com.datastax.driver.core.RequestHandler Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of dse-java-driver-core Show documentation
Show all versions of dse-java-driver-core Show documentation
A driver for DataStax Enterprise (DSE)
and Apache Cassandra 1.2+ clusters that works exclusively with the
Cassandra Query Language version 3 (CQL3) and Cassandra's binary protocol,
supporting DSE-specific features such as geospatial types, DSE Graph and DSE authentication.
/*
* Copyright DataStax, Inc.
*
* This software can be used solely with DataStax Enterprise. Please consult the license at
* http://www.datastax.com/terms/datastax-dse-driver-license-terms
*/
package com.datastax.driver.core;
import com.codahale.metrics.Timer;
import com.datastax.driver.core.exceptions.BootstrappingException;
import com.datastax.driver.core.exceptions.BusyConnectionException;
import com.datastax.driver.core.exceptions.BusyPoolException;
import com.datastax.driver.core.exceptions.ConnectionException;
import com.datastax.driver.core.exceptions.DriverException;
import com.datastax.driver.core.exceptions.DriverInternalError;
import com.datastax.driver.core.exceptions.NoHostAvailableException;
import com.datastax.driver.core.exceptions.OperationTimedOutException;
import com.datastax.driver.core.exceptions.OverloadedException;
import com.datastax.driver.core.exceptions.ReadFailureException;
import com.datastax.driver.core.exceptions.ReadTimeoutException;
import com.datastax.driver.core.exceptions.ServerError;
import com.datastax.driver.core.exceptions.UnavailableException;
import com.datastax.driver.core.exceptions.WriteFailureException;
import com.datastax.driver.core.exceptions.WriteTimeoutException;
import com.datastax.driver.core.policies.RetryPolicy;
import com.datastax.driver.core.policies.RetryPolicy.RetryDecision.Type;
import com.datastax.driver.core.policies.SpeculativeExecutionPolicy.SpeculativeExecutionPlan;
import com.google.common.collect.ImmutableList;
import com.google.common.collect.Iterators;
import com.google.common.collect.Sets;
import com.google.common.util.concurrent.FutureCallback;
import com.google.common.util.concurrent.ListenableFuture;
import io.netty.util.Timeout;
import io.netty.util.TimerTask;
import java.util.Collections;
import java.util.Iterator;
import java.util.List;
import java.util.Set;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ConcurrentMap;
import java.util.concurrent.CopyOnWriteArrayList;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.concurrent.atomic.AtomicReference;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
/**
* Handles a request to cassandra, dealing with host failover and retries on unavailable/timeout.
*/
class RequestHandler {
private static final Logger logger = LoggerFactory.getLogger(RequestHandler.class);
private static final boolean HOST_METRICS_ENABLED =
Boolean.getBoolean("com.datastax.driver.HOST_METRICS_ENABLED");
private static final QueryLogger QUERY_LOGGER = QueryLogger.builder().build();
static final String DISABLE_QUERY_WARNING_LOGS = "com.datastax.driver.DISABLE_QUERY_WARNING_LOGS";
final String id;
private final SessionManager manager;
private final Callback callback;
private final QueryPlan queryPlan;
private final SpeculativeExecutionPlan speculativeExecutionPlan;
private final boolean allowSpeculativeExecutions;
private final Set runningExecutions = Sets.newCopyOnWriteArraySet();
private final Set scheduledExecutions = Sets.newCopyOnWriteArraySet();
private final Statement statement;
private final io.netty.util.Timer scheduler;
private volatile List triedHosts;
private volatile ConcurrentMap errors;
private final Timer.Context timerContext;
private final long startTime;
private final AtomicBoolean isDone = new AtomicBoolean();
private final AtomicInteger executionIndex = new AtomicInteger();
public RequestHandler(SessionManager manager, Callback callback, Statement statement) {
this.id = Long.toString(System.identityHashCode(this));
if (logger.isTraceEnabled()) logger.trace("[{}] {}", id, statement);
this.manager = manager;
this.callback = callback;
this.scheduler = manager.cluster.manager.connectionFactory.timer;
callback.register(this);
// If host is explicitly set on statement, bypass load balancing policy.
if (statement.getHost() != null) {
this.queryPlan = new QueryPlan(Iterators.singletonIterator(statement.getHost()));
} else {
this.queryPlan =
new QueryPlan(
manager.loadBalancingPolicy().newQueryPlan(manager.poolsState.keyspace, statement));
}
this.speculativeExecutionPlan =
manager.speculativeExecutionPolicy().newPlan(manager.poolsState.keyspace, statement);
this.allowSpeculativeExecutions =
statement != Statement.DEFAULT
&& statement.isIdempotentWithDefault(manager.configuration().getQueryOptions());
this.statement = statement;
this.timerContext = metricsEnabled() ? metrics().getRequestsTimer().time() : null;
this.startTime = System.nanoTime();
}
void sendRequest() {
startNewExecution();
}
// Called when the corresponding ResultSetFuture is cancelled by the client
void cancel() {
if (!isDone.compareAndSet(false, true)) return;
cancelPendingExecutions(null);
}
private void startNewExecution() {
if (isDone.get()) return;
Message.Request request = callback.request();
int position = executionIndex.getAndIncrement();
SpeculativeExecution execution = new SpeculativeExecution(request, position);
runningExecutions.add(execution);
execution.findNextHostAndQuery();
}
private void scheduleExecution(long delayMillis) {
if (isDone.get() || delayMillis < 0) return;
if (logger.isTraceEnabled())
logger.trace("[{}] Schedule next speculative execution in {} ms", id, delayMillis);
if (delayMillis == 0) {
// kick off request immediately
scheduleExecutionImmediately();
} else {
scheduledExecutions.add(
scheduler.newTimeout(newExecutionTask, delayMillis, TimeUnit.MILLISECONDS));
}
}
private final TimerTask newExecutionTask =
new TimerTask() {
@Override
public void run(final Timeout timeout) throws Exception {
scheduledExecutions.remove(timeout);
if (!isDone.get()) {
// We're on the timer thread so reschedule to another executor
manager
.executor()
.execute(
new Runnable() {
@Override
public void run() {
scheduleExecutionImmediately();
}
});
}
}
};
private void scheduleExecutionImmediately() {
if (metricsEnabled()) metrics().getErrorMetrics().getSpeculativeExecutions().inc();
startNewExecution();
}
private void cancelPendingExecutions(SpeculativeExecution ignore) {
for (SpeculativeExecution execution : runningExecutions)
if (execution != ignore) // not vital but this produces nicer logs
execution.cancel();
for (Timeout execution : scheduledExecutions) execution.cancel();
}
private void setFinalResult(
SpeculativeExecution execution, Connection connection, Message.Response response) {
if (!isDone.compareAndSet(false, true)) {
if (logger.isTraceEnabled())
logger.trace("[{}] Got beaten to setting the result", execution.id);
return;
}
if (logger.isTraceEnabled()) logger.trace("[{}] Setting final result", execution.id);
cancelPendingExecutions(execution);
try {
if (timerContext != null) timerContext.stop();
ExecutionInfo info;
int speculativeExecutions = executionIndex.get() - 1;
// Avoid creating a new instance if we can reuse the host's default one
if (execution.position == 0
&& speculativeExecutions == 0
&& triedHosts == null
&& execution.retryConsistencyLevel == null
&& response.getCustomPayload() == null) {
info = execution.current.defaultExecutionInfo;
} else {
List hosts;
if (triedHosts == null) {
hosts = ImmutableList.of(execution.current);
} else {
hosts = triedHosts;
hosts.add(execution.current);
}
info =
new ExecutionInfo(
speculativeExecutions,
execution.position,
hosts,
execution.retryConsistencyLevel,
response.getCustomPayload());
}
// if the response from the server has warnings, they'll be set on the ExecutionInfo. Log them
// here, unless they've been disabled.
if (response.warnings != null
&& !response.warnings.isEmpty()
&& !Boolean.getBoolean(RequestHandler.DISABLE_QUERY_WARNING_LOGS)
&& logger.isWarnEnabled()) {
logServerWarnings(response.warnings);
}
callback.onSet(connection, response, info, statement, System.nanoTime() - startTime);
} catch (Exception e) {
callback.onException(
connection,
new DriverInternalError(
"Unexpected exception while setting final result from " + response, e),
System.nanoTime() - startTime, /*unused*/
0);
}
}
private void logServerWarnings(List warnings) {
// truncate the statement query to the DEFAULT_MAX_QUERY_STRING_LENGTH, if necessary
final String queryString = QUERY_LOGGER.statementAsString(statement);
// log each warning separately
for (String warning : warnings) {
logger.warn("Query '{}' generated server side warning(s): {}", queryString, warning);
}
}
private void setFinalException(
SpeculativeExecution execution, Connection connection, Exception exception) {
if (!isDone.compareAndSet(false, true)) {
if (logger.isTraceEnabled())
logger.trace("[{}] Got beaten to setting final exception", execution.id);
return;
}
if (logger.isTraceEnabled()) logger.trace("[{}] Setting final exception", execution.id);
cancelPendingExecutions(execution);
try {
if (timerContext != null) timerContext.stop();
} finally {
callback.onException(connection, exception, System.nanoTime() - startTime, /*unused*/ 0);
}
}
// Triggered when an execution reaches the end of the query plan.
// This is only a failure if there are no other running executions.
private void reportNoMoreHosts(SpeculativeExecution execution) {
runningExecutions.remove(execution);
if (runningExecutions.isEmpty())
setFinalException(
execution,
null,
new NoHostAvailableException(
errors == null ? Collections.emptyMap() : errors));
}
private boolean metricsEnabled() {
return manager.configuration().getMetricsOptions().isEnabled();
}
private boolean hostMetricsEnabled() {
return HOST_METRICS_ENABLED && metricsEnabled();
}
private Metrics metrics() {
return manager.cluster.manager.metrics;
}
private RetryPolicy retryPolicy() {
return statement.getRetryPolicy() == null
? manager.configuration().getPolicies().getRetryPolicy()
: statement.getRetryPolicy();
}
interface Callback extends Connection.ResponseCallback {
void onSet(
Connection connection,
Message.Response response,
ExecutionInfo info,
Statement statement,
long latency);
void register(RequestHandler handler);
}
/**
* An execution of the query against the cluster. There is at least one instance per
* RequestHandler, and possibly more (depending on the SpeculativeExecutionPolicy). Each instance
* may retry on the same host, or on other hosts as defined by the RetryPolicy. All instances run
* concurrently and share the same query plan. There are three ways a SpeculativeExecution can
* stop: - it completes the query (with either a success or a fatal error), and reports the result
* to the RequestHandler - it gets cancelled, either because another execution completed the
* query, or because the RequestHandler was cancelled - it reaches the end of the query plan and
* informs the RequestHandler, which will decide what to do
*/
class SpeculativeExecution implements Connection.ResponseCallback {
final String id;
private final Message.Request request;
private final int position;
private volatile Host current;
private volatile ConsistencyLevel retryConsistencyLevel;
private final AtomicReference queryStateRef;
private final AtomicBoolean nextExecutionScheduled = new AtomicBoolean();
private final long startTime = System.nanoTime();
// This represents the number of times a retry has been triggered by the RetryPolicy (this is
// different from
// queryStateRef.get().retryCount, because some retries don't involve the policy, for example
// after an
// UNPREPARED response).
// This is incremented by one writer at a time, so volatile is good enough.
private volatile int retriesByPolicy;
private volatile Connection.ResponseHandler connectionHandler;
SpeculativeExecution(Message.Request request, int position) {
this.id = RequestHandler.this.id + "-" + position;
this.request = request;
this.position = position;
this.queryStateRef = new AtomicReference(QueryState.INITIAL);
if (logger.isTraceEnabled()) logger.trace("[{}] Starting", id);
}
void findNextHostAndQuery() {
try {
Host host;
while (!isDone.get()
&& (host = queryPlan.next()) != null
&& !queryStateRef.get().isCancelled()) {
if (query(host)) {
if (hostMetricsEnabled()) {
metrics().getRegistry().counter(MetricsUtil.hostMetricName("writes.", host)).inc();
}
return;
} else if (hostMetricsEnabled()) {
metrics()
.getRegistry()
.counter(MetricsUtil.hostMetricName("write-errors.", host))
.inc();
}
}
if (current != null) {
if (triedHosts == null) triedHosts = new CopyOnWriteArrayList();
triedHosts.add(current);
}
reportNoMoreHosts(this);
} catch (Exception e) {
// Shouldn't happen really, but if ever the loadbalancing policy returned iterator throws,
// we don't want to block.
setFinalException(
null,
new DriverInternalError("An unexpected error happened while sending requests", e));
}
}
private boolean query(final Host host) {
HostConnectionPool pool = manager.pools.get(host);
if (pool == null || pool.isClosed()) return false;
if (logger.isTraceEnabled()) logger.trace("[{}] Querying node {}", id, host);
if (allowSpeculativeExecutions && nextExecutionScheduled.compareAndSet(false, true))
scheduleExecution(speculativeExecutionPlan.nextExecution(host));
PoolingOptions poolingOptions = manager.configuration().getPoolingOptions();
ListenableFuture connectionFuture =
pool.borrowConnection(
poolingOptions.getPoolTimeoutMillis(),
TimeUnit.MILLISECONDS,
poolingOptions.getMaxQueueSize());
GuavaCompatibility.INSTANCE.addCallback(
connectionFuture,
new FutureCallback() {
@Override
public void onSuccess(Connection connection) {
if (isDone.get()) {
connection.release();
return;
}
if (current != null) {
if (triedHosts == null) triedHosts = new CopyOnWriteArrayList();
triedHosts.add(current);
}
current = host;
try {
write(connection, SpeculativeExecution.this);
} catch (ConnectionException e) {
// If we have any problem with the connection, move to the next node.
if (metricsEnabled()) metrics().getErrorMetrics().getConnectionErrors().inc();
if (connection != null) connection.release();
logError(host.getEndPoint(), e);
findNextHostAndQuery();
} catch (BusyConnectionException e) {
// The pool shouldn't have give us a busy connection unless we've maxed up the pool,
// so move on to the next host.
connection.release(true);
logError(host.getEndPoint(), e);
findNextHostAndQuery();
} catch (RuntimeException e) {
if (connection != null) connection.release();
logger.error("Unexpected error while querying " + host.getEndPoint(), e);
logError(host.getEndPoint(), e);
findNextHostAndQuery();
}
}
@Override
public void onFailure(Throwable t) {
if (t instanceof BusyPoolException) {
logError(host.getEndPoint(), t);
} else {
logger.error("Unexpected error while querying " + host.getEndPoint(), t);
logError(host.getEndPoint(), t);
}
findNextHostAndQuery();
}
});
return true;
}
private void write(Connection connection, Connection.ResponseCallback responseCallback)
throws ConnectionException, BusyConnectionException {
// Make sure cancel() does not see a stale connectionHandler if it sees the new query state
// before connection.write has completed
connectionHandler = null;
// Ensure query state is "in progress" (can be already if connection.write failed on a
// previous node and we're retrying)
while (true) {
QueryState previous = queryStateRef.get();
if (previous.isCancelled()) {
connection.release();
return;
}
if (previous.inProgress || queryStateRef.compareAndSet(previous, previous.startNext()))
break;
}
connectionHandler =
connection.write(responseCallback, statement.getReadTimeoutMillis(), false, false);
// Only start the timeout when we're sure connectionHandler is set. This avoids an edge case
// where onTimeout() was triggered
// *before* the call to connection.write had returned.
connectionHandler.startTimeout();
// Note that we could have already received the response here (so onSet() / onException()
// would have been called). This is
// why we only test for CANCELLED_WHILE_IN_PROGRESS below.
// If cancel() was called after we set the state to "in progress", but before connection.write
// had completed, it might have
// missed the new value of connectionHandler. So make sure that cancelHandler() gets called
// here (we might call it twice,
// but it knows how to deal with it).
if (queryStateRef.get() == QueryState.CANCELLED_WHILE_IN_PROGRESS
&& connectionHandler.cancelHandler()) connection.release();
}
private RetryPolicy.RetryDecision computeRetryDecisionOnRequestError(
DriverException exception) {
RetryPolicy.RetryDecision decision;
if (statement.isIdempotentWithDefault(manager.cluster.getConfiguration().getQueryOptions())) {
decision =
retryPolicy()
.onRequestError(statement, request().consistency(), exception, retriesByPolicy);
} else {
decision = RetryPolicy.RetryDecision.rethrow();
}
if (metricsEnabled()) {
if (exception instanceof OperationTimedOutException) {
metrics().getErrorMetrics().getClientTimeouts().inc();
if (decision.getType() == Type.RETRY)
metrics().getErrorMetrics().getRetriesOnClientTimeout().inc();
if (decision.getType() == Type.IGNORE)
metrics().getErrorMetrics().getIgnoresOnClientTimeout().inc();
} else if (exception instanceof ConnectionException) {
metrics().getErrorMetrics().getConnectionErrors().inc();
if (decision.getType() == Type.RETRY)
metrics().getErrorMetrics().getRetriesOnConnectionError().inc();
if (decision.getType() == Type.IGNORE)
metrics().getErrorMetrics().getIgnoresOnConnectionError().inc();
} else {
metrics().getErrorMetrics().getOthers().inc();
if (decision.getType() == Type.RETRY)
metrics().getErrorMetrics().getRetriesOnOtherErrors().inc();
if (decision.getType() == Type.IGNORE)
metrics().getErrorMetrics().getIgnoresOnOtherErrors().inc();
}
}
return decision;
}
private void processRetryDecision(
RetryPolicy.RetryDecision retryDecision,
Connection connection,
Exception exceptionToReport) {
switch (retryDecision.getType()) {
case RETRY:
retriesByPolicy++;
if (logger.isDebugEnabled())
logger.debug(
"[{}] Doing retry {} for query {} at consistency {}",
id,
retriesByPolicy,
statement,
retryDecision.getRetryConsistencyLevel());
if (metricsEnabled()) metrics().getErrorMetrics().getRetries().inc();
// log error for the current host if we are switching to another one
if (!retryDecision.isRetryCurrent()) logError(connection.endPoint, exceptionToReport);
retry(retryDecision.isRetryCurrent(), retryDecision.getRetryConsistencyLevel());
break;
case RETHROW:
setFinalException(connection, exceptionToReport);
break;
case IGNORE:
if (metricsEnabled()) metrics().getErrorMetrics().getIgnores().inc();
setFinalResult(connection, new Responses.Result.Void());
break;
}
}
private void retry(final boolean retryCurrent, ConsistencyLevel newConsistencyLevel) {
final Host h = current;
if (newConsistencyLevel != null) this.retryConsistencyLevel = newConsistencyLevel;
if (queryStateRef.get().isCancelled()) return;
if (!retryCurrent || !query(h)) findNextHostAndQuery();
}
private void logError(EndPoint endPoint, Throwable exception) {
logger.debug("[{}] Error querying {} : {}", id, endPoint, exception.toString());
if (errors == null) {
synchronized (RequestHandler.this) {
if (errors == null) {
errors = new ConcurrentHashMap();
}
}
}
errors.put(endPoint, exception);
}
void cancel() {
// Atomically set a special QueryState, that will cause any further operation to abort.
// We want to remember whether a request was in progress when we did this, so there are two
// cancel states.
while (true) {
QueryState previous = queryStateRef.get();
if (previous.isCancelled()) {
return;
} else if (previous.inProgress
&& queryStateRef.compareAndSet(previous, QueryState.CANCELLED_WHILE_IN_PROGRESS)) {
if (logger.isTraceEnabled()) logger.trace("[{}] Cancelled while in progress", id);
// The connectionHandler should be non-null, but we might miss the update if we're racing
// with write().
// If it's still null, this will be handled by re-checking queryStateRef at the end of
// write().
if (connectionHandler != null && connectionHandler.cancelHandler())
connectionHandler.connection.release();
Host queriedHost = current;
if (queriedHost != null && statement != Statement.DEFAULT) {
manager.cluster.manager.reportQuery(
queriedHost,
statement,
CancelledSpeculativeExecutionException.INSTANCE,
System.nanoTime() - startTime);
}
return;
} else if (!previous.inProgress
&& queryStateRef.compareAndSet(previous, QueryState.CANCELLED_WHILE_COMPLETE)) {
if (logger.isTraceEnabled()) logger.trace("[{}] Cancelled while complete", id);
Host queriedHost = current;
if (queriedHost != null && statement != Statement.DEFAULT) {
manager.cluster.manager.reportQuery(
queriedHost,
statement,
CancelledSpeculativeExecutionException.INSTANCE,
System.nanoTime() - startTime);
}
return;
}
}
}
@Override
public Message.Request request() {
if (retryConsistencyLevel != null && retryConsistencyLevel != request.consistency())
return request.copy(retryConsistencyLevel);
else return request;
}
@Override
public void onSet(
Connection connection, Message.Response response, long latency, int retryCount) {
QueryState queryState = queryStateRef.get();
if (!queryState.isInProgressAt(retryCount)
|| !queryStateRef.compareAndSet(queryState, queryState.complete())) {
logger.debug(
"onSet triggered but the response was completed by another thread, cancelling (retryCount = {}, queryState = {}, queryStateRef = {})",
retryCount,
queryState,
queryStateRef.get());
return;
}
Host queriedHost = current;
Exception exceptionToReport = null;
try {
switch (response.type) {
case RESULT:
connection.release();
setFinalResult(connection, response);
break;
case ERROR:
Responses.Error err = (Responses.Error) response;
exceptionToReport = err.asException(connection.endPoint);
RetryPolicy.RetryDecision retry = null;
RetryPolicy retryPolicy = retryPolicy();
switch (err.code) {
case READ_TIMEOUT:
connection.release();
assert err.infos instanceof ReadTimeoutException;
ReadTimeoutException rte = (ReadTimeoutException) err.infos;
retry =
retryPolicy.onReadTimeout(
statement,
rte.getConsistencyLevel(),
rte.getRequiredAcknowledgements(),
rte.getReceivedAcknowledgements(),
rte.wasDataRetrieved(),
retriesByPolicy);
if (metricsEnabled()) {
metrics().getErrorMetrics().getReadTimeouts().inc();
if (retry.getType() == Type.RETRY)
metrics().getErrorMetrics().getRetriesOnReadTimeout().inc();
if (retry.getType() == Type.IGNORE)
metrics().getErrorMetrics().getIgnoresOnReadTimeout().inc();
}
break;
case WRITE_TIMEOUT:
connection.release();
assert err.infos instanceof WriteTimeoutException;
WriteTimeoutException wte = (WriteTimeoutException) err.infos;
if (statement.isIdempotentWithDefault(
manager.cluster.getConfiguration().getQueryOptions()))
retry =
retryPolicy.onWriteTimeout(
statement,
wte.getConsistencyLevel(),
wte.getWriteType(),
wte.getRequiredAcknowledgements(),
wte.getReceivedAcknowledgements(),
retriesByPolicy);
else {
retry = RetryPolicy.RetryDecision.rethrow();
}
if (metricsEnabled()) {
metrics().getErrorMetrics().getWriteTimeouts().inc();
if (retry.getType() == Type.RETRY)
metrics().getErrorMetrics().getRetriesOnWriteTimeout().inc();
if (retry.getType() == Type.IGNORE)
metrics().getErrorMetrics().getIgnoresOnWriteTimeout().inc();
}
break;
case UNAVAILABLE:
connection.release();
assert err.infos instanceof UnavailableException;
UnavailableException ue = (UnavailableException) err.infos;
retry =
retryPolicy.onUnavailable(
statement,
ue.getConsistencyLevel(),
ue.getRequiredReplicas(),
ue.getAliveReplicas(),
retriesByPolicy);
if (metricsEnabled()) {
metrics().getErrorMetrics().getUnavailables().inc();
if (retry.getType() == Type.RETRY)
metrics().getErrorMetrics().getRetriesOnUnavailable().inc();
if (retry.getType() == Type.IGNORE)
metrics().getErrorMetrics().getIgnoresOnUnavailable().inc();
}
break;
case OVERLOADED:
connection.release();
assert exceptionToReport instanceof OverloadedException;
logger.warn("Host {} is overloaded.", connection.endPoint);
retry = computeRetryDecisionOnRequestError((OverloadedException) exceptionToReport);
break;
case SERVER_ERROR:
connection.release();
assert exceptionToReport instanceof ServerError;
logger.warn(
"{} replied with server error ({}), defuncting connection.",
connection.endPoint,
err.message);
// Defunct connection
connection.defunct(exceptionToReport);
retry = computeRetryDecisionOnRequestError((ServerError) exceptionToReport);
break;
case IS_BOOTSTRAPPING:
connection.release();
assert exceptionToReport instanceof BootstrappingException;
logger.error(
"Query sent to {} but it is bootstrapping. This shouldn't happen but trying next host.",
connection.endPoint);
if (metricsEnabled()) {
metrics().getErrorMetrics().getOthers().inc();
}
logError(connection.endPoint, exceptionToReport);
retry(false, null);
return;
case UNPREPARED:
// Do not release connection yet, because we might reuse it to send the PREPARE
// message (see write() call below)
assert err.infos instanceof MD5Digest;
MD5Digest id = (MD5Digest) err.infos;
PreparedStatement toPrepare = manager.cluster.manager.preparedQueries.get(id);
if (toPrepare == null) {
// This shouldn't happen
connection.release();
String msg = String.format("Tried to execute unknown prepared query %s", id);
logger.error(msg);
setFinalException(connection, new DriverInternalError(msg));
return;
}
String currentKeyspace = connection.keyspace();
String prepareKeyspace = toPrepare.getQueryKeyspace();
// Only allow preparing with a different keyspace than configured connection
// keyspace if
// protocol supports it.
if (!ProtocolFeature.KEYSPACE_ON_QUERY.isSupportedBy(err.serverProtocolVersion)
&& prepareKeyspace != null
&& (currentKeyspace == null || !currentKeyspace.equals(prepareKeyspace))) {
// This shouldn't happen in normal use, because a user shouldn't try to execute
// a prepared statement with the wrong keyspace set.
// Fail fast (we can't change the keyspace to reprepare, because we're using a
// pooled connection
// that's shared with other requests).
connection.release();
throw new IllegalStateException(
String.format(
"Statement '%s' was prepared on"
+ " keyspace '%s' which is not the same as the session keyspace"
+ " '%s'. This is not supported with protocol version %s",
toPrepare.getQueryString(),
toPrepare.getQueryKeyspace(),
connection.keyspace(),
err.serverProtocolVersion));
}
logger.info(
"Query '{}' on keyspace '{}' is not prepared on {}, preparing before retrying executing. "
+ "Seeing this message a few times is fine, but seeing it a lot may be source of performance problems",
toPrepare.getQueryString(),
toPrepare.getQueryKeyspace(),
connection.endPoint);
write(connection, prepareAndRetry(toPrepare));
// we're done for now, the prepareAndRetry callback will handle the rest
return;
case READ_FAILURE:
assert exceptionToReport instanceof ReadFailureException;
connection.release();
retry =
computeRetryDecisionOnRequestError((ReadFailureException) exceptionToReport);
break;
case WRITE_FAILURE:
assert exceptionToReport instanceof WriteFailureException;
connection.release();
if (statement.isIdempotentWithDefault(
manager.cluster.getConfiguration().getQueryOptions())) {
retry =
computeRetryDecisionOnRequestError((WriteFailureException) exceptionToReport);
} else {
retry = RetryPolicy.RetryDecision.rethrow();
}
break;
default:
connection.release();
if (metricsEnabled()) metrics().getErrorMetrics().getOthers().inc();
break;
}
if (retry == null) setFinalResult(connection, response);
else {
processRetryDecision(retry, connection, exceptionToReport);
}
break;
default:
connection.release();
setFinalResult(connection, response);
break;
}
} catch (Exception e) {
exceptionToReport = e;
setFinalException(connection, e);
} finally {
if (queriedHost != null && statement != Statement.DEFAULT) {
manager.cluster.manager.reportQuery(queriedHost, statement, exceptionToReport, latency);
}
}
}
private Connection.ResponseCallback prepareAndRetry(final PreparedStatement toPrepare) {
// do not bother inspecting retry policy at this step, no other decision
// makes sense than retry on the same host if the query was prepared,
// or on another host, if an error/timeout occurred.
// The original request hasn't been executed so far, so there is no risk
// of re-executing non-idempotent statements.
return new Connection.ResponseCallback() {
@Override
public Message.Request request() {
Requests.Prepare request =
new Requests.Prepare(toPrepare.getQueryString(), toPrepare.getQueryKeyspace());
// propagate the original custom payload in the prepare request
request.setCustomPayload(statement.getOutgoingPayload());
return request;
}
@Override
public int retryCount() {
return SpeculativeExecution.this.retryCount();
}
@Override
public void onSet(
Connection connection, Message.Response response, long latency, int retryCount) {
QueryState queryState = queryStateRef.get();
if (!queryState.isInProgressAt(retryCount)
|| !queryStateRef.compareAndSet(queryState, queryState.complete())) {
logger.debug(
"onSet triggered but the response was completed by another thread, cancelling (retryCount = {}, queryState = {}, queryStateRef = {})",
retryCount,
queryState,
queryStateRef.get());
return;
}
connection.release();
switch (response.type) {
case RESULT:
if (((Responses.Result) response).kind == Responses.Result.Kind.PREPARED) {
logger.debug("Scheduling retry now that query is prepared");
retry(true, null);
} else {
logError(
connection.endPoint,
new DriverException("Got unexpected response to prepare message: " + response));
retry(false, null);
}
break;
case ERROR:
logError(
connection.endPoint,
new DriverException("Error preparing query, got " + response));
if (metricsEnabled()) metrics().getErrorMetrics().getOthers().inc();
retry(false, null);
break;
default:
// Something's wrong, so we return but we let setFinalResult propagate the exception
SpeculativeExecution.this.setFinalResult(connection, response);
break;
}
}
@Override
public void onException(
Connection connection, Exception exception, long latency, int retryCount) {
SpeculativeExecution.this.onException(connection, exception, latency, retryCount);
}
@Override
public boolean onTimeout(Connection connection, long latency, int retryCount) {
QueryState queryState = queryStateRef.get();
if (!queryState.isInProgressAt(retryCount)
|| !queryStateRef.compareAndSet(queryState, queryState.complete())) {
logger.debug(
"onTimeout triggered but the response was completed by another thread, cancelling (retryCount = {}, queryState = {}, queryStateRef = {})",
retryCount,
queryState,
queryStateRef.get());
return false;
}
connection.release();
logError(
connection.endPoint,
new OperationTimedOutException(
connection.endPoint, "Timed out waiting for response to PREPARE message"));
retry(false, null);
return true;
}
};
}
@Override
public void onException(
Connection connection, Exception exception, long latency, int retryCount) {
QueryState queryState = queryStateRef.get();
if (!queryState.isInProgressAt(retryCount)
|| !queryStateRef.compareAndSet(queryState, queryState.complete())) {
logger.debug(
"onException triggered but the response was completed by another thread, cancelling (retryCount = {}, queryState = {}, queryStateRef = {})",
retryCount,
queryState,
queryStateRef.get());
return;
}
Host queriedHost = current;
try {
connection.release();
if (exception instanceof ConnectionException) {
RetryPolicy.RetryDecision decision =
computeRetryDecisionOnRequestError((ConnectionException) exception);
processRetryDecision(decision, connection, exception);
return;
}
setFinalException(connection, exception);
} catch (Exception e) {
// This shouldn't happen, but if it does, we want to signal the callback, not let it hang
// indefinitely
setFinalException(
null,
new DriverInternalError(
"An unexpected error happened while handling exception " + exception, e));
} finally {
if (queriedHost != null && statement != Statement.DEFAULT)
manager.cluster.manager.reportQuery(queriedHost, statement, exception, latency);
}
}
@Override
public boolean onTimeout(Connection connection, long latency, int retryCount) {
QueryState queryState = queryStateRef.get();
if (!queryState.isInProgressAt(retryCount)
|| !queryStateRef.compareAndSet(queryState, queryState.complete())) {
logger.debug(
"onTimeout triggered but the response was completed by another thread, cancelling (retryCount = {}, queryState = {}, queryStateRef = {})",
retryCount,
queryState,
queryStateRef.get());
return false;
}
Host queriedHost = current;
OperationTimedOutException timeoutException =
new OperationTimedOutException(
connection.endPoint, "Timed out waiting for server response");
try {
connection.release();
RetryPolicy.RetryDecision decision = computeRetryDecisionOnRequestError(timeoutException);
processRetryDecision(decision, connection, timeoutException);
} catch (Exception e) {
// This shouldn't happen, but if it does, we want to signal the callback, not let it hang
// indefinitely
setFinalException(
null,
new DriverInternalError("An unexpected error happened while handling timeout", e));
} finally {
if (queriedHost != null && statement != Statement.DEFAULT)
manager.cluster.manager.reportQuery(queriedHost, statement, timeoutException, latency);
}
return true;
}
@Override
public int retryCount() {
return queryStateRef.get().retryCount;
}
private void setFinalException(Connection connection, Exception exception) {
RequestHandler.this.setFinalException(this, connection, exception);
}
private void setFinalResult(Connection connection, Message.Response response) {
RequestHandler.this.setFinalResult(this, connection, response);
}
}
/**
* The state of a SpeculativeExecution.
*
* This is used to prevent races between request completion (either success or error) and
* timeout. A retry is in progress once we have written the request to the connection and until we
* get back a response (see onSet or onException) or a timeout (see onTimeout). The count
* increments on each retry.
*/
static class QueryState {
static final QueryState INITIAL = new QueryState(-1, false);
static final QueryState CANCELLED_WHILE_IN_PROGRESS = new QueryState(Integer.MIN_VALUE, false);
static final QueryState CANCELLED_WHILE_COMPLETE = new QueryState(Integer.MIN_VALUE + 1, false);
final int retryCount;
final boolean inProgress;
private QueryState(int count, boolean inProgress) {
this.retryCount = count;
this.inProgress = inProgress;
}
boolean isInProgressAt(int retryCount) {
return inProgress && this.retryCount == retryCount;
}
boolean isCompleteAt(int retryCount) {
return !inProgress && this.retryCount == retryCount;
}
QueryState complete() {
assert inProgress;
return new QueryState(retryCount, false);
}
QueryState startNext() {
assert !inProgress;
return new QueryState(retryCount + 1, true);
}
public boolean isCancelled() {
return this == CANCELLED_WHILE_IN_PROGRESS || this == CANCELLED_WHILE_COMPLETE;
}
@Override
public String toString() {
return String.format(
"QueryState(count=%d, inProgress=%s, cancelled=%s)",
retryCount, inProgress, isCancelled());
}
}
/**
* Wraps the iterator return by {@link com.datastax.driver.core.policies.LoadBalancingPolicy} to
* make it safe for concurrent access by multiple threads.
*/
static class QueryPlan {
private final Iterator iterator;
QueryPlan(Iterator iterator) {
this.iterator = iterator;
}
/** @return null if there are no more hosts */
synchronized Host next() {
return iterator.hasNext() ? iterator.next() : null;
}
}
}