wbtiger
/
foundationdb
mirror of https://github.com/apple/foundationdb.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge commit 'f9581de2005e6b085776e81b9fcaa16442b32589' into merge-6.2-to-6.3 

# Conflicts:
#	fdbserver/Knobs.cpp
#	fdbserver/Knobs.h
This commit is contained in:
Steve Atherton 2020-10-27 12:21:26 -07:00
parent 585fdadc8c f9581de200
commit 99c1880a83
8 changed files with 223 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
build/Dockerfile
View File

@ -8,7 +8,7 @@ RUN yum install -y yum-utils &&\
http://opensource.wandisco.com/centos/6/git/x86_64/wandisco-git-release-6-1.noarch.rpm &&\
yum -y install devtoolset-8-8.1-1.el6 java-1.8.0-openjdk-devel \
devtoolset-8-gcc-8.3.1 devtoolset-8-gcc-c++-8.3.1 \
devtoolset-8-libubsan-devel devtoolset-8-valgrind-devel \
devtoolset-8-libubsan-devel devtoolset-8-libasan-devel devtoolset-8-valgrind-devel \
rh-python36-python-devel rh-ruby24 golang python27 rpm-build \
mono-core debbuild python-pip dos2unix valgrind-devel ccache \
distcc wget git lz4 lz4-devel lz4-static &&\
@ -61,8 +61,8 @@ RUN cd /opt/ && curl -L https://github.com/facebook/rocksdb/archive/v6.10.1.tar.
ARG TIMEZONEINFO=America/Los_Angeles
RUN rm -f /etc/localtime && ln -s /usr/share/zoneinfo/${TIMEZONEINFO} /etc/localtime
LABEL version=0.1.18
ENV DOCKER_IMAGEVER=0.1.18
LABEL version=0.1.19
ENV DOCKER_IMAGEVER=0.1.19
ENV JAVA_HOME=/usr/lib/jvm/java-1.8.0
ENV CC=/opt/rh/devtoolset-8/root/usr/bin/gcc
ENV CXX=/opt/rh/devtoolset-8/root/usr/bin/g++

6
build/Dockerfile.devel
View File

@ -1,4 +1,4 @@
FROM foundationdb/foundationdb-build:0.1.18
FROM foundationdb/foundationdb-build:0.1.19
USER root
@ -50,8 +50,8 @@ RUN cp -iv /usr/local/bin/clang++ /usr/local/bin/clang++.deref &&\
ldconfig &&\
rm -rf /mnt/artifacts
LABEL version=0.11.9
ENV DOCKER_IMAGEVER=0.11.9
LABEL version=0.11.10
ENV DOCKER_IMAGEVER=0.11.10
ENV CLANGCC=/usr/local/bin/clang.de8a65ef
ENV CLANGCXX=/usr/local/bin/clang++.de8a65ef

2
build/docker-compose.yaml
View File

@ -2,7 +2,7 @@ version: "3"
services:
common: &common
image: foundationdb/foundationdb-build:0.1.18
image: foundationdb/foundationdb-build:0.1.19
build-setup: &build-setup
<<: *common

5
documentation/sphinx/source/release-notes/release-notes-620.rst
View File

@ -2,6 +2,11 @@
Release Notes
#############
6.2.28
======
* Log detailed team collection information when median available space ratio of all teams is too low. `(PR #3912) <https://github.com/apple/foundationdb/pull/3912>`_
6.2.27
======
* For clusters with a large number of shards, avoid slow tasks in the data distributor by adding yields to the shard map destruction. `(PR #3834) <https://github.com/apple/foundationdb/pull/3834>`_

7
fdbclient/BlobStore.actor.cpp
View File

@ -951,7 +951,12 @@ Future<std::vector<std::string>> BlobStoreEndpoint::listBuckets() {
std::string BlobStoreEndpoint::hmac_sha1(std::string const &msg) {
std::string key = secret;
// First pad the key to 64 bytes.
// Hash key to shorten it if it is longer than SHA1 block size
if(key.size() > 64) {
key = SHA1::from_string(key);
}
// Pad key up to SHA1 block size if needed
key.append(64 - key.size(), '\0');
std::string kipad = key;

201
fdbserver/DataDistribution.actor.cpp
View File

@ -615,6 +615,8 @@ struct DDTeamCollection : ReferenceCounted<DDTeamCollection> {
int lowestUtilizationTeam;
int highestUtilizationTeam;
AsyncTrigger printDetailedTeamsInfo;
void resetLocalitySet() {
storageServerSet = Reference<LocalitySet>(new LocalityMap<UID>());
LocalityMap<UID>* storageServerMap = (LocalityMap<UID>*) storageServerSet.getPtr();
@ -794,6 +796,13 @@ struct DDTeamCollection : ReferenceCounted<DDTeamCollection> {
} else {
self->medianAvailableSpace = SERVER_KNOBS->MIN_AVAILABLE_SPACE_RATIO;
}
if (self->medianAvailableSpace < SERVER_KNOBS->TARGET_AVAILABLE_SPACE_RATIO) {
TraceEvent(SevWarn, "DDTeamMedianAvailableSpaceTooSmall", self->distributorId)
.detail("MedianAvailableSpaceRatio", self->medianAvailableSpace)
.detail("TargetAvailableSpaceRatio", SERVER_KNOBS->TARGET_AVAILABLE_SPACE_RATIO)
.detail("Primary", self->primary);
self->printDetailedTeamsInfo.trigger();
}
}
bool foundSrc = false;
@ -2708,6 +2717,196 @@ ACTOR Future<Void> waitUntilHealthy(DDTeamCollection* self, double extraDelay =
}
}
// Take a snapshot of necessary data structures from `DDTeamCollection` and print them out with yields to avoid slow
// task on the run loop.
ACTOR Future<Void> printSnapshotTeamsInfo(Reference<DDTeamCollection> self) {
state DatabaseConfiguration configuration;
state std::map<UID, Reference<TCServerInfo>> server_info;
state std::map<UID, ServerStatus> server_status;
state vector<Reference<TCTeamInfo>> teams;
state std::map<Standalone<StringRef>, Reference<TCMachineInfo>> machine_info;
state std::vector<Reference<TCMachineTeamInfo>> machineTeams;
// state std::vector<std::string> internedLocalityRecordKeyNameStrings;
// state int machineLocalityMapEntryArraySize;
// state std::vector<Reference<LocalityRecord>> machineLocalityMapRecordArray;
state int traceEventsPrinted = 0;
state std::vector<const UID*> serverIDs;
state double lastPrintTime = 0;
loop {
wait(self->printDetailedTeamsInfo.onTrigger());
if (now() - lastPrintTime < SERVER_KNOBS->DD_TEAMS_INFO_PRINT_INTERVAL) {
continue;
}
lastPrintTime = now();
traceEventsPrinted = 0;
double snapshotStart = now();
configuration = self->configuration;
server_info = self->server_info;
teams = self->teams;
machine_info = self->machine_info;
machineTeams = self->machineTeams;
// internedLocalityRecordKeyNameStrings = self->machineLocalityMap._keymap->_lookuparray;
// machineLocalityMapEntryArraySize = self->machineLocalityMap.size();
// machineLocalityMapRecordArray = self->machineLocalityMap.getRecordArray();
std::vector<const UID*> _uids = self->machineLocalityMap.getObjects();
serverIDs = _uids;
auto const& keys = self->server_status.getKeys();
for (auto const& key : keys) {
server_status.emplace(key, self->server_status.get(key));
}
TraceEvent("DDPrintSnapshotTeasmInfo", self->distributorId)
.detail("SnapshotSpeed", now() - snapshotStart)
.detail("Primary", self->primary);
// Print to TraceEvents
TraceEvent("DDConfig", self->distributorId)
.detail("StorageTeamSize", configuration.storageTeamSize)
.detail("DesiredTeamsPerServer", SERVER_KNOBS->DESIRED_TEAMS_PER_SERVER)
.detail("MaxTeamsPerServer", SERVER_KNOBS->MAX_TEAMS_PER_SERVER)
.detail("Primary", self->primary);
TraceEvent("ServerInfo", self->distributorId)
.detail("Size", server_info.size())
.detail("Primary", self->primary);
state int i;
state std::map<UID, Reference<TCServerInfo>>::iterator server = server_info.begin();
for (i = 0; i < server_info.size(); i++) {
TraceEvent("ServerInfo", self->distributorId)
.detail("ServerInfoIndex", i)
.detail("ServerID", server->first.toString())
.detail("ServerTeamOwned", server->second->teams.size())
.detail("MachineID", server->second->machine->machineID.contents().toString())
.detail("Primary", self->primary);
server++;
if (++traceEventsPrinted % SERVER_KNOBS->DD_TEAMS_INFO_PRINT_YIELD_COUNT == 0) {
wait(yield());
}
}
server = server_info.begin();
for (i = 0; i < server_info.size(); i++) {
const UID& uid = server->first;
TraceEvent("ServerStatus", self->distributorId)
.detail("ServerUID", uid)
.detail("Healthy", !server_status.at(uid).isUnhealthy())
.detail("MachineIsValid", server_info[uid]->machine.isValid())
.detail("MachineTeamSize",
server_info[uid]->machine.isValid() ? server_info[uid]->machine->machineTeams.size() : -1)
.detail("Primary", self->primary);
server++;
if (++traceEventsPrinted % SERVER_KNOBS->DD_TEAMS_INFO_PRINT_YIELD_COUNT == 0) {
wait(yield());
}
}
TraceEvent("ServerTeamInfo", self->distributorId).detail("Size", teams.size()).detail("Primary", self->primary);
for (i = 0; i < teams.size(); i++) {
const auto& team = teams[i];
TraceEvent("ServerTeamInfo", self->distributorId)
.detail("TeamIndex", i)
.detail("Healthy", team->isHealthy())
.detail("TeamSize", team->size())
.detail("MemberIDs", team->getServerIDsStr())
.detail("Primary", self->primary);
if (++traceEventsPrinted % SERVER_KNOBS->DD_TEAMS_INFO_PRINT_YIELD_COUNT == 0) {
wait(yield());
}
}
TraceEvent("MachineInfo", self->distributorId)
.detail("Size", machine_info.size())
.detail("Primary", self->primary);
state std::map<Standalone<StringRef>, Reference<TCMachineInfo>>::iterator machine = machine_info.begin();
state bool isMachineHealthy = false;
for (i = 0; i < machine_info.size(); i++) {
Reference<TCMachineInfo> _machine = machine->second;
if (!_machine.isValid() || machine_info.find(_machine->machineID) == machine_info.end() ||
_machine->serversOnMachine.empty()) {
isMachineHealthy = false;
}
// Healthy machine has at least one healthy server
for (auto& server : _machine->serversOnMachine) {
if (!server_status.at(server->id).isUnhealthy()) {
isMachineHealthy = true;
}
}
isMachineHealthy = false;
TraceEvent("MachineInfo", self->distributorId)
.detail("MachineInfoIndex", i)
.detail("Healthy", isMachineHealthy)
.detail("MachineID", machine->first.contents().toString())
.detail("MachineTeamOwned", machine->second->machineTeams.size())
.detail("ServerNumOnMachine", machine->second->serversOnMachine.size())
.detail("ServersID", machine->second->getServersIDStr())
.detail("Primary", self->primary);
machine++;
if (++traceEventsPrinted % SERVER_KNOBS->DD_TEAMS_INFO_PRINT_YIELD_COUNT == 0) {
wait(yield());
}
}
TraceEvent("MachineTeamInfo", self->distributorId)
.detail("Size", machineTeams.size())
.detail("Primary", self->primary);
for (i = 0; i < machineTeams.size(); i++) {
const auto& team = machineTeams[i];
TraceEvent("MachineTeamInfo", self->distributorId)
.detail("TeamIndex", i)
.detail("MachineIDs", team->getMachineIDsStr())
.detail("ServerTeams", team->serverTeams.size())
.detail("Primary", self->primary);
if (++traceEventsPrinted % SERVER_KNOBS->DD_TEAMS_INFO_PRINT_YIELD_COUNT == 0) {
wait(yield());
}
}
// TODO: re-enable the following logging or remove them.
// TraceEvent("LocalityRecordKeyName", self->distributorId)
// .detail("Size", internedLocalityRecordKeyNameStrings.size())
// .detail("Primary", self->primary);
// for (i = 0; i < internedLocalityRecordKeyNameStrings.size(); i++) {
// TraceEvent("LocalityRecordKeyIndexName", self->distributorId)
// .detail("KeyIndex", i)
// .detail("KeyName", internedLocalityRecordKeyNameStrings[i])
// .detail("Primary", self->primary);
// if (++traceEventsPrinted % SERVER_KNOBS->DD_TEAMS_INFO_PRINT_YIELD_COUNT == 0) {
// wait(yield());
// }
// }
// TraceEvent("MachineLocalityMap", self->distributorId)
// .detail("Size", machineLocalityMapEntryArraySize)
// .detail("Primary", self->primary);
// for (i = 0; i < serverIDs.size(); i++) {
// const auto& serverID = serverIDs[i];
// Reference<LocalityRecord> record = machineLocalityMapRecordArray[i];
// if (record.isValid()) {
// TraceEvent("MachineLocalityMap", self->distributorId)
// .detail("LocalityIndex", i)
// .detail("UID", serverID->toString())
// .detail("LocalityRecord", record->toString())
// .detail("Primary", self->primary);
// } else {
// TraceEvent("MachineLocalityMap", self->distributorId)
// .detail("LocalityIndex", i)
// .detail("UID", serverID->toString())
// .detail("LocalityRecord", "[NotFound]")
// .detail("Primary", self->primary);
// }
// if (++traceEventsPrinted % SERVER_KNOBS->DD_TEAMS_INFO_PRINT_YIELD_COUNT == 0) {
// wait(yield());
// }
// }
}
}
ACTOR Future<Void> removeBadTeams(DDTeamCollection* self) {
wait(self->initialFailureReactionDelay);
wait(waitUntilHealthy(self));
@ -4697,10 +4896,12 @@ ACTOR Future<Void> dataDistribution(Reference<DataDistributorData> self, Promise
teamCollectionsPtrs.push_back(remoteTeamCollection.getPtr());
remoteTeamCollection->teamCollections = teamCollectionsPtrs;
actors.push_back( reportErrorsExcept( dataDistributionTeamCollection( remoteTeamCollection, initData, tcis[1], self->dbInfo ), "DDTeamCollectionSecondary", self->ddId, &normalDDQueueErrors() ) );
actors.push_back(printSnapshotTeamsInfo(remoteTeamCollection));
}
primaryTeamCollection->teamCollections = teamCollectionsPtrs;
self->teamCollection = primaryTeamCollection.getPtr();
actors.push_back( reportErrorsExcept( dataDistributionTeamCollection( primaryTeamCollection, initData, tcis[0], self->dbInfo ), "DDTeamCollectionPrimary", self->ddId, &normalDDQueueErrors() ) );
actors.push_back(printSnapshotTeamsInfo(primaryTeamCollection));
actors.push_back(yieldPromiseStream(output.getFuture(), input));
wait( waitForAll( actors ) );

2
fdbserver/Knobs.cpp
View File

@ -244,6 +244,8 @@ void ServerKnobs::initialize(bool randomize, ClientKnobs* clientKnobs, bool isSi
init( DD_SS_FAILURE_VERSIONLAG, 250000000 );
init( DD_SS_ALLOWED_VERSIONLAG, 200000000 ); if( randomize && BUGGIFY ) { DD_SS_FAILURE_VERSIONLAG = deterministicRandom()->randomInt(15000000, 500000000); DD_SS_ALLOWED_VERSIONLAG = 0.75 * DD_SS_FAILURE_VERSIONLAG; }
init( DD_SS_STUCK_TIME_LIMIT, 300.0 ); if( randomize && BUGGIFY ) { DD_SS_STUCK_TIME_LIMIT = 200.0 + deterministicRandom()->random01() * 100.0; }
init( DD_TEAMS_INFO_PRINT_INTERVAL, 60 ); if( randomize && BUGGIFY ) DD_TEAMS_INFO_PRINT_INTERVAL = 10;
init( DD_TEAMS_INFO_PRINT_YIELD_COUNT, 100 ); if( randomize && BUGGIFY ) DD_TEAMS_INFO_PRINT_YIELD_COUNT = deterministicRandom()->random01() * 1000 + 1;
// TeamRemover
init( TR_FLAG_DISABLE_MACHINE_TEAM_REMOVER, false ); if( randomize && BUGGIFY ) TR_FLAG_DISABLE_MACHINE_TEAM_REMOVER = deterministicRandom()->random01() < 0.1 ? true : false; // false by default. disable the consistency check when it's true

2
fdbserver/Knobs.h
View File

@ -191,6 +191,8 @@ public:
int64_t DD_SS_FAILURE_VERSIONLAG; // Allowed SS version lag from the current read version before marking it as failed.
int64_t DD_SS_ALLOWED_VERSIONLAG; // SS will be marked as healthy if it's version lag goes below this value.
double DD_SS_STUCK_TIME_LIMIT; // If a storage server is not getting new versions for this amount of time, then it becomes undesired.
int DD_TEAMS_INFO_PRINT_INTERVAL;
int DD_TEAMS_INFO_PRINT_YIELD_COUNT;
// TeamRemover to remove redundant teams
bool TR_FLAG_DISABLE_MACHINE_TEAM_REMOVER; // disable the machineTeamRemover actor