@Override
public Result[] call(int timeout) throws IOException {
// If the active replica callable was closed somewhere, invoke the RPC to
// really close it. In the case of regular scanners, this applies. We make couple
// of RPCs to a RegionServer, and when that region is exhausted, we set
// the closed flag. Then an RPC is required to actually close the scanner.
if (currentScannerCallable != null && currentScannerCallable.closed) {
// For closing we target that exact scanner (and not do replica fallback like in
// the case of normal reads)
if (LOG.isDebugEnabled()) {
LOG.debug("Closing scanner " + currentScannerCallable.scannerId);
}
Result[] r = currentScannerCallable.call(timeout);
currentScannerCallable = null;
return r;
}
// We need to do the following:
// 1. When a scan goes out to a certain replica (default or not), we need to
// continue to hit that until there is a failure. So store the last successfully invoked
// replica
// 2. We should close the "losing" scanners (scanners other than the ones we hear back
// from first)
//
RegionLocations rl =
RpcRetryingCallerWithReadReplicas.getRegionLocations(
true,
RegionReplicaUtil.DEFAULT_REPLICA_ID,
cConnection,
tableName,
currentScannerCallable.getRow());
// allocate a boundedcompletion pool of some multiple of number of replicas.
// We want to accomodate some RPCs for redundant replica scans (but are still in progress)
BoundedCompletionService<Pair<Result[], ScannerCallable>> cs =
new BoundedCompletionService<Pair<Result[], ScannerCallable>>(pool, rl.size() * 5);
List<ExecutionException> exceptions = null;
int submitted = 0, completed = 0;
AtomicBoolean done = new AtomicBoolean(false);
replicaSwitched.set(false);
// submit call for the primary replica.
submitted += addCallsForCurrentReplica(cs, rl);
try {
// wait for the timeout to see whether the primary responds back
Future<Pair<Result[], ScannerCallable>> f =
cs.poll(timeBeforeReplicas, TimeUnit.MICROSECONDS); // Yes, microseconds
if (f != null) {
Pair<Result[], ScannerCallable> r = f.get();
if (r != null && r.getSecond() != null) {
updateCurrentlyServingReplica(r.getSecond(), r.getFirst(), done, pool);
}
return r == null ? null : r.getFirst(); // great we got a response
}
} catch (ExecutionException e) {
// the primary call failed with RetriesExhaustedException or DoNotRetryIOException
// but the secondaries might still succeed. Continue on the replica RPCs.
exceptions = new ArrayList<ExecutionException>(rl.size());
exceptions.add(e);
completed++;
} catch (CancellationException e) {
throw new InterruptedIOException(e.getMessage());
} catch (InterruptedException e) {
throw new InterruptedIOException(e.getMessage());
}
// submit call for the all of the secondaries at once
// TODO: this may be an overkill for large region replication
submitted += addCallsForOtherReplicas(cs, rl, 0, rl.size() - 1);
try {
while (completed < submitted) {
try {
Future<Pair<Result[], ScannerCallable>> f = cs.take();
Pair<Result[], ScannerCallable> r = f.get();
if (r != null && r.getSecond() != null) {
updateCurrentlyServingReplica(r.getSecond(), r.getFirst(), done, pool);
}
return r == null ? null : r.getFirst(); // great we got an answer
} catch (ExecutionException e) {
// if not cancel or interrupt, wait until all RPC's are done
// one of the tasks failed. Save the exception for later.
if (exceptions == null) exceptions = new ArrayList<ExecutionException>(rl.size());
exceptions.add(e);
completed++;
}
}
} catch (CancellationException e) {
throw new InterruptedIOException(e.getMessage());
} catch (InterruptedException e) {
throw new InterruptedIOException(e.getMessage());
} finally {
// We get there because we were interrupted or because one or more of the
// calls succeeded or failed. In all case, we stop all our tasks.
cs.cancelAll(true);
}
if (exceptions != null && !exceptions.isEmpty()) {
RpcRetryingCallerWithReadReplicas.throwEnrichedException(
exceptions.get(0), retries); // just rethrow the first exception for now.
}
return null; // unreachable
}