http://git-wip-us.apache.org/repos/asf/hbase-site/blob/495ddb86/devapidocs/src-html/org/apache/hadoop/hbase/filter/FuzzyRowFilter.SatisfiesCode.html ---------------------------------------------------------------------- diff --git a/devapidocs/src-html/org/apache/hadoop/hbase/filter/FuzzyRowFilter.SatisfiesCode.html b/devapidocs/src-html/org/apache/hadoop/hbase/filter/FuzzyRowFilter.SatisfiesCode.html index a310286..e6ec67b 100644 --- a/devapidocs/src-html/org/apache/hadoop/hbase/filter/FuzzyRowFilter.SatisfiesCode.html +++ b/devapidocs/src-html/org/apache/hadoop/hbase/filter/FuzzyRowFilter.SatisfiesCode.html @@ -84,552 +84,560 @@ <span class="sourceLineNo">076</span> private RowTracker tracker;<a name="line.76"></a> <span class="sourceLineNo">077</span><a name="line.77"></a> <span class="sourceLineNo">078</span> public FuzzyRowFilter(List<Pair<byte[], byte[]>> fuzzyKeysData) {<a name="line.78"></a> -<span class="sourceLineNo">079</span> Pair<byte[], byte[]> p;<a name="line.79"></a> -<span class="sourceLineNo">080</span> for (int i = 0; i < fuzzyKeysData.size(); i++) {<a name="line.80"></a> -<span class="sourceLineNo">081</span> p = fuzzyKeysData.get(i);<a name="line.81"></a> -<span class="sourceLineNo">082</span> if (p.getFirst().length != p.getSecond().length) {<a name="line.82"></a> +<span class="sourceLineNo">079</span> List<Pair<byte[], byte[]>> fuzzyKeyDataCopy = new ArrayList<>(fuzzyKeysData.size());<a name="line.79"></a> +<span class="sourceLineNo">080</span><a name="line.80"></a> +<span class="sourceLineNo">081</span> for (Pair<byte[], byte[]> aFuzzyKeysData : fuzzyKeysData) {<a name="line.81"></a> +<span class="sourceLineNo">082</span> if (aFuzzyKeysData.getFirst().length != aFuzzyKeysData.getSecond().length) {<a name="line.82"></a> <span class="sourceLineNo">083</span> Pair<String, String> readable =<a name="line.83"></a> -<span class="sourceLineNo">084</span> new Pair<>(Bytes.toStringBinary(p.getFirst()), Bytes.toStringBinary(p<a name="line.84"></a> -<span class="sourceLineNo">085</span> .getSecond()));<a name="line.85"></a> -<span class="sourceLineNo">086</span> throw new IllegalArgumentException("Fuzzy pair lengths do not match: " + readable);<a name="line.86"></a> -<span class="sourceLineNo">087</span> }<a name="line.87"></a> -<span class="sourceLineNo">088</span> // update mask ( 0 -> -1 (0xff), 1 -> 2)<a name="line.88"></a> -<span class="sourceLineNo">089</span> p.setSecond(preprocessMask(p.getSecond()));<a name="line.89"></a> -<span class="sourceLineNo">090</span> preprocessSearchKey(p);<a name="line.90"></a> -<span class="sourceLineNo">091</span> }<a name="line.91"></a> -<span class="sourceLineNo">092</span> this.fuzzyKeysData = fuzzyKeysData;<a name="line.92"></a> -<span class="sourceLineNo">093</span> this.tracker = new RowTracker();<a name="line.93"></a> -<span class="sourceLineNo">094</span> }<a name="line.94"></a> -<span class="sourceLineNo">095</span><a name="line.95"></a> -<span class="sourceLineNo">096</span> private void preprocessSearchKey(Pair<byte[], byte[]> p) {<a name="line.96"></a> -<span class="sourceLineNo">097</span> if (!UNSAFE_UNALIGNED) {<a name="line.97"></a> -<span class="sourceLineNo">098</span> // do nothing<a name="line.98"></a> -<span class="sourceLineNo">099</span> return;<a name="line.99"></a> -<span class="sourceLineNo">100</span> }<a name="line.100"></a> -<span class="sourceLineNo">101</span> byte[] key = p.getFirst();<a name="line.101"></a> -<span class="sourceLineNo">102</span> byte[] mask = p.getSecond();<a name="line.102"></a> -<span class="sourceLineNo">103</span> for (int i = 0; i < mask.length; i++) {<a name="line.103"></a> -<span class="sourceLineNo">104</span> // set non-fixed part of a search key to 0.<a name="line.104"></a> -<span class="sourceLineNo">105</span> if (mask[i] == 2) {<a name="line.105"></a> -<span class="sourceLineNo">106</span> key[i] = 0;<a name="line.106"></a> -<span class="sourceLineNo">107</span> }<a name="line.107"></a> +<span class="sourceLineNo">084</span> new Pair<>(Bytes.toStringBinary(aFuzzyKeysData.getFirst()), Bytes.toStringBinary(aFuzzyKeysData.getSecond()));<a name="line.84"></a> +<span class="sourceLineNo">085</span> throw new IllegalArgumentException("Fuzzy pair lengths do not match: " + readable);<a name="line.85"></a> +<span class="sourceLineNo">086</span> }<a name="line.86"></a> +<span class="sourceLineNo">087</span><a name="line.87"></a> +<span class="sourceLineNo">088</span> Pair<byte[], byte[]> p = new Pair<>();<a name="line.88"></a> +<span class="sourceLineNo">089</span> // create a copy of pair bytes so that they are not modified by the filter.<a name="line.89"></a> +<span class="sourceLineNo">090</span> p.setFirst(Arrays.copyOf(aFuzzyKeysData.getFirst(), aFuzzyKeysData.getFirst().length));<a name="line.90"></a> +<span class="sourceLineNo">091</span> p.setSecond(Arrays.copyOf(aFuzzyKeysData.getSecond(), aFuzzyKeysData.getSecond().length));<a name="line.91"></a> +<span class="sourceLineNo">092</span><a name="line.92"></a> +<span class="sourceLineNo">093</span> // update mask ( 0 -> -1 (0xff), 1 -> 2)<a name="line.93"></a> +<span class="sourceLineNo">094</span> p.setSecond(preprocessMask(p.getSecond()));<a name="line.94"></a> +<span class="sourceLineNo">095</span> preprocessSearchKey(p);<a name="line.95"></a> +<span class="sourceLineNo">096</span><a name="line.96"></a> +<span class="sourceLineNo">097</span> fuzzyKeyDataCopy.add(p);<a name="line.97"></a> +<span class="sourceLineNo">098</span> }<a name="line.98"></a> +<span class="sourceLineNo">099</span> this.fuzzyKeysData = fuzzyKeyDataCopy;<a name="line.99"></a> +<span class="sourceLineNo">100</span> this.tracker = new RowTracker();<a name="line.100"></a> +<span class="sourceLineNo">101</span> }<a name="line.101"></a> +<span class="sourceLineNo">102</span><a name="line.102"></a> +<span class="sourceLineNo">103</span><a name="line.103"></a> +<span class="sourceLineNo">104</span> private void preprocessSearchKey(Pair<byte[], byte[]> p) {<a name="line.104"></a> +<span class="sourceLineNo">105</span> if (!UNSAFE_UNALIGNED) {<a name="line.105"></a> +<span class="sourceLineNo">106</span> // do nothing<a name="line.106"></a> +<span class="sourceLineNo">107</span> return;<a name="line.107"></a> <span class="sourceLineNo">108</span> }<a name="line.108"></a> -<span class="sourceLineNo">109</span> }<a name="line.109"></a> -<span class="sourceLineNo">110</span><a name="line.110"></a> -<span class="sourceLineNo">111</span> /**<a name="line.111"></a> -<span class="sourceLineNo">112</span> * We need to preprocess mask array, as since we treat 2's as unfixed positions and -1 (0xff) as<a name="line.112"></a> -<span class="sourceLineNo">113</span> * fixed positions<a name="line.113"></a> -<span class="sourceLineNo">114</span> * @param mask<a name="line.114"></a> -<span class="sourceLineNo">115</span> * @return mask array<a name="line.115"></a> -<span class="sourceLineNo">116</span> */<a name="line.116"></a> -<span class="sourceLineNo">117</span> private byte[] preprocessMask(byte[] mask) {<a name="line.117"></a> -<span class="sourceLineNo">118</span> if (!UNSAFE_UNALIGNED) {<a name="line.118"></a> -<span class="sourceLineNo">119</span> // do nothing<a name="line.119"></a> -<span class="sourceLineNo">120</span> return mask;<a name="line.120"></a> -<span class="sourceLineNo">121</span> }<a name="line.121"></a> -<span class="sourceLineNo">122</span> if (isPreprocessedMask(mask)) return mask;<a name="line.122"></a> -<span class="sourceLineNo">123</span> for (int i = 0; i < mask.length; i++) {<a name="line.123"></a> -<span class="sourceLineNo">124</span> if (mask[i] == 0) {<a name="line.124"></a> -<span class="sourceLineNo">125</span> mask[i] = -1; // 0 -> -1<a name="line.125"></a> -<span class="sourceLineNo">126</span> } else if (mask[i] == 1) {<a name="line.126"></a> -<span class="sourceLineNo">127</span> mask[i] = 2;// 1 -> 2<a name="line.127"></a> -<span class="sourceLineNo">128</span> }<a name="line.128"></a> +<span class="sourceLineNo">109</span> byte[] key = p.getFirst();<a name="line.109"></a> +<span class="sourceLineNo">110</span> byte[] mask = p.getSecond();<a name="line.110"></a> +<span class="sourceLineNo">111</span> for (int i = 0; i < mask.length; i++) {<a name="line.111"></a> +<span class="sourceLineNo">112</span> // set non-fixed part of a search key to 0.<a name="line.112"></a> +<span class="sourceLineNo">113</span> if (mask[i] == 2) {<a name="line.113"></a> +<span class="sourceLineNo">114</span> key[i] = 0;<a name="line.114"></a> +<span class="sourceLineNo">115</span> }<a name="line.115"></a> +<span class="sourceLineNo">116</span> }<a name="line.116"></a> +<span class="sourceLineNo">117</span> }<a name="line.117"></a> +<span class="sourceLineNo">118</span><a name="line.118"></a> +<span class="sourceLineNo">119</span> /**<a name="line.119"></a> +<span class="sourceLineNo">120</span> * We need to preprocess mask array, as since we treat 2's as unfixed positions and -1 (0xff) as<a name="line.120"></a> +<span class="sourceLineNo">121</span> * fixed positions<a name="line.121"></a> +<span class="sourceLineNo">122</span> * @param mask<a name="line.122"></a> +<span class="sourceLineNo">123</span> * @return mask array<a name="line.123"></a> +<span class="sourceLineNo">124</span> */<a name="line.124"></a> +<span class="sourceLineNo">125</span> private byte[] preprocessMask(byte[] mask) {<a name="line.125"></a> +<span class="sourceLineNo">126</span> if (!UNSAFE_UNALIGNED) {<a name="line.126"></a> +<span class="sourceLineNo">127</span> // do nothing<a name="line.127"></a> +<span class="sourceLineNo">128</span> return mask;<a name="line.128"></a> <span class="sourceLineNo">129</span> }<a name="line.129"></a> -<span class="sourceLineNo">130</span> return mask;<a name="line.130"></a> -<span class="sourceLineNo">131</span> }<a name="line.131"></a> -<span class="sourceLineNo">132</span><a name="line.132"></a> -<span class="sourceLineNo">133</span> private boolean isPreprocessedMask(byte[] mask) {<a name="line.133"></a> -<span class="sourceLineNo">134</span> for (int i = 0; i < mask.length; i++) {<a name="line.134"></a> -<span class="sourceLineNo">135</span> if (mask[i] != -1 && mask[i] != 2) {<a name="line.135"></a> -<span class="sourceLineNo">136</span> return false;<a name="line.136"></a> -<span class="sourceLineNo">137</span> }<a name="line.137"></a> -<span class="sourceLineNo">138</span> }<a name="line.138"></a> -<span class="sourceLineNo">139</span> return true;<a name="line.139"></a> -<span class="sourceLineNo">140</span> }<a name="line.140"></a> -<span class="sourceLineNo">141</span><a name="line.141"></a> -<span class="sourceLineNo">142</span> @Override<a name="line.142"></a> -<span class="sourceLineNo">143</span> public ReturnCode filterKeyValue(Cell c) {<a name="line.143"></a> -<span class="sourceLineNo">144</span> final int startIndex = lastFoundIndex >= 0 ? lastFoundIndex : 0;<a name="line.144"></a> -<span class="sourceLineNo">145</span> final int size = fuzzyKeysData.size();<a name="line.145"></a> -<span class="sourceLineNo">146</span> for (int i = startIndex; i < size + startIndex; i++) {<a name="line.146"></a> -<span class="sourceLineNo">147</span> final int index = i % size;<a name="line.147"></a> -<span class="sourceLineNo">148</span> Pair<byte[], byte[]> fuzzyData = fuzzyKeysData.get(index);<a name="line.148"></a> -<span class="sourceLineNo">149</span> // This shift is idempotent - always end up with 0 and -1 as mask values.<a name="line.149"></a> -<span class="sourceLineNo">150</span> for (int j = 0; j < fuzzyData.getSecond().length; j++) {<a name="line.150"></a> -<span class="sourceLineNo">151</span> fuzzyData.getSecond()[j] >>= 2;<a name="line.151"></a> -<span class="sourceLineNo">152</span> }<a name="line.152"></a> -<span class="sourceLineNo">153</span> SatisfiesCode satisfiesCode =<a name="line.153"></a> -<span class="sourceLineNo">154</span> satisfies(isReversed(), c.getRowArray(), c.getRowOffset(), c.getRowLength(),<a name="line.154"></a> -<span class="sourceLineNo">155</span> fuzzyData.getFirst(), fuzzyData.getSecond());<a name="line.155"></a> -<span class="sourceLineNo">156</span> if (satisfiesCode == SatisfiesCode.YES) {<a name="line.156"></a> -<span class="sourceLineNo">157</span> lastFoundIndex = index;<a name="line.157"></a> -<span class="sourceLineNo">158</span> return ReturnCode.INCLUDE;<a name="line.158"></a> -<span class="sourceLineNo">159</span> }<a name="line.159"></a> -<span class="sourceLineNo">160</span> }<a name="line.160"></a> -<span class="sourceLineNo">161</span> // NOT FOUND -> seek next using hint<a name="line.161"></a> -<span class="sourceLineNo">162</span> lastFoundIndex = -1;<a name="line.162"></a> -<span class="sourceLineNo">163</span><a name="line.163"></a> -<span class="sourceLineNo">164</span> return ReturnCode.SEEK_NEXT_USING_HINT;<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><a name="line.167"></a> -<span class="sourceLineNo">168</span> @Override<a name="line.168"></a> -<span class="sourceLineNo">169</span> public Cell getNextCellHint(Cell currentCell) {<a name="line.169"></a> -<span class="sourceLineNo">170</span> boolean result = tracker.updateTracker(currentCell);<a name="line.170"></a> -<span class="sourceLineNo">171</span> if (result == false) {<a name="line.171"></a> -<span class="sourceLineNo">172</span> done = true;<a name="line.172"></a> -<span class="sourceLineNo">173</span> return null;<a name="line.173"></a> -<span class="sourceLineNo">174</span> }<a name="line.174"></a> -<span class="sourceLineNo">175</span> byte[] nextRowKey = tracker.nextRow();<a name="line.175"></a> -<span class="sourceLineNo">176</span> return CellUtil.createFirstOnRow(nextRowKey, 0, (short) nextRowKey.length);<a name="line.176"></a> -<span class="sourceLineNo">177</span> }<a name="line.177"></a> -<span class="sourceLineNo">178</span><a name="line.178"></a> -<span class="sourceLineNo">179</span> /**<a name="line.179"></a> -<span class="sourceLineNo">180</span> * If we have multiple fuzzy keys, row tracker should improve overall performance. It calculates<a name="line.180"></a> -<span class="sourceLineNo">181</span> * all next rows (one per every fuzzy key) and put them (the fuzzy key is bundled) into a priority<a name="line.181"></a> -<span class="sourceLineNo">182</span> * queue so that the smallest row key always appears at queue head, which helps to decide the<a name="line.182"></a> -<span class="sourceLineNo">183</span> * "Next Cell Hint". As scanning going on, the number of candidate rows in the RowTracker will<a name="line.183"></a> -<span class="sourceLineNo">184</span> * remain the size of fuzzy keys until some of the fuzzy keys won't possibly have matches any<a name="line.184"></a> -<span class="sourceLineNo">185</span> * more.<a name="line.185"></a> -<span class="sourceLineNo">186</span> */<a name="line.186"></a> -<span class="sourceLineNo">187</span> private class RowTracker {<a name="line.187"></a> -<span class="sourceLineNo">188</span> private final PriorityQueue<Pair<byte[], Pair<byte[], byte[]>>> nextRows;<a name="line.188"></a> -<span class="sourceLineNo">189</span> private boolean initialized = false;<a name="line.189"></a> -<span class="sourceLineNo">190</span><a name="line.190"></a> -<span class="sourceLineNo">191</span> RowTracker() {<a name="line.191"></a> -<span class="sourceLineNo">192</span> nextRows = new PriorityQueue<>(fuzzyKeysData.size(),<a name="line.192"></a> -<span class="sourceLineNo">193</span> new Comparator<Pair<byte[], Pair<byte[], byte[]>>>() {<a name="line.193"></a> -<span class="sourceLineNo">194</span> @Override<a name="line.194"></a> -<span class="sourceLineNo">195</span> public int compare(Pair<byte[], Pair<byte[], byte[]>> o1,<a name="line.195"></a> -<span class="sourceLineNo">196</span> Pair<byte[], Pair<byte[], byte[]>> o2) {<a name="line.196"></a> -<span class="sourceLineNo">197</span> return isReversed()? Bytes.compareTo(o2.getFirst(), o1.getFirst()):<a name="line.197"></a> -<span class="sourceLineNo">198</span> Bytes.compareTo(o1.getFirst(), o2.getFirst());<a name="line.198"></a> -<span class="sourceLineNo">199</span> }<a name="line.199"></a> -<span class="sourceLineNo">200</span> });<a name="line.200"></a> -<span class="sourceLineNo">201</span> }<a name="line.201"></a> -<span class="sourceLineNo">202</span><a name="line.202"></a> -<span class="sourceLineNo">203</span> byte[] nextRow() {<a name="line.203"></a> -<span class="sourceLineNo">204</span> if (nextRows.isEmpty()) {<a name="line.204"></a> -<span class="sourceLineNo">205</span> throw new IllegalStateException(<a name="line.205"></a> -<span class="sourceLineNo">206</span> "NextRows should not be empty, make sure to call nextRow() after updateTracker() return true");<a name="line.206"></a> -<span class="sourceLineNo">207</span> } else {<a name="line.207"></a> -<span class="sourceLineNo">208</span> return nextRows.peek().getFirst();<a name="line.208"></a> -<span class="sourceLineNo">209</span> }<a name="line.209"></a> -<span class="sourceLineNo">210</span> }<a name="line.210"></a> -<span class="sourceLineNo">211</span><a name="line.211"></a> -<span class="sourceLineNo">212</span> boolean updateTracker(Cell currentCell) {<a name="line.212"></a> -<span class="sourceLineNo">213</span> if (!initialized) {<a name="line.213"></a> -<span class="sourceLineNo">214</span> for (Pair<byte[], byte[]> fuzzyData : fuzzyKeysData) {<a name="line.214"></a> -<span class="sourceLineNo">215</span> updateWith(currentCell, fuzzyData);<a name="line.215"></a> -<span class="sourceLineNo">216</span> }<a name="line.216"></a> -<span class="sourceLineNo">217</span> initialized = true;<a name="line.217"></a> -<span class="sourceLineNo">218</span> } else {<a name="line.218"></a> -<span class="sourceLineNo">219</span> while (!nextRows.isEmpty() && !lessThan(currentCell, nextRows.peek().getFirst())) {<a name="line.219"></a> -<span class="sourceLineNo">220</span> Pair<byte[], Pair<byte[], byte[]>> head = nextRows.poll();<a name="line.220"></a> -<span class="sourceLineNo">221</span> Pair<byte[], byte[]> fuzzyData = head.getSecond();<a name="line.221"></a> -<span class="sourceLineNo">222</span> updateWith(currentCell, fuzzyData);<a name="line.222"></a> -<span class="sourceLineNo">223</span> }<a name="line.223"></a> -<span class="sourceLineNo">224</span> }<a name="line.224"></a> -<span class="sourceLineNo">225</span> return !nextRows.isEmpty();<a name="line.225"></a> -<span class="sourceLineNo">226</span> }<a name="line.226"></a> -<span class="sourceLineNo">227</span><a name="line.227"></a> -<span class="sourceLineNo">228</span> boolean lessThan(Cell currentCell, byte[] nextRowKey) {<a name="line.228"></a> -<span class="sourceLineNo">229</span> int compareResult =<a name="line.229"></a> -<span class="sourceLineNo">230</span> CellComparator.COMPARATOR.compareRows(currentCell, nextRowKey, 0, nextRowKey.length);<a name="line.230"></a> -<span class="sourceLineNo">231</span> return (!isReversed() && compareResult < 0) || (isReversed() && compareResult > 0);<a name="line.231"></a> -<span class="sourceLineNo">232</span> }<a name="line.232"></a> -<span class="sourceLineNo">233</span><a name="line.233"></a> -<span class="sourceLineNo">234</span> void updateWith(Cell currentCell, Pair<byte[], byte[]> fuzzyData) {<a name="line.234"></a> -<span class="sourceLineNo">235</span> byte[] nextRowKeyCandidate =<a name="line.235"></a> -<span class="sourceLineNo">236</span> getNextForFuzzyRule(isReversed(), currentCell.getRowArray(), currentCell.getRowOffset(),<a name="line.236"></a> -<span class="sourceLineNo">237</span> currentCell.getRowLength(), fuzzyData.getFirst(), fuzzyData.getSecond());<a name="line.237"></a> -<span class="sourceLineNo">238</span> if (nextRowKeyCandidate != null) {<a name="line.238"></a> -<span class="sourceLineNo">239</span> nextRows.add(new Pair<>(nextRowKeyCandidate, fuzzyData));<a name="line.239"></a> -<span class="sourceLineNo">240</span> }<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> }<a name="line.243"></a> -<span class="sourceLineNo">244</span><a name="line.244"></a> -<span class="sourceLineNo">245</span> @Override<a name="line.245"></a> -<span class="sourceLineNo">246</span> public boolean filterAllRemaining() {<a name="line.246"></a> -<span class="sourceLineNo">247</span> return done;<a name="line.247"></a> -<span class="sourceLineNo">248</span> }<a name="line.248"></a> -<span class="sourceLineNo">249</span><a name="line.249"></a> -<span class="sourceLineNo">250</span> /**<a name="line.250"></a> -<span class="sourceLineNo">251</span> * @return The filter serialized using pb<a name="line.251"></a> -<span class="sourceLineNo">252</span> */<a name="line.252"></a> -<span class="sourceLineNo">253</span> public byte[] toByteArray() {<a name="line.253"></a> -<span class="sourceLineNo">254</span> FilterProtos.FuzzyRowFilter.Builder builder = FilterProtos.FuzzyRowFilter.newBuilder();<a name="line.254"></a> -<span class="sourceLineNo">255</span> for (Pair<byte[], byte[]> fuzzyData : fuzzyKeysData) {<a name="line.255"></a> -<span class="sourceLineNo">256</span> BytesBytesPair.Builder bbpBuilder = BytesBytesPair.newBuilder();<a name="line.256"></a> -<span class="sourceLineNo">257</span> bbpBuilder.setFirst(UnsafeByteOperations.unsafeWrap(fuzzyData.getFirst()));<a name="line.257"></a> -<span class="sourceLineNo">258</span> bbpBuilder.setSecond(UnsafeByteOperations.unsafeWrap(fuzzyData.getSecond()));<a name="line.258"></a> -<span class="sourceLineNo">259</span> builder.addFuzzyKeysData(bbpBuilder);<a name="line.259"></a> -<span class="sourceLineNo">260</span> }<a name="line.260"></a> -<span class="sourceLineNo">261</span> return builder.build().toByteArray();<a name="line.261"></a> -<span class="sourceLineNo">262</span> }<a name="line.262"></a> -<span class="sourceLineNo">263</span><a name="line.263"></a> -<span class="sourceLineNo">264</span> /**<a name="line.264"></a> -<span class="sourceLineNo">265</span> * @param pbBytes A pb serialized {@link FuzzyRowFilter} instance<a name="line.265"></a> -<span class="sourceLineNo">266</span> * @return An instance of {@link FuzzyRowFilter} made from <code>bytes</code><a name="line.266"></a> -<span class="sourceLineNo">267</span> * @throws DeserializationException<a name="line.267"></a> -<span class="sourceLineNo">268</span> * @see #toByteArray<a name="line.268"></a> -<span class="sourceLineNo">269</span> */<a name="line.269"></a> -<span class="sourceLineNo">270</span> public static FuzzyRowFilter parseFrom(final byte[] pbBytes) throws DeserializationException {<a name="line.270"></a> -<span class="sourceLineNo">271</span> FilterProtos.FuzzyRowFilter proto;<a name="line.271"></a> -<span class="sourceLineNo">272</span> try {<a name="line.272"></a> -<span class="sourceLineNo">273</span> proto = FilterProtos.FuzzyRowFilter.parseFrom(pbBytes);<a name="line.273"></a> -<span class="sourceLineNo">274</span> } catch (InvalidProtocolBufferException e) {<a name="line.274"></a> -<span class="sourceLineNo">275</span> throw new DeserializationException(e);<a name="line.275"></a> -<span class="sourceLineNo">276</span> }<a name="line.276"></a> -<span class="sourceLineNo">277</span> int count = proto.getFuzzyKeysDataCount();<a name="line.277"></a> -<span class="sourceLineNo">278</span> ArrayList<Pair<byte[], byte[]>> fuzzyKeysData = new ArrayList<>(count);<a name="line.278"></a> -<span class="sourceLineNo">279</span> for (int i = 0; i < count; ++i) {<a name="line.279"></a> -<span class="sourceLineNo">280</span> BytesBytesPair current = proto.getFuzzyKeysData(i);<a name="line.280"></a> -<span class="sourceLineNo">281</span> byte[] keyBytes = current.getFirst().toByteArray();<a name="line.281"></a> -<span class="sourceLineNo">282</span> byte[] keyMeta = current.getSecond().toByteArray();<a name="line.282"></a> -<span class="sourceLineNo">283</span> fuzzyKeysData.add(new Pair<>(keyBytes, keyMeta));<a name="line.283"></a> +<span class="sourceLineNo">130</span> if (isPreprocessedMask(mask)) return mask;<a name="line.130"></a> +<span class="sourceLineNo">131</span> for (int i = 0; i < mask.length; i++) {<a name="line.131"></a> +<span class="sourceLineNo">132</span> if (mask[i] == 0) {<a name="line.132"></a> +<span class="sourceLineNo">133</span> mask[i] = -1; // 0 -> -1<a name="line.133"></a> +<span class="sourceLineNo">134</span> } else if (mask[i] == 1) {<a name="line.134"></a> +<span class="sourceLineNo">135</span> mask[i] = 2;// 1 -> 2<a name="line.135"></a> +<span class="sourceLineNo">136</span> }<a name="line.136"></a> +<span class="sourceLineNo">137</span> }<a name="line.137"></a> +<span class="sourceLineNo">138</span> return mask;<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> private boolean isPreprocessedMask(byte[] mask) {<a name="line.141"></a> +<span class="sourceLineNo">142</span> for (int i = 0; i < mask.length; i++) {<a name="line.142"></a> +<span class="sourceLineNo">143</span> if (mask[i] != -1 && mask[i] != 2) {<a name="line.143"></a> +<span class="sourceLineNo">144</span> return false;<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> return true;<a name="line.147"></a> +<span class="sourceLineNo">148</span> }<a name="line.148"></a> +<span class="sourceLineNo">149</span><a name="line.149"></a> +<span class="sourceLineNo">150</span> @Override<a name="line.150"></a> +<span class="sourceLineNo">151</span> public ReturnCode filterKeyValue(Cell c) {<a name="line.151"></a> +<span class="sourceLineNo">152</span> final int startIndex = lastFoundIndex >= 0 ? lastFoundIndex : 0;<a name="line.152"></a> +<span class="sourceLineNo">153</span> final int size = fuzzyKeysData.size();<a name="line.153"></a> +<span class="sourceLineNo">154</span> for (int i = startIndex; i < size + startIndex; i++) {<a name="line.154"></a> +<span class="sourceLineNo">155</span> final int index = i % size;<a name="line.155"></a> +<span class="sourceLineNo">156</span> Pair<byte[], byte[]> fuzzyData = fuzzyKeysData.get(index);<a name="line.156"></a> +<span class="sourceLineNo">157</span> // This shift is idempotent - always end up with 0 and -1 as mask values.<a name="line.157"></a> +<span class="sourceLineNo">158</span> for (int j = 0; j < fuzzyData.getSecond().length; j++) {<a name="line.158"></a> +<span class="sourceLineNo">159</span> fuzzyData.getSecond()[j] >>= 2;<a name="line.159"></a> +<span class="sourceLineNo">160</span> }<a name="line.160"></a> +<span class="sourceLineNo">161</span> SatisfiesCode satisfiesCode =<a name="line.161"></a> +<span class="sourceLineNo">162</span> satisfies(isReversed(), c.getRowArray(), c.getRowOffset(), c.getRowLength(),<a name="line.162"></a> +<span class="sourceLineNo">163</span> fuzzyData.getFirst(), fuzzyData.getSecond());<a name="line.163"></a> +<span class="sourceLineNo">164</span> if (satisfiesCode == SatisfiesCode.YES) {<a name="line.164"></a> +<span class="sourceLineNo">165</span> lastFoundIndex = index;<a name="line.165"></a> +<span class="sourceLineNo">166</span> return ReturnCode.INCLUDE;<a name="line.166"></a> +<span class="sourceLineNo">167</span> }<a name="line.167"></a> +<span class="sourceLineNo">168</span> }<a name="line.168"></a> +<span class="sourceLineNo">169</span> // NOT FOUND -> seek next using hint<a name="line.169"></a> +<span class="sourceLineNo">170</span> lastFoundIndex = -1;<a name="line.170"></a> +<span class="sourceLineNo">171</span><a name="line.171"></a> +<span class="sourceLineNo">172</span> return ReturnCode.SEEK_NEXT_USING_HINT;<a name="line.172"></a> +<span class="sourceLineNo">173</span><a name="line.173"></a> +<span class="sourceLineNo">174</span> }<a name="line.174"></a> +<span class="sourceLineNo">175</span><a name="line.175"></a> +<span class="sourceLineNo">176</span> @Override<a name="line.176"></a> +<span class="sourceLineNo">177</span> public Cell getNextCellHint(Cell currentCell) {<a name="line.177"></a> +<span class="sourceLineNo">178</span> boolean result = tracker.updateTracker(currentCell);<a name="line.178"></a> +<span class="sourceLineNo">179</span> if (result == false) {<a name="line.179"></a> +<span class="sourceLineNo">180</span> done = true;<a name="line.180"></a> +<span class="sourceLineNo">181</span> return null;<a name="line.181"></a> +<span class="sourceLineNo">182</span> }<a name="line.182"></a> +<span class="sourceLineNo">183</span> byte[] nextRowKey = tracker.nextRow();<a name="line.183"></a> +<span class="sourceLineNo">184</span> return CellUtil.createFirstOnRow(nextRowKey, 0, (short) nextRowKey.length);<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> /**<a name="line.187"></a> +<span class="sourceLineNo">188</span> * If we have multiple fuzzy keys, row tracker should improve overall performance. It calculates<a name="line.188"></a> +<span class="sourceLineNo">189</span> * all next rows (one per every fuzzy key) and put them (the fuzzy key is bundled) into a priority<a name="line.189"></a> +<span class="sourceLineNo">190</span> * queue so that the smallest row key always appears at queue head, which helps to decide the<a name="line.190"></a> +<span class="sourceLineNo">191</span> * "Next Cell Hint". As scanning going on, the number of candidate rows in the RowTracker will<a name="line.191"></a> +<span class="sourceLineNo">192</span> * remain the size of fuzzy keys until some of the fuzzy keys won't possibly have matches any<a name="line.192"></a> +<span class="sourceLineNo">193</span> * more.<a name="line.193"></a> +<span class="sourceLineNo">194</span> */<a name="line.194"></a> +<span class="sourceLineNo">195</span> private class RowTracker {<a name="line.195"></a> +<span class="sourceLineNo">196</span> private final PriorityQueue<Pair<byte[], Pair<byte[], byte[]>>> nextRows;<a name="line.196"></a> +<span class="sourceLineNo">197</span> private boolean initialized = false;<a name="line.197"></a> +<span class="sourceLineNo">198</span><a name="line.198"></a> +<span class="sourceLineNo">199</span> RowTracker() {<a name="line.199"></a> +<span class="sourceLineNo">200</span> nextRows = new PriorityQueue<>(fuzzyKeysData.size(),<a name="line.200"></a> +<span class="sourceLineNo">201</span> new Comparator<Pair<byte[], Pair<byte[], byte[]>>>() {<a name="line.201"></a> +<span class="sourceLineNo">202</span> @Override<a name="line.202"></a> +<span class="sourceLineNo">203</span> public int compare(Pair<byte[], Pair<byte[], byte[]>> o1,<a name="line.203"></a> +<span class="sourceLineNo">204</span> Pair<byte[], Pair<byte[], byte[]>> o2) {<a name="line.204"></a> +<span class="sourceLineNo">205</span> return isReversed()? Bytes.compareTo(o2.getFirst(), o1.getFirst()):<a name="line.205"></a> +<span class="sourceLineNo">206</span> Bytes.compareTo(o1.getFirst(), o2.getFirst());<a name="line.206"></a> +<span class="sourceLineNo">207</span> }<a name="line.207"></a> +<span class="sourceLineNo">208</span> });<a name="line.208"></a> +<span class="sourceLineNo">209</span> }<a name="line.209"></a> +<span class="sourceLineNo">210</span><a name="line.210"></a> +<span class="sourceLineNo">211</span> byte[] nextRow() {<a name="line.211"></a> +<span class="sourceLineNo">212</span> if (nextRows.isEmpty()) {<a name="line.212"></a> +<span class="sourceLineNo">213</span> throw new IllegalStateException(<a name="line.213"></a> +<span class="sourceLineNo">214</span> "NextRows should not be empty, make sure to call nextRow() after updateTracker() return true");<a name="line.214"></a> +<span class="sourceLineNo">215</span> } else {<a name="line.215"></a> +<span class="sourceLineNo">216</span> return nextRows.peek().getFirst();<a name="line.216"></a> +<span class="sourceLineNo">217</span> }<a name="line.217"></a> +<span class="sourceLineNo">218</span> }<a name="line.218"></a> +<span class="sourceLineNo">219</span><a name="line.219"></a> +<span class="sourceLineNo">220</span> boolean updateTracker(Cell currentCell) {<a name="line.220"></a> +<span class="sourceLineNo">221</span> if (!initialized) {<a name="line.221"></a> +<span class="sourceLineNo">222</span> for (Pair<byte[], byte[]> fuzzyData : fuzzyKeysData) {<a name="line.222"></a> +<span class="sourceLineNo">223</span> updateWith(currentCell, fuzzyData);<a name="line.223"></a> +<span class="sourceLineNo">224</span> }<a name="line.224"></a> +<span class="sourceLineNo">225</span> initialized = true;<a name="line.225"></a> +<span class="sourceLineNo">226</span> } else {<a name="line.226"></a> +<span class="sourceLineNo">227</span> while (!nextRows.isEmpty() && !lessThan(currentCell, nextRows.peek().getFirst())) {<a name="line.227"></a> +<span class="sourceLineNo">228</span> Pair<byte[], Pair<byte[], byte[]>> head = nextRows.poll();<a name="line.228"></a> +<span class="sourceLineNo">229</span> Pair<byte[], byte[]> fuzzyData = head.getSecond();<a name="line.229"></a> +<span class="sourceLineNo">230</span> updateWith(currentCell, fuzzyData);<a name="line.230"></a> +<span class="sourceLineNo">231</span> }<a name="line.231"></a> +<span class="sourceLineNo">232</span> }<a name="line.232"></a> +<span class="sourceLineNo">233</span> return !nextRows.isEmpty();<a name="line.233"></a> +<span class="sourceLineNo">234</span> }<a name="line.234"></a> +<span class="sourceLineNo">235</span><a name="line.235"></a> +<span class="sourceLineNo">236</span> boolean lessThan(Cell currentCell, byte[] nextRowKey) {<a name="line.236"></a> +<span class="sourceLineNo">237</span> int compareResult =<a name="line.237"></a> +<span class="sourceLineNo">238</span> CellComparator.COMPARATOR.compareRows(currentCell, nextRowKey, 0, nextRowKey.length);<a name="line.238"></a> +<span class="sourceLineNo">239</span> return (!isReversed() && compareResult < 0) || (isReversed() && compareResult > 0);<a name="line.239"></a> +<span class="sourceLineNo">240</span> }<a name="line.240"></a> +<span class="sourceLineNo">241</span><a name="line.241"></a> +<span class="sourceLineNo">242</span> void updateWith(Cell currentCell, Pair<byte[], byte[]> fuzzyData) {<a name="line.242"></a> +<span class="sourceLineNo">243</span> byte[] nextRowKeyCandidate =<a name="line.243"></a> +<span class="sourceLineNo">244</span> getNextForFuzzyRule(isReversed(), currentCell.getRowArray(), currentCell.getRowOffset(),<a name="line.244"></a> +<span class="sourceLineNo">245</span> currentCell.getRowLength(), fuzzyData.getFirst(), fuzzyData.getSecond());<a name="line.245"></a> +<span class="sourceLineNo">246</span> if (nextRowKeyCandidate != null) {<a name="line.246"></a> +<span class="sourceLineNo">247</span> nextRows.add(new Pair<>(nextRowKeyCandidate, fuzzyData));<a name="line.247"></a> +<span class="sourceLineNo">248</span> }<a name="line.248"></a> +<span class="sourceLineNo">249</span> }<a name="line.249"></a> +<span class="sourceLineNo">250</span><a name="line.250"></a> +<span class="sourceLineNo">251</span> }<a name="line.251"></a> +<span class="sourceLineNo">252</span><a name="line.252"></a> +<span class="sourceLineNo">253</span> @Override<a name="line.253"></a> +<span class="sourceLineNo">254</span> public boolean filterAllRemaining() {<a name="line.254"></a> +<span class="sourceLineNo">255</span> return done;<a name="line.255"></a> +<span class="sourceLineNo">256</span> }<a name="line.256"></a> +<span class="sourceLineNo">257</span><a name="line.257"></a> +<span class="sourceLineNo">258</span> /**<a name="line.258"></a> +<span class="sourceLineNo">259</span> * @return The filter serialized using pb<a name="line.259"></a> +<span class="sourceLineNo">260</span> */<a name="line.260"></a> +<span class="sourceLineNo">261</span> public byte[] toByteArray() {<a name="line.261"></a> +<span class="sourceLineNo">262</span> FilterProtos.FuzzyRowFilter.Builder builder = FilterProtos.FuzzyRowFilter.newBuilder();<a name="line.262"></a> +<span class="sourceLineNo">263</span> for (Pair<byte[], byte[]> fuzzyData : fuzzyKeysData) {<a name="line.263"></a> +<span class="sourceLineNo">264</span> BytesBytesPair.Builder bbpBuilder = BytesBytesPair.newBuilder();<a name="line.264"></a> +<span class="sourceLineNo">265</span> bbpBuilder.setFirst(UnsafeByteOperations.unsafeWrap(fuzzyData.getFirst()));<a name="line.265"></a> +<span class="sourceLineNo">266</span> bbpBuilder.setSecond(UnsafeByteOperations.unsafeWrap(fuzzyData.getSecond()));<a name="line.266"></a> +<span class="sourceLineNo">267</span> builder.addFuzzyKeysData(bbpBuilder);<a name="line.267"></a> +<span class="sourceLineNo">268</span> }<a name="line.268"></a> +<span class="sourceLineNo">269</span> return builder.build().toByteArray();<a name="line.269"></a> +<span class="sourceLineNo">270</span> }<a name="line.270"></a> +<span class="sourceLineNo">271</span><a name="line.271"></a> +<span class="sourceLineNo">272</span> /**<a name="line.272"></a> +<span class="sourceLineNo">273</span> * @param pbBytes A pb serialized {@link FuzzyRowFilter} instance<a name="line.273"></a> +<span class="sourceLineNo">274</span> * @return An instance of {@link FuzzyRowFilter} made from <code>bytes</code><a name="line.274"></a> +<span class="sourceLineNo">275</span> * @throws DeserializationException<a name="line.275"></a> +<span class="sourceLineNo">276</span> * @see #toByteArray<a name="line.276"></a> +<span class="sourceLineNo">277</span> */<a name="line.277"></a> +<span class="sourceLineNo">278</span> public static FuzzyRowFilter parseFrom(final byte[] pbBytes) throws DeserializationException {<a name="line.278"></a> +<span class="sourceLineNo">279</span> FilterProtos.FuzzyRowFilter proto;<a name="line.279"></a> +<span class="sourceLineNo">280</span> try {<a name="line.280"></a> +<span class="sourceLineNo">281</span> proto = FilterProtos.FuzzyRowFilter.parseFrom(pbBytes);<a name="line.281"></a> +<span class="sourceLineNo">282</span> } catch (InvalidProtocolBufferException e) {<a name="line.282"></a> +<span class="sourceLineNo">283</span> throw new DeserializationException(e);<a name="line.283"></a> <span class="sourceLineNo">284</span> }<a name="line.284"></a> -<span class="sourceLineNo">285</span> return new FuzzyRowFilter(fuzzyKeysData);<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 String toString() {<a name="line.289"></a> -<span class="sourceLineNo">290</span> final StringBuilder sb = new StringBuilder();<a name="line.290"></a> -<span class="sourceLineNo">291</span> sb.append("FuzzyRowFilter");<a name="line.291"></a> -<span class="sourceLineNo">292</span> sb.append("{fuzzyKeysData=");<a name="line.292"></a> -<span class="sourceLineNo">293</span> for (Pair<byte[], byte[]> fuzzyData : fuzzyKeysData) {<a name="line.293"></a> -<span class="sourceLineNo">294</span> sb.append('{').append(Bytes.toStringBinary(fuzzyData.getFirst())).append(":");<a name="line.294"></a> -<span class="sourceLineNo">295</span> sb.append(Bytes.toStringBinary(fuzzyData.getSecond())).append('}');<a name="line.295"></a> -<span class="sourceLineNo">296</span> }<a name="line.296"></a> -<span class="sourceLineNo">297</span> sb.append("}, ");<a name="line.297"></a> -<span class="sourceLineNo">298</span> return sb.toString();<a name="line.298"></a> -<span class="sourceLineNo">299</span> }<a name="line.299"></a> -<span class="sourceLineNo">300</span><a name="line.300"></a> -<span class="sourceLineNo">301</span> // Utility methods<a name="line.301"></a> -<span class="sourceLineNo">302</span><a name="line.302"></a> -<span class="sourceLineNo">303</span> static enum SatisfiesCode {<a name="line.303"></a> -<span class="sourceLineNo">304</span> /** row satisfies fuzzy rule */<a name="line.304"></a> -<span class="sourceLineNo">305</span> YES,<a name="line.305"></a> -<span class="sourceLineNo">306</span> /** row doesn't satisfy fuzzy rule, but there's possible greater row that does */<a name="line.306"></a> -<span class="sourceLineNo">307</span> NEXT_EXISTS,<a name="line.307"></a> -<span class="sourceLineNo">308</span> /** row doesn't satisfy fuzzy rule and there's no greater row that does */<a name="line.308"></a> -<span class="sourceLineNo">309</span> NO_NEXT<a name="line.309"></a> -<span class="sourceLineNo">310</span> }<a name="line.310"></a> -<span class="sourceLineNo">311</span><a name="line.311"></a> -<span class="sourceLineNo">312</span> @VisibleForTesting<a name="line.312"></a> -<span class="sourceLineNo">313</span> static SatisfiesCode satisfies(byte[] row, byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) {<a name="line.313"></a> -<span class="sourceLineNo">314</span> return satisfies(false, row, 0, row.length, fuzzyKeyBytes, fuzzyKeyMeta);<a name="line.314"></a> -<span class="sourceLineNo">315</span> }<a name="line.315"></a> -<span class="sourceLineNo">316</span><a name="line.316"></a> -<span class="sourceLineNo">317</span> @VisibleForTesting<a name="line.317"></a> -<span class="sourceLineNo">318</span> static SatisfiesCode satisfies(boolean reverse, byte[] row, byte[] fuzzyKeyBytes,<a name="line.318"></a> -<span class="sourceLineNo">319</span> byte[] fuzzyKeyMeta) {<a name="line.319"></a> -<span class="sourceLineNo">320</span> return satisfies(reverse, row, 0, row.length, fuzzyKeyBytes, fuzzyKeyMeta);<a name="line.320"></a> -<span class="sourceLineNo">321</span> }<a name="line.321"></a> -<span class="sourceLineNo">322</span><a name="line.322"></a> -<span class="sourceLineNo">323</span> static SatisfiesCode satisfies(boolean reverse, byte[] row, int offset, int length,<a name="line.323"></a> -<span class="sourceLineNo">324</span> byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) {<a name="line.324"></a> -<span class="sourceLineNo">325</span><a name="line.325"></a> -<span class="sourceLineNo">326</span> if (!UNSAFE_UNALIGNED) {<a name="line.326"></a> -<span class="sourceLineNo">327</span> return satisfiesNoUnsafe(reverse, row, offset, length, fuzzyKeyBytes, fuzzyKeyMeta);<a name="line.327"></a> -<span class="sourceLineNo">328</span> }<a name="line.328"></a> -<span class="sourceLineNo">329</span><a name="line.329"></a> -<span class="sourceLineNo">330</span> if (row == null) {<a name="line.330"></a> -<span class="sourceLineNo">331</span> // do nothing, let scan to proceed<a name="line.331"></a> -<span class="sourceLineNo">332</span> return SatisfiesCode.YES;<a name="line.332"></a> -<span class="sourceLineNo">333</span> }<a name="line.333"></a> -<span class="sourceLineNo">334</span> length = Math.min(length, fuzzyKeyBytes.length);<a name="line.334"></a> -<span class="sourceLineNo">335</span> int numWords = length / Bytes.SIZEOF_LONG;<a name="line.335"></a> -<span class="sourceLineNo">336</span><a name="line.336"></a> -<span class="sourceLineNo">337</span> int j = numWords << 3; // numWords * SIZEOF_LONG;<a name="line.337"></a> -<span class="sourceLineNo">338</span><a name="line.338"></a> -<span class="sourceLineNo">339</span> for (int i = 0; i < j; i += Bytes.SIZEOF_LONG) {<a name="line.339"></a> -<span class="sourceLineNo">340</span> long fuzzyBytes = UnsafeAccess.toLong(fuzzyKeyBytes, i);<a name="line.340"></a> -<span class="sourceLineNo">341</span> long fuzzyMeta = UnsafeAccess.toLong(fuzzyKeyMeta, i);<a name="line.341"></a> -<span class="sourceLineNo">342</span> long rowValue = UnsafeAccess.toLong(row, offset + i);<a name="line.342"></a> -<span class="sourceLineNo">343</span> if ((rowValue & fuzzyMeta) != (fuzzyBytes)) {<a name="line.343"></a> -<span class="sourceLineNo">344</span> // We always return NEXT_EXISTS<a name="line.344"></a> -<span class="sourceLineNo">345</span> return SatisfiesCode.NEXT_EXISTS;<a name="line.345"></a> -<span class="sourceLineNo">346</span> }<a name="line.346"></a> -<span class="sourceLineNo">347</span> }<a name="line.347"></a> -<span class="sourceLineNo">348</span><a name="line.348"></a> -<span class="sourceLineNo">349</span> int off = j;<a name="line.349"></a> -<span class="sourceLineNo">350</span><a name="line.350"></a> -<span class="sourceLineNo">351</span> if (length - off >= Bytes.SIZEOF_INT) {<a name="line.351"></a> -<span class="sourceLineNo">352</span> int fuzzyBytes = UnsafeAccess.toInt(fuzzyKeyBytes, off);<a name="line.352"></a> -<span class="sourceLineNo">353</span> int fuzzyMeta = UnsafeAccess.toInt(fuzzyKeyMeta, off);<a name="line.353"></a> -<span class="sourceLineNo">354</span> int rowValue = UnsafeAccess.toInt(row, offset + off);<a name="line.354"></a> -<span class="sourceLineNo">355</span> if ((rowValue & fuzzyMeta) != (fuzzyBytes)) {<a name="line.355"></a> -<span class="sourceLineNo">356</span> // We always return NEXT_EXISTS<a name="line.356"></a> -<span class="sourceLineNo">357</span> return SatisfiesCode.NEXT_EXISTS;<a name="line.357"></a> -<span class="sourceLineNo">358</span> }<a name="line.358"></a> -<span class="sourceLineNo">359</span> off += Bytes.SIZEOF_INT;<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> if (length - off >= Bytes.SIZEOF_SHORT) {<a name="line.362"></a> -<span class="sourceLineNo">363</span> short fuzzyBytes = UnsafeAccess.toShort(fuzzyKeyBytes, off);<a name="line.363"></a> -<span class="sourceLineNo">364</span> short fuzzyMeta = UnsafeAccess.toShort(fuzzyKeyMeta, off);<a name="line.364"></a> -<span class="sourceLineNo">365</span> short rowValue = UnsafeAccess.toShort(row, offset + off);<a name="line.365"></a> -<span class="sourceLineNo">366</span> if ((rowValue & fuzzyMeta) != (fuzzyBytes)) {<a name="line.366"></a> -<span class="sourceLineNo">367</span> // We always return NEXT_EXISTS<a name="line.367"></a> -<span class="sourceLineNo">368</span> // even if it does not (in this case getNextForFuzzyRule<a name="line.368"></a> -<span class="sourceLineNo">369</span> // will return null)<a name="line.369"></a> -<span class="sourceLineNo">370</span> return SatisfiesCode.NEXT_EXISTS;<a name="line.370"></a> -<span class="sourceLineNo">371</span> }<a name="line.371"></a> -<span class="sourceLineNo">372</span> off += Bytes.SIZEOF_SHORT;<a name="line.372"></a> -<span class="sourceLineNo">373</span> }<a name="line.373"></a> -<span class="sourceLineNo">374</span><a name="line.374"></a> -<span class="sourceLineNo">375</span> if (length - off >= Bytes.SIZEOF_BYTE) {<a name="line.375"></a> -<span class="sourceLineNo">376</span> int fuzzyBytes = fuzzyKeyBytes[off] & 0xff;<a name="line.376"></a> -<span class="sourceLineNo">377</span> int fuzzyMeta = fuzzyKeyMeta[off] & 0xff;<a name="line.377"></a> -<span class="sourceLineNo">378</span> int rowValue = row[offset + off] & 0xff;<a name="line.378"></a> -<span class="sourceLineNo">379</span> if ((rowValue & fuzzyMeta) != (fuzzyBytes)) {<a name="line.379"></a> -<span class="sourceLineNo">380</span> // We always return NEXT_EXISTS<a name="line.380"></a> -<span class="sourceLineNo">381</span> return SatisfiesCode.NEXT_EXISTS;<a name="line.381"></a> -<span class="sourceLineNo">382</span> }<a name="line.382"></a> -<span class="sourceLineNo">383</span> }<a name="line.383"></a> -<span class="sourceLineNo">384</span> return SatisfiesCode.YES;<a name="line.384"></a> -<span class="sourceLineNo">385</span> }<a name="line.385"></a> -<span class="sourceLineNo">386</span><a name="line.386"></a> -<span class="sourceLineNo">387</span> static SatisfiesCode satisfiesNoUnsafe(boolean reverse, byte[] row, int offset, int length,<a name="line.387"></a> -<span class="sourceLineNo">388</span> byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) {<a name="line.388"></a> -<span class="sourceLineNo">389</span> if (row == null) {<a name="line.389"></a> -<span class="sourceLineNo">390</span> // do nothing, let scan to proceed<a name="line.390"></a> -<span class="sourceLineNo">391</span> return SatisfiesCode.YES;<a name="line.391"></a> -<span class="sourceLineNo">392</span> }<a name="line.392"></a> -<span class="sourceLineNo">393</span><a name="line.393"></a> -<span class="sourceLineNo">394</span> Order order = Order.orderFor(reverse);<a name="line.394"></a> -<span class="sourceLineNo">395</span> boolean nextRowKeyCandidateExists = false;<a name="line.395"></a> -<span class="sourceLineNo">396</span><a name="line.396"></a> -<span class="sourceLineNo">397</span> for (int i = 0; i < fuzzyKeyMeta.length && i < length; i++) {<a name="line.397"></a> -<span class="sourceLineNo">398</span> // First, checking if this position is fixed and not equals the given one<a name="line.398"></a> -<span class="sourceLineNo">399</span> boolean byteAtPositionFixed = fuzzyKeyMeta[i] == 0;<a name="line.399"></a> -<span class="sourceLineNo">400</span> boolean fixedByteIncorrect = byteAtPositionFixed && fuzzyKeyBytes[i] != row[i + offset];<a name="line.400"></a> -<span class="sourceLineNo">401</span> if (fixedByteIncorrect) {<a name="line.401"></a> -<span class="sourceLineNo">402</span> // in this case there's another row that satisfies fuzzy rule and bigger than this row<a name="line.402"></a> -<span class="sourceLineNo">403</span> if (nextRowKeyCandidateExists) {<a name="line.403"></a> -<span class="sourceLineNo">404</span> return SatisfiesCode.NEXT_EXISTS;<a name="line.404"></a> -<span class="sourceLineNo">405</span> }<a name="line.405"></a> -<span class="sourceLineNo">406</span><a name="line.406"></a> -<span class="sourceLineNo">407</span> // If this row byte is less than fixed then there's a byte array bigger than<a name="line.407"></a> -<span class="sourceLineNo">408</span> // this row and which satisfies the fuzzy rule. Otherwise there's no such byte array:<a name="line.408"></a> -<span class="sourceLineNo">409</span> // this row is simply bigger than any byte array that satisfies the fuzzy rule<a name="line.409"></a> -<span class="sourceLineNo">410</span> boolean rowByteLessThanFixed = (row[i + offset] & 0xFF) < (fuzzyKeyBytes[i] & 0xFF);<a name="line.410"></a> -<span class="sourceLineNo">411</span> if (rowByteLessThanFixed && !reverse) {<a name="line.411"></a> +<span class="sourceLineNo">285</span> int count = proto.getFuzzyKeysDataCount();<a name="line.285"></a> +<span class="sourceLineNo">286</span> ArrayList<Pair<byte[], byte[]>> fuzzyKeysData = new ArrayList<>(count);<a name="line.286"></a> +<span class="sourceLineNo">287</span> for (int i = 0; i < count; ++i) {<a name="line.287"></a> +<span class="sourceLineNo">288</span> BytesBytesPair current = proto.getFuzzyKeysData(i);<a name="line.288"></a> +<span class="sourceLineNo">289</span> byte[] keyBytes = current.getFirst().toByteArray();<a name="line.289"></a> +<span class="sourceLineNo">290</span> byte[] keyMeta = current.getSecond().toByteArray();<a name="line.290"></a> +<span class="sourceLineNo">291</span> fuzzyKeysData.add(new Pair<>(keyBytes, keyMeta));<a name="line.291"></a> +<span class="sourceLineNo">292</span> }<a name="line.292"></a> +<span class="sourceLineNo">293</span> return new FuzzyRowFilter(fuzzyKeysData);<a name="line.293"></a> +<span class="sourceLineNo">294</span> }<a name="line.294"></a> +<span class="sourceLineNo">295</span><a name="line.295"></a> +<span class="sourceLineNo">296</span> @Override<a name="line.296"></a> +<span class="sourceLineNo">297</span> public String toString() {<a name="line.297"></a> +<span class="sourceLineNo">298</span> final StringBuilder sb = new StringBuilder();<a name="line.298"></a> +<span class="sourceLineNo">299</span> sb.append("FuzzyRowFilter");<a name="line.299"></a> +<span class="sourceLineNo">300</span> sb.append("{fuzzyKeysData=");<a name="line.300"></a> +<span class="sourceLineNo">301</span> for (Pair<byte[], byte[]> fuzzyData : fuzzyKeysData) {<a name="line.301"></a> +<span class="sourceLineNo">302</span> sb.append('{').append(Bytes.toStringBinary(fuzzyData.getFirst())).append(":");<a name="line.302"></a> +<span class="sourceLineNo">303</span> sb.append(Bytes.toStringBinary(fuzzyData.getSecond())).append('}');<a name="line.303"></a> +<span class="sourceLineNo">304</span> }<a name="line.304"></a> +<span class="sourceLineNo">305</span> sb.append("}, ");<a name="line.305"></a> +<span class="sourceLineNo">306</span> return sb.toString();<a name="line.306"></a> +<span class="sourceLineNo">307</span> }<a name="line.307"></a> +<span class="sourceLineNo">308</span><a name="line.308"></a> +<span class="sourceLineNo">309</span> // Utility methods<a name="line.309"></a> +<span class="sourceLineNo">310</span><a name="line.310"></a> +<span class="sourceLineNo">311</span> static enum SatisfiesCode {<a name="line.311"></a> +<span class="sourceLineNo">312</span> /** row satisfies fuzzy rule */<a name="line.312"></a> +<span class="sourceLineNo">313</span> YES,<a name="line.313"></a> +<span class="sourceLineNo">314</span> /** row doesn't satisfy fuzzy rule, but there's possible greater row that does */<a name="line.314"></a> +<span class="sourceLineNo">315</span> NEXT_EXISTS,<a name="line.315"></a> +<span class="sourceLineNo">316</span> /** row doesn't satisfy fuzzy rule and there's no greater row that does */<a name="line.316"></a> +<span class="sourceLineNo">317</span> NO_NEXT<a name="line.317"></a> +<span class="sourceLineNo">318</span> }<a name="line.318"></a> +<span class="sourceLineNo">319</span><a name="line.319"></a> +<span class="sourceLineNo">320</span> @VisibleForTesting<a name="line.320"></a> +<span class="sourceLineNo">321</span> static SatisfiesCode satisfies(byte[] row, byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) {<a name="line.321"></a> +<span class="sourceLineNo">322</span> return satisfies(false, row, 0, row.length, fuzzyKeyBytes, fuzzyKeyMeta);<a name="line.322"></a> +<span class="sourceLineNo">323</span> }<a name="line.323"></a> +<span class="sourceLineNo">324</span><a name="line.324"></a> +<span class="sourceLineNo">325</span> @VisibleForTesting<a name="line.325"></a> +<span class="sourceLineNo">326</span> static SatisfiesCode satisfies(boolean reverse, byte[] row, byte[] fuzzyKeyBytes,<a name="line.326"></a> +<span class="sourceLineNo">327</span> byte[] fuzzyKeyMeta) {<a name="line.327"></a> +<span class="sourceLineNo">328</span> return satisfies(reverse, row, 0, row.length, fuzzyKeyBytes, fuzzyKeyMeta);<a name="line.328"></a> +<span class="sourceLineNo">329</span> }<a name="line.329"></a> +<span class="sourceLineNo">330</span><a name="line.330"></a> +<span class="sourceLineNo">331</span> static SatisfiesCode satisfies(boolean reverse, byte[] row, int offset, int length,<a name="line.331"></a> +<span class="sourceLineNo">332</span> byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) {<a name="line.332"></a> +<span class="sourceLineNo">333</span><a name="line.333"></a> +<span class="sourceLineNo">334</span> if (!UNSAFE_UNALIGNED) {<a name="line.334"></a> +<span class="sourceLineNo">335</span> return satisfiesNoUnsafe(reverse, row, offset, length, fuzzyKeyBytes, fuzzyKeyMeta);<a name="line.335"></a> +<span class="sourceLineNo">336</span> }<a name="line.336"></a> +<span class="sourceLineNo">337</span><a name="line.337"></a> +<span class="sourceLineNo">338</span> if (row == null) {<a name="line.338"></a> +<span class="sourceLineNo">339</span> // do nothing, let scan to proceed<a name="line.339"></a> +<span class="sourceLineNo">340</span> return SatisfiesCode.YES;<a name="line.340"></a> +<span class="sourceLineNo">341</span> }<a name="line.341"></a> +<span class="sourceLineNo">342</span> length = Math.min(length, fuzzyKeyBytes.length);<a name="line.342"></a> +<span class="sourceLineNo">343</span> int numWords = length / Bytes.SIZEOF_LONG;<a name="line.343"></a> +<span class="sourceLineNo">344</span><a name="line.344"></a> +<span class="sourceLineNo">345</span> int j = numWords << 3; // numWords * SIZEOF_LONG;<a name="line.345"></a> +<span class="sourceLineNo">346</span><a name="line.346"></a> +<span class="sourceLineNo">347</span> for (int i = 0; i < j; i += Bytes.SIZEOF_LONG) {<a name="line.347"></a> +<span class="sourceLineNo">348</span> long fuzzyBytes = UnsafeAccess.toLong(fuzzyKeyBytes, i);<a name="line.348"></a> +<span class="sourceLineNo">349</span> long fuzzyMeta = UnsafeAccess.toLong(fuzzyKeyMeta, i);<a name="line.349"></a> +<span class="sourceLineNo">350</span> long rowValue = UnsafeAccess.toLong(row, offset + i);<a name="line.350"></a> +<span class="sourceLineNo">351</span> if ((rowValue & fuzzyMeta) != (fuzzyBytes)) {<a name="line.351"></a> +<span class="sourceLineNo">352</span> // We always return NEXT_EXISTS<a name="line.352"></a> +<span class="sourceLineNo">353</span> return SatisfiesCode.NEXT_EXISTS;<a name="line.353"></a> +<span class="sourceLineNo">354</span> }<a name="line.354"></a> +<span class="sourceLineNo">355</span> }<a name="line.355"></a> +<span class="sourceLineNo">356</span><a name="line.356"></a> +<span class="sourceLineNo">357</span> int off = j;<a name="line.357"></a> +<span class="sourceLineNo">358</span><a name="line.358"></a> +<span class="sourceLineNo">359</span> if (length - off >= Bytes.SIZEOF_INT) {<a name="line.359"></a> +<span class="sourceLineNo">360</span> int fuzzyBytes = UnsafeAccess.toInt(fuzzyKeyBytes, off);<a name="line.360"></a> +<span class="sourceLineNo">361</span> int fuzzyMeta = UnsafeAccess.toInt(fuzzyKeyMeta, off);<a name="line.361"></a> +<span class="sourceLineNo">362</span> int rowValue = UnsafeAccess.toInt(row, offset + off);<a name="line.362"></a> +<span class="sourceLineNo">363</span> if ((rowValue & fuzzyMeta) != (fuzzyBytes)) {<a name="line.363"></a> +<span class="sourceLineNo">364</span> // We always return NEXT_EXISTS<a name="line.364"></a> +<span class="sourceLineNo">365</span> return SatisfiesCode.NEXT_EXISTS;<a name="line.365"></a> +<span class="sourceLineNo">366</span> }<a name="line.366"></a> +<span class="sourceLineNo">367</span> off += Bytes.SIZEOF_INT;<a name="line.367"></a> +<span class="sourceLineNo">368</span> }<a name="line.368"></a> +<span class="sourceLineNo">369</span><a name="line.369"></a> +<span class="sourceLineNo">370</span> if (length - off >= Bytes.SIZEOF_SHORT) {<a name="line.370"></a> +<span class="sourceLineNo">371</span> short fuzzyBytes = UnsafeAccess.toShort(fuzzyKeyBytes, off);<a name="line.371"></a> +<span class="sourceLineNo">372</span> short fuzzyMeta = UnsafeAccess.toShort(fuzzyKeyMeta, off);<a name="line.372"></a> +<span class="sourceLineNo">373</span> short rowValue = UnsafeAccess.toShort(row, offset + off);<a name="line.373"></a> +<span class="sourceLineNo">374</span> if ((rowValue & fuzzyMeta) != (fuzzyBytes)) {<a name="line.374"></a> +<span class="sourceLineNo">375</span> // We always return NEXT_EXISTS<a name="line.375"></a> +<span class="sourceLineNo">376</span> // even if it does not (in this case getNextForFuzzyRule<a name="line.376"></a> +<span class="sourceLineNo">377</span> // will return null)<a name="line.377"></a> +<span class="sourceLineNo">378</span> return SatisfiesCode.NEXT_EXISTS;<a name="line.378"></a> +<span class="sourceLineNo">379</span> }<a name="line.379"></a> +<span class="sourceLineNo">380</span> off += Bytes.SIZEOF_SHORT;<a name="line.380"></a> +<span class="sourceLineNo">381</span> }<a name="line.381"></a> +<span class="sourceLineNo">382</span><a name="line.382"></a> +<span class="sourceLineNo">383</span> if (length - off >= Bytes.SIZEOF_BYTE) {<a name="line.383"></a> +<span class="sourceLineNo">384</span> int fuzzyBytes = fuzzyKeyBytes[off] & 0xff;<a name="line.384"></a> +<span class="sourceLineNo">385</span> int fuzzyMeta = fuzzyKeyMeta[off] & 0xff;<a name="line.385"></a> +<span class="sourceLineNo">386</span> int rowValue = row[offset + off] & 0xff;<a name="line.386"></a> +<span class="sourceLineNo">387</span> if ((rowValue & fuzzyMeta) != (fuzzyBytes)) {<a name="line.387"></a> +<span class="sourceLineNo">388</span> // We always return NEXT_EXISTS<a name="line.388"></a> +<span class="sourceLineNo">389</span> return SatisfiesCode.NEXT_EXISTS;<a name="line.389"></a> +<span class="sourceLineNo">390</span> }<a name="line.390"></a> +<span class="sourceLineNo">391</span> }<a name="line.391"></a> +<span class="sourceLineNo">392</span> return SatisfiesCode.YES;<a name="line.392"></a> +<span class="sourceLineNo">393</span> }<a name="line.393"></a> +<span class="sourceLineNo">394</span><a name="line.394"></a> +<span class="sourceLineNo">395</span> static SatisfiesCode satisfiesNoUnsafe(boolean reverse, byte[] row, int offset, int length,<a name="line.395"></a> +<span class="sourceLineNo">396</span> byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) {<a name="line.396"></a> +<span class="sourceLineNo">397</span> if (row == null) {<a name="line.397"></a> +<span class="sourceLineNo">398</span> // do nothing, let scan to proceed<a name="line.398"></a> +<span class="sourceLineNo">399</span> return SatisfiesCode.YES;<a name="line.399"></a> +<span class="sourceLineNo">400</span> }<a name="line.400"></a> +<span class="sourceLineNo">401</span><a name="line.401"></a> +<span class="sourceLineNo">402</span> Order order = Order.orderFor(reverse);<a name="line.402"></a> +<span class="sourceLineNo">403</span> boolean nextRowKeyCandidateExists = false;<a name="line.403"></a> +<span class="sourceLineNo">404</span><a name="line.404"></a> +<span class="sourceLineNo">405</span> for (int i = 0; i < fuzzyKeyMeta.length && i < length; i++) {<a name="line.405"></a> +<span class="sourceLineNo">406</span> // First, checking if this position is fixed and not equals the given one<a name="line.406"></a> +<span class="sourceLineNo">407</span> boolean byteAtPositionFixed = fuzzyKeyMeta[i] == 0;<a name="line.407"></a> +<span class="sourceLineNo">408</span> boolean fixedByteIncorrect = byteAtPositionFixed && fuzzyKeyBytes[i] != row[i + offset];<a name="line.408"></a> +<span class="sourceLineNo">409</span> if (fixedByteIncorrect) {<a name="line.409"></a> +<span class="sourceLineNo">410</span> // in this case there's another row that satisfies fuzzy rule and bigger than this row<a name="line.410"></a> +<span class="sourceLineNo">411</span> if (nextRowKeyCandidateExists) {<a name="line.411"></a> <span class="sourceLineNo">412</span> return SatisfiesCode.NEXT_EXISTS;<a name="line.412"></a> -<span class="sourceLineNo">413</span> } else if (!rowByteLessThanFixed && reverse) {<a name="line.413"></a> -<span class="sourceLineNo">414</span> return SatisfiesCode.NEXT_EXISTS;<a name="line.414"></a> -<span class="sourceLineNo">415</span> } else {<a name="line.415"></a> -<span class="sourceLineNo">416</span> return SatisfiesCode.NO_NEXT;<a name="line.416"></a> -<span class="sourceLineNo">417</span> }<a name="line.417"></a> -<span class="sourceLineNo">418</span> }<a name="line.418"></a> -<span class="sourceLineNo">419</span><a name="line.419"></a> -<span class="sourceLineNo">420</span> // Second, checking if this position is not fixed and byte value is not the biggest. In this<a name="line.420"></a> -<span class="sourceLineNo">421</span> // case there's a byte array bigger than this row and which satisfies the fuzzy rule. To get<a name="line.421"></a> -<span class="sourceLineNo">422</span> // bigger byte array that satisfies the rule we need to just increase this byte<a name="line.422"></a> -<span class="sourceLineNo">423</span> // (see the code of getNextForFuzzyRule below) by one.<a name="line.423"></a> -<span class="sourceLineNo">424</span> // Note: if non-fixed byte is already at biggest value, this doesn't allow us to say there's<a name="line.424"></a> -<span class="sourceLineNo">425</span> // bigger one that satisfies the rule as it can't be increased.<a name="line.425"></a> -<span class="sourceLineNo">426</span> if (fuzzyKeyMeta[i] == 1 && !order.isMax(fuzzyKeyBytes[i])) {<a name="line.426"></a> -<span class="sourceLineNo">427</span> nextRowKeyCandidateExists = true;<a name="line.427"></a> -<span class="sourceLineNo">428</span> }<a name="line.428"></a> -<span class="sourceLineNo">429</span> }<a name="line.429"></a> -<span class="sourceLineNo">430</span> return SatisfiesCode.YES;<a name="line.430"></a> -<span class="sourceLineNo">431</span> }<a name="line.431"></a> -<span class="sourceLineNo">432</span><a name="line.432"></a> -<span class="sourceLineNo">433</span> @VisibleForTesting<a name="line.433"></a> -<span class="sourceLineNo">434</span> static byte[] getNextForFuzzyRule(byte[] row, byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) {<a name="line.434"></a> -<span class="sourceLineNo">435</span> return getNextForFuzzyRule(false, row, 0, row.length, fuzzyKeyBytes, fuzzyKeyMeta);<a name="line.435"></a> -<span class="sourceLineNo">436</span> }<a name="line.436"></a> -<span class="sourceLineNo">437</span><a name="line.437"></a> -<span class="sourceLineNo">438</span> @VisibleForTesting<a name="line.438"></a> -<span class="sourceLineNo">439</span> static byte[] getNextForFuzzyRule(boolean reverse, byte[] row, byte[] fuzzyKeyBytes,<a name="line.439"></a> -<span class="sourceLineNo">440</span> byte[] fuzzyKeyMeta) {<a name="line.440"></a> -<span class="sourceLineNo">441</span> return getNextForFuzzyRule(reverse, row, 0, row.length, fuzzyKeyBytes, fuzzyKeyMeta);<a name="line.441"></a> -<span class="sourceLineNo">442</span> }<a name="line.442"></a> -<span class="sourceLineNo">443</span><a name="line.443"></a> -<span class="sourceLineNo">444</span> /** Abstracts directional comparisons based on scan direction. */<a name="line.444"></a> -<span class="sourceLineNo">445</span> private enum Order {<a name="line.445"></a> -<span class="sourceLineNo">446</span> ASC {<a name="line.446"></a> -<span class="sourceLineNo">447</span> public boolean lt(int lhs, int rhs) {<a name="line.447"></a> -<span class="sourceLineNo">448</span> return lhs < rhs;<a name="line.448"></a> -<span class="sourceLineNo">449</span> }<a name="line.449"></a> -<span class="sourceLineNo">450</span><a name="line.450"></a> -<span class="sourceLineNo">451</span> public boolean gt(int lhs, int rhs) {<a name="line.451"></a> -<span class="sourceLineNo">452</span> return lhs > rhs;<a name="line.452"></a> -<span class="sourceLineNo">453</span> }<a name="line.453"></a> -<span class="sourceLineNo">454</span><a name="line.454"></a> -<span class="sourceLineNo">455</span> public byte inc(byte val) {<a name="line.455"></a> -<span class="sourceLineNo">456</span> // TODO: what about over/underflow?<a name="line.456"></a> -<span class="sourceLineNo">457</span> return (byte) (val + 1);<a name="line.457"></a> -<span class="sourceLineNo">458</span> }<a name="line.458"></a> -<span class="sourceLineNo">459</span><a name="line.459"></a> -<span class="sourceLineNo">460</span> public boolean isMax(byte val) {<a name="line.460"></a> -<span class="sourceLineNo">461</span> return val == (byte) 0xff;<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> public byte min() {<a name="line.464"></a> -<span class="sourceLineNo">465</span> return 0;<a name="line.465"></a> +<span class="sourceLineNo">413</span> }<a name="line.413"></a> +<span class="sourceLineNo">414</span><a name="line.414"></a> +<span class="sourceLineNo">415</span> // If this row byte is less than fixed then there's a byte array bigger than<a name="line.415"></a> +<span class="sourceLineNo">416</span> // this row and which satisfies the fuzzy rule. Otherwise there's no such byte array:<a name="line.416"></a> +<span class="sourceLineNo">417</span> // this row is simply bigger than any byte array that satisfies the fuzzy rule<a name="line.417"></a> +<span class="sourceLineNo">418</span> boolean rowByteLessThanFixed = (row[i + offset] & 0xFF) < (fuzzyKeyBytes[i] & 0xFF);<a name="line.418"></a> +<span class="sourceLineNo">419</span> if (rowByteLessThanFixed && !reverse) {<a name="line.419"></a> +<span class="sourceLineNo">420</span> return SatisfiesCode.NEXT_EXISTS;<a name="line.420"></a> +<span class="sourceLineNo">421</span> } else if (!rowByteLessThanFixed && reverse) {<a name="line.421"></a> +<span class="sourceLineNo">422</span> return SatisfiesCode.NEXT_EXISTS;<a name="line.422"></a> +<span class="sourceLineNo">423</span> } else {<a name="line.423"></a> +<span class="sourceLineNo">424</span> return SatisfiesCode.NO_NEXT;<a name="line.424"></a> +<span class="sourceLineNo">425</span> }<a name="line.425"></a> +<span class="sourceLineNo">426</span> }<a name="line.426"></a> +<span class="sourceLineNo">427</span><a name="line.427"></a> +<span class="sourceLineNo">428</span> // Second, checking if this position is not fixed and byte value is not the biggest. In this<a name="line.428"></a> +<span class="sourceLineNo">429</span> // case there's a byte array bigger than this row and which satisfies the fuzzy rule. To get<a name="line.429"></a> +<span class="sourceLineNo">430</span> // bigger byte array that satisfies the rule we need to just increase this byte<a name="line.430"></a> +<span class="sourceLineNo">431</span> // (see the code of getNextForFuzzyRule below) by one.<a name="line.431"></a> +<span class="sourceLineNo">432</span> // Note: if non-fixed byte is already at biggest value, this doesn't allow us to say there's<a name="line.432"></a> +<span class="sourceLineNo">433</span> // bigger one that satisfies the rule as it can't be increased.<a name="line.433"></a> +<span class="sourceLineNo">434</span> if (fuzzyKeyMeta[i] == 1 && !order.isMax(fuzzyKeyBytes[i])) {<a name="line.434"></a> +<span class="sourceLineNo">435</span> nextRowKeyCandidateExists = true;<a name="line.435"></a> +<span class="sourceLineNo">436</span> }<a name="line.436"></a> +<span class="sourceLineNo">437</span> }<a name="line.437"></a> +<span class="sourceLineNo">438</span> return SatisfiesCode.YES;<a name="line.438"></a> +<span class="sourceLineNo">439</span> }<a name="line.439"></a> +<span class="sourceLineNo">440</span><a name="line.440"></a> +<span class="sourceLineNo">441</span> @VisibleForTesting<a name="line.441"></a> +<span class="sourceLineNo">442</span> static byte[] getNextForFuzzyRule(byte[] row, byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) {<a name="line.442"></a> +<span class="sourceLineNo">443</span> return getNextForFuzzyRule(false, row, 0, row.length, fuzzyKeyBytes, fuzzyKeyMeta);<a name="line.443"></a> +<span class="sourceLineNo">444</span> }<a name="line.444"></a> +<span class="sourceLineNo">445</span><a name="line.445"></a> +<span class="sourceLineNo">446</span> @VisibleForTesting<a name="line.446"></a> +<span class="sourceLineNo">447</span> static byte[] getNextForFuzzyRule(boolean reverse, byte[] row, byte[] fuzzyKeyBytes,<a name="line.447"></a> +<span class="sourceLineNo">448</span> byte[] fuzzyKeyMeta) {<a name="line.448"></a> +<span class="sourceLineNo">449</span> return getNextForFuzzyRule(reverse, row, 0, row.length, fuzzyKeyBytes, fuzzyKeyMeta);<a name="line.449"></a> +<span class="sourceLineNo">450</span> }<a name="line.450"></a> +<span class="sourceLineNo">451</span><a name="line.451"></a> +<span class="sourceLineNo">452</span> /** Abstracts directional comparisons based on scan direction. */<a name="line.452"></a> +<span class="sourceLineNo">453</span> private enum Order {<a name="line.453"></a> +<span class="sourceLineNo">454</span> ASC {<a name="line.454"></a> +<span class="sourceLineNo">455</span> public boolean lt(int lhs, int rhs) {<a name="line.455"></a> +<span class="sourceLineNo">456</span> return lhs < rhs;<a name="line.456"></a> +<span class="sourceLineNo">457</span> }<a name="line.457"></a> +<span class="sourceLineNo">458</span><a name="line.458"></a> +<span class="sourceLineNo">459</span> public boolean gt(int lhs, int rhs) {<a name="line.459"></a> +<span class="sourceLineNo">460</span> return lhs > rhs;<a name="line.460"></a> +<span class="sourceLineNo">461</span> }<a name="line.461"></a> +<span class="sourceLineNo">462</span><a name="line.462"></a> +<span class="sourceLineNo">463</span> public byte inc(byte val) {<a name="line.463"></a> +<span class="sourceLineNo">464</span> // TODO: what about over/underflow?<a name="line.464"></a> +<span class="sourceLineNo">465</span> return (byte) (val + 1);<a name="line.465"></a> <span class="sourceLineNo">466</span> }<a name="line.466"></a> -<span class="sourceLineNo">467</span> },<a name="line.467"></a> -<span class="sourceLineNo">468</span> DESC {<a name="line.468"></a> -<span class="sourceLineNo">469</span> public boolean lt(int lhs, int rhs) {<a name="line.469"></a> -<span class="sourceLineNo">470</span> return lhs > rhs;<a name="line.470"></a> -<span class="sourceLineNo">471</span> }<a name="line.471"></a> -<span class="sourceLineNo">472</span><a name="line.472"></a> -<span class="sourceLineNo">473</span> public boolean gt(int lhs, int rhs) {<a name="line.473"></a> -<span class="sourceLineNo">474</span> return lhs < rhs;<a name="line.474"></a> -<span class="sourceLineNo">475</span> }<a name="line.475"></a> -<span class="sourceLineNo">476</span><a name="line.476"></a> -<span class="sourceLineNo">477</span> public byte inc(byte val) {<a name="line.477"></a> -<span class="sourceLineNo">478</span> // TODO: what about over/underflow?<a name="line.478"></a> -<span class="sourceLineNo">479</span> return (byte) (val - 1);<a name="line.479"></a> -<span class="sourceLineNo">480</span> }<a name="line.480"></a> -<span class="sourceLineNo">481</span><a name="line.481"></a> -<span class="sourceLineNo">482</span> public boolean isMax(byte val) {<a name="line.482"></a> -<span class="sourceLineNo">483</span> return val == 0;<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> public byte min() {<a name="line.486"></a> -<span class="sourceLineNo">487</span> return (byte) 0xFF;<a name="line.487"></a> +<span class="sourceLineNo">467</span><a name="line.467"></a> +<span class="sourceLineNo">468</span> public boolean isMax(byte val) {<a name="line.468"></a> +<span class="sourceLineNo">469</span> return val == (byte) 0xff;<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> public byte min() {<a name="line.472"></a> +<span class="sourceLineNo">473</span> return 0;<a name="line.473"></a> +<span class="sourceLineNo">474</span> }<a name="line.474"></a> +<span class="sourceLineNo">475</span> },<a name="line.475"></a> +<span class="sourceLineNo">476</span> DESC {<a name="line.476"></a> +<span class="sourceLineNo">477</span> public boolean lt(int lhs, int rhs) {<a name="line.477"></a> +<span class="sourceLineNo">478</span> return lhs > rhs;<a name="line.478"></a> +<span class="sourceLineNo">479</span> }<a name="line.479"></a> +<span class="sourceLineNo">480</span><a name="line.480"></a> +<span class="sourceLineNo">481</span> public boolean gt(int lhs, int rhs) {<a name="line.481"></a> +<span class="sourceLineNo">482</span> return lhs < rhs;<a name="line.482"></a> +<span class="sourceLineNo">483</span> }<a name="line.483"></a> +<span class="sourceLineNo">484</span><a name="line.484"></a> +<span class="sourceLineNo">485</span> public byte inc(byte val) {<a name="line.485"></a> +<span class="sourceLineNo">486</span> // TODO: what about over/underflow?<a name="line.486"></a> +<span class="sourceLineNo">487</span> return (byte) (val - 1);<a name="line.487"></a> <span class="sourceLineNo">488</span> }<a name="line.488"></a> -<span class="sourceLineNo">489</span> };<a name="line.489"></a> -<span class="sourceLineNo">490</span><a name="line.490"></a> -<span class="sourceLineNo">491</span> public static Order orderFor(boolean reverse) {<a name="line.491"></a> -<span class="sourceLineNo">492</span> return reverse ? DESC : ASC;<a name="line.492"></a> -<span class="sourceLineNo">493</span> }<a name="line.493"></a> -<span class="sourceLineNo">494</span><a name="line.494"></a> -<span class="sourceLineNo">495</span> /** Returns true when {@code lhs < rhs}. */<a name="line.495"></a> -<span class="sourceLineNo">496</span> public abstract boolean lt(int lhs, int rhs);<a name="line.496"></a> -<span class="sourceLineNo">497</span><a name="line.497"></a> -<span class="sourceLineNo">498</span> /** Returns true when {@code lhs > rhs}. */<a name="line.498"></a> -<span class="sourceLineNo">499</span> public abstract boolean gt(int lhs, int rhs);<a name="line.499"></a> -<span class="sourceLineNo">500</span><a name="line.500"></a> -<span class="sourceLineNo">501</span> /** Returns {@code val} incremented by 1. */<a name="line.501"></a> -<span class="sourceLineNo">502</span> public abstract byte inc(byte val);<a name="line.502"></a> -<span class="sourceLineNo">503</span><a name="line.503"></a> -<span class="sourceLineNo">504</span> /** Return true when {@code val} is the maximum value */<a name="line.504"></a> -<span class="sourceLineNo">505</span> public abstract boolean isMax(byte val);<a name="line.505"></a> -<span class="sourceLineNo">506</span><a name="line.506"></a> -<span class="sourceLineNo">507</span> /** Return the minimum value according to this ordering scheme. */<a name="line.507"></a> -<span class="sourceLineNo">508</span> public abstract byte min();<a name="line.508"></a> -<span class="sourceLineNo">509</span> }<a name="line.509"></a> -<span class="sourceLineNo">510</span><a name="line.510"></a> -<span class="sourceLineNo">511</span> /**<a name="line.511"></a> -<span class="sourceLineNo">512</span> * @return greater byte array than given (row) which satisfies the fuzzy rule if it exists, null<a name="line.512"></a> -<span class="sourceLineNo">513</span> * otherwise<a name="line.513"></a> -<span class="sourceLineNo">514</span> */<a name="line.514"></a> -<span class="sourceLineNo">515</span> @VisibleForTesting<a name="line.515"></a> -<span class="sourceLineNo">516</span> static byte[] getNextForFuzzyRule(boolean reverse, byte[] row, int offset, int length,<a name="line.516"></a> -<span class="sourceLineNo">517</span> byte[] fuzzyKeyBytes, byte[] fuzzyKeyMeta) {<a name="line.517"></a> -<span class="sourceLineNo">518</span> // To find out the next "smallest" byte array that satisfies fuzzy rule and "greater" than<a name="line.518"></a> -<span class="sourceLineNo">519</span> // the given one we do the following:<a name="line.519"></a> -<span class="sourceLineNo">520</span> // 1. setting values on all "fixed" positions to the values from fuzzyKeyBytes<a name="line.520"></a> -<span class="sourceLineNo">521</span> // 2. if during the first step given row did not increase, then we increase the value at<a name="line.521"></a> -<span class="sourceLineNo">522</span> // the first "non-fixed" position (where it is not maximum already)<a name="line.522"></a> -<span class="sourceLineNo">523</span><a name="line.523"></a> -<span class="sourceLineNo">524</span> // It is easier to perform this by using fuzzyKeyBytes copy and setting "non-fixed" position<a name="line.524"></a> -<span class="sourceLineNo">525</span> // values than otherwise.<a name="line.525"></a> -<span class="sourceLineNo">526</span> byte[] result =<a name="line.526"></a> -<span class="sourceLineNo">527</span> Arrays.copyOf(fuzzyKeyBytes, length > fuzzyKeyBytes.length ? length : fuzzyKeyBytes.length);<a name="line.527"></a> -<span class="sourceLineNo">528</span> if (reverse && length > fuzzyKeyBytes.length) {<a name="line.528"></a> -<span class="sourceLineNo">529</span> // we need trailing 0xff's instead of trailing 0x00's<a name="line.529"></a> -<span class="sourceLineNo">530</span> for (int i = fuzzyKeyBytes.length; i < result.length; i++) {<a name="line.530"></a> -<span class="sourceLineNo">531</span> result[i] = (byte) 0xFF;<a name="line.531"></a> -<span class="sourceLineNo">532</span> }<a name="line.532"></a> -<span class="sourceLineNo">533</span> }<a name="line.533"></a> -<span class="sourceLineNo">534</span> int toInc = -1;<a name="line.534"></a> -<span class="sourceLineNo">535</span> final Order order = Order.orderFor(reverse);<a name="line.535"></a> -<span class="sourceLineNo">536</span><a name="line.536"></a> -<span class="sourceLineNo">537</span> boolean increased = false;<a name="line.537"></a> -<span class="sourceLineNo">538</span> for (int i = 0; i < result.length; i++) {<a name="line.538"></a> -<span class="sourceLineNo">539</span> if (i >= fuzzyKeyMeta.length || fuzzyKeyMeta[i] == 0 /* non-fixed */) {<a name="line.539"></a> -<span class="sourceLineNo">540</span> result[i] = row[offset + i];<a name="line.540"></a> -<span class="sourceLineNo">541</span> if (!order.isMax(row[offset + i])) {<a name="line.541"></a> -<span class="sourceLineNo">542</span> // this is "non-fixed" position and is not at max value, hence we can increase it<a name="line.542"></a> -<span class="sourceLineNo">543</span> toInc = i;<a name="line.543"></a> -<span class="sourceLineNo">544</span> }<a name="line.544"></a> -<span class="sourceLineNo">545</span> } else if (i < fuzzyKeyMeta.length && fuzzyKeyMeta[i] == -1 /* fixed */) {<a name="line.545"></a> -<span class="sourceLineNo">546</span> if (order.lt((row[i + offset] & 0xFF), (fuzzyKeyBytes[i] & 0xFF))) {<a name="line.546"></a> -<span class="sourceLineNo">547</span> // if setting value for any fixed position increased the original array,<a name="line.547"></a> -<span class="sourceLineNo">548</span> // we are OK<a name="line.548"></a> -<span class="sourceLineNo">549</span> increased = true;<a name="line.549"></a> -<span class="sourceLineNo">550</span> break;<a name="line.550"></a> -<span class="sourceLineNo">551</span> }<a name="line.551"></a> -<span class="sourceLineNo">552</span><a name="line.552"></a> -<span class="sourceLineNo">553</span> if (order.gt((row[i + offset] & 0xFF), (fuzzyKeyBytes[i] & 0xFF))) {<a name="line.553"></a> -<span class="sourceLineNo">554</span> // if setting value for any fixed position makes array "smaller", then just stop:<a name="line.554"></a> -<span class="sourceLineNo">555</span> // in case we found some non-fixed position to increase we will do it, otherwise<a name="line.555"></a> -<span class="sourceLineNo">556</span> // there's no "next" row key that satisfies fuzzy rule and "greater" than given row<a name="line.556"></a> -<span class="sourceLineNo">557</span> break;<a name="line.557"></a> -<span class="sourceLineNo">558</span> }<a name="line.558"></a> -<span class="sourceLineNo">559</span> }<a name="line.559"></a> -<span class="sourceLineNo">560</span> }<a name="line.560"></a> -<span class="sourceLineNo">561</span><a name="line.561"></a> -<span class="sourceLineNo">562</span> if (!increased) {<a name="line.562"></a> -<span class="sourceLineNo">563</span> if (toInc < 0) {<a name="line.563"></a> -<span class="sourceLineNo">564</span> return null;<a name="line.564"></a> -<span class="sourceLineNo">565</span> }<a name="line.565"></a> -<span class="sourceLineNo">566</span> result[toInc] = order.inc(result[toInc]);<a name="line.566"></a> -<span class="sourceLineNo">567</span><a name="line.567"></a> -<span class="sourceLineNo">568</span> // Setting all "non-fixed" positions to zeroes to the right of the one we increased so<a name="line.568"></a> -<span class="sourceLineNo">569</span> // that found "next" row key is the smallest possible<a name="line.569"></a> -<span class="sourceLineNo">570</span> for (int i = toInc + 1; i < result.length; i++) {<a name="line.570"></a> -<span class="sourceLineNo">571</span> if (i >= fuzzyKeyMeta.length || fuzzyKeyMeta[i] == 0 /* non-fixed */) {<a name="line.571"></a> -<span class="sourceLineNo">572</span> result[i] = order.min();<a name="line.572"></a> -<span class="sourceLineNo">573</span> }<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><a name="line.576"></a> -<span class="sourceLineNo">577</span> return reverse? result: trimTrailingZeroes(result, fuzzyKeyMeta, toInc);<a name="line.577"></a> -<span class="sourceLineNo">578</span> }<a name="line.578"></a> -<span class="sourceLineNo">579</span><a name="line.579"></a> -<span class="sourceLineNo">580</span> /**<a name="line.580"></a> -<span class="sourceLineNo">581</span> * For forward scanner, next cell hint should not contain any trailing zeroes<a name="line.581"></a> -<span class="sourceLineNo">582</span> * unless they are part of fuzzyKeyMeta<a name="line.582"></a> -<span class="sourceLineNo">583</span> * hint = '\x01\x01\x01\x00\x00'<a name="line.583"></a> -<span class="sourceLineNo">584</span> * will skip valid row '\x01\x01\x01'<a name="line.584"></a>
<TRUNCATED>
