org.apache.hadoop.hbase.client.AsyncScanSingleRegionRpcRetryingCaller Maven / Gradle / Ivy
/**
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.hadoop.hbase.client;
import static org.apache.hadoop.hbase.client.ConnectionUtils.SLEEP_DELTA_NS;
import static org.apache.hadoop.hbase.client.ConnectionUtils.getPauseTime;
import static org.apache.hadoop.hbase.client.ConnectionUtils.incRPCCallsMetrics;
import static org.apache.hadoop.hbase.client.ConnectionUtils.incRPCRetriesMetrics;
import static org.apache.hadoop.hbase.client.ConnectionUtils.noMoreResultsForReverseScan;
import static org.apache.hadoop.hbase.client.ConnectionUtils.noMoreResultsForScan;
import static org.apache.hadoop.hbase.client.ConnectionUtils.resetController;
import static org.apache.hadoop.hbase.client.ConnectionUtils.translateException;
import static org.apache.hadoop.hbase.client.ConnectionUtils.updateResultsMetrics;
import static org.apache.hadoop.hbase.client.ConnectionUtils.updateServerSideMetrics;
import java.io.IOException;
import java.util.ArrayList;
import java.util.List;
import java.util.Optional;
import java.util.concurrent.CompletableFuture;
import java.util.concurrent.TimeUnit;
import org.apache.hadoop.hbase.DoNotRetryIOException;
import org.apache.hadoop.hbase.HRegionLocation;
import org.apache.hadoop.hbase.NotServingRegionException;
import org.apache.hadoop.hbase.UnknownScannerException;
import org.apache.hadoop.hbase.client.AdvancedScanResultConsumer.ScanResumer;
import org.apache.hadoop.hbase.client.metrics.ScanMetrics;
import org.apache.hadoop.hbase.exceptions.OutOfOrderScannerNextException;
import org.apache.hadoop.hbase.exceptions.ScannerResetException;
import org.apache.hadoop.hbase.ipc.HBaseRpcController;
import org.apache.hadoop.hbase.regionserver.RegionServerStoppedException;
import org.apache.hadoop.hbase.util.EnvironmentEdgeManager;
import org.apache.yetus.audience.InterfaceAudience;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.apache.hbase.thirdparty.com.google.common.base.Preconditions;
import org.apache.hbase.thirdparty.io.netty.util.Timeout;
import org.apache.hbase.thirdparty.io.netty.util.Timer;
import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;
import org.apache.hadoop.hbase.shaded.protobuf.RequestConverter;
import org.apache.hadoop.hbase.shaded.protobuf.ResponseConverter;
import org.apache.hadoop.hbase.shaded.protobuf.generated.ClientProtos.ClientService;
import org.apache.hadoop.hbase.shaded.protobuf.generated.ClientProtos.ClientService.Interface;
import org.apache.hadoop.hbase.shaded.protobuf.generated.ClientProtos.ScanRequest;
import org.apache.hadoop.hbase.shaded.protobuf.generated.ClientProtos.ScanResponse;
/**
* Retry caller for scanning a region.
*
* We will modify the {@link Scan} object passed in directly. The upper layer should store the
* reference of this object and use it to open new single region scanners.
*/
@InterfaceAudience.Private
class AsyncScanSingleRegionRpcRetryingCaller {
private static final Logger LOG =
LoggerFactory.getLogger(AsyncScanSingleRegionRpcRetryingCaller.class);
private final Timer retryTimer;
private final Scan scan;
private final ScanMetrics scanMetrics;
private final long scannerId;
private final ScanResultCache resultCache;
private final AdvancedScanResultConsumer consumer;
private final ClientService.Interface stub;
private final HRegionLocation loc;
private final boolean regionServerRemote;
private final long scannerLeaseTimeoutPeriodNs;
private final long pauseNs;
private final int maxAttempts;
private final long scanTimeoutNs;
private final long rpcTimeoutNs;
private final int startLogErrorsCnt;
private final Runnable completeWhenNoMoreResultsInRegion;
private final CompletableFuture future;
private final HBaseRpcController controller;
private byte[] nextStartRowWhenError;
private boolean includeNextStartRowWhenError;
private long nextCallStartNs;
private int tries;
private final List exceptions;
private long nextCallSeq = -1L;
private enum ScanControllerState {
INITIALIZED, SUSPENDED, TERMINATED, DESTROYED
}
// Since suspend and terminate should only be called within onNext or onHeartbeat(see the comments
// of RawScanResultConsumer.onNext and onHeartbeat), we need to add some check to prevent invalid
// usage. We use two things to prevent invalid usage:
// 1. Record the thread that construct the ScanControllerImpl instance. We will throw an
// IllegalStateException if the caller thread is not this thread.
// 2. The ControllerState. The initial state is INITIALIZED, if you call suspend, the state will
// be transformed to SUSPENDED, and if you call terminate, the state will be transformed to
// TERMINATED. And when we are back from onNext or onHeartbeat in the onComplete method, we will
// call destroy to get the current state and set the state to DESTROYED. And when user calls
// suspend or terminate, we will check if the current state is INITIALIZED, if not we will throw
// an IllegalStateException. Notice that the DESTROYED state is necessary as you may not call
// suspend or terminate so the state will still be INITIALIZED when back from onNext or
// onHeartbeat. We need another state to replace the INITIALIZED state to prevent the controller
// to be used in the future.
// Notice that, the public methods of this class is supposed to be called by upper layer only, and
// package private methods can only be called within the implementation of
// AsyncScanSingleRegionRpcRetryingCaller.
private final class ScanControllerImpl implements AdvancedScanResultConsumer.ScanController {
// Make sure the methods are only called in this thread.
private final Thread callerThread;
private final Optional cursor;
// INITIALIZED -> SUSPENDED -> DESTROYED
// INITIALIZED -> TERMINATED -> DESTROYED
// INITIALIZED -> DESTROYED
// If the state is incorrect we will throw IllegalStateException.
private ScanControllerState state = ScanControllerState.INITIALIZED;
private ScanResumerImpl resumer;
public ScanControllerImpl(Optional cursor) {
this.callerThread = Thread.currentThread();
this.cursor = cursor;
}
private void preCheck() {
Preconditions.checkState(Thread.currentThread() == callerThread,
"The current thread is %s, expected thread is %s, " +
"you should not call this method outside onNext or onHeartbeat",
Thread.currentThread(), callerThread);
Preconditions.checkState(state.equals(ScanControllerState.INITIALIZED),
"Invalid Stopper state %s", state);
}
@Override
public ScanResumer suspend() {
preCheck();
state = ScanControllerState.SUSPENDED;
ScanResumerImpl resumer = new ScanResumerImpl();
this.resumer = resumer;
return resumer;
}
@Override
public void terminate() {
preCheck();
state = ScanControllerState.TERMINATED;
}
// return the current state, and set the state to DESTROYED.
ScanControllerState destroy() {
ScanControllerState state = this.state;
this.state = ScanControllerState.DESTROYED;
return state;
}
@Override
public Optional cursor() {
return cursor;
}
}
private enum ScanResumerState {
INITIALIZED, SUSPENDED, RESUMED
}
// The resume method is allowed to be called in another thread so here we also use the
// ResumerState to prevent race. The initial state is INITIALIZED, and in most cases, when back
// from onNext or onHeartbeat, we will call the prepare method to change the state to SUSPENDED,
// and when user calls resume method, we will change the state to RESUMED. But the resume method
// could be called in other thread, and in fact, user could just do this:
// controller.suspend().resume()
// This is strange but valid. This means the scan could be resumed before we call the prepare
// method to do the actual suspend work. So in the resume method, we will check if the state is
// INTIALIZED, if it is, then we will just set the state to RESUMED and return. And in prepare
// method, if the state is RESUMED already, we will just return an let the scan go on.
// Notice that, the public methods of this class is supposed to be called by upper layer only, and
// package private methods can only be called within the implementation of
// AsyncScanSingleRegionRpcRetryingCaller.
private final class ScanResumerImpl implements AdvancedScanResultConsumer.ScanResumer {
// INITIALIZED -> SUSPENDED -> RESUMED
// INITIALIZED -> RESUMED
private ScanResumerState state = ScanResumerState.INITIALIZED;
private ScanResponse resp;
private int numberOfCompleteRows;
// If the scan is suspended successfully, we need to do lease renewal to prevent it being closed
// by RS due to lease expire. It is a one-time timer task so we need to schedule a new task
// every time when the previous task is finished. There could also be race as the renewal is
// executed in the timer thread, so we also need to check the state before lease renewal. If the
// state is RESUMED already, we will give up lease renewal and also not schedule the next lease
// renewal task.
private Timeout leaseRenewer;
@Override
public void resume() {
// just used to fix findbugs warnings. In fact, if resume is called before prepare, then we
// just return at the first if condition without loading the resp and numValidResuls field. If
// resume is called after suspend, then it is also safe to just reference resp and
// numValidResults after the synchronized block as no one will change it anymore.
ScanResponse localResp;
int localNumberOfCompleteRows;
synchronized (this) {
if (state == ScanResumerState.INITIALIZED) {
// user calls this method before we call prepare, so just set the state to
// RESUMED, the implementation will just go on.
state = ScanResumerState.RESUMED;
return;
}
if (state == ScanResumerState.RESUMED) {
// already resumed, give up.
return;
}
state = ScanResumerState.RESUMED;
if (leaseRenewer != null) {
leaseRenewer.cancel();
}
localResp = this.resp;
localNumberOfCompleteRows = this.numberOfCompleteRows;
}
completeOrNext(localResp, localNumberOfCompleteRows);
}
private void scheduleRenewLeaseTask() {
leaseRenewer = retryTimer.newTimeout(t -> tryRenewLease(), scannerLeaseTimeoutPeriodNs / 2,
TimeUnit.NANOSECONDS);
}
private synchronized void tryRenewLease() {
// the scan has already been resumed, give up
if (state == ScanResumerState.RESUMED) {
return;
}
renewLease();
// schedule the next renew lease task again as this is a one-time task.
scheduleRenewLeaseTask();
}
// return false if the scan has already been resumed. See the comment above for ScanResumerImpl
// for more details.
synchronized boolean prepare(ScanResponse resp, int numberOfCompleteRows) {
if (state == ScanResumerState.RESUMED) {
// user calls resume before we actually suspend the scan, just continue;
return false;
}
state = ScanResumerState.SUSPENDED;
this.resp = resp;
this.numberOfCompleteRows = numberOfCompleteRows;
// if there are no more results in region then the scanner at RS side will be closed
// automatically so we do not need to renew lease.
if (resp.getMoreResultsInRegion()) {
// schedule renew lease task
scheduleRenewLeaseTask();
}
return true;
}
}
public AsyncScanSingleRegionRpcRetryingCaller(Timer retryTimer,
AsyncConnectionImpl conn, Scan scan, ScanMetrics scanMetrics, long scannerId,
ScanResultCache resultCache, AdvancedScanResultConsumer consumer, Interface stub,
HRegionLocation loc, boolean isRegionServerRemote, long scannerLeaseTimeoutPeriodNs,
long pauseNs, int maxAttempts, long scanTimeoutNs, long rpcTimeoutNs, int startLogErrorsCnt) {
this.retryTimer = retryTimer;
this.scan = scan;
this.scanMetrics = scanMetrics;
this.scannerId = scannerId;
this.resultCache = resultCache;
this.consumer = consumer;
this.stub = stub;
this.loc = loc;
this.regionServerRemote = isRegionServerRemote;
this.scannerLeaseTimeoutPeriodNs = scannerLeaseTimeoutPeriodNs;
this.pauseNs = pauseNs;
this.maxAttempts = maxAttempts;
this.scanTimeoutNs = scanTimeoutNs;
this.rpcTimeoutNs = rpcTimeoutNs;
this.startLogErrorsCnt = startLogErrorsCnt;
if (scan.isReversed()) {
completeWhenNoMoreResultsInRegion = this::completeReversedWhenNoMoreResultsInRegion;
} else {
completeWhenNoMoreResultsInRegion = this::completeWhenNoMoreResultsInRegion;
}
this.future = new CompletableFuture<>();
this.controller = conn.rpcControllerFactory.newController();
this.exceptions = new ArrayList<>();
}
private long elapsedMs() {
return TimeUnit.NANOSECONDS.toMillis(System.nanoTime() - nextCallStartNs);
}
private long remainingTimeNs() {
return scanTimeoutNs - (System.nanoTime() - nextCallStartNs);
}
private void closeScanner() {
incRPCCallsMetrics(scanMetrics, regionServerRemote);
resetController(controller, rpcTimeoutNs);
ScanRequest req = RequestConverter.buildScanRequest(this.scannerId, 0, true, false);
stub.scan(controller, req, resp -> {
if (controller.failed()) {
LOG.warn("Call to " + loc.getServerName() + " for closing scanner id = " + scannerId +
" for " + loc.getRegion().getEncodedName() + " of " +
loc.getRegion().getTable() + " failed, ignore, probably already closed",
controller.getFailed());
}
});
}
private void completeExceptionally(boolean closeScanner) {
resultCache.clear();
if (closeScanner) {
closeScanner();
}
future.completeExceptionally(new RetriesExhaustedException(tries - 1, exceptions));
}
private void completeNoMoreResults() {
future.complete(false);
}
private void completeWithNextStartRow(byte[] row, boolean inclusive) {
scan.withStartRow(row, inclusive);
future.complete(true);
}
private void completeWhenError(boolean closeScanner) {
incRPCRetriesMetrics(scanMetrics, closeScanner);
resultCache.clear();
if (closeScanner) {
closeScanner();
}
if (nextStartRowWhenError != null) {
scan.withStartRow(nextStartRowWhenError, includeNextStartRowWhenError);
}
future.complete(true);
}
private void onError(Throwable error) {
error = translateException(error);
if (tries > startLogErrorsCnt) {
LOG.warn("Call to " + loc.getServerName() + " for scanner id = " + scannerId + " for " +
loc.getRegion().getEncodedName() + " of " + loc.getRegion().getTable() +
" failed, , tries = " + tries + ", maxAttempts = " + maxAttempts + ", timeout = " +
TimeUnit.NANOSECONDS.toMillis(scanTimeoutNs) + " ms, time elapsed = " + elapsedMs() +
" ms",
error);
}
boolean scannerClosed = error instanceof UnknownScannerException ||
error instanceof NotServingRegionException || error instanceof RegionServerStoppedException;
RetriesExhaustedException.ThrowableWithExtraContext qt =
new RetriesExhaustedException.ThrowableWithExtraContext(error,
EnvironmentEdgeManager.currentTime(), "");
exceptions.add(qt);
if (tries >= maxAttempts) {
completeExceptionally(!scannerClosed);
return;
}
long delayNs;
if (scanTimeoutNs > 0) {
long maxDelayNs = remainingTimeNs() - SLEEP_DELTA_NS;
if (maxDelayNs <= 0) {
completeExceptionally(!scannerClosed);
return;
}
delayNs = Math.min(maxDelayNs, getPauseTime(pauseNs, tries - 1));
} else {
delayNs = getPauseTime(pauseNs, tries - 1);
}
if (scannerClosed) {
completeWhenError(false);
return;
}
if (error instanceof OutOfOrderScannerNextException || error instanceof ScannerResetException) {
completeWhenError(true);
return;
}
if (error instanceof DoNotRetryIOException) {
completeExceptionally(true);
return;
}
tries++;
retryTimer.newTimeout(t -> call(), delayNs, TimeUnit.NANOSECONDS);
}
private void updateNextStartRowWhenError(Result result) {
nextStartRowWhenError = result.getRow();
includeNextStartRowWhenError = result.mayHaveMoreCellsInRow();
}
private void completeWhenNoMoreResultsInRegion() {
if (noMoreResultsForScan(scan, loc.getRegion())) {
completeNoMoreResults();
} else {
completeWithNextStartRow(loc.getRegion().getEndKey(), true);
}
}
private void completeReversedWhenNoMoreResultsInRegion() {
if (noMoreResultsForReverseScan(scan, loc.getRegion())) {
completeNoMoreResults();
} else {
completeWithNextStartRow(loc.getRegion().getStartKey(), false);
}
}
private void completeOrNext(ScanResponse resp, int numberOfCompleteRows) {
if (resp.hasMoreResults() && !resp.getMoreResults()) {
// RS tells us there is no more data for the whole scan
completeNoMoreResults();
return;
}
if (scan.getLimit() > 0) {
// The RS should have set the moreResults field in ScanResponse to false when we have reached
// the limit, so we add an assert here.
int newLimit = scan.getLimit() - numberOfCompleteRows;
assert newLimit > 0;
scan.setLimit(newLimit);
}
// as in 2.0 this value will always be set
if (!resp.getMoreResultsInRegion()) {
completeWhenNoMoreResultsInRegion.run();
return;
}
next();
}
private void onComplete(HBaseRpcController controller, ScanResponse resp) {
if (controller.failed()) {
onError(controller.getFailed());
return;
}
updateServerSideMetrics(scanMetrics, resp);
boolean isHeartbeatMessage = resp.hasHeartbeatMessage() && resp.getHeartbeatMessage();
Result[] rawResults;
Result[] results;
int numberOfCompleteRowsBefore = resultCache.numberOfCompleteRows();
try {
rawResults = ResponseConverter.getResults(controller.cellScanner(), resp);
updateResultsMetrics(scanMetrics, rawResults, isHeartbeatMessage);
results = resultCache.addAndGet(
Optional.ofNullable(rawResults).orElse(ScanResultCache.EMPTY_RESULT_ARRAY),
isHeartbeatMessage);
} catch (IOException e) {
// We can not retry here. The server has responded normally and the call sequence has been
// increased so a new scan with the same call sequence will cause an
// OutOfOrderScannerNextException. Let the upper layer open a new scanner.
LOG.warn("decode scan response failed", e);
completeWhenError(true);
return;
}
ScanControllerImpl scanController;
if (results.length > 0) {
scanController = new ScanControllerImpl(
resp.hasCursor() ? Optional.of(ProtobufUtil.toCursor(resp.getCursor()))
: Optional.empty());
updateNextStartRowWhenError(results[results.length - 1]);
consumer.onNext(results, scanController);
} else {
Optional cursor = Optional.empty();
if (resp.hasCursor()) {
cursor = Optional.of(ProtobufUtil.toCursor(resp.getCursor()));
} else if (scan.isNeedCursorResult() && rawResults.length > 0) {
// It is size limit exceed and we need to return the last Result's row.
// When user setBatch and the scanner is reopened, the server may return Results that
// user has seen and the last Result can not be seen because the number is not enough.
// So the row keys of results may not be same, we must use the last one.
cursor = Optional.of(new Cursor(rawResults[rawResults.length - 1].getRow()));
}
scanController = new ScanControllerImpl(cursor);
if (isHeartbeatMessage || cursor.isPresent()) {
// only call onHeartbeat if server tells us explicitly this is a heartbeat message, or we
// want to pass a cursor to upper layer.
consumer.onHeartbeat(scanController);
}
}
ScanControllerState state = scanController.destroy();
if (state == ScanControllerState.TERMINATED) {
if (resp.getMoreResultsInRegion()) {
// we have more results in region but user request to stop the scan, so we need to close the
// scanner explicitly.
closeScanner();
}
completeNoMoreResults();
return;
}
int numberOfCompleteRows = resultCache.numberOfCompleteRows() - numberOfCompleteRowsBefore;
if (state == ScanControllerState.SUSPENDED) {
if (scanController.resumer.prepare(resp, numberOfCompleteRows)) {
return;
}
}
completeOrNext(resp, numberOfCompleteRows);
}
private void call() {
// As we have a call sequence for scan, it is useless to have a different rpc timeout which is
// less than the scan timeout. If the server does not respond in time(usually this will not
// happen as we have heartbeat now), we will get an OutOfOrderScannerNextException when
// resending the next request and the only way to fix this is to close the scanner and open a
// new one.
long callTimeoutNs;
if (scanTimeoutNs > 0) {
long remainingNs = scanTimeoutNs - (System.nanoTime() - nextCallStartNs);
if (remainingNs <= 0) {
completeExceptionally(true);
return;
}
callTimeoutNs = remainingNs;
} else {
callTimeoutNs = 0L;
}
incRPCCallsMetrics(scanMetrics, regionServerRemote);
if (tries > 1) {
incRPCRetriesMetrics(scanMetrics, regionServerRemote);
}
resetController(controller, callTimeoutNs);
ScanRequest req = RequestConverter.buildScanRequest(scannerId, scan.getCaching(), false,
nextCallSeq, false, false, scan.getLimit());
stub.scan(controller, req, resp -> onComplete(controller, resp));
}
private void next() {
nextCallSeq++;
tries = 1;
exceptions.clear();
nextCallStartNs = System.nanoTime();
call();
}
private void renewLease() {
incRPCCallsMetrics(scanMetrics, regionServerRemote);
nextCallSeq++;
resetController(controller, rpcTimeoutNs);
ScanRequest req =
RequestConverter.buildScanRequest(scannerId, 0, false, nextCallSeq, false, true, -1);
stub.scan(controller, req, resp -> {
});
}
/**
* Now we will also fetch some cells along with the scanner id when opening a scanner, so we also
* need to process the ScanResponse for the open scanner request. The HBaseRpcController for the
* open scanner request is also needed because we may have some data in the CellScanner which is
* contained in the controller.
* @return {@code true} if we should continue, otherwise {@code false}.
*/
public CompletableFuture start(HBaseRpcController controller,
ScanResponse respWhenOpen) {
onComplete(controller, respWhenOpen);
return future;
}
}