http://git-wip-us.apache.org/repos/asf/hbase-site/blob/22cff34f/apidocs/src-html/org/apache/hadoop/hbase/mapreduce/TableInputFormatBase.html ---------------------------------------------------------------------- diff --git a/apidocs/src-html/org/apache/hadoop/hbase/mapreduce/TableInputFormatBase.html b/apidocs/src-html/org/apache/hadoop/hbase/mapreduce/TableInputFormatBase.html index 7003255..a2b0cb0 100644 --- a/apidocs/src-html/org/apache/hadoop/hbase/mapreduce/TableInputFormatBase.html +++ b/apidocs/src-html/org/apache/hadoop/hbase/mapreduce/TableInputFormatBase.html @@ -145,523 +145,522 @@ <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> /** The reverse DNS lookup cache mapping: IPAddress => HostName */<a name="line.139"></a> -<span class="sourceLineNo">140</span> private HashMap<InetAddress, String> reverseDNSCacheMap =<a name="line.140"></a> -<span class="sourceLineNo">141</span> new HashMap<InetAddress, String>();<a name="line.141"></a> -<span class="sourceLineNo">142</span><a name="line.142"></a> -<span class="sourceLineNo">143</span> /**<a name="line.143"></a> -<span class="sourceLineNo">144</span> * Builds a {@link TableRecordReader}. If no {@link TableRecordReader} was provided, uses<a name="line.144"></a> -<span class="sourceLineNo">145</span> * the default.<a name="line.145"></a> -<span class="sourceLineNo">146</span> *<a name="line.146"></a> -<span class="sourceLineNo">147</span> * @param split The split to work with.<a name="line.147"></a> -<span class="sourceLineNo">148</span> * @param context The current context.<a name="line.148"></a> -<span class="sourceLineNo">149</span> * @return The newly created record reader.<a name="line.149"></a> -<span class="sourceLineNo">150</span> * @throws IOException When creating the reader fails.<a name="line.150"></a> -<span class="sourceLineNo">151</span> * @see org.apache.hadoop.mapreduce.InputFormat#createRecordReader(<a name="line.151"></a> -<span class="sourceLineNo">152</span> * org.apache.hadoop.mapreduce.InputSplit,<a name="line.152"></a> -<span class="sourceLineNo">153</span> * org.apache.hadoop.mapreduce.TaskAttemptContext)<a name="line.153"></a> -<span class="sourceLineNo">154</span> */<a name="line.154"></a> -<span class="sourceLineNo">155</span> @Override<a name="line.155"></a> -<span class="sourceLineNo">156</span> public RecordReader<ImmutableBytesWritable, Result> createRecordReader(<a name="line.156"></a> -<span class="sourceLineNo">157</span> InputSplit split, TaskAttemptContext context)<a name="line.157"></a> -<span class="sourceLineNo">158</span> throws IOException {<a name="line.158"></a> -<span class="sourceLineNo">159</span> // Just in case a subclass is relying on JobConfigurable magic.<a name="line.159"></a> -<span class="sourceLineNo">160</span> if (table == null) {<a name="line.160"></a> -<span class="sourceLineNo">161</span> initialize(context);<a name="line.161"></a> -<span class="sourceLineNo">162</span> }<a name="line.162"></a> -<span class="sourceLineNo">163</span> // null check in case our child overrides getTable to not throw.<a name="line.163"></a> -<span class="sourceLineNo">164</span> try {<a name="line.164"></a> -<span class="sourceLineNo">165</span> if (getTable() == null) {<a name="line.165"></a> -<span class="sourceLineNo">166</span> // initialize() must not have been implemented in the subclass.<a name="line.166"></a> -<span class="sourceLineNo">167</span> throw new IOException(INITIALIZATION_ERROR);<a name="line.167"></a> -<span class="sourceLineNo">168</span> }<a name="line.168"></a> -<span class="sourceLineNo">169</span> } catch (IllegalStateException exception) {<a name="line.169"></a> -<span class="sourceLineNo">170</span> throw new IOException(INITIALIZATION_ERROR, exception);<a name="line.170"></a> -<span class="sourceLineNo">171</span> }<a name="line.171"></a> -<span class="sourceLineNo">172</span> TableSplit tSplit = (TableSplit) split;<a name="line.172"></a> -<span class="sourceLineNo">173</span> LOG.info("Input split length: " + StringUtils.humanReadableInt(tSplit.getLength()) + " bytes.");<a name="line.173"></a> -<span class="sourceLineNo">174</span> final TableRecordReader trr =<a name="line.174"></a> -<span class="sourceLineNo">175</span> this.tableRecordReader != null ? this.tableRecordReader : new TableRecordReader();<a name="line.175"></a> -<span class="sourceLineNo">176</span> Scan sc = new Scan(this.scan);<a name="line.176"></a> -<span class="sourceLineNo">177</span> sc.setStartRow(tSplit.getStartRow());<a name="line.177"></a> -<span class="sourceLineNo">178</span> sc.setStopRow(tSplit.getEndRow());<a name="line.178"></a> -<span class="sourceLineNo">179</span> trr.setScan(sc);<a name="line.179"></a> -<span class="sourceLineNo">180</span> trr.setTable(getTable());<a name="line.180"></a> -<span class="sourceLineNo">181</span> return new RecordReader<ImmutableBytesWritable, Result>() {<a name="line.181"></a> -<span class="sourceLineNo">182</span><a name="line.182"></a> -<span class="sourceLineNo">183</span> @Override<a name="line.183"></a> -<span class="sourceLineNo">184</span> public void close() throws IOException {<a name="line.184"></a> -<span class="sourceLineNo">185</span> trr.close();<a name="line.185"></a> -<span class="sourceLineNo">186</span> closeTable();<a name="line.186"></a> -<span class="sourceLineNo">187</span> }<a name="line.187"></a> -<span class="sourceLineNo">188</span><a name="line.188"></a> -<span class="sourceLineNo">189</span> @Override<a name="line.189"></a> -<span class="sourceLineNo">190</span> public ImmutableBytesWritable getCurrentKey() throws IOException, InterruptedException {<a name="line.190"></a> -<span class="sourceLineNo">191</span> return trr.getCurrentKey();<a name="line.191"></a> -<span class="sourceLineNo">192</span> }<a name="line.192"></a> -<span class="sourceLineNo">193</span><a name="line.193"></a> -<span class="sourceLineNo">194</span> @Override<a name="line.194"></a> -<span class="sourceLineNo">195</span> public Result getCurrentValue() throws IOException, InterruptedException {<a name="line.195"></a> -<span class="sourceLineNo">196</span> return trr.getCurrentValue();<a name="line.196"></a> -<span class="sourceLineNo">197</span> }<a name="line.197"></a> -<span class="sourceLineNo">198</span><a name="line.198"></a> -<span class="sourceLineNo">199</span> @Override<a name="line.199"></a> -<span class="sourceLineNo">200</span> public float getProgress() throws IOException, InterruptedException {<a name="line.200"></a> -<span class="sourceLineNo">201</span> return trr.getProgress();<a name="line.201"></a> -<span class="sourceLineNo">202</span> }<a name="line.202"></a> -<span class="sourceLineNo">203</span><a name="line.203"></a> -<span class="sourceLineNo">204</span> @Override<a name="line.204"></a> -<span class="sourceLineNo">205</span> public void initialize(InputSplit inputsplit, TaskAttemptContext context) throws IOException,<a name="line.205"></a> -<span class="sourceLineNo">206</span> InterruptedException {<a name="line.206"></a> -<span class="sourceLineNo">207</span> trr.initialize(inputsplit, context);<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> @Override<a name="line.210"></a> -<span class="sourceLineNo">211</span> public boolean nextKeyValue() throws IOException, InterruptedException {<a name="line.211"></a> -<span class="sourceLineNo">212</span> return trr.nextKeyValue();<a name="line.212"></a> -<span class="sourceLineNo">213</span> }<a name="line.213"></a> -<span class="sourceLineNo">214</span> };<a name="line.214"></a> -<span class="sourceLineNo">215</span> }<a name="line.215"></a> -<span class="sourceLineNo">216</span><a name="line.216"></a> -<span class="sourceLineNo">217</span> protected Pair<byte[][],byte[][]> getStartEndKeys() throws IOException {<a name="line.217"></a> -<span class="sourceLineNo">218</span> return getRegionLocator().getStartEndKeys();<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> /**<a name="line.221"></a> -<span class="sourceLineNo">222</span> * Calculates the splits that will serve as input for the map tasks. The<a name="line.222"></a> -<span class="sourceLineNo">223</span> * number of splits matches the number of regions in a table.<a name="line.223"></a> -<span class="sourceLineNo">224</span> *<a name="line.224"></a> -<span class="sourceLineNo">225</span> * @param context The current job context.<a name="line.225"></a> -<span class="sourceLineNo">226</span> * @return The list of input splits.<a name="line.226"></a> -<span class="sourceLineNo">227</span> * @throws IOException When creating the list of splits fails.<a name="line.227"></a> -<span class="sourceLineNo">228</span> * @see org.apache.hadoop.mapreduce.InputFormat#getSplits(<a name="line.228"></a> -<span class="sourceLineNo">229</span> * org.apache.hadoop.mapreduce.JobContext)<a name="line.229"></a> -<span class="sourceLineNo">230</span> */<a name="line.230"></a> -<span class="sourceLineNo">231</span> @Override<a name="line.231"></a> -<span class="sourceLineNo">232</span> public List<InputSplit> getSplits(JobContext context) throws IOException {<a name="line.232"></a> -<span class="sourceLineNo">233</span> boolean closeOnFinish = false;<a name="line.233"></a> -<span class="sourceLineNo">234</span><a name="line.234"></a> -<span class="sourceLineNo">235</span> // Just in case a subclass is relying on JobConfigurable magic.<a name="line.235"></a> -<span class="sourceLineNo">236</span> if (table == null) {<a name="line.236"></a> -<span class="sourceLineNo">237</span> initialize(context);<a name="line.237"></a> -<span class="sourceLineNo">238</span> closeOnFinish = true;<a name="line.238"></a> -<span class="sourceLineNo">239</span> }<a name="line.239"></a> -<span class="sourceLineNo">240</span><a name="line.240"></a> -<span class="sourceLineNo">241</span> // null check in case our child overrides getTable to not throw.<a name="line.241"></a> -<span class="sourceLineNo">242</span> try {<a name="line.242"></a> -<span class="sourceLineNo">243</span> if (getTable() == null) {<a name="line.243"></a> -<span class="sourceLineNo">244</span> // initialize() must not have been implemented in the subclass.<a name="line.244"></a> -<span class="sourceLineNo">245</span> throw new IOException(INITIALIZATION_ERROR);<a name="line.245"></a> -<span class="sourceLineNo">246</span> }<a name="line.246"></a> -<span class="sourceLineNo">247</span> } catch (IllegalStateException exception) {<a name="line.247"></a> -<span class="sourceLineNo">248</span> throw new IOException(INITIALIZATION_ERROR, exception);<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> try {<a name="line.251"></a> -<span class="sourceLineNo">252</span> RegionSizeCalculator sizeCalculator =<a name="line.252"></a> -<span class="sourceLineNo">253</span> new RegionSizeCalculator(getRegionLocator(), getAdmin());<a name="line.253"></a> -<span class="sourceLineNo">254</span> <a name="line.254"></a> -<span class="sourceLineNo">255</span> TableName tableName = getTable().getName();<a name="line.255"></a> -<span class="sourceLineNo">256</span> <a name="line.256"></a> -<span class="sourceLineNo">257</span> Pair<byte[][], byte[][]> keys = getStartEndKeys();<a name="line.257"></a> -<span class="sourceLineNo">258</span> if (keys == null || keys.getFirst() == null ||<a name="line.258"></a> -<span class="sourceLineNo">259</span> keys.getFirst().length == 0) {<a name="line.259"></a> -<span class="sourceLineNo">260</span> HRegionLocation regLoc =<a name="line.260"></a> -<span class="sourceLineNo">261</span> getRegionLocator().getRegionLocation(HConstants.EMPTY_BYTE_ARRAY, false);<a name="line.261"></a> -<span class="sourceLineNo">262</span> if (null == regLoc) {<a name="line.262"></a> -<span class="sourceLineNo">263</span> throw new IOException("Expecting at least one region.");<a name="line.263"></a> -<span class="sourceLineNo">264</span> }<a name="line.264"></a> -<span class="sourceLineNo">265</span> List<InputSplit> splits = new ArrayList<InputSplit>(1);<a name="line.265"></a> -<span class="sourceLineNo">266</span> long regionSize = sizeCalculator.getRegionSize(regLoc.getRegionInfo().getRegionName());<a name="line.266"></a> -<span class="sourceLineNo">267</span> TableSplit split = new TableSplit(tableName, scan,<a name="line.267"></a> -<span class="sourceLineNo">268</span> HConstants.EMPTY_BYTE_ARRAY, HConstants.EMPTY_BYTE_ARRAY, regLoc<a name="line.268"></a> -<span class="sourceLineNo">269</span> .getHostnamePort().split(Addressing.HOSTNAME_PORT_SEPARATOR)[0], regionSize);<a name="line.269"></a> -<span class="sourceLineNo">270</span> splits.add(split);<a name="line.270"></a> -<span class="sourceLineNo">271</span> return splits;<a name="line.271"></a> -<span class="sourceLineNo">272</span> }<a name="line.272"></a> -<span class="sourceLineNo">273</span> List<InputSplit> splits = new ArrayList<InputSplit>(keys.getFirst().length);<a name="line.273"></a> -<span class="sourceLineNo">274</span> for (int i = 0; i < keys.getFirst().length; i++) {<a name="line.274"></a> -<span class="sourceLineNo">275</span> if (!includeRegionInSplit(keys.getFirst()[i], keys.getSecond()[i])) {<a name="line.275"></a> -<span class="sourceLineNo">276</span> continue;<a name="line.276"></a> -<span class="sourceLineNo">277</span> }<a name="line.277"></a> -<span class="sourceLineNo">278</span><a name="line.278"></a> -<span class="sourceLineNo">279</span> byte[] startRow = scan.getStartRow();<a name="line.279"></a> -<span class="sourceLineNo">280</span> byte[] stopRow = scan.getStopRow();<a name="line.280"></a> -<span class="sourceLineNo">281</span> // determine if the given start an stop key fall into the region<a name="line.281"></a> -<span class="sourceLineNo">282</span> if ((startRow.length == 0 || keys.getSecond()[i].length == 0 ||<a name="line.282"></a> -<span class="sourceLineNo">283</span> Bytes.compareTo(startRow, keys.getSecond()[i]) < 0) &&<a name="line.283"></a> -<span class="sourceLineNo">284</span> (stopRow.length == 0 ||<a name="line.284"></a> -<span class="sourceLineNo">285</span> Bytes.compareTo(stopRow, keys.getFirst()[i]) > 0)) {<a name="line.285"></a> -<span class="sourceLineNo">286</span> byte[] splitStart = startRow.length == 0 ||<a name="line.286"></a> -<span class="sourceLineNo">287</span> Bytes.compareTo(keys.getFirst()[i], startRow) >= 0 ?<a name="line.287"></a> -<span class="sourceLineNo">288</span> keys.getFirst()[i] : startRow;<a name="line.288"></a> -<span class="sourceLineNo">289</span> byte[] splitStop = (stopRow.length == 0 ||<a name="line.289"></a> -<span class="sourceLineNo">290</span> Bytes.compareTo(keys.getSecond()[i], stopRow) <= 0) &&<a name="line.290"></a> -<span class="sourceLineNo">291</span> keys.getSecond()[i].length > 0 ?<a name="line.291"></a> -<span class="sourceLineNo">292</span> keys.getSecond()[i] : stopRow;<a name="line.292"></a> -<span class="sourceLineNo">293</span><a name="line.293"></a> -<span class="sourceLineNo">294</span> HRegionLocation location = getRegionLocator().getRegionLocation(keys.getFirst()[i], false);<a name="line.294"></a> -<span class="sourceLineNo">295</span> // The below InetSocketAddress creation does a name resolution.<a name="line.295"></a> -<span class="sourceLineNo">296</span> InetSocketAddress isa = new InetSocketAddress(location.getHostname(), location.getPort());<a name="line.296"></a> -<span class="sourceLineNo">297</span> if (isa.isUnresolved()) {<a name="line.297"></a> -<span class="sourceLineNo">298</span> LOG.warn("Failed resolve " + isa);<a name="line.298"></a> -<span class="sourceLineNo">299</span> }<a name="line.299"></a> -<span class="sourceLineNo">300</span> InetAddress regionAddress = isa.getAddress();<a name="line.300"></a> -<span class="sourceLineNo">301</span> String regionLocation;<a name="line.301"></a> -<span class="sourceLineNo">302</span> regionLocation = reverseDNS(regionAddress);<a name="line.302"></a> -<span class="sourceLineNo">303</span><a name="line.303"></a> -<span class="sourceLineNo">304</span> byte[] regionName = location.getRegionInfo().getRegionName();<a name="line.304"></a> -<span class="sourceLineNo">305</span> String encodedRegionName = location.getRegionInfo().getEncodedName();<a name="line.305"></a> -<span class="sourceLineNo">306</span> long regionSize = sizeCalculator.getRegionSize(regionName);<a name="line.306"></a> -<span class="sourceLineNo">307</span> TableSplit split = new TableSplit(tableName, scan,<a name="line.307"></a> -<span class="sourceLineNo">308</span> splitStart, splitStop, regionLocation, encodedRegionName, regionSize);<a name="line.308"></a> -<span class="sourceLineNo">309</span> splits.add(split);<a name="line.309"></a> -<span class="sourceLineNo">310</span> if (LOG.isDebugEnabled()) {<a name="line.310"></a> -<span class="sourceLineNo">311</span> LOG.debug("getSplits: split -> " + i + " -> " + split);<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> }<a name="line.314"></a> -<span class="sourceLineNo">315</span> //The default value of "hbase.mapreduce.input.autobalance" is false, which means not enabled.<a name="line.315"></a> -<span class="sourceLineNo">316</span> boolean enableAutoBalance = context.getConfiguration()<a name="line.316"></a> -<span class="sourceLineNo">317</span> .getBoolean(MAPREDUCE_INPUT_AUTOBALANCE, false);<a name="line.317"></a> -<span class="sourceLineNo">318</span> if (enableAutoBalance) {<a name="line.318"></a> -<span class="sourceLineNo">319</span> long totalRegionSize=0;<a name="line.319"></a> -<span class="sourceLineNo">320</span> for (int i = 0; i < splits.size(); i++){<a name="line.320"></a> -<span class="sourceLineNo">321</span> TableSplit ts = (TableSplit)splits.get(i);<a name="line.321"></a> -<span class="sourceLineNo">322</span> totalRegionSize += ts.getLength();<a name="line.322"></a> -<span class="sourceLineNo">323</span> }<a name="line.323"></a> -<span class="sourceLineNo">324</span> long averageRegionSize = totalRegionSize / splits.size();<a name="line.324"></a> -<span class="sourceLineNo">325</span> // the averageRegionSize must be positive.<a name="line.325"></a> -<span class="sourceLineNo">326</span> if (averageRegionSize <= 0) {<a name="line.326"></a> -<span class="sourceLineNo">327</span> LOG.warn("The averageRegionSize is not positive: "+ averageRegionSize + ", " +<a name="line.327"></a> -<span class="sourceLineNo">328</span> "set it to 1.");<a name="line.328"></a> -<span class="sourceLineNo">329</span> averageRegionSize = 1;<a name="line.329"></a> -<span class="sourceLineNo">330</span> }<a name="line.330"></a> -<span class="sourceLineNo">331</span> return calculateRebalancedSplits(splits, context, averageRegionSize);<a name="line.331"></a> -<span class="sourceLineNo">332</span> } else {<a name="line.332"></a> -<span class="sourceLineNo">333</span> return splits;<a name="line.333"></a> -<span class="sourceLineNo">334</span> }<a name="line.334"></a> -<span class="sourceLineNo">335</span> } finally {<a name="line.335"></a> -<span class="sourceLineNo">336</span> if (closeOnFinish) {<a name="line.336"></a> -<span class="sourceLineNo">337</span> closeTable();<a name="line.337"></a> -<span class="sourceLineNo">338</span> }<a name="line.338"></a> -<span class="sourceLineNo">339</span> }<a name="line.339"></a> -<span class="sourceLineNo">340</span> }<a name="line.340"></a> -<span class="sourceLineNo">341</span><a name="line.341"></a> -<span class="sourceLineNo">342</span> String reverseDNS(InetAddress ipAddress) throws UnknownHostException {<a name="line.342"></a> -<span class="sourceLineNo">343</span> String hostName = this.reverseDNSCacheMap.get(ipAddress);<a name="line.343"></a> -<span class="sourceLineNo">344</span> if (hostName == null) {<a name="line.344"></a> -<span class="sourceLineNo">345</span> String ipAddressString = null;<a name="line.345"></a> -<span class="sourceLineNo">346</span> try {<a name="line.346"></a> -<span class="sourceLineNo">347</span> ipAddressString = DNS.reverseDns(ipAddress, null);<a name="line.347"></a> -<span class="sourceLineNo">348</span> } catch (Exception e) {<a name="line.348"></a> -<span class="sourceLineNo">349</span> // We can use InetAddress in case the jndi failed to pull up the reverse DNS entry from the<a name="line.349"></a> -<span class="sourceLineNo">350</span> // name service. Also, in case of ipv6, we need to use the InetAddress since resolving<a name="line.350"></a> -<span class="sourceLineNo">351</span> // reverse DNS using jndi doesn't work well with ipv6 addresses.<a name="line.351"></a> -<span class="sourceLineNo">352</span> ipAddressString = InetAddress.getByName(ipAddress.getHostAddress()).getHostName();<a name="line.352"></a> -<span class="sourceLineNo">353</span> }<a name="line.353"></a> -<span class="sourceLineNo">354</span> if (ipAddressString == null) throw new UnknownHostException("No host found for " + ipAddress);<a name="line.354"></a> -<span class="sourceLineNo">355</span> hostName = Strings.domainNamePointerToHostName(ipAddressString);<a name="line.355"></a> -<span class="sourceLineNo">356</span> this.reverseDNSCacheMap.put(ipAddress, hostName);<a name="line.356"></a> -<span class="sourceLineNo">357</span> }<a name="line.357"></a> -<span class="sourceLineNo">358</span> return hostName;<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> * Calculates the number of MapReduce input splits for the map tasks. The number of<a name="line.362"></a> -<span class="sourceLineNo">363</span> * MapReduce input splits depends on the average region size and the "data skew ratio" user set in<a name="line.363"></a> -<span class="sourceLineNo">364</span> * configuration.<a name="line.364"></a> -<span class="sourceLineNo">365</span> *<a name="line.365"></a> -<span class="sourceLineNo">366</span> * @param list The list of input splits before balance.<a name="line.366"></a> -<span class="sourceLineNo">367</span> * @param context The current job context.<a name="line.367"></a> -<span class="sourceLineNo">368</span> * @param average The average size of all regions .<a name="line.368"></a> -<span class="sourceLineNo">369</span> * @return The list of input splits.<a name="line.369"></a> -<span class="sourceLineNo">370</span> * @throws IOException When creating the list of splits fails.<a name="line.370"></a> -<span class="sourceLineNo">371</span> * @see org.apache.hadoop.mapreduce.InputFormat#getSplits(<a name="line.371"></a> -<span class="sourceLineNo">372</span> * org.apache.hadoop.mapreduce.JobContext)<a name="line.372"></a> -<span class="sourceLineNo">373</span> */<a name="line.373"></a> -<span class="sourceLineNo">374</span> private List<InputSplit> calculateRebalancedSplits(List<InputSplit> list, JobContext context,<a name="line.374"></a> -<span class="sourceLineNo">375</span> long average) throws IOException {<a name="line.375"></a> -<span class="sourceLineNo">376</span> List<InputSplit> resultList = new ArrayList<InputSplit>();<a name="line.376"></a> -<span class="sourceLineNo">377</span> Configuration conf = context.getConfiguration();<a name="line.377"></a> -<span class="sourceLineNo">378</span> //The default data skew ratio is 3<a name="line.378"></a> -<span class="sourceLineNo">379</span> long dataSkewRatio = conf.getLong(INPUT_AUTOBALANCE_MAXSKEWRATIO, 3);<a name="line.379"></a> -<span class="sourceLineNo">380</span> //It determines which mode to use: text key mode or binary key mode. The default is text mode.<a name="line.380"></a> -<span class="sourceLineNo">381</span> boolean isTextKey = context.getConfiguration().getBoolean(TABLE_ROW_TEXTKEY, true);<a name="line.381"></a> -<span class="sourceLineNo">382</span> long dataSkewThreshold = dataSkewRatio * average;<a name="line.382"></a> -<span class="sourceLineNo">383</span> int count = 0;<a name="line.383"></a> -<span class="sourceLineNo">384</span> while (count < list.size()) {<a name="line.384"></a> -<span class="sourceLineNo">385</span> TableSplit ts = (TableSplit)list.get(count);<a name="line.385"></a> -<span class="sourceLineNo">386</span> TableName tableName = ts.getTable();<a name="line.386"></a> -<span class="sourceLineNo">387</span> String regionLocation = ts.getRegionLocation();<a name="line.387"></a> -<span class="sourceLineNo">388</span> String encodedRegionName = ts.getEncodedRegionName();<a name="line.388"></a> -<span class="sourceLineNo">389</span> long regionSize = ts.getLength();<a name="line.389"></a> -<span class="sourceLineNo">390</span> if (regionSize >= dataSkewThreshold) {<a name="line.390"></a> -<span class="sourceLineNo">391</span> // if the current region size is large than the data skew threshold,<a name="line.391"></a> -<span class="sourceLineNo">392</span> // split the region into two MapReduce input splits.<a name="line.392"></a> -<span class="sourceLineNo">393</span> byte[] splitKey = getSplitKey(ts.getStartRow(), ts.getEndRow(), isTextKey);<a name="line.393"></a> -<span class="sourceLineNo">394</span> if (Arrays.equals(ts.getEndRow(), splitKey)) {<a name="line.394"></a> -<span class="sourceLineNo">395</span> // Not splitting since the end key is the same as the split key<a name="line.395"></a> -<span class="sourceLineNo">396</span> resultList.add(ts);<a name="line.396"></a> -<span class="sourceLineNo">397</span> } else {<a name="line.397"></a> -<span class="sourceLineNo">398</span> //Set the size of child TableSplit as 1/2 of the region size. The exact size of the<a name="line.398"></a> -<span class="sourceLineNo">399</span> // MapReduce input splits is not far off.<a name="line.399"></a> -<span class="sourceLineNo">400</span> TableSplit t1 = new TableSplit(tableName, scan, ts.getStartRow(), splitKey,<a name="line.400"></a> -<span class="sourceLineNo">401</span> regionLocation, regionSize / 2);<a name="line.401"></a> -<span class="sourceLineNo">402</span> TableSplit t2 = new TableSplit(tableName, scan, splitKey, ts.getEndRow(), regionLocation,<a name="line.402"></a> -<span class="sourceLineNo">403</span> regionSize - regionSize / 2);<a name="line.403"></a> -<span class="sourceLineNo">404</span> resultList.add(t1);<a name="line.404"></a> -<span class="sourceLineNo">405</span> resultList.add(t2);<a name="line.405"></a> -<span class="sourceLineNo">406</span> }<a name="line.406"></a> -<span class="sourceLineNo">407</span> count++;<a name="line.407"></a> -<span class="sourceLineNo">408</span> } else if (regionSize >= average) {<a name="line.408"></a> -<span class="sourceLineNo">409</span> // if the region size between average size and data skew threshold size,<a name="line.409"></a> -<span class="sourceLineNo">410</span> // make this region as one MapReduce input split.<a name="line.410"></a> -<span class="sourceLineNo">411</span> resultList.add(ts);<a name="line.411"></a> -<span class="sourceLineNo">412</span> count++;<a name="line.412"></a> -<span class="sourceLineNo">413</span> } else {<a name="line.413"></a> -<span class="sourceLineNo">414</span> // if the total size of several small continuous regions less than the average region size,<a name="line.414"></a> -<span class="sourceLineNo">415</span> // combine them into one MapReduce input split.<a name="line.415"></a> -<span class="sourceLineNo">416</span> long totalSize = regionSize;<a name="line.416"></a> -<span class="sourceLineNo">417</span> byte[] splitStartKey = ts.getStartRow();<a name="line.417"></a> -<span class="sourceLineNo">418</span> byte[] splitEndKey = ts.getEndRow();<a name="line.418"></a> -<span class="sourceLineNo">419</span> count++;<a name="line.419"></a> -<span class="sourceLineNo">420</span> for (; count < list.size(); count++) {<a name="line.420"></a> -<span class="sourceLineNo">421</span> TableSplit nextRegion = (TableSplit)list.get(count);<a name="line.421"></a> -<span class="sourceLineNo">422</span> long nextRegionSize = nextRegion.getLength();<a name="line.422"></a> -<span class="sourceLineNo">423</span> if (totalSize + nextRegionSize <= dataSkewThreshold) {<a name="line.423"></a> -<span class="sourceLineNo">424</span> totalSize = totalSize + nextRegionSize;<a name="line.424"></a> -<span class="sourceLineNo">425</span> splitEndKey = nextRegion.getEndRow();<a name="line.425"></a> -<span class="sourceLineNo">426</span> } else {<a name="line.426"></a> -<span class="sourceLineNo">427</span> break;<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> TableSplit t = new TableSplit(tableName, scan, splitStartKey, splitEndKey,<a name="line.430"></a> -<span class="sourceLineNo">431</span> regionLocation, encodedRegionName, totalSize);<a name="line.431"></a> -<span class="sourceLineNo">432</span> resultList.add(t);<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 resultList;<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> /**<a name="line.438"></a> -<span class="sourceLineNo">439</span> * select a split point in the region. The selection of the split point is based on an uniform<a name="line.439"></a> -<span class="sourceLineNo">440</span> * distribution assumption for the keys in a region.<a name="line.440"></a> -<span class="sourceLineNo">441</span> * Here are some examples:<a name="line.441"></a> -<span class="sourceLineNo">442</span> *<a name="line.442"></a> -<span class="sourceLineNo">443</span> * <table><a name="line.443"></a> -<span class="sourceLineNo">444</span> * <tr><a name="line.444"></a> -<span class="sourceLineNo">445</span> * <th>start key</th><a name="line.445"></a> -<span class="sourceLineNo">446</span> * <th>end key</th><a name="line.446"></a> -<span class="sourceLineNo">447</span> * <th>is text</th><a name="line.447"></a> -<span class="sourceLineNo">448</span> * <th>split point</th><a name="line.448"></a> -<span class="sourceLineNo">449</span> * </tr><a name="line.449"></a> -<span class="sourceLineNo">450</span> * <tr><a name="line.450"></a> -<span class="sourceLineNo">451</span> * <td>'a', 'a', 'a', 'b', 'c', 'd', 'e', 'f', 'g'</td><a name="line.451"></a> -<span class="sourceLineNo">452</span> * <td>'a', 'a', 'a', 'f', 'f', 'f'</td><a name="line.452"></a> -<span class="sourceLineNo">453</span> * <td>true</td><a name="line.453"></a> -<span class="sourceLineNo">454</span> * <td>'a', 'a', 'a', 'd', 'd', -78, 50, -77, 51</td><a name="line.454"></a> -<span class="sourceLineNo">455</span> * </tr><a name="line.455"></a> -<span class="sourceLineNo">456</span> * <tr><a name="line.456"></a> -<span class="sourceLineNo">457</span> * <td>'1', '1', '1', '0', '0', '0'</td><a name="line.457"></a> -<span class="sourceLineNo">458</span> * <td>'1', '1', '2', '5', '7', '9', '0'</td><a name="line.458"></a> -<span class="sourceLineNo">459</span> * <td>true</td><a name="line.459"></a> -<span class="sourceLineNo">460</span> * <td>'1', '1', '1', -78, -77, -76, -104</td><a name="line.460"></a> -<span class="sourceLineNo">461</span> * </tr><a name="line.461"></a> -<span class="sourceLineNo">462</span> * <tr><a name="line.462"></a> -<span class="sourceLineNo">463</span> * <td>'1', '1', '1', '0'</td><a name="line.463"></a> -<span class="sourceLineNo">464</span> * <td>'1', '1', '2', '0'</td><a name="line.464"></a> -<span class="sourceLineNo">465</span> * <td>true</td><a name="line.465"></a> -<span class="sourceLineNo">466</span> * <td>'1', '1', '1', -80</td><a name="line.466"></a> -<span class="sourceLineNo">467</span> * </tr><a name="line.467"></a> -<span class="sourceLineNo">468</span> * <tr><a name="line.468"></a> -<span class="sourceLineNo">469</span> * <td>13, -19, 126, 127</td><a name="line.469"></a> -<span class="sourceLineNo">470</span> * <td>13, -19, 127, 0</td><a name="line.470"></a> -<span class="sourceLineNo">471</span> * <td>false</td><a name="line.471"></a> -<span class="sourceLineNo">472</span> * <td>13, -19, 126, -65</td><a name="line.472"></a> -<span class="sourceLineNo">473</span> * </tr><a name="line.473"></a> -<span class="sourceLineNo">474</span> * </table><a name="line.474"></a> -<span class="sourceLineNo">475</span> *<a name="line.475"></a> -<span class="sourceLineNo">476</span> * Set this function as "public static", make it easier for test.<a name="line.476"></a> -<span class="sourceLineNo">477</span> *<a name="line.477"></a> -<span class="sourceLineNo">478</span> * @param start Start key of the region<a name="line.478"></a> -<span class="sourceLineNo">479</span> * @param end End key of the region<a name="line.479"></a> -<span class="sourceLineNo">480</span> * @param isText It determines to use text key mode or binary key mode<a name="line.480"></a> -<span class="sourceLineNo">481</span> * @return The split point in the region.<a name="line.481"></a> -<span class="sourceLineNo">482</span> */<a name="line.482"></a> -<span class="sourceLineNo">483</span> @InterfaceAudience.Private<a name="line.483"></a> -<span class="sourceLineNo">484</span> public static byte[] getSplitKey(byte[] start, byte[] end, boolean isText) {<a name="line.484"></a> -<span class="sourceLineNo">485</span> byte upperLimitByte;<a name="line.485"></a> -<span class="sourceLineNo">486</span> byte lowerLimitByte;<a name="line.486"></a> -<span class="sourceLineNo">487</span> //Use text mode or binary mode.<a name="line.487"></a> -<span class="sourceLineNo">488</span> if (isText) {<a name="line.488"></a> -<span class="sourceLineNo">489</span> //The range of text char set in ASCII is [32,126], the lower limit is space and the upper<a name="line.489"></a> -<span class="sourceLineNo">490</span> // limit is '~'.<a name="line.490"></a> -<span class="sourceLineNo">491</span> upperLimitByte = '~';<a name="line.491"></a> -<span class="sourceLineNo">492</span> lowerLimitByte = ' ';<a name="line.492"></a> -<span class="sourceLineNo">493</span> } else {<a name="line.493"></a> -<span class="sourceLineNo">494</span> upperLimitByte = -1;<a name="line.494"></a> -<span class="sourceLineNo">495</span> lowerLimitByte = 0;<a name="line.495"></a> -<span class="sourceLineNo">496</span> }<a name="line.496"></a> -<span class="sourceLineNo">497</span> // For special case<a name="line.497"></a> -<span class="sourceLineNo">498</span> // Example 1 : startkey=null, endkey="hhhqqqwww", splitKey="h"<a name="line.498"></a> -<span class="sourceLineNo">499</span> // Example 2 (text key mode): startKey="ffffaaa", endKey=null, splitkey="f~~~~~~"<a name="line.499"></a> -<span class="sourceLineNo">500</span> if (start.length == 0 && end.length == 0){<a name="line.500"></a> -<span class="sourceLineNo">501</span> return new byte[]{(byte) ((lowerLimitByte + upperLimitByte) / 2)};<a name="line.501"></a> -<span class="sourceLineNo">502</span> }<a name="line.502"></a> -<span class="sourceLineNo">503</span> if (start.length == 0 && end.length != 0){<a name="line.503"></a> -<span class="sourceLineNo">504</span> return new byte[]{ end[0] };<a name="line.504"></a> -<span class="sourceLineNo">505</span> }<a name="line.505"></a> -<span class="sourceLineNo">506</span> if (start.length != 0 && end.length == 0){<a name="line.506"></a> -<span class="sourceLineNo">507</span> byte[] result =new byte[start.length];<a name="line.507"></a> -<span class="sourceLineNo">508</span> result[0]=start[0];<a name="line.508"></a> -<span class="sourceLineNo">509</span> for (int k = 1; k < start.length; k++){<a name="line.509"></a> -<span class="sourceLineNo">510</span> result[k] = upperLimitByte;<a name="line.510"></a> -<span class="sourceLineNo">511</span> }<a name="line.511"></a> -<span class="sourceLineNo">512</span> return result;<a name="line.512"></a> -<span class="sourceLineNo">513</span> }<a name="line.513"></a> -<span class="sourceLineNo">514</span> return Bytes.split(start, end, false, 1)[1];<a name="line.514"></a> -<span class="sourceLineNo">515</span> }<a name="line.515"></a> -<span class="sourceLineNo">516</span><a name="line.516"></a> -<span class="sourceLineNo">517</span> /**<a name="line.517"></a> -<span class="sourceLineNo">518</span> * Test if the given region is to be included in the InputSplit while splitting<a name="line.518"></a> -<span class="sourceLineNo">519</span> * the regions of a table.<a name="line.519"></a> -<span class="sourceLineNo">520</span> * <p><a name="line.520"></a> -<span class="sourceLineNo">521</span> * This optimization is effective when there is a specific reasoning to exclude an entire region from the M-R job,<a name="line.521"></a> -<span class="sourceLineNo">522</span> * (and hence, not contributing to the InputSplit), given the start and end keys of the same. <br><a name="line.522"></a> -<span class="sourceLineNo">523</span> * Useful when we need to remember the last-processed top record and revisit the [last, current) interval for M-R processing,<a name="line.523"></a> -<span class="sourceLineNo">524</span> * continuously. In addition to reducing InputSplits, reduces the load on the region server as well, due to the ordering of the keys.<a name="line.524"></a> +<span class="sourceLineNo">140</span> private HashMap<InetAddress, String> reverseDNSCacheMap = new HashMap<>();<a name="line.140"></a> +<span class="sourceLineNo">141</span><a name="line.141"></a> +<span class="sourceLineNo">142</span> /**<a name="line.142"></a> +<span class="sourceLineNo">143</span> * Builds a {@link TableRecordReader}. If no {@link TableRecordReader} was provided, uses<a name="line.143"></a> +<span class="sourceLineNo">144</span> * the default.<a name="line.144"></a> +<span class="sourceLineNo">145</span> *<a name="line.145"></a> +<span class="sourceLineNo">146</span> * @param split The split to work with.<a name="line.146"></a> +<span class="sourceLineNo">147</span> * @param context The current context.<a name="line.147"></a> +<span class="sourceLineNo">148</span> * @return The newly created record reader.<a name="line.148"></a> +<span class="sourceLineNo">149</span> * @throws IOException When creating the reader fails.<a name="line.149"></a> +<span class="sourceLineNo">150</span> * @see org.apache.hadoop.mapreduce.InputFormat#createRecordReader(<a name="line.150"></a> +<span class="sourceLineNo">151</span> * org.apache.hadoop.mapreduce.InputSplit,<a name="line.151"></a> +<span class="sourceLineNo">152</span> * org.apache.hadoop.mapreduce.TaskAttemptContext)<a name="line.152"></a> +<span class="sourceLineNo">153</span> */<a name="line.153"></a> +<span class="sourceLineNo">154</span> @Override<a name="line.154"></a> +<span class="sourceLineNo">155</span> public RecordReader<ImmutableBytesWritable, Result> createRecordReader(<a name="line.155"></a> +<span class="sourceLineNo">156</span> InputSplit split, TaskAttemptContext context)<a name="line.156"></a> +<span class="sourceLineNo">157</span> throws IOException {<a name="line.157"></a> +<span class="sourceLineNo">158</span> // Just in case a subclass is relying on JobConfigurable magic.<a name="line.158"></a> +<span class="sourceLineNo">159</span> if (table == null) {<a name="line.159"></a> +<span class="sourceLineNo">160</span> initialize(context);<a name="line.160"></a> +<span class="sourceLineNo">161</span> }<a name="line.161"></a> +<span class="sourceLineNo">162</span> // null check in case our child overrides getTable to not throw.<a name="line.162"></a> +<span class="sourceLineNo">163</span> try {<a name="line.163"></a> +<span class="sourceLineNo">164</span> if (getTable() == null) {<a name="line.164"></a> +<span class="sourceLineNo">165</span> // initialize() must not have been implemented in the subclass.<a name="line.165"></a> +<span class="sourceLineNo">166</span> throw new IOException(INITIALIZATION_ERROR);<a name="line.166"></a> +<span class="sourceLineNo">167</span> }<a name="line.167"></a> +<span class="sourceLineNo">168</span> } catch (IllegalStateException exception) {<a name="line.168"></a> +<span class="sourceLineNo">169</span> throw new IOException(INITIALIZATION_ERROR, exception);<a name="line.169"></a> +<span class="sourceLineNo">170</span> }<a name="line.170"></a> +<span class="sourceLineNo">171</span> TableSplit tSplit = (TableSplit) split;<a name="line.171"></a> +<span class="sourceLineNo">172</span> LOG.info("Input split length: " + StringUtils.humanReadableInt(tSplit.getLength()) + " bytes.");<a name="line.172"></a> +<span class="sourceLineNo">173</span> final TableRecordReader trr =<a name="line.173"></a> +<span class="sourceLineNo">174</span> this.tableRecordReader != null ? this.tableRecordReader : new TableRecordReader();<a name="line.174"></a> +<span class="sourceLineNo">175</span> Scan sc = new Scan(this.scan);<a name="line.175"></a> +<span class="sourceLineNo">176</span> sc.setStartRow(tSplit.getStartRow());<a name="line.176"></a> +<span class="sourceLineNo">177</span> sc.setStopRow(tSplit.getEndRow());<a name="line.177"></a> +<span class="sourceLineNo">178</span> trr.setScan(sc);<a name="line.178"></a> +<span class="sourceLineNo">179</span> trr.setTable(getTable());<a name="line.179"></a> +<span class="sourceLineNo">180</span> return new RecordReader<ImmutableBytesWritable, Result>() {<a name="line.180"></a> +<span class="sourceLineNo">181</span><a name="line.181"></a> +<span class="sourceLineNo">182</span> @Override<a name="line.182"></a> +<span class="sourceLineNo">183</span> public void close() throws IOException {<a name="line.183"></a> +<span class="sourceLineNo">184</span> trr.close();<a name="line.184"></a> +<span class="sourceLineNo">185</span> closeTable();<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> @Override<a name="line.188"></a> +<span class="sourceLineNo">189</span> public ImmutableBytesWritable getCurrentKey() throws IOException, InterruptedException {<a name="line.189"></a> +<span class="sourceLineNo">190</span> return trr.getCurrentKey();<a name="line.190"></a> +<span class="sourceLineNo">191</span> }<a name="line.191"></a> +<span class="sourceLineNo">192</span><a name="line.192"></a> +<span class="sourceLineNo">193</span> @Override<a name="line.193"></a> +<span class="sourceLineNo">194</span> public Result getCurrentValue() throws IOException, InterruptedException {<a name="line.194"></a> +<span class="sourceLineNo">195</span> return trr.getCurrentValue();<a name="line.195"></a> +<span class="sourceLineNo">196</span> }<a name="line.196"></a> +<span class="sourceLineNo">197</span><a name="line.197"></a> +<span class="sourceLineNo">198</span> @Override<a name="line.198"></a> +<span class="sourceLineNo">199</span> public float getProgress() throws IOException, InterruptedException {<a name="line.199"></a> +<span class="sourceLineNo">200</span> return trr.getProgress();<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> @Override<a name="line.203"></a> +<span class="sourceLineNo">204</span> public void initialize(InputSplit inputsplit, TaskAttemptContext context) throws IOException,<a name="line.204"></a> +<span class="sourceLineNo">205</span> InterruptedException {<a name="line.205"></a> +<span class="sourceLineNo">206</span> trr.initialize(inputsplit, context);<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> @Override<a name="line.209"></a> +<span class="sourceLineNo">210</span> public boolean nextKeyValue() throws IOException, InterruptedException {<a name="line.210"></a> +<span class="sourceLineNo">211</span> return trr.nextKeyValue();<a name="line.211"></a> +<span class="sourceLineNo">212</span> }<a name="line.212"></a> +<span class="sourceLineNo">213</span> };<a name="line.213"></a> +<span class="sourceLineNo">214</span> }<a name="line.214"></a> +<span class="sourceLineNo">215</span><a name="line.215"></a> +<span class="sourceLineNo">216</span> protected Pair<byte[][],byte[][]> getStartEndKeys() throws IOException {<a name="line.216"></a> +<span class="sourceLineNo">217</span> return getRegionLocator().getStartEndKeys();<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> /**<a name="line.220"></a> +<span class="sourceLineNo">221</span> * Calculates the splits that will serve as input for the map tasks. The<a name="line.221"></a> +<span class="sourceLineNo">222</span> * number of splits matches the number of regions in a table.<a name="line.222"></a> +<span class="sourceLineNo">223</span> *<a name="line.223"></a> +<span class="sourceLineNo">224</span> * @param context The current job context.<a name="line.224"></a> +<span class="sourceLineNo">225</span> * @return The list of input splits.<a name="line.225"></a> +<span class="sourceLineNo">226</span> * @throws IOException When creating the list of splits fails.<a name="line.226"></a> +<span class="sourceLineNo">227</span> * @see org.apache.hadoop.mapreduce.InputFormat#getSplits(<a name="line.227"></a> +<span class="sourceLineNo">228</span> * org.apache.hadoop.mapreduce.JobContext)<a name="line.228"></a> +<span class="sourceLineNo">229</span> */<a name="line.229"></a> +<span class="sourceLineNo">230</span> @Override<a name="line.230"></a> +<span class="sourceLineNo">231</span> public List<InputSplit> getSplits(JobContext context) throws IOException {<a name="line.231"></a> +<span class="sourceLineNo">232</span> boolean closeOnFinish = false;<a name="line.232"></a> +<span class="sourceLineNo">233</span><a name="line.233"></a> +<span class="sourceLineNo">234</span> // Just in case a subclass is relying on JobConfigurable magic.<a name="line.234"></a> +<span class="sourceLineNo">235</span> if (table == null) {<a name="line.235"></a> +<span class="sourceLineNo">236</span> initialize(context);<a name="line.236"></a> +<span class="sourceLineNo">237</span> closeOnFinish = true;<a name="line.237"></a> +<span class="sourceLineNo">238</span> }<a name="line.238"></a> +<span class="sourceLineNo">239</span><a name="line.239"></a> +<span class="sourceLineNo">240</span> // null check in case our child overrides getTable to not throw.<a name="line.240"></a> +<span class="sourceLineNo">241</span> try {<a name="line.241"></a> +<span class="sourceLineNo">242</span> if (getTable() == null) {<a name="line.242"></a> +<span class="sourceLineNo">243</span> // initialize() must not have been implemented in the subclass.<a name="line.243"></a> +<span class="sourceLineNo">244</span> throw new IOException(INITIALIZATION_ERROR);<a name="line.244"></a> +<span class="sourceLineNo">245</span> }<a name="line.245"></a> +<span class="sourceLineNo">246</span> } catch (IllegalStateException exception) {<a name="line.246"></a> +<span class="sourceLineNo">247</span> throw new IOException(INITIALIZATION_ERROR, exception);<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> try {<a name="line.250"></a> +<span class="sourceLineNo">251</span> RegionSizeCalculator sizeCalculator =<a name="line.251"></a> +<span class="sourceLineNo">252</span> new RegionSizeCalculator(getRegionLocator(), getAdmin());<a name="line.252"></a> +<span class="sourceLineNo">253</span> <a name="line.253"></a> +<span class="sourceLineNo">254</span> TableName tableName = getTable().getName();<a name="line.254"></a> +<span class="sourceLineNo">255</span> <a name="line.255"></a> +<span class="sourceLineNo">256</span> Pair<byte[][], byte[][]> keys = getStartEndKeys();<a name="line.256"></a> +<span class="sourceLineNo">257</span> if (keys == null || keys.getFirst() == null ||<a name="line.257"></a> +<span class="sourceLineNo">258</span> keys.getFirst().length == 0) {<a name="line.258"></a> +<span class="sourceLineNo">259</span> HRegionLocation regLoc =<a name="line.259"></a> +<span class="sourceLineNo">260</span> getRegionLocator().getRegionLocation(HConstants.EMPTY_BYTE_ARRAY, false);<a name="line.260"></a> +<span class="sourceLineNo">261</span> if (null == regLoc) {<a name="line.261"></a> +<span class="sourceLineNo">262</span> throw new IOException("Expecting at least one region.");<a name="line.262"></a> +<span class="sourceLineNo">263</span> }<a name="line.263"></a> +<span class="sourceLineNo">264</span> List<InputSplit> splits = new ArrayList<>(1);<a name="line.264"></a> +<span class="sourceLineNo">265</span> long regionSize = sizeCalculator.getRegionSize(regLoc.getRegionInfo().getRegionName());<a name="line.265"></a> +<span class="sourceLineNo">266</span> TableSplit split = new TableSplit(tableName, scan,<a name="line.266"></a> +<span class="sourceLineNo">267</span> HConstants.EMPTY_BYTE_ARRAY, HConstants.EMPTY_BYTE_ARRAY, regLoc<a name="line.267"></a> +<span class="sourceLineNo">268</span> .getHostnamePort().split(Addressing.HOSTNAME_PORT_SEPARATOR)[0], regionSize);<a name="line.268"></a> +<span class="sourceLineNo">269</span> splits.add(split);<a name="line.269"></a> +<span class="sourceLineNo">270</span> return splits;<a name="line.270"></a> +<span class="sourceLineNo">271</span> }<a name="line.271"></a> +<span class="sourceLineNo">272</span> List<InputSplit> splits = new ArrayList<>(keys.getFirst().length);<a name="line.272"></a> +<span class="sourceLineNo">273</span> for (int i = 0; i < keys.getFirst().length; i++) {<a name="line.273"></a> +<span class="sourceLineNo">274</span> if (!includeRegionInSplit(keys.getFirst()[i], keys.getSecond()[i])) {<a name="line.274"></a> +<span class="sourceLineNo">275</span> continue;<a name="line.275"></a> +<span class="sourceLineNo">276</span> }<a name="line.276"></a> +<span class="sourceLineNo">277</span><a name="line.277"></a> +<span class="sourceLineNo">278</span> byte[] startRow = scan.getStartRow();<a name="line.278"></a> +<span class="sourceLineNo">279</span> byte[] stopRow = scan.getStopRow();<a name="line.279"></a> +<span class="sourceLineNo">280</span> // determine if the given start an stop key fall into the region<a name="line.280"></a> +<span class="sourceLineNo">281</span> if ((startRow.length == 0 || keys.getSecond()[i].length == 0 ||<a name="line.281"></a> +<span class="sourceLineNo">282</span> Bytes.compareTo(startRow, keys.getSecond()[i]) < 0) &&<a name="line.282"></a> +<span class="sourceLineNo">283</span> (stopRow.length == 0 ||<a name="line.283"></a> +<span class="sourceLineNo">284</span> Bytes.compareTo(stopRow, keys.getFirst()[i]) > 0)) {<a name="line.284"></a> +<span class="sourceLineNo">285</span> byte[] splitStart = startRow.length == 0 ||<a name="line.285"></a> +<span class="sourceLineNo">286</span> Bytes.compareTo(keys.getFirst()[i], startRow) >= 0 ?<a name="line.286"></a> +<span class="sourceLineNo">287</span> keys.getFirst()[i] : startRow;<a name="line.287"></a> +<span class="sourceLineNo">288</span> byte[] splitStop = (stopRow.length == 0 ||<a name="line.288"></a> +<span class="sourceLineNo">289</span> Bytes.compareTo(keys.getSecond()[i], stopRow) <= 0) &&<a name="line.289"></a> +<span class="sourceLineNo">290</span> keys.getSecond()[i].length > 0 ?<a name="line.290"></a> +<span class="sourceLineNo">291</span> keys.getSecond()[i] : stopRow;<a name="line.291"></a> +<span class="sourceLineNo">292</span><a name="line.292"></a> +<span class="sourceLineNo">293</span> HRegionLocation location = getRegionLocator().getRegionLocation(keys.getFirst()[i], false);<a name="line.293"></a> +<span class="sourceLineNo">294</span> // The below InetSocketAddress creation does a name resolution.<a name="line.294"></a> +<span class="sourceLineNo">295</span> InetSocketAddress isa = new InetSocketAddress(location.getHostname(), location.getPort());<a name="line.295"></a> +<span class="sourceLineNo">296</span> if (isa.isUnresolved()) {<a name="line.296"></a> +<span class="sourceLineNo">297</span> LOG.warn("Failed resolve " + isa);<a name="line.297"></a> +<span class="sourceLineNo">298</span> }<a name="line.298"></a> +<span class="sourceLineNo">299</span> InetAddress regionAddress = isa.getAddress();<a name="line.299"></a> +<span class="sourceLineNo">300</span> String regionLocation;<a name="line.300"></a> +<span class="sourceLineNo">301</span> regionLocation = reverseDNS(regionAddress);<a name="line.301"></a> +<span class="sourceLineNo">302</span><a name="line.302"></a> +<span class="sourceLineNo">303</span> byte[] regionName = location.getRegionInfo().getRegionName();<a name="line.303"></a> +<span class="sourceLineNo">304</span> String encodedRegionName = location.getRegionInfo().getEncodedName();<a name="line.304"></a> +<span class="sourceLineNo">305</span> long regionSize = sizeCalculator.getRegionSize(regionName);<a name="line.305"></a> +<span class="sourceLineNo">306</span> TableSplit split = new TableSplit(tableName, scan,<a name="line.306"></a> +<span class="sourceLineNo">307</span> splitStart, splitStop, regionLocation, encodedRegionName, regionSize);<a name="line.307"></a> +<span class="sourceLineNo">308</span> splits.add(split);<a name="line.308"></a> +<span class="sourceLineNo">309</span> if (LOG.isDebugEnabled()) {<a name="line.309"></a> +<span class="sourceLineNo">310</span> LOG.debug("getSplits: split -> " + i + " -> " + split);<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> //The default value of "hbase.mapreduce.input.autobalance" is false, which means not enabled.<a name="line.314"></a> +<span class="sourceLineNo">315</span> boolean enableAutoBalance = context.getConfiguration()<a name="line.315"></a> +<span class="sourceLineNo">316</span> .getBoolean(MAPREDUCE_INPUT_AUTOBALANCE, false);<a name="line.316"></a> +<span class="sourceLineNo">317</span> if (enableAutoBalance) {<a name="line.317"></a> +<span class="sourceLineNo">318</span> long totalRegionSize=0;<a name="line.318"></a> +<span class="sourceLineNo">319</span> for (int i = 0; i < splits.size(); i++){<a name="line.319"></a> +<span class="sourceLineNo">320</span> TableSplit ts = (TableSplit)splits.get(i);<a name="line.320"></a> +<span class="sourceLineNo">321</span> totalRegionSize += ts.getLength();<a name="line.321"></a> +<span class="sourceLineNo">322</span> }<a name="line.322"></a> +<span class="sourceLineNo">323</span> long averageRegionSize = totalRegionSize / splits.size();<a name="line.323"></a> +<span class="sourceLineNo">324</span> // the averageRegionSize must be positive.<a name="line.324"></a> +<span class="sourceLineNo">325</span> if (averageRegionSize <= 0) {<a name="line.325"></a> +<span class="sourceLineNo">326</span> LOG.warn("The averageRegionSize is not positive: "+ averageRegionSize + ", " +<a name="line.326"></a> +<span class="sourceLineNo">327</span> "set it to 1.");<a name="line.327"></a> +<span class="sourceLineNo">328</span> averageRegionSize = 1;<a name="line.328"></a> +<span class="sourceLineNo">329</span> }<a name="line.329"></a> +<span class="sourceLineNo">330</span> return calculateRebalancedSplits(splits, context, averageRegionSize);<a name="line.330"></a> +<span class="sourceLineNo">331</span> } else {<a name="line.331"></a> +<span class="sourceLineNo">332</span> return splits;<a name="line.332"></a> +<span class="sourceLineNo">333</span> }<a name="line.333"></a> +<span class="sourceLineNo">334</span> } finally {<a name="line.334"></a> +<span class="sourceLineNo">335</span> if (closeOnFinish) {<a name="line.335"></a> +<span class="sourceLineNo">336</span> closeTable();<a name="line.336"></a> +<span class="sourceLineNo">337</span> }<a name="line.337"></a> +<span class="sourceLineNo">338</span> }<a name="line.338"></a> +<span class="sourceLineNo">339</span> }<a name="line.339"></a> +<span class="sourceLineNo">340</span><a name="line.340"></a> +<span class="sourceLineNo">341</span> String reverseDNS(InetAddress ipAddress) throws UnknownHostException {<a name="line.341"></a> +<span class="sourceLineNo">342</span> String hostName = this.reverseDNSCacheMap.get(ipAddress);<a name="line.342"></a> +<span class="sourceLineNo">343</span> if (hostName == null) {<a name="line.343"></a> +<span class="sourceLineNo">344</span> String ipAddressString = null;<a name="line.344"></a> +<span class="sourceLineNo">345</span> try {<a name="line.345"></a> +<span class="sourceLineNo">346</span> ipAddressString = DNS.reverseDns(ipAddress, null);<a name="line.346"></a> +<span class="sourceLineNo">347</span> } catch (Exception e) {<a name="line.347"></a> +<span class="sourceLineNo">348</span> // We can use InetAddress in case the jndi failed to pull up the reverse DNS entry from the<a name="line.348"></a> +<span class="sourceLineNo">349</span> // name service. Also, in case of ipv6, we need to use the InetAddress since resolving<a name="line.349"></a> +<span class="sourceLineNo">350</span> // reverse DNS using jndi doesn't work well with ipv6 addresses.<a name="line.350"></a> +<span class="sourceLineNo">351</span> ipAddressString = InetAddress.getByName(ipAddress.getHostAddress()).getHostName();<a name="line.351"></a> +<span class="sourceLineNo">352</span> }<a name="line.352"></a> +<span class="sourceLineNo">353</span> if (ipAddressString == null) throw new UnknownHostException("No host found for " + ipAddress);<a name="line.353"></a> +<span class="sourceLineNo">354</span> hostName = Strings.domainNamePointerToHostName(ipAddressString);<a name="line.354"></a> +<span class="sourceLineNo">355</span> this.reverseDNSCacheMap.put(ipAddress, hostName);<a name="line.355"></a> +<span class="sourceLineNo">356</span> }<a name="line.356"></a> +<span class="sourceLineNo">357</span> return hostName;<a name="line.357"></a> +<span class="sourceLineNo">358</span> }<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> * Calculates the number of MapReduce input splits for the map tasks. The number of<a name="line.361"></a> +<span class="sourceLineNo">362</span> * MapReduce input splits depends on the average region size and the "data skew ratio" user set in<a name="line.362"></a> +<span class="sourceLineNo">363</span> * configuration.<a name="line.363"></a> +<span class="sourceLineNo">364</span> *<a name="line.364"></a> +<span class="sourceLineNo">365</span> * @param list The list of input splits before balance.<a name="line.365"></a> +<span class="sourceLineNo">366</span> * @param context The current job context.<a name="line.366"></a> +<span class="sourceLineNo">367</span> * @param average The average size of all regions .<a name="line.367"></a> +<span class="sourceLineNo">368</span> * @return The list of input splits.<a name="line.368"></a> +<span class="sourceLineNo">369</span> * @throws IOException When creating the list of splits fails.<a name="line.369"></a> +<span class="sourceLineNo">370</span> * @see org.apache.hadoop.mapreduce.InputFormat#getSplits(<a name="line.370"></a> +<span class="sourceLineNo">371</span> * org.apache.hadoop.mapreduce.JobContext)<a name="line.371"></a> +<span class="sourceLineNo">372</span> */<a name="line.372"></a> +<span class="sourceLineNo">373</span> private List<InputSplit> calculateRebalancedSplits(List<InputSplit> list, JobContext context,<a name="line.373"></a> +<span class="sourceLineNo">374</span> long average) throws IOException {<a name="line.374"></a> +<span class="sourceLineNo">375</span> List<InputSplit> resultList = new ArrayList<>();<a name="line.375"></a> +<span class="sourceLineNo">376</span> Configuration conf = context.getConfiguration();<a name="line.376"></a> +<span class="sourceLineNo">377</span> //The default data skew ratio is 3<a name="line.377"></a> +<span class="sourceLineNo">378</span> long dataSkewRatio = conf.getLong(INPUT_AUTOBALANCE_MAXSKEWRATIO, 3);<a name="line.378"></a> +<span class="sourceLineNo">379</span> //It determines which mode to use: text key mode or binary key mode. The default is text mode.<a name="line.379"></a> +<span class="sourceLineNo">380</span> boolean isTextKey = context.getConfiguration().getBoolean(TABLE_ROW_TEXTKEY, true);<a name="line.380"></a> +<span class="sourceLineNo">381</span> long dataSkewThreshold = dataSkewRatio * average;<a name="line.381"></a> +<span class="sourceLineNo">382</span> int count = 0;<a name="line.382"></a> +<span class="sourceLineNo">383</span> while (count < list.size()) {<a name="line.383"></a> +<span class="sourceLineNo">384</span> TableSplit ts = (TableSplit)list.get(count);<a name="line.384"></a> +<span class="sourceLineNo">385</span> TableName tableName = ts.getTable();<a name="line.385"></a> +<span class="sourceLineNo">386</span> String regionLocation = ts.getRegionLocation();<a name="line.386"></a> +<span class="sourceLineNo">387</span> String encodedRegionName = ts.getEncodedRegionName();<a name="line.387"></a> +<span class="sourceLineNo">388</span> long regionSize = ts.getLength();<a name="line.388"></a> +<span class="sourceLineNo">389</span> if (regionSize >= dataSkewThreshold) {<a name="line.389"></a> +<span class="sourceLineNo">390</span> // if the current region size is large than the data skew threshold,<a name="line.390"></a> +<span class="sourceLineNo">391</span> // split the region into two MapReduce input splits.<a name="line.391"></a> +<span class="sourceLineNo">392</span> byte[] splitKey = getSplitKey(ts.getStartRow(), ts.getEndRow(), isTextKey);<a name="line.392"></a> +<span class="sourceLineNo">393</span> if (Arrays.equals(ts.getEndRow(), splitKey)) {<a name="line.393"></a> +<span class="sourceLineNo">394</span> // Not splitting since the end key is the same as the split key<a name="line.394"></a> +<span class="sourceLineNo">395</span> resultList.add(ts);<a name="line.395"></a> +<span class="sourceLineNo">396</span> } else {<a name="line.396"></a> +<span class="sourceLineNo">397</span> //Set the size of child TableSplit as 1/2 of the region size. The exact size of the<a name="line.397"></a> +<span class="sourceLineNo">398</span> // MapReduce input splits is not far off.<a name="line.398"></a> +<span class="sourceLineNo">399</span> TableSplit t1 = new TableSplit(tableName, scan, ts.getStartRow(), splitKey,<a name="line.399"></a> +<span class="sourceLineNo">400</span> regionLocation, regionSize / 2);<a name="line.400"></a> +<span class="sourceLineNo">401</span> TableSplit t2 = new TableSplit(tableName, scan, splitKey, ts.getEndRow(), regionLocation,<a name="line.401"></a> +<span class="sourceLineNo">402</span> regionSize - regionSize / 2);<a name="line.402"></a> +<span class="sourceLineNo">403</span> resultList.add(t1);<a name="line.403"></a> +<span class="sourceLineNo">404</span> resultList.add(t2);<a name="line.404"></a> +<span class="sourceLineNo">405</span> }<a name="line.405"></a> +<span class="sourceLineNo">406</span> count++;<a name="line.406"></a> +<span class="sourceLineNo">407</span> } else if (regionSize >= average) {<a name="line.407"></a> +<span class="sourceLineNo">408</span> // if the region size between average size and data skew threshold size,<a name="line.408"></a> +<span class="sourceLineNo">409</span> // make this region as one MapReduce input split.<a name="line.409"></a> +<span class="sourceLineNo">410</span> resultList.add(ts);<a name="line.410"></a> +<span class="sourceLineNo">411</span> count++;<a name="line.411"></a> +<span class="sourceLineNo">412</span> } else {<a name="line.412"></a> +<span class="sourceLineNo">413</span> // if the total size of several small continuous regions less than the average region size,<a name="line.413"></a> +<span class="sourceLineNo">414</span> // combine them into one MapReduce input split.<a name="line.414"></a> +<span class="sourceLineNo">415</span> long totalSize = regionSize;<a name="line.415"></a> +<span class="sourceLineNo">416</span> byte[] splitStartKey = ts.getStartRow();<a name="line.416"></a> +<span class="sourceLineNo">417</span> byte[] splitEndKey = ts.getEndRow();<a name="line.417"></a> +<span class="sourceLineNo">418</span> count++;<a name="line.418"></a> +<span class="sourceLineNo">419</span> for (; count < list.size(); count++) {<a name="line.419"></a> +<span class="sourceLineNo">420</span> TableSplit nextRegion = (TableSplit)list.get(count);<a name="line.420"></a> +<span class="sourceLineNo">421</span> long nextRegionSize = nextRegion.getLength();<a name="line.421"></a> +<span class="sourceLineNo">422</span> if (totalSize + nextRegionSize <= dataSkewThreshold) {<a name="line.422"></a> +<span class="sourceLineNo">423</span> totalSize = totalSize + nextRegionSize;<a name="line.423"></a> +<span class="sourceLineNo">424</span> splitEndKey = nextRegion.getEndRow();<a name="line.424"></a> +<span class="sourceLineNo">425</span> } else {<a name="line.425"></a> +<span class="sourceLineNo">426</span> break;<a name="line.426"></a> +<span class="sourceLineNo">427</span> }<a name="line.427"></a> +<span class="sourceLineNo">428</span> }<a name="line.428"></a> +<span class="sourceLineNo">429</span> TableSplit t = new TableSplit(tableName, scan, splitStartKey, splitEndKey,<a name="line.429"></a> +<span class="sourceLineNo">430</span> regionLocation, encodedRegionName, totalSize);<a name="line.430"></a> +<span class="sourceLineNo">431</span> resultList.add(t);<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> return resultList;<a name="line.434"></a> +<span class="sourceLineNo">435</span> }<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> * select a split point in the region. The selection of the split point is based on an uniform<a name="line.438"></a> +<span class="sourceLineNo">439</span> * distribution assumption for the keys in a region.<a name="line.439"></a> +<span class="sourceLineNo">440</span> * Here are some examples:<a name="line.440"></a> +<span class="sourceLineNo">441</span> *<a name="line.441"></a> +<span class="sourceLineNo">442</span> * <table><a name="line.442"></a> +<span class="sourceLineNo">443</span> * <tr><a name="line.443"></a> +<span class="sourceLineNo">444</span> * <th>start key</th><a name="line.444"></a> +<span class="sourceLineNo">445</span> * <th>end key</th><a name="line.445"></a> +<span class="sourceLineNo">446</span> * <th>is text</th><a name="line.446"></a> +<span class="sourceLineNo">447</span> * <th>split point</th><a name="line.447"></a> +<span class="sourceLineNo">448</span> * </tr><a name="line.448"></a> +<span class="sourceLineNo">449</span> * <tr><a name="line.449"></a> +<span class="sourceLineNo">450</span> * <td>'a', 'a', 'a', 'b', 'c', 'd', 'e', 'f', 'g'</td><a name="line.450"></a> +<span class="sourceLineNo">451</span> * <td>'a', 'a', 'a', 'f', 'f', 'f'</td><a name="line.451"></a> +<span class="sourceLineNo">452</span> * <td>true</td><a name="line.452"></a> +<span class="sourceLineNo">453</span> * <td>'a', 'a', 'a', 'd', 'd', -78, 50, -77, 51</td><a name="line.453"></a> +<span class="sourceLineNo">454</span> * </tr><a name="line.454"></a> +<span class="sourceLineNo">455</span> * <tr><a name="line.455"></a> +<span class="sourceLineNo">456</span> * <td>'1', '1', '1', '0', '0', '0'</td><a name="line.456"></a> +<span class="sourceLineNo">457</span> * <td>'1', '1', '2', '5', '7', '9', '0'</td><a name="line.457"></a> +<span class="sourceLineNo">458</span> * <td>true</td><a name="line.458"></a> +<span class="sourceLineNo">459</span> * <td>'1', '1', '1', -78, -77, -76, -104</td><a name="line.459"></a> +<span class="sourceLineNo">460</span> * </tr><a name="line.460"></a> +<span class="sourceLineNo">461</span> * <tr><a name="line.461"></a> +<span class="sourceLineNo">462</span> * <td>'1', '1', '1', '0'</td><a name="line.462"></a> +<span class="sourceLineNo">463</span> * <td>'1', '1', '2', '0'</td><a name="line.463"></a> +<span class="sourceLineNo">464</span> * <td>true</td><a name="line.464"></a> +<span class="sourceLineNo">465</span> * <td>'1', '1', '1', -80</td><a name="line.465"></a> +<span class="sourceLineNo">466</span> * </tr><a name="line.466"></a> +<span class="sourceLineNo">467</span> * <tr><a name="line.467"></a> +<span class="sourceLineNo">468</span> * <td>13, -19, 126, 127</td><a name="line.468"></a> +<span class="sourceLineNo">469</span> * <td>13, -19, 127, 0</td><a name="line.469"></a> +<span class="sourceLineNo">470</span> * <td>false</td><a name="line.470"></a> +<span class="sourceLineNo">471</span> * <td>13, -19, 126, -65</td><a name="line.471"></a> +<span class="sourceLineNo">472</span> * </tr><a name="line.472"></a> +<span class="sourceLineNo">473</span> * </table><a name="line.473"></a> +<span class="sourceLineNo">474</span> *<a name="line.474"></a> +<span class="sourceLineNo">475</span> * Set this function as "public static", make it easier for test.<a name="line.475"></a> +<span class="sourceLineNo">476</span> *<a name="line.476"></a> +<span class="sourceLineNo">477</span> * @param start Start key of the region<a name="line.477"></a> +<span class="sourceLineNo">478</span> * @param end End key of the region<a name="line.478"></a> +<span class="sourceLineNo">479</span> * @param isText It determines to use text key mode or binary key mode<a name="line.479"></a> +<span class="sourceLineNo">480</span> * @return The split point in the region.<a name="line.480"></a> +<span class="sourceLineNo">481</span> */<a name="line.481"></a> +<span class="sourceLineNo">482</span> @InterfaceAudience.Private<a name="line.482"></a> +<span class="sourceLineNo">483</span> public static byte[] getSplitKey(byte[] start, byte[] end, boolean isText) {<a name="line.483"></a> +<span class="sourceLineNo">484</span> byte upperLimitByte;<a name="line.484"></a> +<span class="sourceLineNo">485</span> byte lowerLimitByte;<a name="line.485"></a> +<span class="sourceLineNo">486</span> //Use text mode or binary mode.<a name="line.486"></a> +<span class="sourceLineNo">487</span> if (isText) {<a name="line.487"></a> +<span class="sourceLineNo">488</span> //The range of text char set in ASCII is [32,126], the lower limit is space and the upper<a name="line.488"></a> +<span class="sourceLineNo">489</span> // limit is '~'.<a name="line.489"></a> +<span class="sourceLineNo">490</span> upperLimitByte = '~';<a name="line.490"></a> +<span class="sourceLineNo">491</span> lowerLimitByte = ' ';<a name="line.491"></a> +<span class="sourceLineNo">492</span> } else {<a name="line.492"></a> +<span class="sourceLineNo">493</span> upperLimitByte = -1;<a name="line.493"></a> +<span class="sourceLineNo">494</span> lowerLimitByte = 0;<a name="line.494"></a> +<span class="sourceLineNo">495</span> }<a name="line.495"></a> +<span class="sourceLineNo">496</span> // For special case<a name="line.496"></a> +<span class="sourceLineNo">497</span> // Example 1 : startkey=null, endkey="hhhqqqwww", splitKey="h"<a name="line.497"></a> +<span class="sourceLineNo">498</span> // Example 2 (text key mode): startKey="ffffaaa", endKey=null, splitkey="f~~~~~~"<a name="line.498"></a> +<span class="sourceLineNo">499</span> if (start.length == 0 && end.length == 0){<a name="line.499"></a> +<span class="sourceLineNo">500</span> return new byte[]{(byte) ((lowerLimitByte + upperLimitByte) / 2)};<a name="line.500"></a> +<span class="sourceLineNo">501</span> }<a name="line.501"></a> +<span class="sourceLineNo">502</span> if (start.length == 0 && end.length != 0){<a name="line.502"></a> +<span class="sourceLineNo">503</span> return new byte[]{ end[0] };<a name="line.503"></a> +<span class="sourceLineNo">504</span> }<a name="line.504"></a> +<span class="sourceLineNo">505</span> if (start.length != 0 && end.length == 0){<a name="line.505"></a> +<span class="sourceLineNo">506</span> byte[] result =new byte[start.length];<a name="line.506"></a> +<span class="sourceLineNo">507</span> result[0]=start[0];<a name="line.507"></a> +<span class="sourceLineNo">508</span> for (int k = 1; k < start.length; k++){<a name="line.508"></a> +<span class="sourceLineNo">509</span> result[k] = upperLimitByte;<a name="line.509"></a> +<span class="sourceLineNo">510</span> }<a name="line.510"></a> +<span class="sourceLineNo">511</span> return result;<a name="line.511"></a> +<span class="sourceLineNo">512</span> }<a name="line.512"></a> +<span class="sourceLineNo">513</span> return Bytes.split(start, end, false, 1)[1];<a name="line.513"></a> +<span class="sourceLineNo">514</span> }<a name="line.514"></a> +<span class="sourceLineNo">515</span><a name="line.515"></a> +<span class="sourceLineNo">516</span> /**<a name="line.516"></a> +<span class="sourceLineNo">517</span> * Test if the given region is to be included in the InputSplit while splitting<a name="line.517"></a> +<span class="sourceLineNo">518</span> * the regions of a table.<a name="line.518"></a> +<span class="sourceLineNo">519</span> * <p><a name="line.519"></a> +<span class="sourceLineNo">520</span> * This optimization is effective when there is a specific reasoning to exclude an entire region from the M-R job,<a name="line.520"></a> +<span class="sourceLineNo">521</span> * (and hence, not contributing to the InputSplit), given the start and end keys of the same. <br><a name="line.521"></a> +<span class="sourceLineNo">522</span> * Useful when we need to remember the last-processed top record and revisit the [last, current) interval for M-R processing,<a name="line.522"></a> +<span class="sourceLineNo">523</span> * continuously. In addition to reducing InputSplits, reduces the load on the region server as well, due to the ordering of the keys.<a name="line.523"></a> +<span class="sourceLineNo">524</span> * <br><a name="line.524"></a> <span class="sourceLineNo">525</span> * <br><a name="line.525"></a> -<span class="sourceLineNo">526</span> * <br><a name="line.526"></a> -<span class="sourceLineNo">527</span> * Note: It is possible that <code>endKey.length() == 0 </code> , for the last (recent) region.<a name="line.527"></a> -<span class="sourceLineNo">528</span> * <br><a name="line.528"></a> -<span class="sourceLineNo">529</span> * Override this method, if you want to bulk exclude regions altogether from M-R. By default, no region is excluded( i.e. all regions are included).<a name="line.529"></a> +<span class="sourceLineNo">526</span> * Note: It is possible that <code>endKey.length() == 0 </code> , for the last (recent) region.<a name="line.526"></a> +<span class="sourceLineNo">527</span> * <br><a name="line.527"></a> +<span class="sourceLineNo">528</span> * Override this method, if you want to bulk exclude regions altogether from M-R. By default, no region is excluded( i.e. all regions are included).<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> *<a name="line.531"></a> -<span class="sourceLineNo">532</span> * @param startKey Start key of the region<a name="line.532"></a> -<span class="sourceLineNo">533</span> * @param endKey End key of the region<a name="line.533"></a> -<span class="sourceLineNo">534</span> * @return true, if this region needs to be included as part of the input (default).<a name="line.534"></a> -<span class="sourceLineNo">535</span> *<a name="line.535"></a> -<span class="sourceLineNo">536</span> */<a name="line.536"></a> -<span class="sourceLineNo">537</span> protected boolean includeRegionInSplit(final byte[] startKey, final byte [] endKey) {<a name="line.537"></a> -<span class="sourceLineNo">538</span> return true;<a name="line.538"></a> -<span class="sourceLineNo">539</span> }<a name="line.539"></a> -<span class="sourceLineNo">540</span><a name="line.540"></a> -<span class="sourceLineNo">541</span> /**<a name="line.541"></a> -<span class="sourceLineNo">542</span> * Allows subclasses to get the {@link RegionLocator}.<a name="line.542"></a> -<span class="sourceLineNo">543</span> */<a name="line.543"></a> -<span class="sourceLineNo">544</span> protected RegionLocator getRegionLocator() {<a name="line.544"></a> -<span class="sourceLineNo">545</span> if (regionLocator == null) {<a name="line.545"></a> -<span class="sourceLineNo">546</span> throw new IllegalStateException(NOT_INITIALIZED);<a name="line.546"></a> -<span class="sourceLineNo">547</span> }<a name="line.547"></a> -<span class="sourceLineNo">548</span> return regionLocator;<a name="line.548"></a> -<span class="sourceLineNo">549</span> }<a name="line.549"></a> -<span class="sourceLineNo">550</span> <a name="line.550"></a> -<span class="sourceLineNo">551</span> /**<a name="line.551"></a> -<span class="sourceLineNo">552</span> * Allows subclasses to get the {@link Table}.<a name="line.552"></a> -<span class="sourceLineNo">553</span> */<a name="line.553"></a> -<span class="sourceLineNo">554</span> protected Table getTable() {<a name="line.554"></a> -<span class="sourceLineNo">555</span> if (table == null) {<a name="line.555"></a> -<span class="sourceLineNo">556</span> throw new IllegalStateException(NOT_INITIALIZED);<a name="line.556"></a> -<span class="sourceLineNo">557</span> }<a name="line.557"></a> -<span class="sourceLineNo">558</span> return table;<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> * Allows subclasses to get the {@link Admin}.<a name="line.562"></a> -<span class="sourceLineNo">563</span> */<a name="line.563"></a> -<span class="sourceLineNo">564</span> protected Admin getAdmin() {<a name="line.564"></a> -<span class="sourceLineNo">565</span> if (admin == null) {<a name="line.565"></a> -<span class="sourceLineNo">566</span> throw new IllegalStateException(NOT_INITIALIZED);<a name="line.566"></a> -<span class="sourceLineNo">567</span> }<a name="line.567"></a> -<span class="sourceLineNo">568</span> return admin;<a name="line.568"></a> -<span class="sourceLineNo">569</span> }<a name="line.569"></a> -<span class="sourceLineNo">570</span><a name="line.570"></a> -<span class="sourceLineNo">571</span> /**<a name="line.571"></a> -<span class="sourceLineNo">572</span> * Allows subclasses to initialize the table information.<a name="line.572"></a> -<span class="sourceLineNo">573</span> *<a name="line.573"></a> -<span class="sourceLineNo">574</span> * @param connection The Connection to the HBase cluster. MUST be unmanaged. We will close.<a name="line.574"></a> -<span class="sourceLineNo">575</span> * @param tableName The {@link TableName} of the table to process. <a name="line.575"></a> -<span class="sourceLineNo">576</span> * @throws IOException <a name="line.576"></a> -<span class="sourceLineNo">577</span> */<a name="line.577"></a> -<span class="sourceLineNo">578</span> protected void initializeTable(Connection connection, TableName tableName) throws IOException {<a name="line.578"></a> -<span class="sourceLineNo">579</span> if (this.table != null || this.connection != null) {<a name="line.579"></a> -<span class="sourceLineNo">580</span> LOG.warn("initializeTable called multiple times. Overwriting connection and table " +<a name="line.580"></a> -<span class="sourceLineNo">581</span> "reference; TableInputFormatBase will not close these old references when done.");<a name="line.581"></a> -<span class="sourceLineNo">582</span> }<a name="line.582"></a> -<span class="sourceLineNo">583</span> this.table = connection.getTable(tableName);<a name="line.583"></a> -<span class="sourceLineNo">584</span> this.regionLocator = connection.getRegionLocator(tableName);<a name="line.584"></a> -<span class="sourceLineNo">585</span> this.admin = connection.getAdmin();<a name="line.585"></a> -<span class="sourceLineNo">586</span> this.connection = connection;<a name="line.586"></a> -<span class="sourceLineNo">587</span> }<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> * Gets the scan defining the actual details like columns etc.<a name="line.590"></a> -<span class="sourceLineNo">591</span> *<a name="line.591"></a> -<span class="sourceLineNo">592</span> * @return The internal scan instance.<a name="line.592"></a> -<span class="sourceLineNo">593</span> */<a name="line.593"></a> -<span class="sourceLineNo">594</span> public Scan getScan() {<a name="line.594"></a> -<span class="sourceLineNo">595</span> if (this.scan == null) this.scan = new Scan();<a name="line.595"></a> -<span class="sourceLineNo">596</span> return scan;<a name="line.596"></a> -<span class="sourceLineNo">597</span> }<a name="line.597"></a> -<span class="sourceLineNo">598</span><a name="line.598"></a> -<span class="sourceLineNo">599</span> /**<a name="line.599"></a> -<span class="sourceLineNo">600</span> * Sets the scan defining the actual details like columns etc.<a name="line.600"></a> -<span class="sourceLineNo">601</span> *<a name="line.601"></a> -<span class="sourceLineNo">602</span> * @param scan The scan to set.<a name="line.602"></a> -<span class="sourceLineNo">603</span> */<a name="line.603"></a> -<span class="sourceLineNo">604</span> public void setScan(Scan scan) {<a name="line.604"></a> -<span class="sourceLineNo">605</span> this.scan = scan;<a name="line.605"></a> -<span class="sourceLineNo">606</span> }<a name="line.606"></a> -<span class="sourceLineNo">607</span><a name="line.607"></a> -<span class="sourceLineNo">608</span> /**<a name="line.608"></a> -<span class="sourceLineNo">609</span> * Allows subclasses to set the {@link TableRecordReader}.<a name="line.609"></a> -<span class="sourceLineNo">610</span> *<a name="line.610"></a> -<span class="sourceLineNo">611</span> * @param tableRecordReader A different {@link TableRecordReader}<a name="line.611"></a> -<span class="sourceLineNo">612</span> * implementation.<a name="line.612"></a> -<span class="sourceLineNo">613</span> */<a name="line.613"></a> -<span class="sourceLineNo">614</span> protected void setTableRecordReader(TableRecordReader tableRecordReader) {<a name="line.614"></a> -<span class="sourceLineNo">615</span> this.tableRecordReader = tableRecordReader;<a name="line.615"></a> -<span class="sourceLineNo">616</span> }<a name="line.616"></a> -<span class="sourceLineNo">617</span> <a name="line.617"></a> -<span class="sourceLineNo">618</span> /**<a name="line.618"></a> -<span class="sourceLineNo">619</span> * Handle subclass specific set up.<a name="line.619"></a> -<span class="sourceLineNo">620</span> * Each of the entry points used by the MapReduce framework,<a name="line.620"></a> -<span class="sourceLineNo">621</span> * {@link #createRecordReader(InputSplit, TaskAttemptContext)} and {@link #getSplits(JobContext)},<a name="line.621"></a> -<span class="sourceLineNo">622</span> * will call {@link #initialize(JobContext)} as a convenient centralized location
<TRUNCATED>