.io.IOUtils;<a name="line.54"></a> +<span class="sourceLineNo">055</span><a name="line.55"></a> +<span class="sourceLineNo">056</span>import org.apache.hbase.thirdparty.com.google.common.annotations.VisibleForTesting;<a name="line.56"></a> +<span class="sourceLineNo">057</span>import org.apache.hbase.thirdparty.com.google.common.base.Preconditions;<a name="line.57"></a> +<span class="sourceLineNo">058</span><a name="line.58"></a> +<span class="sourceLineNo">059</span>/**<a name="line.59"></a> +<span class="sourceLineNo">060</span> * Cacheable Blocks of an {@link HFile} version 2 file.<a name="line.60"></a> +<span class="sourceLineNo">061</span> * Version 2 was introduced in hbase-0.92.0.<a name="line.61"></a> +<span class="sourceLineNo">062</span> *<a name="line.62"></a> +<span class="sourceLineNo">063</span> * <p>Version 1 was the original file block. Version 2 was introduced when we changed the hbase file<a name="line.63"></a> +<span class="sourceLineNo">064</span> * format to support multi-level block indexes and compound bloom filters (HBASE-3857). Support<a name="line.64"></a> +<span class="sourceLineNo">065</span> * for Version 1 was removed in hbase-1.3.0.<a name="line.65"></a> +<span class="sourceLineNo">066</span> *<a name="line.66"></a> +<span class="sourceLineNo">067</span> * <h3>HFileBlock: Version 2</h3><a name="line.67"></a> +<span class="sourceLineNo">068</span> * In version 2, a block is structured as follows:<a name="line.68"></a> +<span class="sourceLineNo">069</span> * <ul><a name="line.69"></a> +<span class="sourceLineNo">070</span> * <li><b>Header:</b> See Writer#putHeader() for where header is written; header total size is<a name="line.70"></a> +<span class="sourceLineNo">071</span> * HFILEBLOCK_HEADER_SIZE<a name="line.71"></a> +<span class="sourceLineNo">072</span> * <ul><a name="line.72"></a> +<span class="sourceLineNo">073</span> * <li>0. blockType: Magic record identifying the {@link BlockType} (8 bytes):<a name="line.73"></a> +<span class="sourceLineNo">074</span> * e.g. <code>DATABLK*</code><a name="line.74"></a> +<span class="sourceLineNo">075</span> * <li>1. onDiskSizeWithoutHeader: Compressed -- a.k.a 'on disk' -- block size, excluding header,<a name="line.75"></a> +<span class="sourceLineNo">076</span> * but including tailing checksum bytes (4 bytes)<a name="line.76"></a> +<span class="sourceLineNo">077</span> * <li>2. uncompressedSizeWithoutHeader: Uncompressed block size, excluding header, and excluding<a name="line.77"></a> +<span class="sourceLineNo">078</span> * checksum bytes (4 bytes)<a name="line.78"></a> +<span class="sourceLineNo">079</span> * <li>3. prevBlockOffset: The offset of the previous block of the same type (8 bytes). This is<a name="line.79"></a> +<span class="sourceLineNo">080</span> * used to navigate to the previous block without having to go to the block index<a name="line.80"></a> +<span class="sourceLineNo">081</span> * <li>4: For minorVersions &gt;=1, the ordinal describing checksum type (1 byte)<a name="line.81"></a> +<span class="sourceLineNo">082</span> * <li>5: For minorVersions &gt;=1, the number of data bytes/checksum chunk (4 bytes)<a name="line.82"></a> +<span class="sourceLineNo">083</span> * <li>6: onDiskDataSizeWithHeader: For minorVersions &gt;=1, the size of data 'on disk', including<a name="line.83"></a> +<span class="sourceLineNo">084</span> * header, excluding checksums (4 bytes)<a name="line.84"></a> +<span class="sourceLineNo">085</span> * </ul><a name="line.85"></a> +<span class="sourceLineNo">086</span> * </li><a name="line.86"></a> +<span class="sourceLineNo">087</span> * <li><b>Raw/Compressed/Encrypted/Encoded data:</b> The compression<a name="line.87"></a> +<span class="sourceLineNo">088</span> * algorithm is the same for all the blocks in an {@link HFile}. If compression is NONE, this is<a name="line.88"></a> +<span class="sourceLineNo">089</span> * just raw, serialized Cells.<a name="line.89"></a> +<span class="sourceLineNo">090</span> * <li><b>Tail:</b> For minorVersions &gt;=1, a series of 4 byte checksums, one each for<a name="line.90"></a> +<span class="sourceLineNo">091</span> * the number of bytes specified by bytesPerChecksum.<a name="line.91"></a> +<span class="sourceLineNo">092</span> * </ul><a name="line.92"></a> +<span class="sourceLineNo">093</span> *<a name="line.93"></a> +<span class="sourceLineNo">094</span> * <h3>Caching</h3><a name="line.94"></a> +<span class="sourceLineNo">095</span> * Caches cache whole blocks with trailing checksums if any. We then tag on some metadata, the<a name="line.95"></a> +<span class="sourceLineNo">096</span> * content of BLOCK_METADATA_SPACE which will be flag on if we are doing 'hbase'<a name="line.96"></a> +<span class="sourceLineNo">097</span> * checksums and then the offset into the file which is needed when we re-make a cache key<a name="line.97"></a> +<span class="sourceLineNo">098</span> * when we return the block to the cache as 'done'.<a name="line.98"></a> +<span class="sourceLineNo">099</span> * See {@link Cacheable#serialize(ByteBuffer, boolean)} and {@link Cacheable#getDeserializer()}.<a name="line.99"></a> +<span class="sourceLineNo">100</span> *<a name="line.100"></a> +<span class="sourceLineNo">101</span> * <p>TODO: Should we cache the checksums? Down in Writer#getBlockForCaching(CacheConfig) where<a name="line.101"></a> +<span class="sourceLineNo">102</span> * we make a block to cache-on-write, there is an attempt at turning off checksums. This is not the<a name="line.102"></a> +<span class="sourceLineNo">103</span> * only place we get blocks to cache. We also will cache the raw return from an hdfs read. In this<a name="line.103"></a> +<span class="sourceLineNo">104</span> * case, the checksums may be present. If the cache is backed by something that doesn't do ECC,<a name="line.104"></a> +<span class="sourceLineNo">105</span> * say an SSD, we might want to preserve checksums. For now this is open question.<a name="line.105"></a> +<span class="sourceLineNo">106</span> * <p>TODO: Over in BucketCache, we save a block allocation by doing a custom serialization.<a name="line.106"></a> +<span class="sourceLineNo">107</span> * Be sure to change it if serialization changes in here. Could we add a method here that takes an<a name="line.107"></a> +<span class="sourceLineNo">108</span> * IOEngine and that then serializes to it rather than expose our internals over in BucketCache?<a name="line.108"></a> +<span class="sourceLineNo">109</span> * IOEngine is in the bucket subpackage. Pull it up? Then this class knows about bucketcache. Ugh.<a name="line.109"></a> +<span class="sourceLineNo">110</span> */<a name="line.110"></a> +<span class="sourceLineNo">111</span>@InterfaceAudience.Private<a name="line.111"></a> +<span class="sourceLineNo">112</span>public class HFileBlock implements Cacheable {<a name="line.112"></a> +<span class="sourceLineNo">113</span> private static final Logger LOG = LoggerFactory.getLogger(HFileBlock.class);<a name="line.113"></a> +<span class="sourceLineNo">114</span><a name="line.114"></a> +<span class="sourceLineNo">115</span> // Block Header fields.<a name="line.115"></a> +<span class="sourceLineNo">116</span><a name="line.116"></a> +<span class="sourceLineNo">117</span> // TODO: encapsulate Header related logic in this inner class.<a name="line.117"></a> +<span class="sourceLineNo">118</span> static class Header {<a name="line.118"></a> +<span class="sourceLineNo">119</span> // Format of header is:<a name="line.119"></a> +<span class="sourceLineNo">120</span> // 8 bytes - block magic<a name="line.120"></a> +<span class="sourceLineNo">121</span> // 4 bytes int - onDiskSizeWithoutHeader<a name="line.121"></a> +<span class="sourceLineNo">122</span> // 4 bytes int - uncompressedSizeWithoutHeader<a name="line.122"></a> +<span class="sourceLineNo">123</span> // 8 bytes long - prevBlockOffset<a name="line.123"></a> +<span class="sourceLineNo">124</span> // The following 3 are only present if header contains checksum information<a name="line.124"></a> +<span class="sourceLineNo">125</span> // 1 byte - checksum type<a name="line.125"></a> +<span class="sourceLineNo">126</span> // 4 byte int - bytes per checksum<a name="line.126"></a> +<span class="sourceLineNo">127</span> // 4 byte int - onDiskDataSizeWithHeader<a name="line.127"></a> +<span class="sourceLineNo">128</span> static int BLOCK_MAGIC_INDEX = 0;<a name="line.128"></a> +<span class="sourceLineNo">129</span> static int ON_DISK_SIZE_WITHOUT_HEADER_INDEX = 8;<a name="line.129"></a> +<span class="sourceLineNo">130</span> static int UNCOMPRESSED_SIZE_WITHOUT_HEADER_INDEX = 12;<a name="line.130"></a> +<span class="sourceLineNo">131</span> static int PREV_BLOCK_OFFSET_INDEX = 16;<a name="line.131"></a> +<span class="sourceLineNo">132</span> static int CHECKSUM_TYPE_INDEX = 24;<a name="line.132"></a> +<span class="sourceLineNo">133</span> static int BYTES_PER_CHECKSUM_INDEX = 25;<a name="line.133"></a> +<span class="sourceLineNo">134</span> static int ON_DISK_DATA_SIZE_WITH_HEADER_INDEX = 29;<a name="line.134"></a> +<span class="sourceLineNo">135</span> }<a name="line.135"></a> +<span class="sourceLineNo">136</span><a name="line.136"></a> +<span class="sourceLineNo">137</span> /** Type of block. Header field 0. */<a name="line.137"></a> +<span class="sourceLineNo">138</span> private BlockType blockType;<a name="line.138"></a> +<span class="sourceLineNo">139</span><a name="line.139"></a> +<span class="sourceLineNo">140</span> /**<a name="line.140"></a> +<span class="sourceLineNo">141</span> * Size on disk excluding header, including checksum. Header field 1.<a name="line.141"></a> +<span class="sourceLineNo">142</span> * @see Writer#putHeader(byte[], int, int, int, int)<a name="line.142"></a> +<span class="sourceLineNo">143</span> */<a name="line.143"></a> +<span class="sourceLineNo">144</span> private int onDiskSizeWithoutHeader;<a name="line.144"></a> +<span class="sourceLineNo">145</span><a name="line.145"></a> +<span class="sourceLineNo">146</span> /**<a name="line.146"></a> +<span class="sourceLineNo">147</span> * Size of pure data. Does not include header or checksums. Header field 2.<a name="line.147"></a> +<span class="sourceLineNo">148</span> * @see Writer#putHeader(byte[], int, int, int, int)<a name="line.148"></a> +<span class="sourceLineNo">149</span> */<a name="line.149"></a> +<span class="sourceLineNo">150</span> private int uncompressedSizeWithoutHeader;<a name="line.150"></a> +<span class="sourceLineNo">151</span><a name="line.151"></a> +<span class="sourceLineNo">152</span> /**<a name="line.152"></a> +<span class="sourceLineNo">153</span> * The offset of the previous block on disk. Header field 3.<a name="line.153"></a> +<span class="sourceLineNo">154</span> * @see Writer#putHeader(byte[], int, int, int, int)<a name="line.154"></a> +<span class="sourceLineNo">155</span> */<a name="line.155"></a> +<span class="sourceLineNo">156</span> private long prevBlockOffset;<a name="line.156"></a> +<span class="sourceLineNo">157</span><a name="line.157"></a> +<span class="sourceLineNo">158</span> /**<a name="line.158"></a> +<span class="sourceLineNo">159</span> * Size on disk of header + data. Excludes checksum. Header field 6,<a name="line.159"></a> +<span class="sourceLineNo">160</span> * OR calculated from {@link #onDiskSizeWithoutHeader} when using HDFS checksum.<a name="line.160"></a> +<span class="sourceLineNo">161</span> * @see Writer#putHeader(byte[], int, int, int, int)<a name="line.161"></a> +<span class="sourceLineNo">162</span> */<a name="line.162"></a> +<span class="sourceLineNo">163</span> private int onDiskDataSizeWithHeader;<a name="line.163"></a> +<span class="sourceLineNo">164</span> // End of Block Header fields.<a name="line.164"></a> +<span class="sourceLineNo">165</span><a name="line.165"></a> +<span class="sourceLineNo">166</span> /**<a name="line.166"></a> +<span class="sourceLineNo">167</span> * The in-memory representation of the hfile block. Can be on or offheap. Can be backed by<a name="line.167"></a> +<span class="sourceLineNo">168</span> * a single ByteBuffer or by many. Make no assumptions.<a name="line.168"></a> +<span class="sourceLineNo">169</span> *<a name="line.169"></a> +<span class="sourceLineNo">170</span> * <p>Be careful reading from this <code>buf</code>. Duplicate and work on the duplicate or if<a name="line.170"></a> +<span class="sourceLineNo">171</span> * not, be sure to reset position and limit else trouble down the road.<a name="line.171"></a> +<span class="sourceLineNo">172</span> *<a name="line.172"></a> +<span class="sourceLineNo">173</span> * <p>TODO: Make this read-only once made.<a name="line.173"></a> +<span class="sourceLineNo">174</span> *<a name="line.174"></a> +<span class="sourceLineNo">175</span> * <p>We are using the ByteBuff type. ByteBuffer is not extensible yet we need to be able to have<a name="line.175"></a> +<span class="sourceLineNo">176</span> * a ByteBuffer-like API across multiple ByteBuffers reading from a cache such as BucketCache.<a name="line.176"></a> +<span class="sourceLineNo">177</span> * So, we have this ByteBuff type. Unfortunately, it is spread all about HFileBlock. Would be<a name="line.177"></a> +<span class="sourceLineNo">178</span> * good if could be confined to cache-use only but hard-to-do.<a name="line.178"></a> +<span class="sourceLineNo">179</span> */<a name="line.179"></a> +<span class="sourceLineNo">180</span> private ByteBuff buf;<a name="line.180"></a> +<span class="sourceLineNo">181</span><a name="line.181"></a> +<span class="sourceLineNo">182</span> /** Meta data that holds meta information on the hfileblock.<a name="line.182"></a> +<span class="sourceLineNo">183</span> */<a name="line.183"></a> +<span class="sourceLineNo">184</span> private HFileContext fileContext;<a name="line.184"></a> +<span class="sourceLineNo">185</span><a name="line.185"></a> +<span class="sourceLineNo">186</span> /**<a name="line.186"></a> +<span class="sourceLineNo">187</span> * The offset of this block in the file. Populated by the reader for<a name="line.187"></a> +<span class="sourceLineNo">188</span> * convenience of access. This offset is not part of the block header.<a name="line.188"></a> +<span class="sourceLineNo">189</span> */<a name="line.189"></a> +<span class="sourceLineNo">190</span> private long offset = UNSET;<a name="line.190"></a> +<span class="sourceLineNo">191</span><a name="line.191"></a> +<span class="sourceLineNo">192</span> private MemoryType memType = MemoryType.EXCLUSIVE;<a name="line.192"></a> +<span class="sourceLineNo">193</span><a name="line.193"></a> +<span class="sourceLineNo">194</span> /**<a name="line.194"></a> +<span class="sourceLineNo">195</span> * The on-disk size of the next block, including the header and checksums if present.<a name="line.195"></a> +<span class="sourceLineNo">196</span> * UNSET if unknown.<a name="line.196"></a> +<span class="sourceLineNo">197</span> *<a name="line.197"></a> +<span class="sourceLineNo">198</span> * Blocks try to carry the size of the next block to read in this data member. Usually<a name="line.198"></a> +<span class="sourceLineNo">199</span> * we get block sizes from the hfile index but sometimes the index is not available:<a name="line.199"></a> +<span class="sourceLineNo">200</span> * e.g. when we read the indexes themselves (indexes are stored in blocks, we do not<a name="line.200"></a> +<span class="sourceLineNo">201</span> * have an index for the indexes). Saves seeks especially around file open when<a name="line.201"></a> +<span class="sourceLineNo">202</span> * there is a flurry of reading in hfile metadata.<a name="line.202"></a> +<span class="sourceLineNo">203</span> */<a name="line.203"></a> +<span class="sourceLineNo">204</span> private int nextBlockOnDiskSize = UNSET;<a name="line.204"></a> +<span class="sourceLineNo">205</span><a name="line.205"></a> +<span class="sourceLineNo">206</span> /**<a name="line.206"></a> +<span class="sourceLineNo">207</span> * On a checksum failure, do these many succeeding read requests using hdfs checksums before<a name="line.207"></a> +<span class="sourceLineNo">208</span> * auto-reenabling hbase checksum verification.<a name="line.208"></a> +<span class="sourceLineNo">209</span> */<a name="line.209"></a> +<span class="sourceLineNo">210</span> static final int CHECKSUM_VERIFICATION_NUM_IO_THRESHOLD = 3;<a name="line.210"></a> +<span class="sourceLineNo">211</span><a name="line.211"></a> +<span class="sourceLineNo">212</span> private static int UNSET = -1;<a name="line.212"></a> +<span class="sourceLineNo">213</span> public static final boolean FILL_HEADER = true;<a name="line.213"></a> +<span class="sourceLineNo">214</span> public static final boolean DONT_FILL_HEADER = false;<a name="line.214"></a> +<span class="sourceLineNo">215</span><a name="line.215"></a> +<span class="sourceLineNo">216</span> // How to get the estimate correctly? if it is a singleBB?<a name="line.216"></a> +<span class="sourceLineNo">217</span> public static final int MULTI_BYTE_BUFFER_HEAP_SIZE =<a name="line.217"></a> +<span class="sourceLineNo">218</span> (int)ClassSize.estimateBase(MultiByteBuff.class, false);<a name="line.218"></a> +<span class="sourceLineNo">219</span><a name="line.219"></a> +<span class="sourceLineNo">220</span> /**<a name="line.220"></a> +<span class="sourceLineNo">221</span> * Space for metadata on a block that gets stored along with the block when we cache it.<a name="line.221"></a> +<span class="sourceLineNo">222</span> * There are a few bytes stuck on the end of the HFileBlock that we pull in from HDFS.<a name="line.222"></a> +<span class="sourceLineNo">223</span> * 8 bytes are for the offset of this block (long) in the file. Offset is important because is is<a name="line.223"></a> +<span class="sourceLineNo">224</span> * used when we remake the CacheKey when we return block to the cache when done. There is also<a name="line.224"></a> +<span class="sourceLineNo">225</span> * a flag on whether checksumming is being done by hbase or not. See class comment for note on<a name="line.225"></a> +<span class="sourceLineNo">226</span> * uncertain state of checksumming of blocks that come out of cache (should we or should we not?).<a name="line.226"></a> +<span class="sourceLineNo">227</span> * Finally there are 4 bytes to hold the length of the next block which can save a seek on<a name="line.227"></a> +<span class="sourceLineNo">228</span> * occasion if available.<a name="line.228"></a> +<span class="sourceLineNo">229</span> * (This EXTRA info came in with original commit of the bucketcache, HBASE-7404. It was<a name="line.229"></a> +<span class="sourceLineNo">230</span> * formerly known as EXTRA_SERIALIZATION_SPACE).<a name="line.230"></a> +<span class="sourceLineNo">231</span> */<a name="line.231"></a> +<span class="sourceLineNo">232</span> static final int BLOCK_METADATA_SPACE = Bytes.SIZEOF_BYTE + Bytes.SIZEOF_LONG + Bytes.SIZEOF_INT;<a name="line.232"></a> +<span class="sourceLineNo">233</span><a name="line.233"></a> +<span class="sourceLineNo">234</span> /**<a name="line.234"></a> +<span class="sourceLineNo">235</span> * Each checksum value is an integer that can be stored in 4 bytes.<a name="line.235"></a> +<span class="sourceLineNo">236</span> */<a name="line.236"></a> +<span class="sourceLineNo">237</span> static final int CHECKSUM_SIZE = Bytes.SIZEOF_INT;<a name="line.237"></a> +<span class="sourceLineNo">238</span><a name="line.238"></a> +<span class="sourceLineNo">239</span> static final byte[] DUMMY_HEADER_NO_CHECKSUM =<a name="line.239"></a> +<span class="sourceLineNo">240</span> new byte[HConstants.HFILEBLOCK_HEADER_SIZE_NO_CHECKSUM];<a name="line.240"></a> +<span class="sourceLineNo">241</span><a name="line.241"></a> +<span class="sourceLineNo">242</span> /**<a name="line.242"></a> +<span class="sourceLineNo">243</span> * Used deserializing blocks from Cache.<a name="line.243"></a> +<span class="sourceLineNo">244</span> *<a name="line.244"></a> +<span class="sourceLineNo">245</span> * <code><a name="line.245"></a> +<span class="sourceLineNo">246</span> * ++++++++++++++<a name="line.246"></a> +<span class="sourceLineNo">247</span> * + HFileBlock +<a name="line.247"></a> +<span class="sourceLineNo">248</span> * ++++++++++++++<a name="line.248"></a> +<span class="sourceLineNo">249</span> * + Checksums + <= Optional<a name="line.249"></a> +<span class="sourceLineNo">250</span> * ++++++++++++++<a name="line.250"></a> +<span class="sourceLineNo">251</span> * + Metadata! + <= See note on BLOCK_METADATA_SPACE above.<a name="line.251"></a> +<span class="sourceLineNo">252</span> * ++++++++++++++<a name="line.252"></a> +<span class="sourceLineNo">253</span> * </code><a name="line.253"></a> +<span class="sourceLineNo">254</span> * @see #serialize(ByteBuffer, boolean)<a name="line.254"></a> +<span class="sourceLineNo">255</span> */<a name="line.255"></a> +<span class="sourceLineNo">256</span> public static final CacheableDeserializer<Cacheable> BLOCK_DESERIALIZER = new BlockDeserializer();<a name="line.256"></a> +<span class="sourceLineNo">257</span><a name="line.257"></a> +<span class="sourceLineNo">258</span> public static final class BlockDeserializer implements CacheableDeserializer<Cacheable> {<a name="line.258"></a> +<span class="sourceLineNo">259</span> private BlockDeserializer() {<a name="line.259"></a> +<span class="sourceLineNo">260</span> }<a name="line.260"></a> +<span class="sourceLineNo">261</span><a name="line.261"></a> +<span class="sourceLineNo">262</span> @Override<a name="line.262"></a> +<span class="sourceLineNo">263</span> public HFileBlock deserialize(ByteBuff buf, boolean reuse, MemoryType memType)<a name="line.263"></a> +<span class="sourceLineNo">264</span> throws IOException {<a name="line.264"></a> +<span class="sourceLineNo">265</span> // The buf has the file block followed by block metadata.<a name="line.265"></a> +<span class="sourceLineNo">266</span> // Set limit to just before the BLOCK_METADATA_SPACE then rewind.<a name="line.266"></a> +<span class="sourceLineNo">267</span> buf.limit(buf.limit() - BLOCK_METADATA_SPACE).rewind();<a name="line.267"></a> +<span class="sourceLineNo">268</span> // Get a new buffer to pass the HFileBlock for it to 'own'.<a name="line.268"></a> +<span class="sourceLineNo">269</span> ByteBuff newByteBuff;<a name="line.269"></a> +<span class="sourceLineNo">270</span> if (reuse) {<a name="line.270"></a> +<span class="sourceLineNo">271</span> newByteBuff = buf.slice();<a name="line.271"></a> +<span class="sourceLineNo">272</span> } else {<a name="line.272"></a> +<span class="sourceLineNo">273</span> int len = buf.limit();<a name="line.273"></a> +<span class="sourceLineNo">274</span> newByteBuff = new SingleByteBuff(ByteBuffer.allocate(len));<a name="line.274"></a> +<span class="sourceLineNo">275</span> newByteBuff.put(0, buf, buf.position(), len);<a name="line.275"></a> +<span class="sourceLineNo">276</span> }<a name="line.276"></a> +<span class="sourceLineNo">277</span> // Read out the BLOCK_METADATA_SPACE content and shove into our HFileBlock.<a name="line.277"></a> +<span class="sourceLineNo">278</span> buf.position(buf.limit());<a name="line.278"></a> +<span class="sourceLineNo">279</span> buf.limit(buf.limit() + HFileBlock.BLOCK_METADATA_SPACE);<a name="line.279"></a> +<span class="sourceLineNo">280</span> boolean usesChecksum = buf.get() == (byte) 1;<a name="line.280"></a> +<span class="sourceLineNo">281</span> long offset = buf.getLong();<a name="line.281"></a> +<span class="sourceLineNo">282</span> int nextBlockOnDiskSize = buf.getInt();<a name="line.282"></a> +<span class="sourceLineNo">283</span> HFileBlock hFileBlock =<a name="line.283"></a> +<span class="sourceLineNo">284</span> new HFileBlock(newByteBuff, usesChecksum, memType, offset, nextBlockOnDiskSize, null);<a name="line.284"></a> +<span class="sourceLineNo">285</span> return hFileBlock;<a name="line.285"></a> +<span class="sourceLineNo">286</span> }<a name="line.286"></a> +<span class="sourceLineNo">287</span><a name="line.287"></a> +<span class="sourceLineNo">288</span> @Override<a name="line.288"></a> +<span class="sourceLineNo">289</span> public int getDeserialiserIdentifier() {<a name="line.289"></a> +<span class="sourceLineNo">290</span> return DESERIALIZER_IDENTIFIER;<a name="line.290"></a> +<span class="sourceLineNo">291</span> }<a name="line.291"></a> +<span class="sourceLineNo">292</span><a name="line.292"></a> +<span class="sourceLineNo">293</span> @Override<a name="line.293"></a> +<span class="sourceLineNo">294</span> public HFileBlock deserialize(ByteBuff b) throws IOException {<a name="line.294"></a> +<span class="sourceLineNo">295</span> // Used only in tests<a name="line.295"></a> +<span class="sourceLineNo">296</span> return deserialize(b, false, MemoryType.EXCLUSIVE);<a name="line.296"></a> +<span class="sourceLineNo">297</span> }<a name="line.297"></a> +<span class="sourceLineNo">298</span> }<a name="line.298"></a> +<span class="sourceLineNo">299</span><a name="line.299"></a> +<span class="sourceLineNo">300</span> private static final int DESERIALIZER_IDENTIFIER;<a name="line.300"></a> +<span class="sourceLineNo">301</span> static {<a name="line.301"></a> +<span class="sourceLineNo">302</span> DESERIALIZER_IDENTIFIER =<a name="line.302"></a> +<span class="sourceLineNo">303</span> CacheableDeserializerIdManager.registerDeserializer(BLOCK_DESERIALIZER);<a name="line.303"></a> +<span class="sourceLineNo">304</span> }<a name="line.304"></a> +<span class="sourceLineNo">305</span><a name="line.305"></a> +<span class="sourceLineNo">306</span> /**<a name="line.306"></a> +<span class="sourceLineNo">307</span> * Copy constructor. Creates a shallow copy of {@code that}'s buffer.<a name="line.307"></a> +<span class="sourceLineNo">308</span> */<a name="line.308"></a> +<span class="sourceLineNo">309</span> private HFileBlock(HFileBlock that) {<a name="line.309"></a> +<span class="sourceLineNo">310</span> this(that, false);<a name="line.310"></a> +<span class="sourceLineNo">311</span> }<a name="line.311"></a> +<span class="sourceLineNo">312</span><a name="line.312"></a> +<span class="sourceLineNo">313</span> /**<a name="line.313"></a> +<span class="sourceLineNo">314</span> * Copy constructor. Creates a shallow/deep copy of {@code that}'s buffer as per the boolean<a name="line.314"></a> +<span class="sourceLineNo">315</span> * param.<a name="line.315"></a> +<span class="sourceLineNo">316</span> */<a name="line.316"></a> +<span class="sourceLineNo">317</span> private HFileBlock(HFileBlock that, boolean bufCopy) {<a name="line.317"></a> +<span class="sourceLineNo">318</span> init(that.blockType, that.onDiskSizeWithoutHeader,<a name="line.318"></a> +<span class="sourceLineNo">319</span> that.uncompressedSizeWithoutHeader, that.prevBlockOffset,<a name="line.319"></a> +<span class="sourceLineNo">320</span> that.offset, that.onDiskDataSizeWithHeader, that.nextBlockOnDiskSize, that.fileContext);<a name="line.320"></a> +<span class="sourceLineNo">321</span> if (bufCopy) {<a name="line.321"></a> +<span class="sourceLineNo">322</span> this.buf = new SingleByteBuff(ByteBuffer.wrap(that.buf.toBytes(0, that.buf.limit())));<a name="line.322"></a> +<span class="sourceLineNo">323</span> } else {<a name="line.323"></a> +<span class="sourceLineNo">324</span> this.buf = that.buf.duplicate();<a name="line.324"></a> +<span class="sourceLineNo">325</span> }<a name="line.325"></a> +<span class="sourceLineNo">326</span> }<a name="line.326"></a> +<span class="sourceLineNo">327</span><a name="line.327"></a> +<span class="sourceLineNo">328</span> /**<a name="line.328"></a> +<span class="sourceLineNo">329</span> * Creates a new {@link HFile} block from the given fields. This constructor<a name="line.329"></a> +<span class="sourceLineNo">330</span> * is used only while writing blocks and caching,<a name="line.330"></a> +<span class="sourceLineNo">331</span> * and is sitting in a byte buffer and we want to stuff the block into cache.<a name="line.331"></a> +<span class="sourceLineNo">332</span> *<a name="line.332"></a> +<span class="sourceLineNo">333</span> * <p>TODO: The caller presumes no checksumming<a name="line.333"></a> +<span class="sourceLineNo">334</span> * required of this block instance since going into cache; checksum already verified on<a name="line.334"></a> +<span class="sourceLineNo">335</span> * underlying block data pulled in from filesystem. Is that correct? What if cache is SSD?<a name="line.335"></a> +<span class="sourceLineNo">336</span> *<a name="line.336"></a> +<span class="sourceLineNo">337</span> * @param blockType the type of this block, see {@link BlockType}<a name="line.337"></a> +<span class="sourceLineNo">338</span> * @param onDiskSizeWithoutHeader see {@link #onDiskSizeWithoutHeader}<a name="line.338"></a> +<span class="sourceLineNo">339</span> * @param uncompressedSizeWithoutHeader see {@link #uncompressedSizeWithoutHeader}<a name="line.339"></a> +<span class="sourceLineNo">340</span> * @param prevBlockOffset see {@link #prevBlockOffset}<a name="line.340"></a> +<span class="sourceLineNo">341</span> * @param b block header ({@link HConstants#HFILEBLOCK_HEADER_SIZE} bytes)<a name="line.341"></a> +<span class="sourceLineNo">342</span> * @param fillHeader when true, write the first 4 header fields into passed buffer.<a name="line.342"></a> +<span class="sourceLineNo">343</span> * @param offset the file offset the block was read from<a name="line.343"></a> +<span class="sourceLineNo">344</span> * @param onDiskDataSizeWithHeader see {@link #onDiskDataSizeWithHeader}<a name="line.344"></a> +<span class="sourceLineNo">345</span> * @param fileContext HFile meta data<a name="line.345"></a> +<span class="sourceLineNo">346</span> */<a name="line.346"></a> +<span class="sourceLineNo">347</span> @VisibleForTesting<a name="line.347"></a> +<span class="sourceLineNo">348</span> public HFileBlock(BlockType blockType, int onDiskSizeWithoutHeader,<a name="line.348"></a> +<span class="sourceLineNo">349</span> int uncompressedSizeWithoutHeader, long prevBlockOffset, ByteBuffer b, boolean fillHeader,<a name="line.349"></a> +<span class="sourceLineNo">350</span> long offset, final int nextBlockOnDiskSize, int onDiskDataSizeWithHeader,<a name="line.350"></a> +<span class="sourceLineNo">351</span> HFileContext fileContext) {<a name="line.351"></a> +<span class="sourceLineNo">352</span> init(blockType, onDiskSizeWithoutHeader, uncompressedSizeWithoutHeader,<a name="line.352"></a> +<span class="sourceLineNo">353</span> prevBlockOffset, offset, onDiskDataSizeWithHeader, nextBlockOnDiskSize, fileContext);<a name="line.353"></a> +<span class="sourceLineNo">354</span> this.buf = new SingleByteBuff(b);<a name="line.354"></a> +<span class="sourceLineNo">355</span> if (fillHeader) {<a name="line.355"></a> +<span class="sourceLineNo">356</span> overwriteHeader();<a name="line.356"></a> +<span class="sourceLineNo">357</span> }<a name="line.357"></a> +<span class="sourceLineNo">358</span> this.buf.rewind();<a name="line.358"></a> +<span class="sourceLineNo">359</span> }<a name="line.359"></a> +<span class="sourceLineNo">360</span><a name="line.360"></a> +<span class="sourceLineNo">361</span> /**<a name="line.361"></a> +<span class="sourceLineNo">362</span> * Creates a block from an existing buffer starting with a header. Rewinds<a name="line.362"></a> +<span class="sourceLineNo">363</span> * and takes ownership of the buffer. By definition of rewind, ignores the<a name="line.363"></a> +<span class="sourceLineNo">364</span> * buffer position, but if you slice the buffer beforehand, it will rewind<a name="line.364"></a> +<span class="sourceLineNo">365</span> * to that point.<a name="line.365"></a> +<span class="sourceLineNo">366</span> * @param buf Has header, content, and trailing checksums if present.<a name="line.366"></a> +<span class="sourceLineNo">367</span> */<a name="line.367"></a> +<span class="sourceLineNo">368</span> HFileBlock(ByteBuff buf, boolean usesHBaseChecksum, MemoryType memType, final long offset,<a name="line.368"></a> +<span class="sourceLineNo">369</span> final int nextBlockOnDiskSize, HFileContext fileContext) throws IOException {<a name="line.369"></a> +<span class="sourceLineNo">370</span> buf.rewind();<a name="line.370"></a> +<span class="sourceLineNo">371</span> final BlockType blockType = BlockType.read(buf);<a name="line.371"></a> +<span class="sourceLineNo">372</span> final int onDiskSizeWithoutHeader = buf.getInt(Header.ON_DISK_SIZE_WITHOUT_HEADER_INDEX);<a name="line.372"></a> +<span class="sourceLineNo">373</span> final int uncompressedSizeWithoutHeader =<a name="line.373"></a> +<span class="sourceLineNo">374</span> buf.getInt(Header.UNCOMPRESSED_SIZE_WITHOUT_HEADER_INDEX);<a name="line.374"></a> +<span class="sourceLineNo">375</span> final long prevBlockOffset = buf.getLong(Header.PREV_BLOCK_OFFSET_INDEX);<a name="line.375"></a> +<span class="sourceLineNo">376</span> // This constructor is called when we deserialize a block from cache and when we read a block in<a name="line.376"></a> +<span class="sourceLineNo">377</span> // from the fs. fileCache is null when deserialized from cache so need to make up one.<a name="line.377"></a> +<span class="sourceLineNo">378</span> HFileContextBuilder fileContextBuilder = fileContext != null?<a name="line.378"></a> +<span class="sourceLineNo">379</span> new HFileContextBuilder(fileContext): new HFileContextBuilder();<a name="line.379"></a> +<span class="sourceLineNo">380</span> fileContextBuilder.withHBaseCheckSum(usesHBaseChecksum);<a name="line.380"></a> +<span class="sourceLineNo">381</span> int onDiskDataSizeWithHeader;<a name="line.381"></a> +<span class="sourceLineNo">382</span> if (usesHBaseChecksum) {<a name="line.382"></a> +<span class="sourceLineNo">383</span> byte checksumType = buf.get(Header.CHECKSUM_TYPE_INDEX);<a name="line.383"></a> +<span class="sourceLineNo">384</span> int bytesPerChecksum = buf.getInt(Header.BYTES_PER_CHECKSUM_INDEX);<a name="line.384"></a> +<span class="sourceLineNo">385</span> onDiskDataSizeWithHeader = buf.getInt(Header.ON_DISK_DATA_SIZE_WITH_HEADER_INDEX);<a name="line.385"></a> +<span class="sourceLineNo">386</span> // Use the checksum type and bytes per checksum from header, not from filecontext.<a name="line.386"></a> +<span class="sourceLineNo">387</span> fileContextBuilder.withChecksumType(ChecksumType.codeToType(checksumType));<a name="line.387"></a> +<span class="sourceLineNo">388</span> fileContextBuilder.withBytesPerCheckSum(bytesPerChecksum);<a name="line.388"></a> +<span class="sourceLineNo">389</span> } else {<a name="line.389"></a> +<span class="sourceLineNo">390</span> fileContextBuilder.withChecksumType(ChecksumType.NULL);<a name="line.390"></a> +<span class="sourceLineNo">391</span> fileContextBuilder.withBytesPerCheckSum(0);<a name="line.391"></a> +<span class="sourceLineNo">392</span> // Need to fix onDiskDataSizeWithHeader; there are not checksums after-block-data<a name="line.392"></a> +<span class="sourceLineNo">393</span> onDiskDataSizeWithHeader = onDiskSizeWithoutHeader + headerSize(usesHBaseChecksum);<a name="line.393"></a> +<span class="sourceLineNo">394</span> }<a name="line.394"></a> +<span class="sourceLineNo">395</span> fileContext = fileContextBuilder.build();<a name="line.395"></a> +<span class="sourceLineNo">396</span> assert usesHBaseChecksum == fileContext.isUseHBaseChecksum();<a name="line.396"></a> +<span class="sourceLineNo">397</span> init(blockType, onDiskSizeWithoutHeader, uncompressedSizeWithoutHeader,<a name="line.397"></a> +<span class="sourceLineNo">398</span> prevBlockOffset, offset, onDiskDataSizeWithHeader, nextBlockOnDiskSize, fileContext);<a name="line.398"></a> +<span class="sourceLineNo">399</span> this.memType = memType;<a name="line.399"></a> +<span class="sourceLineNo">400</span> this.offset = offset;<a name="line.400"></a> +<span class="sourceLineNo">401</span> this.buf = buf;<a name="line.401"></a> +<span class="sourceLineNo">402</span> this.buf.rewind();<a name="line.402"></a> +<span class="sourceLineNo">403</span> }<a name="line.403"></a> +<span class="sourceLineNo">404</span><a name="line.404"></a> +<span class="sourceLineNo">405</span> /**<a name="line.405"></a> +<span class="sourceLineNo">406</span> * Called from constructors.<a name="line.406"></a> +<span class="sourceLineNo">407</span> */<a name="line.407"></a> +<span class="sourceLineNo">408</span> private void init(BlockType blockType, int onDiskSizeWithoutHeader,<a name="line.408"></a> +<span class="sourceLineNo">409</span> int uncompressedSizeWithoutHeader, long prevBlockOffset,<a name="line.409"></a> +<span class="sourceLineNo">410</span> long offset, int onDiskDataSizeWithHeader, final int nextBlockOnDiskSize,<a name="line.410"></a> +<span class="sourceLineNo">411</span> HFileContext fileContext) {<a name="line.411"></a> +<span class="sourceLineNo">412</span> this.blockType = blockType;<a name="line.412"></a> +<span class="sourceLineNo">413</span> this.onDiskSizeWithoutHeader = onDiskSizeWithoutHeader;<a name="line.413"></a> +<span class="sourceLineNo">414</span> this.uncompressedSizeWithoutHeader = uncompressedSizeWithoutHeader;<a name="line.414"></a> +<span class="sourceLineNo">415</span> this.prevBlockOffset = prevBlockOffset;<a name="line.415"></a> +<span class="sourceLineNo">416</span> this.offset = offset;<a name="line.416"></a> +<span class="sourceLineNo">417</span> this.onDiskDataSizeWithHeader = onDiskDataSizeWithHeader;<a name="line.417"></a> +<span class="sourceLineNo">418</span> this.nextBlockOnDiskSize = nextBlockOnDiskSize;<a name="line.418"></a> +<span class="sourceLineNo">419</span> this.fileContext = fileContext;<a name="line.419"></a> +<span class="sourceLineNo">420</span> }<a name="line.420"></a> +<span class="sourceLineNo">421</span><a name="line.421"></a> +<span class="sourceLineNo">422</span> /**<a name="line.422"></a> +<span class="sourceLineNo">423</span> * Parse total on disk size including header and checksum.<a name="line.423"></a> +<span class="sourceLineNo">424</span> * @param headerBuf Header ByteBuffer. Presumed exact size of header.<a name="line.424"></a> +<span class="sourceLineNo">425</span> * @param verifyChecksum true if checksum verification is in use.<a name="line.425"></a> +<span class="sourceLineNo">426</span> * @return Size of the block with header included.<a name="line.426"></a> +<span class="sourceLineNo">427</span> */<a name="line.427"></a> +<span class="sourceLineNo">428</span> private static int getOnDiskSizeWithHeader(final ByteBuffer headerBuf,<a name="line.428"></a> +<span class="sourceLineNo">429</span> boolean verifyChecksum) {<a name="line.429"></a> +<span class="sourceLineNo">430</span> return headerBuf.getInt(Header.ON_DISK_SIZE_WITHOUT_HEADER_INDEX) +<a name="line.430"></a> +<span class="sourceLineNo">431</span> headerSize(verifyChecksum);<a name="line.431"></a> +<span class="sourceLineNo">432</span> }<a name="line.432"></a> +<span class="sourceLineNo">433</span><a name="line.433"></a> +<span class="sourceLineNo">434</span> /**<a name="line.434"></a> +<span class="sourceLineNo">435</span> * @return the on-disk size of the next block (including the header size and any checksums if<a name="line.435"></a> +<span class="sourceLineNo">436</span> * present) read by peeking into the next block's header; use as a hint when doing<a name="line.436"></a> +<span class="sourceLineNo">437</span> * a read of the next block when scanning or running over a file.<a name="line.437"></a> +<span class="sourceLineNo">438</span> */<a name="line.438"></a> +<span class="sourceLineNo">439</span> int getNextBlockOnDiskSize() {<a name="line.439"></a> +<span class="sourceLineNo">440</span> return nextBlockOnDiskSize;<a name="line.440"></a> +<span class="sourceLineNo">441</span> }<a name="line.441"></a> +<span class="sourceLineNo">442</span><a name="line.442"></a> +<span class="sourceLineNo">443</span> @Override<a name="line.443"></a> +<span class="sourceLineNo">444</span> public BlockType getBlockType() {<a name="line.444"></a> +<span class="sourceLineNo">445</span> return blockType;<a name="line.445"></a> +<span class="sourceLineNo">446</span> }<a name="line.446"></a> +<span class="sourceLineNo">447</span><a name="line.447"></a> +<span class="sourceLineNo">448</span> /** @return get data block encoding id that was used to encode this block */<a name="line.448"></a> +<span class="sourceLineNo">449</span> short getDataBlockEncodingId() {<a name="line.449"></a> +<span class="sourceLineNo">450</span> if (blockType != BlockType.ENCODED_DATA) {<a name="line.450"></a> +<span class="sourceLineNo">451</span> throw new IllegalArgumentException("Querying encoder ID of a block " +<a name="line.451"></a> +<span class="sourceLineNo">452</span> "of type other than " + BlockType.ENCODED_DATA + ": " + blockType);<a name="line.452"></a> +<span class="sourceLineNo">453</span> }<a name="line.453"></a> +<span class="sourceLineNo">454</span> return buf.getShort(headerSize());<a name="line.454"></a> +<span class="sourceLineNo">455</span> }<a name="line.455"></a> +<span class="sourceLineNo">456</span><a name="line.456"></a> +<span class="sourceLineNo">457</span> /**<a name="line.457"></a> +<span class="sourceLineNo">458</span> * @return the on-disk size of header + data part + checksum.<a name="line.458"></a> +<span class="sourceLineNo">459</span> */<a name="line.459"></a> +<span class="sourceLineNo">460</span> public int getOnDiskSizeWithHeader() {<a name="line.460"></a> +<span class="sourceLineNo">461</span> return onDiskSizeWithoutHeader + headerSize();<a name="line.461"></a> +<span class="sourceLineNo">462</span> }<a name="line.462"></a> +<span class="sourceLineNo">463</span><a name="line.463"></a> +<span class="sourceLineNo">464</span> /**<a name="line.464"></a> +<span class="sourceLineNo">465</span> * @return the on-disk size of the data part + checksum (header excluded).<a name="line.465"></a> +<span class="sourceLineNo">466</span> */<a name="line.466"></a> +<span class="sourceLineNo">467</span> int getOnDiskSizeWithoutHeader() {<a name="line.467"></a> +<span class="sourceLineNo">468</span> return onDiskSizeWithoutHeader;<a name="line.468"></a> +<span class="sourceLineNo">469</span> }<a name="line.469"></a> +<span class="sourceLineNo">470</span><a name="line.470"></a> +<span class="sourceLineNo">471</span> /**<a name="line.471"></a> +<span class="sourceLineNo">472</span> * @return the uncompressed size of data part (header and checksum excluded).<a name="line.472"></a> +<span class="sourceLineNo">473</span> */<a name="line.473"></a> +<span class="sourceLineNo">474</span> int getUncompressedSizeWithoutHeader() {<a name="line.474"></a> +<span class="sourceLineNo">475</span> return uncompressedSizeWithoutHeader;<a name="line.475"></a> +<span class="sourceLineNo">476</span> }<a name="line.476"></a> +<span class="sourceLineNo">477</span><a name="line.477"></a> +<span class="sourceLineNo">478</span> /**<a name="line.478"></a> +<span class="sourceLineNo">479</span> * @return the offset of the previous block of the same type in the file, or<a name="line.479"></a> +<span class="sourceLineNo">480</span> * -1 if unknown<a name="line.480"></a> +<span class="sourceLineNo">481</span> */<a name="line.481"></a> +<span class="sourceLineNo">482</span> long getPrevBlockOffset() {<a name="line.482"></a> +<span class="sourceLineNo">483</span> return prevBlockOffset;<a name="line.483"></a> +<span class="sourceLineNo">484</span> }<a name="line.484"></a> +<span class="sourceLineNo">485</span><a name="line.485"></a> +<span class="sourceLineNo">486</span> /**<a name="line.486"></a> +<span class="sourceLineNo">487</span> * Rewinds {@code buf} and writes first 4 header fields. {@code buf} position<a name="line.487"></a> +<span class="sourceLineNo">488</span> * is modified as side-effect.<a name="line.488"></a> +<span class="sourceLineNo">489</span> */<a name="line.489"></a> +<span class="sourceLineNo">490</span> private void overwriteHeader() {<a name="line.490"></a> +<span class="sourceLineNo">491</span> buf.rewind();<a name="line.491"></a> +<span class="sourceLineNo">492</span> blockType.write(buf);<a name="line.492"></a> +<span class="sourceLineNo">493</span> buf.putInt(onDiskSizeWithoutHeader);<a name="line.493"></a> +<span class="sourceLineNo">494</span> buf.putInt(uncompressedSizeWithoutHeader);<a name="line.494"></a> +<span class="sourceLineNo">495</span> buf.putLong(prevBlockOffset);<a name="line.495"></a> +<span class="sourceLineNo">496</span> if (this.fileContext.isUseHBaseChecksum()) {<a name="line.496"></a> +<span class="sourceLineNo">497</span> buf.put(fileContext.getChecksumType().getCode());<a name="line.497"></a> +<span class="sourceLineNo">498</span> buf.putInt(fileContext.getBytesPerChecksum());<a name="line.498"></a> +<span class="sourceLineNo">499</span> buf.putInt(onDiskDataSizeWithHeader);<a name="line.499"></a> +<span class="sourceLineNo">500</span> }<a name="line.500"></a> +<span class="sourceLineNo">501</span> }<a name="line.501"></a> +<span class="sourceLineNo">502</span><a name="line.502"></a> +<span class="sourceLineNo">503</span> /**<a name="line.503"></a> +<span class="sourceLineNo">504</span> * Returns a buffer that does not include the header or checksum.<a name="line.504"></a> +<span class="sourceLineNo">505</span> *<a name="line.505"></a> +<span class="sourceLineNo">506</span> * @return the buffer with header skipped and checksum omitted.<a name="line.506"></a> +<span class="sourceLineNo">507</span> */<a name="line.507"></a> +<span class="sourceLineNo">508</span> public ByteBuff getBufferWithoutHeader() {<a name="line.508"></a> +<span class="sourceLineNo">509</span> ByteBuff dup = getBufferReadOnly();<a name="line.509"></a> +<span class="sourceLineNo">510</span> // Now set it up so Buffer spans content only -- no header or no checksums.<a name="line.510"></a> +<span class="sourceLineNo">511</span> return dup.position(headerSize()).limit(buf.limit() - totalChecksumBytes()).slice();<a name="line.511"></a> +<span class="sourceLineNo">512</span> }<a name="line.512"></a> +<span class="sourceLineNo">513</span><a name="line.513"></a> +<span class="sourceLineNo">514</span> /**<a name="line.514"></a> +<span class="sourceLineNo">515</span> * Returns a read-only duplicate of the buffer this block stores internally ready to be read.<a name="line.515"></a> +<span class="sourceLineNo">516</span> * Clients must not modify the buffer object though they may set position and limit on the<a name="line.516"></a> +<span class="sourceLineNo">517</span> * returned buffer since we pass back a duplicate. This method has to be public because it is used<a name="line.517"></a> +<span class="sourceLineNo">518</span> * in {@link CompoundBloomFilter} to avoid object creation on every Bloom<a name="line.518"></a> +<span class="sourceLineNo">519</span> * filter lookup, but has to be used with caution. Buffer holds header, block content,<a name="line.519"></a> +<span class="sourceLineNo">520</span> * and any follow-on checksums if present.<a name="line.520"></a> +<span class="sourceLineNo">521</span> *<a name="line.521"></a> +<span class="sourceLineNo">522</span> * @return the buffer of this block for read-only operations<a name="line.522"></a> +<span class="sourceLineNo">523</span> */<a name="line.523"></a> +<span class="sourceLineNo">524</span> public ByteBuff getBufferReadOnly() {<a name="line.524"></a> +<span class="sourceLineNo">525</span> // TODO: ByteBuf does not support asReadOnlyBuffer(). Fix.<a name="line.525"></a> +<span class="sourceLineNo">526</span> ByteBuff dup = this.buf.duplicate();<a name="line.526"></a> +<span class="sourceLineNo">527</span> assert dup.position() == 0;<a name="line.527"></a> +<span class="sourceLineNo">528</span> return dup;<a name="line.528"></a> +<span class="sourceLineNo">529</span> }<a name="line.529"></a> +<span class="sourceLineNo">530</span><a name="line.530"></a> +<span class="sourceLineNo">531</span> @VisibleForTesting<a name="line.531"></a> +<span class="sourceLineNo">532</span> private void sanityCheckAssertion(long valueFromBuf, long valueFromField,<a name="line.532"></a> +<span class="sourceLineNo">533</span> String fieldName) throws IOException {<a name="line.533"></a> +<span class="sourceLineNo">534</span> if (valueFromBuf != valueFromField) {<a name="line.534"></a> +<span class="sourceLineNo">535</span> throw new AssertionError(fieldName + " in the buffer (" + valueFromBuf<a name="line.535"></a> +<span class="sourceLineNo">536</span> + ") is different from that in the field (" + valueFromField + ")");<a name="line.536"></a> +<span class="sourceLineNo">537</span> }<a name="line.537"></a> +<span class="sourceLineNo">538</span> }<a name="line.538"></a> +<span class="sourceLineNo">539</span><a name="line.539"></a> +<span class="sourceLineNo">540</span> @VisibleForTesting<a name="line.540"></a> +<span class="sourceLineNo">541</span> private void sanityCheckAssertion(BlockType valueFromBuf, BlockType valueFromField)<a name="line.541"></a> +<span class="sourceLineNo">542</span> throws IOException {<a name="line.542"></a> +<span class="sourceLineNo">543</span> if (valueFromBuf != valueFromField) {<a name="line.543"></a> +<span class="sourceLineNo">544</span> throw new IOException("Block type stored in the buffer: " +<a name="line.544"></a> +<span class="sourceLineNo">545</span> valueFromBuf + ", block type field: " + valueFromField);<a name="line.545"></a> +<span class="sourceLineNo">546</span> }<a name="line.546"></a> +<span class="sourceLineNo">547</span> }<a name="line.547"></a> +<span class="sourceLineNo">548</span><a name="line.548"></a> +<span class="sourceLineNo">549</span> /**<a name="line.549"></a> +<span class="sourceLineNo">550</span> * Checks if the block is internally consistent, i.e. the first<a name="line.550"></a> +<span class="sourceLineNo">551</span> * {@link HConstants#HFILEBLOCK_HEADER_SIZE} bytes of the buffer contain a<a name="line.551"></a> +<span class="sourceLineNo">552</span> * valid header consistent with the fields. Assumes a packed block structure.<a name="line.552"></a> +<span class="sourceLineNo">553</span> * This function is primary for testing and debugging, and is not<a name="line.553"></a> +<span class="sourceLineNo">554</span> * thread-safe, because it alters the internal buffer pointer.<a name="line.554"></a> +<span class="sourceLineNo">555</span> * Used by tests only.<a name="line.555"></a> +<span class="sourceLineNo">556</span> */<a name="line.556"></a> +<span class="sourceLineNo">557</span> @VisibleForTesting<a name="line.557"></a> +<span class="sourceLineNo">558</span> void sanityCheck() throws IOException {<a name="line.558"></a> +<span class="sourceLineNo">559</span> // Duplicate so no side-effects<a name="line.559"></a> +<span class="sourceLineNo">560</span> ByteBuff dup = this.buf.duplicate().rewind();<a name="line.560"></a> +<span class="sourceLineNo">561</span> sanityCheckAssertion(BlockType.read(dup), blockType);<a name="line.561"></a> +<span class="sourceLineNo">562</span><a name="line.562"></a> +<span class="sourceLineNo">563</span> sanityCheckAssertion(dup.getInt(), onDiskSizeWithoutHeader, "onDiskSizeWithoutHeader");<a name="line.563"></a> +<span class="sourceLineNo">564</span><a name="line.564"></a> +<span class="sourceLineNo">565</span> sanityCheckAssertion(dup.getInt(), uncompressedSizeWithoutHeader,<a name="line.565"></a> +<span class="sourceLineNo">566</span> "uncompressedSizeWithoutHeader");<a name="line.566"></a> +<span class="sourceLineNo">567</span><a name="line.567"></a> +<span class="sourceLineNo">568</span> sanityCheckAssertion(dup.getLong(), prevBlockOffset, "prevBlockOffset");<a name="line.568"></a> +<span class="sourceLineNo">569</span> if (this.fileContext.isUseHBaseChecksum()) {<a name="line.569"></a> +<span class="sourceLineNo">570</span> sanityCheckAssertion(dup.get(), this.fileContext.getChecksumType().getCode(), "checksumType");<a name="line.570"></a> +<span class="sourceLineNo">571</span> sanityCheckAssertion(dup.getInt(), this.fileContext.getBytesPerChecksum(),<a name="line.571"></a> +<span class="sourceLineNo">572</span> "bytesPerChecksum");<a name="line.572"></a> +<span class="sourceLineNo">573</span> sanityCheckAssertion(dup.getInt(), onDiskDataSizeWithHeader, "onDiskDataSizeWithHeader");<a name="line.573"></a> +<span class="sourceLineNo">574</span> }<a name="line.574"></a> +<span class="sourceLineNo">575</span><a name="line.575"></a> +<span class="sourceLineNo">576</span> int cksumBytes = totalChecksumBytes();<a name="line.576"></a> +<span class="sourceLineNo">577</span> int expectedBufLimit = onDiskDataSizeWithHeader + cksumBytes;<a name="line.577"></a> +<span class="sourceLineNo">578</span> if (dup.limit() != expectedBufLimit) {<a name="line.578"></a> +<span class="sourceLineNo">579</span> throw new AssertionError("Expected limit " + expectedBufLimit + ", got " + dup.limit());<a name="line.579"></a> +<span class="sourceLineNo">580</span> }<a name="line.580"></a> +<span class="sourceLineNo">581</span><a name="line.581"></a> +<span class="sourceLineNo">582</span> // We might optionally allocate HFILEBLOCK_HEADER_SIZE more bytes to read the next<a name="line.582"></a> +<span class="sourceLineNo">583</span> // block's header, so there are two sensible values for buffer capacity.<a name="line.583"></a> +<span class="sourceLineNo">584</span> int hdrSize = headerSize();<a name="line.584"></a> +<span class="sourceLineNo">585</span> if (dup.capacity() != expectedBufLimit && dup.capacity() != expectedBufLimit + hdrSize) {<a name="line.585"></a> +<span class="sourceLineNo">586</span> throw new AssertionError("Invalid buffer capacity: " + dup.capacity() +<a name="line.586"></a> +<span class="sourceLineNo">587</span> ", expected " + expectedBufLimit + " or " + (expectedBufLimit + hdrSize));<a name="line.587"></a> +<span class="sourceLineNo">588</span> }<a name="line.588"></a> +<span class="sourceLineNo">589</span> }<a name="line.589"></a> +<span class="sourceLineNo">590</span><a name="line.590"></a> +<span class="sourceLineNo">591</span> @Override<a name="line.591"></a> +<span class="sourceLineNo">592</span> public String toString() {<a name="line.592"></a> +<span class="sourceLineNo">593</span> StringBuilder sb = new StringBuilder()<a name="line.593"></a> +<span class="sourceLineNo">594</span> .append("[")<a name="line.594"></a> +<span class="sourceLineNo">595</span> .append("blockType=").append(blockType)<a name="line.595"></a> +<span class="sourceLineNo">596</span> .append(", fileOffset=").append(offset)<a name="line.596"></a> +<span class="sourceLineNo">597</span> .append(", headerSize=").append(headerSize())<a name="line.597"></a> +<span class="sourceLineNo">598</span> .append(", onDiskSizeWithoutHeader=").append(onDiskSizeWithoutHeader)<a name="line.598"></a> +<span class="sourceLineNo">599</span> .append(", uncompressedSizeWithoutHeader=").append(uncompressedSizeWithoutHeader)<a name="line.599"></a> +<span class="sourceLineNo">600</span> .append(", prevBlockOffset=").append(prevBlockOffset)<a name="line.600"></a> +<span class="sourceLineNo">601</span> .append(", isUseHBaseChecksum=").append(fileContext.isUseHBaseChecksum());<a name="line.601"></a> +<span class="sourceLineNo">602</span> if (fileContext.isUseHBaseChecksum()) {<a name="line.602"></a> +<span class="sourceLineNo">603</span> sb.append(", checksumType=").append(ChecksumType.codeToType(this.buf.get(24)))<a name="line.603"></a> +<span class="sourceLineNo">604</span> .append(", bytesPerChecksum=").append(this.buf.getInt(24 + 1))<a name="line.604"></a> +<span class="sourceLineNo">605</span> .append(", onDiskDataSizeWithHeader=").append(onDiskDataSizeWithHeader);<a name="line.605"></a> +<span class="sourceLineNo">606</span> } else {<a name="line.606"></a> +<span class="sourceLineNo">607</span> sb.append(", onDiskDataSizeWithHeader=").append(onDiskDataSizeWithHeader)<a name="line.607"></a> +<span class="sourceLineNo">608</span> .append("(").append(onDiskSizeWithoutHeader)<a name="line.608"></a> +<span class="sourceLineNo">609</span> .append("+").append(HConstants.HFILEBLOCK_HEADER_SIZE_NO_CHECKSUM).append(")");<a name="line.609"></a> +<span class="sourceLineNo">610</span> }<a name="line.610"></a> +<span class="sourceLineNo">611</span> String dataBegin = null;<a name="line.611"></a> +<span class="sourceLineNo">612</span> if (buf.hasArray()) {<a name="line.612"></a> +<span class="sourceLineNo">613</span> dataBegin = Bytes.toStringBinary(buf.array(), buf.arrayOffset() + headerSize(),<a name="line.613"></a> +<span class="sourceLineNo">614</span> Math.min(32, buf.limit() - buf.arrayOffset() - headerSize()));<a name="line.614"></a> +<span class="sourceLineNo">615</span> } else {<a name="line.615"></a> +<span class="sourceLineNo">616</span> ByteBuff bufWithoutHeader = getBufferWithoutHeader();<a name="line.616"></a> +<span class="sourceLineNo">617</span> byte[] dataBeginBytes = new byte[Math.min(32,<a name="line.617"></a> +<span class="sourceLineNo">618</span> bufWithoutHeader.limit() - bufWithoutHeader.position())];<a name="line.618"></a> +<span class="sourceLineNo">619</span> bufWithoutHeader.get(dataBeginBytes);<a name="line.619"></a> +<span class="sourceLineNo">620</span> dataBegin = Bytes.toStringBinary(dataBeginBytes);<a name="line.620"></a> +<span class="sourceLineNo">621</span> }<a name="line.621"></a> +<span class="sourceLineNo">622</span> sb.append(", getOnDiskSizeWithHeader=").append(getOnDiskSizeWithHeader())<a name="line.622"></a> +<span class="sourceLineNo">623</span> .append(", totalChecksumBytes=").append(totalChecksumBytes())<a name="line.623"></a> +<span class="sourceLineNo">624</span> .append(", isUnpacked=").append(isUnpacked())<a name="line.624"></a> +<span class="sourceLineNo">625</span> .append(", buf=[").append(buf).append("]")<a name="line.625"></a> +<span class="sourceLineNo">626</span> .append(", dataBeginsWith=").append(dataBegin)<a name="line.626"></a> +<span class="sourceLineNo">627</span> .append(", fileContext=").append(fileContext)<a name="line.627"></a> +<span class="sourceLineNo">628</span> .append(", nextBlockOnDiskSize=").append(nextBlockOnDiskSize)<a name="line.628"></a> +<span class="sourceLineNo">629</span> .append("]");<a name="line.629"></a> +<span class="sourceLineNo">630</span> return sb.toString();<a name="line.630"></a> +<span class="sourceLineNo">631</span> }<a name="line.631"></a> +<span class="sourceLineNo">632</span><a name="line.632"></a> +<span class="sourceLineNo">633</span> /**<a name="line.633"></a> +<span class="sourceLineNo">634</span> * Retrieves the decompressed/decrypted view of this block. An encoded block remains in its<a name="line.634"></a> +<span class="sourceLineNo">635</span> * encoded structure. Internal structures are shared between instances where applicable.<a name="line.635"></a> +<span class="sourceLineNo">636</span> */<a name="line.636"></a> +<span class="sourceLineNo">637</span> HFileBlock unpack(HFileContext fileContext, FSReader reader) throws IOException {<a name="line.637"></a> +<span class="sourceLineNo">638</span> if (!fileContext.isCompressedOrEncrypted()) {<a name="line.638"></a> +<span class="sourceLineNo">639</span> // TODO: cannot use our own fileContext here because HFileBlock(ByteBuffer, boolean),<a name="line.639"></a> +<span class="sourceLineNo">640</span> // which is used for block serialization to L2 cache, does not preserve encoding and<a name="line.640"></a> +<span class="sourceLineNo">641</span> // encryption details.<a name="line.641"></a> +<span class="sourceLineNo">642</span> return this;<a name="line.642"></a> +<span class="sourceLineNo">643</span> }<a name="line.643"></a> +<span class="sourceLineNo">644</span><a name="line.644"></a> +<span class="sourceLineNo">645</span> HFileBlock unpacked = new HFileBlock(this);<a name="line.645"></a> +<span class="sourceLineNo">646</span> unpacked.allocateBuffer(); // allocates space for the decompressed block<a name="line.646"></a> +<span class="sourceLineNo">647</span><a name="line.647"></a> +<span class="sourceLineNo">648</span> HFileBlockDecodingContext ctx = blockType == BlockType.ENCODED_DATA ?<a name="line.648"></a> +<span class="sourceLineNo">649</span> reader.getBlockDecodingContext() : reader.getDefaultBlockDecodingContext();<a name="line.649"></a> +<span class="sourceLineNo">650</span><a name="line.650"></a> +<span class="sourceLineNo">651</span> ByteBuff dup = this.buf.duplicate();<a name="line.651"></a> +<span class="sourceLineNo">652</span> dup.position(this.headerSize());<a name="line.652"></a> +<span class="sourceLineNo">653</span> dup = dup.slice();<a name="line.653"></a> +<span class="sourceLineNo">654</span> ctx.prepareDecoding(unpacked.getOnDiskSizeWithoutHeader(),<a name="line.654"></a> +<span class="sourceLineNo">655</span> unpacked.getUncompressedSizeWithoutHeader(), unpacked.getBufferWithoutHeader(),<a name="line.655"></a> +<span class="sourceLineNo">656</span> dup);<a name="line.656"></a> +<span class="sourceLineNo">657</span> return unpacked;<a name="line.657"></a> +<span class="sourceLineNo">658</span> }<a name="line.658"></a> +<span class="sourceLineNo">659</span><a name="line.659"></a> +<span class="sourceLineNo">660</span> /**<a name="line.660"></a> +<span class="sourceLineNo">661</span> * Always allocates a new buffer of the correct size. Copies header bytes<a name="line.661"></a> +<span class="sourceLineNo">662</span> * from the existing buffer. Does not change header fields.<a name="line.662"></a> +<span class="sourceLineNo">663</span> * Reserve room to keep checksum bytes too.<a name="line.663"></a> +<span class="sourceLineNo">664</span> */<a name="line.664"></a> +<span class="sourceLineNo">665</span> private void allocateBuffer() {<a name="line.665"></a> +<span class="sourceLineNo">666</span> int cksumBytes = totalChecksumBytes();<a name="line.666"></a> +<span class="sourceLineNo">667</span> int headerSize = headerSize();<a name="line.667"></a> +<span class="sourceLineNo">668</span> int capacityNeeded = headerSize + uncompressedSizeWithoutHeader + cksumBytes;<a name="line.668"></a> +<span class="sourceLineNo">669</span><a name="line.669"></a> +<span class="sourceLineNo">670</span> // TODO we need consider allocating offheap here?<a name="line.670"></a> +<span class="sourceLineNo">671</span> ByteBuffer newBuf = ByteBuffer.allocate(capacityNeeded);<a name="line.671"></a> +<span class="sourceLineNo">672</span><a name="line.672"></a> +<span class="sourceLineNo">673</span> // Copy header bytes into newBuf.<a name="line.673"></a> +<span class="sourceLineNo">674</span> // newBuf is HBB so no issue in calling array()<a name="line.674"></a> +<span class="sourceLineNo">675</span> buf.position(0);<a name="line.675"></a> +<span class="sourceLineNo">676</span> buf.get(newBuf.array(), newBuf.arrayOffset(), headerSize);<a name="line.676"></a> +<span class="sourceLineNo">677</span><a name="line.677"></a> +<span class="sourceLineNo">678</span> buf = new SingleByteBuff(newBuf);<a name="line.678"></a> +<span class="sourceLineNo">679</span> // set limit to exclude next block's header<a name="line.679"></a> +<span class="sourceLineNo">680</span> buf.limit(headerSize + uncompressedSizeWithoutHeader + cksumBytes);<a name="line.680"></a> +<span class="sourceLineNo">681</span> }<a name="line.681"></a> +<span class="sourceLineNo">682</span><a name="line.682"></a> +<span class="sourceLineNo">683</span> /**<a name="line.683"></a> +<span class="sourceLineNo">684</span> * Return true when this block's buffer has been unpacked, false otherwise. Note this is a<a name="line.684"></a> +<span class="sourceLineNo">685</span> * calculated heuristic, not tracked attribute of the block.<a name="line.685"></a> +<span class="sourceLineNo">686</span> */<a name="line.686"></a> +<span class="sourceLineNo">687</span> public boolean isUnpacked() {<a name="line.687"></a> +<span class="sourceLineNo">688</span> final int cksumBytes = totalChecksumBytes();<a name="line.688"></a> +<span class="sourceLineNo">689</span> final int headerSize = headerSize();<a name="line.689"></a> +<span class="sourceLineNo">690</span> final int expectedCapacity = headerSize + uncompressedSizeWithoutHeader + cksumBytes;<a name="line.690"></a> +<span class="sourceLineNo">691</span> final int bufCapacity = buf.capacity();<a name="line.691"></a> +<span class="sourceLineNo">692</span> return bufCapacity == expectedCapacity || bufCapacity == expectedCapacity + headerSize;<a name="line.692"></a> +<span class="sourceLineNo">693</span> }<a name="line.693"></a> +<span class="sourceLineNo">694</span><a name="line.694"></a> +<span class="sourceLineNo">695</span> /** An additional sanity-check in case no compression or encryption is being used. */<a name="line.695"></a> +<span class="sourceLineNo">696</span> @VisibleForTesting<a name="line.696"></a> +<span class="sourceLineNo">697</span> void sanityCheckUncompressedSize() throws IOException {<a name="line.697"></a> +<span class="sourceLineNo">698</span> if (onDiskSizeWithoutHeader != uncompressedSizeWithoutHeader + totalChecksumBytes()) {<a name="line.698"></a> +<span class="sourceLineNo">699</span> throw new IOException("Using no compression but "<a name="line.699"></a> +<span class="sourceLineNo">700</span> + "onDiskSizeWithoutHeader=" + onDiskSizeWithoutHeader + ", "<a name="line.700"></a> +<span class="sourceLineNo">701</span> + "uncompressedSizeWithoutHeader=" + uncompressedSizeWithoutHeader<a name="line.701"></a> +<span class="sourceLineNo">702</span> + ", numChecksumbytes=" + totalChecksumBytes());<a name="line.702"></a> +<span class="sourceLineNo">703</span> }<a name="line.703"></a> +<span class="sourceLineNo">704</span> }<a name="line.704"></a> +<span class="sourceLineNo">705</span><a name="line.705"></a> +<span class="sourceLineNo">706</span> /**<a name="line.706"></a> +<span class="sourceLineNo">707</span> * Cannot be {@link #UNSET}. Must be a legitimate value. Used re-making the {@link BlockCacheKey} when<a name="line.707"></a> +<span class="sourceLineNo">708</span> * block is returned to the cache.<a name="line.708"></a> +<span class="sourceLineNo">709</span> * @return the offset of this block in the file it was read from<a name="line.709"></a> +<span class="sourceLineNo">710</span> */<a name="line.710"></a> +<span class="sourceLineNo">711</span> long getOffset() {<a name="line.711"></a> +<span class="sourceLineNo">712</span> if (offset < 0) {<a name="line.712"></a> +<span class="sourceLineNo">713</span> throw new IllegalStateException("HFile block offset not initialized properly");<a name="line.713"></a> +<span class="sourceLineNo">714</span> }<a name="line.714"></a> +<span class="sourceLineNo">715</span> return offset;<a name="line.715"></a> +<span class="sourceLineNo">716</span> }<a name="line.716"></a> +<span class="sourceLineNo">717</span><a name="line.717"></a> +<span class="sourceLineNo">718</span> /**<a name="line.718"></a> +<span class="sourceLineNo">719</span> * @return a byte stream reading the data + checksum of this block<a name="line.719"></a> +<span class="sourceLineNo">720</span> */<a name="line.720"></a> +<span class="sourceLineNo">721</span> DataInputStream getByteStream() {<a name="line.721"></a> +<span class="sourceLineNo">722</span> ByteBuff dup = this.buf.duplicate();<a name="line.722"></a> +<span class="sourceLineNo">723</span> dup.position(this.headerSize());<a name="line.723"></a> +<span class="sourceLineNo">724</span> return new DataInputStream(new ByteBuffInputStream(dup));<a name="line.724"></a> +<span class="sourceLineNo">725</span> }<a name="line.725"></a> +<span class="sourceLineNo">726</span><a name="line.726"></a> +<span class="sourceLineNo">727</span> @Override<a name="line.727"></a> +<span class="sourceLineNo">728</span> public long heapSize() {<a name="line.728"></a> +<span class="sourceLineNo">729</span> long size = ClassSize.align(<a name="line.729"></a> +<span class="sourceLineNo">730</span> ClassSize.OBJECT +<a name="line.730"></a> +<span class="sourceLineNo">731</span> // Block type, multi byte buffer, MemoryType and meta references<a name="line.731"></a> +<span class="sourceLineNo">732</span> 4 * ClassSize.REFERENCE +<a name="line.732"></a> +<span class="sourceLineNo">733</span> // On-disk size, uncompressed size, and next block's on-disk size<a name="line.733"></a> +<span class="sourceLineNo">734</span> // bytePerChecksum and onDiskDataSize<a name="line.734"></a> +<span class="sourceLineNo">735</span> 4 * Bytes.SIZEOF_INT +<a name="line.735"></a> +<span class="sourceLineNo">736</span> // This and previous block offset<a name="line.736"></a> +<span class="sourceLineNo">737</span> 2 * Bytes.SIZEOF_LONG +<a name="line.737"></a> +<span class="sourceLineNo">738</span> // Heap size of the meta object. meta will be always not null.<a name="line.738"></a> +<span class="sourceLineNo">739</span> fileContext.heapSize()<a name="line.739"></a> +<span class="sourceLineNo">740</span> );<a name="line.740"></a> +<span class="sourceLineNo">741</span><a name="line.741"></a> +<span class="sourceLineNo">742</span> if (buf != null) {<a name="line.742"></a> +<span class="sourceLineNo">743</span> // Deep overhead of the byte buffer. Needs to be aligned separately.<a name="line.743"></a> +<span class="sourceLineNo">744</span> size += ClassSize.align(buf.capacity() + MULTI_BYTE_BUFFER_HEAP_SIZE);<a name="line.744"></a> +<span class="sourceLineNo">745</span> }<a name="line.745"></a> +<span class="sourceLineNo">746</span><a name="line.746"></a> +<span class="sourceLineNo">747</span> return ClassSize.align(size);<a name="line.747"></a> +<span class="sourceLineNo">748</span> }<a name="line.748"></a> +<span class="sourceLineNo">749</span><a name="line.749"></a> +<span class="sourceLineNo">750</span> /**<a name="line.750"></a> +<span class="sourceLineNo">751</span> * Read from an input stream at least <code>necessaryLen</code> and if possible,<a name="line.751"></a> +<span class="sourceLineNo">752</span> * <code>extraLen</code> also if available. Analogous to<a name="line.752"></a> +<span class="sourceLineNo">753</span> * {@link IOUtils#readFully(InputStream, byte[], int, int)}, but specifies a<a name="line.753"></a> +<span class="sourceLineNo">754</span> * number of "extra" bytes to also optionally read.<a name="line.754"></a> +<span class="sourceLineNo">755</span> *<a name="line.755"></a> +<span class="sourceLineNo">756</span> * @param in the input stream to read from<a name="line.756"></a> +<span class="sourceLineNo">757</span> * @param buf the buffer to read into<a name="line.757"></a> +<span class="sourceLineNo">758</span> * @param bufOffset the destination offset in the buffer<a name="line.758"></a> +<span class="sourceLineNo">759</span> * @param necessaryLen the number of bytes that are absolutely necessary to read<a name="line.759"></a> +<span class="sourceLineNo">760</span> * @param extraLen the number of extra bytes that would be nice to read<a name="line.760"></a> +<span class="sourceLineNo">761</span> * @return true if succeeded reading the extra bytes<a name="line.761"></a> +<span class="sourceLineNo">762</span> * @throws IOException if failed to read the necessary bytes<a name="line.762"></a> +<span class="sourceLineNo">763</span> */<a name="line.763"></a> +<span class="sourceLineNo">764</span> static boolean readWithExtra(InputStream in, byte[] buf,<a name="line.764"></a> +<span class="sourceLineNo">765</span> int bufOffset, int necessaryLen, int extraLen) throws IOException {<a name="line.765"></a> +<span class="sourceLineNo">766</span> int bytesRemaining = necessaryLen + extraLen;<a name="line.766"></a> +<span class="sourceLineNo">767</span> while (bytesRemaining > 0) {<a name="line.767"></a> +<span class="sourceLineNo">768</span> int ret = in.read(buf, bufOffset, bytesRemaining);<a name="line.768"></a> +<span class="sourceLineNo">769</span> if (ret == -1 && bytesRemaining <= extraLen) {<a name="line.769"></a> +<span class="sourceLineNo">770</span> // We could not read the "extra data", but that is OK.<a name="line.770"></a> +<span class="sourceLineNo">771</span> break;<a name="line.771"></a> +<span class="sourceLineNo">772</span> }<a name="line.772"></a> +<span class="sourceLineNo">773</span> if (ret < 0) {<a name="line.773"></a> +<span class="sourceLineNo">774</span> throw new IOException("Premature EOF from inputStream (read "<a name="line.774"></a> +<span class="sourceLineNo">775</span> + "returned " + ret + ", was trying to read " + necessaryLen<a name="line.775"></a> +<span class="sourceLineNo">776</span> + " necessary bytes and " + extraLen + " extra bytes, "<a name="line.776"></a> +<span class="sourceLineNo">777</span> + "successfully read "<a name="line.777"></a> +<span class="sourceLineNo">778</span> + (necessaryLen + extraLen - bytesRemaining));<a name="line.778"></a> +<span class="sourceLineNo">779</span> }<a name="line.779"></a> +<span class="sourceLineNo">780</span> bufOffset += ret;<a name="line.780"></a> +<span class="sourceLineNo">781</span> bytesRemaining -= ret;<a name="line.781"></a> +<span class="sourceLineNo">782</span> }<a name="line.782"></a> +<span class="sourceLineNo">783</span> return bytesRemaining <= 0;<a name="line.783"></a> +<span class="sourceLineNo">784</span> }<a name="line.784"></a> +<span class="sourceLineNo">785</span><a name="line.785"></a> +<span class="sourceLineNo">786</span> /**<a name="line.786"></a> +<span class="sourceLineNo">787</span> * Read from an input stream at least <code>necessaryLen</code> and if possible,<a name="line.787"></a> +<span class="sourceLineNo">788</span> * <code>extraLen</code> also if available. Analogous to<a name="line.788"></a> +<span class="sourceLineNo">789</span> * {@link IOUtils#readFully(InputStream, byte[], int, int)}, but uses<a name="line.789"></a> +<span class="sourceLineNo">790</span> * positional read and specifies a number of "extra" bytes that would be<a name="line.790"></a> +<span class="sourceLineNo">791</span> * desirable but not absolutely necessary to read.<a name="line.791"></a> +<span class="sourceLineNo">792</span> *<a name="line.792"></a> +<span class="sourceLineNo">793</span> * @param in the input stream to read from<a name="line.793"></a> +<span class="sourceLineNo">794</span> * @param position the position within the stream from which to start reading<a name="line.794"></a> +<span class="sourceLineNo">795</span> * @param buf the buffer to read into<a name="line.795"></a> +<span class="sourceLineNo">796</span> * @param bufOffset the destination offset in the buffer<a name="line.796"></a> +<span class="sourceLineNo">797</span> * @param necessaryLen the number of bytes that are absolutely necessary to<a name="line.797"></a> +<span class="sourceLineNo">798</span> * read<a name="line.798"></a> +<span class="sourceLineNo">799</span> * @param extraLen the number of extra bytes that would be nice to read<a name="line.799"></a> +<span class="sourceLineNo">800</span> * @return true if and only if extraLen is > 0 and reading those extra bytes<a name="line.800"></a> +<span class="sourceLineNo">801</span> * was successful<a name="line.801"></a> +<span class="sourceLineNo">802</span> * @throws IOException if failed to read the necessary bytes<a name="line.802"></a> +<span class="sourceLineNo">803</span> */<a name="line.803"></a> +<span class="sourceLineNo">804</span> @VisibleForTesting<a name="line.804"></a> +<span class="sourceLineNo">805</span> static boolean positionalReadWithExtra(FSDataInputStream in,<a name="line.805"></a> +<span class="sourceLineNo">806</span> long position, byte[] buf, int bufOffset, int necessaryLen, int extraLen)<a name="line.806"></a> +<span class="sourceLineNo">807</span> throws IOException {<a name="line.807"></a> +<span class="sourceLineNo">808</span> int bytesRemaining = necessaryLen + extraLen;<a name="line.808"></a> +<span class="sourceLineNo">809</span> int bytesRead = 0;<a name="line.809"></a> +<span class="sourceLineNo">810</span> while (bytesRead < necessaryLen) {<a name="line.810"></a> +<span class="sourceLineNo">811</span> int ret = in.read(position, buf, bufOffset, bytesRemaining);<a name="line.811"></a> +<span class="sourceLineNo">812</span> if (ret < 0) {<a name="line.812"></a> +<span class="sourceLineNo">813</span> throw new IOException("Premature EOF from inputStream (positional read "<a name="line.813"></a> +<span class="sourceLineNo">814</span> + "returned " + ret + ", was trying to read " + necessaryLen<a name="line.814"></a> +<span class="sourceLineNo">815</span> + " necessary bytes and " + extraLen + " extra bytes, "<a name="line.815"></a> +<span class="sourceLineNo">816</span> + "successfully read " + bytesRead);<a name="line.816"></a> +<span class="sourceLineNo">817</span> }<a name="line.817"></a> +<span class="sourceLineNo">818</span> position += ret;<a name="line.818"></a> +<span class="sourceLineNo">819</span> bufOffset += ret;<a name="line.819"></a> +<span class="sourceLineNo">820</span> bytesRemaining -= ret;<a name="line.820"></a> +<span class="sourceLineNo">821</span> bytesRead += ret;<a name="line.821"></a> +<span class="sourceLineNo">822</span> }<a name="line.822"></a> +<span class="sourceLineNo">823</span> return bytesRead != necessaryLen && bytesRemaining <= 0;<a name="line.823"></a> +<span class="sourceLineNo">824</span> }<a name="line.824"></a> +<span class="sourceLineNo">825</span><a name="line.825"></a> +<span class="sourceLineNo">826</span> /**<a name="line.826"></a> +<span class="sourceLineNo">827</span> * Unified version 2 {@link HFile} block writer. The intended usage pattern<a name="line.827"></a> +<span class="sourceLineNo">828</span> * is as follows:<a name="line.828"></a> +<span class="sourceLineNo">829</span> * <ol><a name="line.829"></a> +<span class="sourceLineNo">830</span> * <li>Construct an {@link HFileBlock.Writer}, providing a compression algorithm.<a name="line.830"></a> +<span class="sourceLineNo">831</span> * <li>Call {@link Writer#startWriting} and get a data stream to write to.<a name="line.831"></a> +<span class="sourceLineNo">832</span> * <li>Write your data into the stream.<a name="line.832"></a> +<span class="sourceLineNo">833</span> * <li>Call Writer#writeHeaderAndData(FSDataOutputStream) as many times as you need to.<a name="line.833"></a> +<span class="sourceLineNo">834</span> * store the serialized block into an external stream.<a name="line.834"></a> +<span class="sourceLineNo">835</span> * <li>Repeat to write more blocks.<a name="line.835"></a> +<span class="sourceLineNo">836</span> * </ol><a name="line.836"></a> +<span class="sourceLineNo">837</span> * <p><a name="line.837"></a> +<span class="sourceLineNo">838</span> */<a name="line.838"></a> +<span class="sourceLineNo">839</span> static class Writer {<a name="line.839"></a> +<span class="sourceLineNo">840</span> private enum State {<a name="line.840"></a> +<span class="sourceLineNo">841</span> INIT,<a name="line.841"></a> +<span class="sourceLineNo">842</span> WRITING,<a name="line.842"></a> +<span class="sourceLineNo">843</span> BLOCK_READY<a name="line.843"></a> +<span class="sourceLineNo">844</span> }<a name="line.844"></a> +<span class="sourceLineNo">845</span><a name="line.845"></a> +<span class="sourceLineNo">846</span> /** Writer state. Used to ensure the correct usage protocol. */<a name="line.846"></a> +<span class="sourceLineNo">847</span> private State state = State.INIT;<a name="line.847"></a> +<span class="sourceLineNo">848</span><a name="line.848"></a> +<span class="sourceLineNo">849</span> /** Data block encoder used for data blocks */<a name="line.849"></a> +<span class="sourceLineNo">850</span> private final HFileDataBlockEncoder dataBlockEncoder;<a name="line.850"></a> +<span class="sourceLineNo">851</span><a name="line.851"></a> +<span class="sourceLineNo">852</span> private HFileBlockEncodingContext dataBlockEncodingCtx;<a name="line.852"></a> +<span class="sourceLineNo">853</span><a name="line.853"></a> +<span class="sourceLineNo">854</span> /** block encoding context for non-data blocks*/<a name="line.854"></a> +<span class="sourceLineNo">855</span> private HFileBlockDefaultEncodingContext defaultBlockEncodingCtx;<a name="line.855"></a> +<span class="sourceLineNo">856</span><a name="line.856"></a> +<span class="sourceLineNo">857</span> /**<a name="line.857"></a> +<span class="sourceLineNo">858</span> * The stream we use to accumulate data into a blo
<TRUNCATED>
