/**
* Gets the splits of the tables that have been set on the job by reading the metadata table for
* the specified ranges.
*
* @return the splits from the tables based on the ranges.
* @throws java.io.IOException if a table set on the job doesn't exist or an error occurs
* initializing the tablet locator
*/
@Override
public List<InputSplit> getSplits(JobContext context) throws IOException {
Level logLevel = getLogLevel(context);
log.setLevel(logLevel);
validateOptions(context);
Random random = new Random();
LinkedList<InputSplit> splits = new LinkedList<InputSplit>();
Map<String, InputTableConfig> tableConfigs = getInputTableConfigs(context);
for (Map.Entry<String, InputTableConfig> tableConfigEntry : tableConfigs.entrySet()) {
String tableName = tableConfigEntry.getKey();
InputTableConfig tableConfig = tableConfigEntry.getValue();
Instance instance = getInstance(context);
String tableId;
// resolve table name to id once, and use id from this point forward
if (DeprecationUtil.isMockInstance(instance)) {
tableId = "";
} else {
try {
tableId = Tables.getTableId(instance, tableName);
} catch (TableNotFoundException e) {
throw new IOException(e);
}
}
Authorizations auths = getScanAuthorizations(context);
String principal = getPrincipal(context);
AuthenticationToken token = getAuthenticationToken(context);
boolean batchScan = InputConfigurator.isBatchScan(CLASS, context.getConfiguration());
boolean supportBatchScan =
!(tableConfig.isOfflineScan()
|| tableConfig.shouldUseIsolatedScanners()
|| tableConfig.shouldUseLocalIterators());
if (batchScan && !supportBatchScan)
throw new IllegalArgumentException(
"BatchScanner optimization not available for offline scan, isolated, or local iterators");
boolean autoAdjust = tableConfig.shouldAutoAdjustRanges();
if (batchScan && !autoAdjust)
throw new IllegalArgumentException(
"AutoAdjustRanges must be enabled when using BatchScanner optimization");
List<Range> ranges =
autoAdjust ? Range.mergeOverlapping(tableConfig.getRanges()) : tableConfig.getRanges();
if (ranges.isEmpty()) {
ranges = new ArrayList<Range>(1);
ranges.add(new Range());
}
// get the metadata information for these ranges
Map<String, Map<KeyExtent, List<Range>>> binnedRanges =
new HashMap<String, Map<KeyExtent, List<Range>>>();
TabletLocator tl;
try {
if (tableConfig.isOfflineScan()) {
binnedRanges = binOfflineTable(context, tableId, ranges);
while (binnedRanges == null) {
// Some tablets were still online, try again
// sleep randomly between 100 and 200 ms
sleepUninterruptibly(100 + random.nextInt(100), TimeUnit.MILLISECONDS);
binnedRanges = binOfflineTable(context, tableId, ranges);
}
} else {
tl = InputConfigurator.getTabletLocator(CLASS, context.getConfiguration(), tableId);
// its possible that the cache could contain complete, but old information about a tables
// tablets... so clear it
tl.invalidateCache();
ClientContext clientContext =
new ClientContext(
getInstance(context),
new Credentials(getPrincipal(context), getAuthenticationToken(context)),
getClientConfiguration(context));
while (!tl.binRanges(clientContext, ranges, binnedRanges).isEmpty()) {
if (!DeprecationUtil.isMockInstance(instance)) {
if (!Tables.exists(instance, tableId)) throw new TableDeletedException(tableId);
if (Tables.getTableState(instance, tableId) == TableState.OFFLINE)
throw new TableOfflineException(instance, tableId);
}
binnedRanges.clear();
log.warn("Unable to locate bins for specified ranges. Retrying.");
// sleep randomly between 100 and 200 ms
sleepUninterruptibly(100 + random.nextInt(100), TimeUnit.MILLISECONDS);
tl.invalidateCache();
}
}
} catch (Exception e) {
throw new IOException(e);
}
// all of this code will add either range per each locations or split ranges and add
// range-location split
// Map from Range to Array of Locations, we only use this if we're don't split
HashMap<Range, ArrayList<String>> splitsToAdd = null;
if (!autoAdjust) splitsToAdd = new HashMap<Range, ArrayList<String>>();
HashMap<String, String> hostNameCache = new HashMap<String, String>();
for (Map.Entry<String, Map<KeyExtent, List<Range>>> tserverBin : binnedRanges.entrySet()) {
String ip = tserverBin.getKey().split(":", 2)[0];
String location = hostNameCache.get(ip);
if (location == null) {
InetAddress inetAddress = InetAddress.getByName(ip);
location = inetAddress.getCanonicalHostName();
hostNameCache.put(ip, location);
}
for (Map.Entry<KeyExtent, List<Range>> extentRanges : tserverBin.getValue().entrySet()) {
Range ke = extentRanges.getKey().toDataRange();
if (batchScan) {
// group ranges by tablet to be read by a BatchScanner
ArrayList<Range> clippedRanges = new ArrayList<Range>();
for (Range r : extentRanges.getValue()) clippedRanges.add(ke.clip(r));
BatchInputSplit split =
new BatchInputSplit(tableName, tableId, clippedRanges, new String[] {location});
SplitUtils.updateSplit(split, instance, tableConfig, principal, token, auths, logLevel);
splits.add(split);
} else {
// not grouping by tablet
for (Range r : extentRanges.getValue()) {
if (autoAdjust) {
// divide ranges into smaller ranges, based on the tablets
RangeInputSplit split =
new RangeInputSplit(tableName, tableId, ke.clip(r), new String[] {location});
SplitUtils.updateSplit(
split, instance, tableConfig, principal, token, auths, logLevel);
split.setOffline(tableConfig.isOfflineScan());
split.setIsolatedScan(tableConfig.shouldUseIsolatedScanners());
split.setUsesLocalIterators(tableConfig.shouldUseLocalIterators());
splits.add(split);
} else {
// don't divide ranges
ArrayList<String> locations = splitsToAdd.get(r);
if (locations == null) locations = new ArrayList<String>(1);
locations.add(location);
splitsToAdd.put(r, locations);
}
}
}
}
}
if (!autoAdjust)
for (Map.Entry<Range, ArrayList<String>> entry : splitsToAdd.entrySet()) {
RangeInputSplit split =
new RangeInputSplit(
tableName, tableId, entry.getKey(), entry.getValue().toArray(new String[0]));
SplitUtils.updateSplit(split, instance, tableConfig, principal, token, auths, logLevel);
split.setOffline(tableConfig.isOfflineScan());
split.setIsolatedScan(tableConfig.shouldUseIsolatedScanners());
split.setUsesLocalIterators(tableConfig.shouldUseLocalIterators());
splits.add(split);
}
}
return splits;
}
@Override
public void initialize(InputSplit inSplit, TaskAttemptContext attempt) throws IOException {
split = (RangeInputSplit) inSplit;
log.debug("Initializing input split: " + split.toString());
Instance instance = split.getInstance(getClientConfiguration(attempt));
if (null == instance) {
instance = getInstance(attempt);
}
String principal = split.getPrincipal();
if (null == principal) {
principal = getPrincipal(attempt);
}
AuthenticationToken token = split.getToken();
if (null == token) {
token = getAuthenticationToken(attempt);
}
Authorizations authorizations = split.getAuths();
if (null == authorizations) {
authorizations = getScanAuthorizations(attempt);
}
String classLoaderContext = getClassLoaderContext(attempt);
String table = split.getTableName();
// in case the table name changed, we can still use the previous name for terms of
// configuration,
// but the scanner will use the table id resolved at job setup time
InputTableConfig tableConfig = getInputTableConfig(attempt, split.getTableName());
log.debug("Creating connector with user: "******"Creating scanner for table: " + table);
log.debug("Authorizations are: " + authorizations);
if (split instanceof BatchInputSplit) {
BatchInputSplit batchSplit = (BatchInputSplit) split;
BatchScanner scanner;
try {
// Note: BatchScanner will use at most one thread per tablet, currently BatchInputSplit
// will not span tablets
int scanThreads = 1;
scanner =
instance
.getConnector(principal, token)
.createBatchScanner(split.getTableName(), authorizations, scanThreads);
setupIterators(attempt, scanner, split.getTableName(), split);
if (null != classLoaderContext) {
scanner.setClassLoaderContext(classLoaderContext);
}
} catch (Exception e) {
e.printStackTrace();
throw new IOException(e);
}
scanner.setRanges(batchSplit.getRanges());
scannerBase = scanner;
} else {
Scanner scanner;
Boolean isOffline = split.isOffline();
if (null == isOffline) {
isOffline = tableConfig.isOfflineScan();
}
Boolean isIsolated = split.isIsolatedScan();
if (null == isIsolated) {
isIsolated = tableConfig.shouldUseIsolatedScanners();
}
Boolean usesLocalIterators = split.usesLocalIterators();
if (null == usesLocalIterators) {
usesLocalIterators = tableConfig.shouldUseLocalIterators();
}
try {
if (isOffline) {
scanner =
new OfflineScanner(
instance,
new Credentials(principal, token),
split.getTableId(),
authorizations);
} else if (DeprecationUtil.isMockInstance(instance)) {
scanner =
instance
.getConnector(principal, token)
.createScanner(split.getTableName(), authorizations);
} else {
ClientConfiguration clientConf = getClientConfiguration(attempt);
ClientContext context =
new ClientContext(instance, new Credentials(principal, token), clientConf);
scanner = new ScannerImpl(context, split.getTableId(), authorizations);
}
if (isIsolated) {
log.info("Creating isolated scanner");
scanner = new IsolatedScanner(scanner);
}
if (usesLocalIterators) {
log.info("Using local iterators");
scanner = new ClientSideIteratorScanner(scanner);
}
setupIterators(attempt, scanner, split.getTableName(), split);
} catch (Exception e) {
throw new IOException(e);
}
scanner.setRange(split.getRange());
scannerBase = scanner;
}
Collection<Pair<Text, Text>> columns = split.getFetchedColumns();
if (null == columns) {
columns = tableConfig.getFetchedColumns();
}
// setup a scanner within the bounds of this split
for (Pair<Text, Text> c : columns) {
if (c.getSecond() != null) {
log.debug("Fetching column " + c.getFirst() + ":" + c.getSecond());
scannerBase.fetchColumn(c.getFirst(), c.getSecond());
} else {
log.debug("Fetching column family " + c.getFirst());
scannerBase.fetchColumnFamily(c.getFirst());
}
}
SamplerConfiguration samplerConfig = split.getSamplerConfiguration();
if (null == samplerConfig) {
samplerConfig = tableConfig.getSamplerConfiguration();
}
if (samplerConfig != null) {
scannerBase.setSamplerConfiguration(samplerConfig);
}
scannerIterator = scannerBase.iterator();
numKeysRead = 0;
}