Merge pull request #13 from 1inker/ow-fork-7.1

Bump to pre-release 7.1.21

CMakeLists.txt

@@ -23,7 +23,7 @@ else()
 endif()
 
 project(foundationdb
-  VERSION 7.1.19
+  VERSION 7.1.21
   DESCRIPTION "FoundationDB is a scalable, fault-tolerant, ordered key-value store with full ACID transactions."
   HOMEPAGE_URL "http://www.foundationdb.org/"
   LANGUAGES C CXX ASM)

contrib/transaction_profiling_analyzer/transaction_profiling_analyzer.py

@@ -277,6 +277,12 @@ def __init__(self, bb, protocol_version, full_output=True):
         if protocol_version >= PROTOCOL_VERSION_6_3:
             self.report_conflicting_keys = bb.get_bool()
 
+        if protocol_version >= PROTOCOL_VERSION_7_1:
+            lock_aware = bb.get_bool()
+            if bb.get_bool():
+                spanId = bb.get_bytes(16)
+
+
 class UnsupportedProtocolVersionError(Exception):
     def __init__(self, protocol_version):
         super().__init__("Unsupported protocol version 0x%0.2X" % protocol_version)

documentation/sphinx/source/release-notes/release-notes-710.rst

@@ -1,9 +1,32 @@
-.. _release-notes:
-
 #############
 Release Notes
 #############
 
+=======
+7.1.21
+======
+* Same as 7.1.20 release with AVX enabled.
+
+7.1.20
+======
+* Released with AVX disabled.
+* Fixed missing localities for fdbserver that can cause cross DC calls among storage servers. `(PR #7995) <https://github.com/apple/foundationdb/pull/7995>`_
+* Removed extremely spammy trace event in FetchKeys and fixed transaction_profiling_analyzer.py. `(PR #7934) <https://github.com/apple/foundationdb/pull/7934>`_
+* Fixed bugs when GRV proxy returns an error. `(PR #7860) <https://github.com/apple/foundationdb/pull/7860>`_
+
+7.1.19
+======
+* Same as 7.1.18 release with AVX enabled.
+
+7.1.18
+======
+* Released with AVX disabled.
+* Added knobs for the minimum and the maximum of the Ratekeeper's default priority. `(PR #7820) <https://github.com/apple/foundationdb/pull/7820>`_
+* Fixed bugs in ``getRange`` of the special key space. `(PR #7778) <https://github.com/apple/foundationdb/pull/7778>`_, `(PR #7720) <https://github.com/apple/foundationdb/pull/7720>`_
+* Added debug ID for secondary queries in index prefetching. `(PR #7755) <https://github.com/apple/foundationdb/pull/7755>`_
+* Changed hostname resolving to prefer IPv6 addresses. `(PR #7750) <https://github.com/apple/foundationdb/pull/7750>`_
+* Added more transaction debug events for prefetch queries. `(PR #7732) <https://github.com/apple/foundationdb/pull/7732>`_
+
 7.1.17
 ======
 * Same as 7.1.16 release with AVX enabled.
@@ -17,7 +40,7 @@ Release Notes
 * Fixed ScopeEventFieldTypeMismatch error for TLogMetrics. `(PR #7640) <https://github.com/apple/foundationdb/pull/7640>`_
 * Added getMappedRange latency metrics. `(PR #7632) <https://github.com/apple/foundationdb/pull/7632>`_
 * Fixed a version vector performance bug due to not updating client side tag cache. `(PR #7616) <https://github.com/apple/foundationdb/pull/7616>`_
-* Fixed DiskReadSeconds and DiskWriteSeconds calculaion in ProcessMetrics. `(PR #7609) <https://github.com/apple/foundationdb/pull/7609>`_
+* Fixed DiskReadSeconds and DiskWriteSeconds calculation in ProcessMetrics. `(PR #7609) <https://github.com/apple/foundationdb/pull/7609>`_
 * Added Rocksdb compression and data size stats. `(PR #7596) <https://github.com/apple/foundationdb/pull/7596>`_
 
 7.1.15
@@ -76,7 +99,7 @@ Release Notes
 * Added support of the reboot command in go bindings. `(PR #7270) <https://github.com/apple/foundationdb/pull/7270>`_
 * Fixed several issues in profiling special keys using GlobalConfig. `(PR #7120) <https://github.com/apple/foundationdb/pull/7120>`_
 * Fixed a stuck transaction system bug due to inconsistent recovery transaction version. `(PR #7261) <https://github.com/apple/foundationdb/pull/7261>`_
-* Fixed a unknown_error crash due to not resolving hostnames. `(PR #7254) <https://github.com/apple/foundationdb/pull/7254>`_
+* Fixed an unknown_error crash due to not resolving hostnames. `(PR #7254) <https://github.com/apple/foundationdb/pull/7254>`_
 * Fixed a heap-use-after-free bug. `(PR #7250) <https://github.com/apple/foundationdb/pull/7250>`_
 * Fixed a performance issue that remote TLogs are sending too many pops to log routers. `(PR #7235) <https://github.com/apple/foundationdb/pull/7235>`_
 * Fixed an issue that SharedTLogs are not displaced and leaking disk space. `(PR #7246) <https://github.com/apple/foundationdb/pull/7246>`_

fdbclient/ClientKnobs.cpp

@@ -63,6 +63,7 @@ void ClientKnobs::initialize(Randomize randomize) {
 
 	init( WRONG_SHARD_SERVER_DELAY,                .01 ); if( randomize && BUGGIFY ) WRONG_SHARD_SERVER_DELAY = deterministicRandom()->random01(); // FLOW_KNOBS->PREVENT_FAST_SPIN_DELAY; // SOMEDAY: This delay can limit performance of retrieving data when the cache is mostly wrong (e.g. dumping the database after a test)
 	init( FUTURE_VERSION_RETRY_DELAY,              .01 ); if( randomize && BUGGIFY ) FUTURE_VERSION_RETRY_DELAY = deterministicRandom()->random01();// FLOW_KNOBS->PREVENT_FAST_SPIN_DELAY;
+	init( GRV_ERROR_RETRY_DELAY,                   5.0 ); if( randomize && BUGGIFY ) GRV_ERROR_RETRY_DELAY = 0.01 + 5 * deterministicRandom()->random01();
 	init( UNKNOWN_TENANT_RETRY_DELAY,              0.0 ); if( randomize && BUGGIFY ) UNKNOWN_TENANT_RETRY_DELAY = deterministicRandom()->random01();
 	init( REPLY_BYTE_LIMIT,                      80000 );
 	init( DEFAULT_BACKOFF,                         .01 ); if( randomize && BUGGIFY ) DEFAULT_BACKOFF = deterministicRandom()->random01();

fdbclient/ClientKnobs.h

@@ -61,6 +61,7 @@ class ClientKnobs : public KnobsImpl<ClientKnobs> {
 	double WRONG_SHARD_SERVER_DELAY; // SOMEDAY: This delay can limit performance of retrieving data when the cache is
 	                                 // mostly wrong (e.g. dumping the database after a test)
 	double FUTURE_VERSION_RETRY_DELAY;
+	double GRV_ERROR_RETRY_DELAY;
 	double UNKNOWN_TENANT_RETRY_DELAY;
 	int REPLY_BYTE_LIMIT;
 	double DEFAULT_BACKOFF;

fdbclient/DatabaseContext.h

@@ -528,6 +528,7 @@ class DatabaseContext : public ReferenceCounted<DatabaseContext>, public FastAll
 	Counter transactionsExpensiveClearCostEstCount;
 	Counter transactionGrvFullBatches;
 	Counter transactionGrvTimedOutBatches;
+	Counter transactionCommitVersionNotFoundForSS;
 
 	ContinuousSample<double> latencies, readLatencies, commitLatencies, GRVLatencies, mutationsPerCommit,
 	    bytesPerCommit, bgLatencies, bgGranulesPerRequest;

fdbclient/NativeAPI.actor.cpp

@@ -240,7 +240,7 @@ void DatabaseContext::getLatestCommitVersions(const Reference<LocationInfo>& loc
 		return;
 	}
 
-	if (ssVersionVectorCache.getMaxVersion() != invalidVersion && readVersion > ssVersionVectorCache.getMaxVersion()) {
+	if (readVersion > ssVersionVectorCache.getMaxVersion()) {
 		if (!CLIENT_KNOBS->FORCE_GRV_CACHE_OFF && !info->options.skipGrvCache && info->options.useGrvCache) {
 			return;
 		} else {
@@ -253,16 +253,32 @@ void DatabaseContext::getLatestCommitVersions(const Reference<LocationInfo>& loc
 
 	std::map<Version, std::set<Tag>> versionMap; // order the versions to be returned
 	for (int i = 0; i < locationInfo->locations()->size(); i++) {
-		UID uid = locationInfo->locations()->getId(i);
-		if (ssidTagMapping.find(uid) != ssidTagMapping.end()) {
-			Tag tag = ssidTagMapping[uid];
+		bool updatedVersionMap = false;
+		Version commitVersion = invalidVersion;
+		Tag tag = invalidTag;
+		auto iter = ssidTagMapping.find(locationInfo->locations()->getId(i));
+		if (iter != ssidTagMapping.end()) {
+			tag = iter->second;
 			if (ssVersionVectorCache.hasVersion(tag)) {
-				Version commitVersion = ssVersionVectorCache.getVersion(tag); // latest commit version
+				commitVersion = ssVersionVectorCache.getVersion(tag); // latest commit version
 				if (commitVersion < readVersion) {
+					updatedVersionMap = true;
 					versionMap[commitVersion].insert(tag);
 				}
 			}
 		}
+		// do not log if commitVersion == readVersion as it is common.
+		if (!updatedVersionMap && commitVersion != readVersion) {
+			TraceEvent(SevDebug, "CommitVersionNotFoundForSS")
+			    .detail("InSSIDMap", iter != ssidTagMapping.end() ? 1 : 0)
+			    .detail("Tag", tag)
+			    .detail("CommitVersion", commitVersion)
+			    .detail("ReadVersion", readVersion)
+			    .detail("VersionVector", ssVersionVectorCache.toString())
+			    .setMaxEventLength(11000)
+			    .setMaxFieldLength(10000);
+			++transactionCommitVersionNotFoundForSS;
+		}
 	}
 
 	// insert the commit versions in the version vector.
@@ -1445,13 +1461,13 @@ DatabaseContext::DatabaseContext(Reference<AsyncVar<Reference<IClusterConnection
     transactionsProcessBehind("ProcessBehind", cc), transactionsThrottled("Throttled", cc),
     transactionsExpensiveClearCostEstCount("ExpensiveClearCostEstCount", cc),
     transactionGrvFullBatches("NumGrvFullBatches", cc), transactionGrvTimedOutBatches("NumGrvTimedOutBatches", cc),
-    latencies(1000), readLatencies(1000), commitLatencies(1000), GRVLatencies(1000), mutationsPerCommit(1000),
-    bytesPerCommit(1000), bgLatencies(1000), bgGranulesPerRequest(1000), outstandingWatches(0), sharedStatePtr(nullptr),
-    lastGrvTime(0.0), cachedReadVersion(0), lastRkBatchThrottleTime(0.0), lastRkDefaultThrottleTime(0.0),
-    lastProxyRequestTime(0.0), transactionTracingSample(false), taskID(taskID), clientInfo(clientInfo),
-    clientInfoMonitor(clientInfoMonitor), coordinator(coordinator), apiVersion(apiVersion), mvCacheInsertLocation(0),
-    healthMetricsLastUpdated(0), detailedHealthMetricsLastUpdated(0),
-    smoothMidShardSize(CLIENT_KNOBS->SHARD_STAT_SMOOTH_AMOUNT),
+    transactionCommitVersionNotFoundForSS("CommitVersionNotFoundForSS", cc), latencies(1000), readLatencies(1000),
+    commitLatencies(1000), GRVLatencies(1000), mutationsPerCommit(1000), bytesPerCommit(1000), bgLatencies(1000),
+    bgGranulesPerRequest(1000), outstandingWatches(0), sharedStatePtr(nullptr), lastGrvTime(0.0), cachedReadVersion(0),
+    lastRkBatchThrottleTime(0.0), lastRkDefaultThrottleTime(0.0), lastProxyRequestTime(0.0),
+    transactionTracingSample(false), taskID(taskID), clientInfo(clientInfo), clientInfoMonitor(clientInfoMonitor),
+    coordinator(coordinator), apiVersion(apiVersion), mvCacheInsertLocation(0), healthMetricsLastUpdated(0),
+    detailedHealthMetricsLastUpdated(0), smoothMidShardSize(CLIENT_KNOBS->SHARD_STAT_SMOOTH_AMOUNT),
     specialKeySpace(std::make_unique<SpecialKeySpace>(specialKeys.begin, specialKeys.end, /* test */ false)),
     connectToDatabaseEventCacheHolder(format("ConnectToDatabase/%s", dbId.toString().c_str())) {
 
@@ -1723,8 +1739,9 @@ DatabaseContext::DatabaseContext(const Error& err)
     transactionsProcessBehind("ProcessBehind", cc), transactionsThrottled("Throttled", cc),
     transactionsExpensiveClearCostEstCount("ExpensiveClearCostEstCount", cc),
     transactionGrvFullBatches("NumGrvFullBatches", cc), transactionGrvTimedOutBatches("NumGrvTimedOutBatches", cc),
-    latencies(1000), readLatencies(1000), commitLatencies(1000), GRVLatencies(1000), mutationsPerCommit(1000),
-    bytesPerCommit(1000), bgLatencies(1000), bgGranulesPerRequest(1000), transactionTracingSample(false),
+    transactionCommitVersionNotFoundForSS("CommitVersionNotFoundForSS", cc), latencies(1000), readLatencies(1000),
+    commitLatencies(1000), GRVLatencies(1000), mutationsPerCommit(1000), bytesPerCommit(1000), bgLatencies(1000),
+    bgGranulesPerRequest(1000), transactionTracingSample(false),
     smoothMidShardSize(CLIENT_KNOBS->SHARD_STAT_SMOOTH_AMOUNT),
     connectToDatabaseEventCacheHolder(format("ConnectToDatabase/%s", dbId.toString().c_str())) {}
 
@@ -3460,8 +3477,8 @@ ACTOR Future<Key> getKey(Reference<TransactionState> trState,
 
 ACTOR Future<Version> waitForCommittedVersion(Database cx, Version version, SpanID spanContext) {
 	state Span span("NAPI:waitForCommittedVersion"_loc, { spanContext });
-	try {
-		loop {
+	loop {
+		try {
 			choose {
 				when(wait(cx->onProxiesChanged())) {}
 				when(GetReadVersionReply v = wait(basicLoadBalance(
@@ -3487,10 +3504,16 @@ ACTOR Future<Version> waitForCommittedVersion(Database cx, Version version, Span
 					wait(delay(CLIENT_KNOBS->FUTURE_VERSION_RETRY_DELAY, cx->taskID));
 				}
 			}
+		} catch (Error& e) {
+			if (e.code() == error_code_batch_transaction_throttled ||
+			    e.code() == error_code_proxy_memory_limit_exceeded) {
+				// GRV Proxy returns an error
+				wait(delayJittered(CLIENT_KNOBS->GRV_ERROR_RETRY_DELAY));
+			} else {
+				TraceEvent(SevError, "WaitForCommittedVersionError").error(e);
+				throw;
+			}
 		}
-	} catch (Error& e) {
-		TraceEvent(SevError, "WaitForCommittedVersionError").error(e);
-		throw;
 	}
 }
 
@@ -6632,9 +6655,12 @@ ACTOR Future<GetReadVersionReply> getConsistentReadVersion(SpanID parentSpan,
 				}
 			}
 		} catch (Error& e) {
-			if (e.code() != error_code_broken_promise && e.code() != error_code_batch_transaction_throttled)
+			if (e.code() != error_code_broken_promise && e.code() != error_code_batch_transaction_throttled &&
+			    e.code() != error_code_proxy_memory_limit_exceeded)
 				TraceEvent(SevError, "GetConsistentReadVersionError").error(e);
-			if (e.code() == error_code_batch_transaction_throttled && !cx->apiVersionAtLeast(630)) {
+			if ((e.code() == error_code_batch_transaction_throttled ||
+			     e.code() == error_code_proxy_memory_limit_exceeded) &&
+			    !cx->apiVersionAtLeast(630)) {
 				wait(delayJittered(5.0));
 			} else {
 				throw;

fdbrpc/LoadBalance.actor.h

@@ -486,6 +486,12 @@ Future<REPLY_TYPE(Request)> loadBalance(
 				// server count is within "LOAD_BALANCE_MAX_BAD_OPTIONS". We
 				// do not need to consider any remote servers.
 				break;
+			} else if (badServers == alternatives->countBest() && i == badServers) {
+				TraceEvent("AllLocalAlternativesFailed")
+				    .suppressFor(1.0)
+				    .detail("Alternatives", alternatives->description())
+				    .detail("Total", alternatives->size())
+				    .detail("Best", alternatives->countBest());
 			}
 
 			RequestStream<Request> const* thisStream = &alternatives->get(i, channel);
@@ -587,6 +593,7 @@ Future<REPLY_TYPE(Request)> loadBalance(
 		// nextAlt. This logic matters only if model == nullptr. Otherwise, the
 		// bestAlt and nextAlt have been decided.
 		state RequestStream<Request> const* stream = nullptr;
+		state LBDistance::Type distance;
 		for (int alternativeNum = 0; alternativeNum < alternatives->size(); alternativeNum++) {
 			int useAlt = nextAlt;
 			if (nextAlt == startAlt)
@@ -595,13 +602,15 @@ Future<REPLY_TYPE(Request)> loadBalance(
 				useAlt = (nextAlt + alternatives->size() - 1) % alternatives->size();
 
 			stream = &alternatives->get(useAlt, channel);
+			distance = alternatives->getDistance(useAlt);
 			if (!IFailureMonitor::failureMonitor().getState(stream->getEndpoint()).failed &&
 			    (!firstRequestEndpoint.present() || stream->getEndpoint().token.first() != firstRequestEndpoint.get()))
 				break;
 			nextAlt = (nextAlt + 1) % alternatives->size();
 			if (nextAlt == startAlt)
 				triedAllOptions = TriedAllOptions::True;
 			stream = nullptr;
+			distance = LBDistance::DISTANT;
 		}
 
 		if (!stream && !firstRequestData.isValid()) {
@@ -637,6 +646,18 @@ Future<REPLY_TYPE(Request)> loadBalance(
 			firstRequestEndpoint = Optional<uint64_t>();
 		} else if (firstRequestData.isValid()) {
 			// Issue a second request, the first one is taking a long time.
+			if (distance == LBDistance::DISTANT) {
+				TraceEvent("LBDistant2nd")
+				    .suppressFor(0.1)
+				    .detail("Distance", (int)distance)
+				    .detail("BackOff", backoff)
+				    .detail("TriedAllOptions", triedAllOptions)
+				    .detail("Alternatives", alternatives->description())
+				    .detail("Token", stream->getEndpoint().token)
+				    .detail("Total", alternatives->size())
+				    .detail("Best", alternatives->countBest())
+				    .detail("Attempts", numAttempts);
+			}
 			secondRequestData.startRequest(backoff, triedAllOptions, stream, request, model, alternatives, channel);
 			state bool firstFinished = false;
 
@@ -666,6 +687,18 @@ Future<REPLY_TYPE(Request)> loadBalance(
 			}
 		} else {
 			// Issue a request, if it takes too long to get a reply, go around the loop
+			if (distance == LBDistance::DISTANT) {
+				TraceEvent("LBDistant")
+				    .suppressFor(0.1)
+				    .detail("Distance", (int)distance)
+				    .detail("BackOff", backoff)
+				    .detail("TriedAllOptions", triedAllOptions)
+				    .detail("Alternatives", alternatives->description())
+				    .detail("Token", stream->getEndpoint().token)
+				    .detail("Total", alternatives->size())
+				    .detail("Best", alternatives->countBest())
+				    .detail("Attempts", numAttempts);
+			}
 			firstRequestData.startRequest(backoff, triedAllOptions, stream, request, model, alternatives, channel);
 			firstRequestEndpoint = stream->getEndpoint().token.first();
 

fdbrpc/MultiInterface.h

@@ -226,6 +226,7 @@ class MultiInterface<ReferencedInterface<T>> : public ReferenceCounted<MultiInte
 	}
 
 	T const& getInterface(int index) { return alternatives[index]->interf; }
+	LBDistance::Type getDistance(int index) const { return (LBDistance::Type)alternatives[index]->distance; }
 	UID getId(int index) const { return alternatives[index]->interf.id(); }
 	bool hasInterface(UID id) const {
 		for (const auto& ref : alternatives) {

fdbserver/BackupWorker.actor.cpp

@@ -428,20 +428,30 @@ struct BackupData {
 	ACTOR static Future<Version> _getMinKnownCommittedVersion(BackupData* self) {
 		state Span span("BA:GetMinCommittedVersion"_loc);
 		loop {
-			GetReadVersionRequest request(span.context,
-			                              0,
-			                              TransactionPriority::DEFAULT,
-			                              invalidVersion,
-			                              GetReadVersionRequest::FLAG_USE_MIN_KNOWN_COMMITTED_VERSION);
-			choose {
-				when(wait(self->cx->onProxiesChanged())) {}
-				when(GetReadVersionReply reply =
-				         wait(basicLoadBalance(self->cx->getGrvProxies(UseProvisionalProxies::False),
-				                               &GrvProxyInterface::getConsistentReadVersion,
-				                               request,
-				                               self->cx->taskID))) {
-					self->cx->ssVersionVectorCache.applyDelta(reply.ssVersionVectorDelta);
-					return reply.version;
+			try {
+				GetReadVersionRequest request(span.context,
+				                              0,
+				                              TransactionPriority::DEFAULT,
+				                              invalidVersion,
+				                              GetReadVersionRequest::FLAG_USE_MIN_KNOWN_COMMITTED_VERSION);
+				choose {
+					when(wait(self->cx->onProxiesChanged())) {}
+					when(GetReadVersionReply reply =
+					         wait(basicLoadBalance(self->cx->getGrvProxies(UseProvisionalProxies::False),
+					                               &GrvProxyInterface::getConsistentReadVersion,
+					                               request,
+					                               self->cx->taskID))) {
+						self->cx->ssVersionVectorCache.applyDelta(reply.ssVersionVectorDelta);
+						return reply.version;
+					}
+				}
+			} catch (Error& e) {
+				if (e.code() == error_code_batch_transaction_throttled ||
+				    e.code() == error_code_proxy_memory_limit_exceeded) {
+					// GRV Proxy returns an error
+					wait(delayJittered(CLIENT_KNOBS->GRV_ERROR_RETRY_DELAY));
+				} else {
+					throw;
 				}
 			}
 		}

fdbserver/GrvProxyServer.actor.cpp

@@ -29,6 +29,7 @@
 #include "fdbserver/WaitFailure.h"
 #include "fdbserver/WorkerInterface.actor.h"
 #include "fdbrpc/sim_validation.h"
+#include "flow/IRandom.h"
 #include "flow/flow.h"
 #include "flow/actorcompiler.h" // This must be the last #include.
 
@@ -425,7 +426,9 @@ ACTOR Future<Void> queueGetReadVersionRequests(Reference<AsyncVar<ServerDBInfo>
 			// WARNING: this code is run at a high priority, so it needs to do as little work as possible
 			bool canBeQueued = true;
 			if (stats->txnRequestIn.getValue() - stats->txnRequestOut.getValue() >
-			    SERVER_KNOBS->START_TRANSACTION_MAX_QUEUE_SIZE) {
+			        SERVER_KNOBS->START_TRANSACTION_MAX_QUEUE_SIZE ||
+			    (g_network->isSimulated() && !g_simulator.speedUpSimulation &&
+			     deterministicRandom()->random01() < 0.01)) {
 				// When the limit is hit, try to drop requests from the lower priority queues.
 				if (req.priority == TransactionPriority::BATCH) {
 					canBeQueued = false;