2017-05-26 04:48:44 +08:00
|
|
|
/*
|
|
|
|
* QuietDatabase.actor.cpp
|
|
|
|
*
|
|
|
|
* This source file is part of the FoundationDB open source project
|
|
|
|
*
|
2022-03-22 04:36:23 +08:00
|
|
|
* Copyright 2013-2022 Apple Inc. and the FoundationDB project authors
|
2018-02-22 02:25:11 +08:00
|
|
|
*
|
2017-05-26 04:48:44 +08:00
|
|
|
* Licensed 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
|
2018-02-22 02:25:11 +08:00
|
|
|
*
|
2017-05-26 04:48:44 +08:00
|
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
2018-02-22 02:25:11 +08:00
|
|
|
*
|
2017-05-26 04:48:44 +08:00
|
|
|
* 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.
|
|
|
|
*/
|
|
|
|
|
2019-05-05 01:52:02 +08:00
|
|
|
#include <cinttypes>
|
2021-10-20 23:54:19 +08:00
|
|
|
#include <vector>
|
|
|
|
|
2022-02-26 03:43:20 +08:00
|
|
|
#include "fdbclient/FDBOptions.g.h"
|
2021-09-21 03:42:20 +08:00
|
|
|
#include "fdbclient/SystemData.h"
|
2017-05-26 04:48:44 +08:00
|
|
|
#include "flow/ActorCollection.h"
|
|
|
|
#include "fdbrpc/simulator.h"
|
|
|
|
#include "flow/Trace.h"
|
|
|
|
#include "fdbclient/DatabaseContext.h"
|
2019-06-20 05:02:15 +08:00
|
|
|
#include "fdbclient/NativeAPI.actor.h"
|
|
|
|
#include "fdbclient/ReadYourWrites.h"
|
|
|
|
#include "fdbclient/RunTransaction.actor.h"
|
2021-05-31 02:51:47 +08:00
|
|
|
#include "fdbserver/Knobs.h"
|
2019-02-18 11:25:16 +08:00
|
|
|
#include "fdbserver/TesterInterface.actor.h"
|
2019-02-18 11:13:26 +08:00
|
|
|
#include "fdbserver/WorkerInterface.actor.h"
|
2018-10-20 01:30:13 +08:00
|
|
|
#include "fdbserver/ServerDBInfo.h"
|
|
|
|
#include "fdbserver/Status.h"
|
2019-02-18 09:38:13 +08:00
|
|
|
#include "fdbclient/ManagementAPI.actor.h"
|
2019-05-22 01:53:36 +08:00
|
|
|
#include <boost/lexical_cast.hpp>
|
2021-03-11 02:06:03 +08:00
|
|
|
#include "flow/actorcompiler.h" // This must be the last #include.
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-09-17 08:42:34 +08:00
|
|
|
ACTOR Future<std::vector<WorkerDetails>> getWorkers(Reference<AsyncVar<ServerDBInfo> const> dbInfo, int flags = 0) {
|
2017-05-26 04:48:44 +08:00
|
|
|
loop {
|
|
|
|
choose {
|
2021-09-17 08:42:34 +08:00
|
|
|
when(std::vector<WorkerDetails> w = wait(brokenPromiseToNever(
|
2021-03-11 02:06:03 +08:00
|
|
|
dbInfo->get().clusterInterface.getWorkers.getReply(GetWorkersRequest(flags))))) {
|
2017-05-26 04:48:44 +08:00
|
|
|
return w;
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
when(wait(dbInfo->onChange())) {}
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
// Gets the WorkerInterface representing the Master server.
|
2021-07-12 13:04:38 +08:00
|
|
|
ACTOR Future<WorkerInterface> getMasterWorker(Database cx, Reference<AsyncVar<ServerDBInfo> const> dbInfo) {
|
2018-08-17 01:24:12 +08:00
|
|
|
TraceEvent("GetMasterWorker").detail("Stage", "GettingWorkers");
|
2017-05-26 04:48:44 +08:00
|
|
|
|
|
|
|
loop {
|
2021-09-17 08:42:34 +08:00
|
|
|
state std::vector<WorkerDetails> workers = wait(getWorkers(dbInfo));
|
2021-03-11 02:06:03 +08:00
|
|
|
|
|
|
|
for (int i = 0; i < workers.size(); i++) {
|
|
|
|
if (workers[i].interf.address() == dbInfo->get().master.address()) {
|
|
|
|
TraceEvent("GetMasterWorker")
|
|
|
|
.detail("Stage", "GotWorkers")
|
|
|
|
.detail("MasterId", dbInfo->get().master.id())
|
|
|
|
.detail("WorkerId", workers[i].interf.id());
|
2019-03-09 00:25:07 +08:00
|
|
|
return workers[i].interf;
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TraceEvent(SevWarn, "GetMasterWorkerError")
|
2021-03-11 02:06:03 +08:00
|
|
|
.detail("Error", "MasterWorkerNotFound")
|
|
|
|
.detail("Master", dbInfo->get().master.id())
|
|
|
|
.detail("MasterAddress", dbInfo->get().master.address())
|
|
|
|
.detail("WorkerCount", workers.size());
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(delay(1.0));
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-12-14 05:31:37 +08:00
|
|
|
// Gets the WorkerInterface representing the data distributor.
|
2021-07-12 13:04:38 +08:00
|
|
|
ACTOR Future<WorkerInterface> getDataDistributorWorker(Database cx, Reference<AsyncVar<ServerDBInfo> const> dbInfo) {
|
2018-12-14 05:31:37 +08:00
|
|
|
TraceEvent("GetDataDistributorWorker").detail("Stage", "GettingWorkers");
|
|
|
|
|
2019-01-19 03:30:18 +08:00
|
|
|
loop {
|
2021-09-17 08:42:34 +08:00
|
|
|
state std::vector<WorkerDetails> workers = wait(getWorkers(dbInfo));
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!dbInfo->get().distributor.present())
|
|
|
|
continue;
|
|
|
|
|
|
|
|
for (int i = 0; i < workers.size(); i++) {
|
|
|
|
if (workers[i].interf.address() == dbInfo->get().distributor.get().address()) {
|
|
|
|
TraceEvent("GetDataDistributorWorker")
|
|
|
|
.detail("Stage", "GotWorkers")
|
|
|
|
.detail("DataDistributorId", dbInfo->get().distributor.get().id())
|
|
|
|
.detail("WorkerId", workers[i].interf.id());
|
2019-03-09 00:25:07 +08:00
|
|
|
return workers[i].interf;
|
2018-12-14 05:31:37 +08:00
|
|
|
}
|
|
|
|
}
|
2019-01-19 03:30:18 +08:00
|
|
|
|
|
|
|
TraceEvent(SevWarn, "GetDataDistributorWorker")
|
2021-03-11 02:06:03 +08:00
|
|
|
.detail("Error", "DataDistributorWorkerNotFound")
|
|
|
|
.detail("DataDistributorId", dbInfo->get().distributor.get().id())
|
|
|
|
.detail("DataDistributorAddress", dbInfo->get().distributor.get().address())
|
|
|
|
.detail("WorkerCount", workers.size());
|
2018-12-14 05:31:37 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-01-23 13:19:31 +08:00
|
|
|
// Gets the number of bytes in flight from the data distributor.
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<int64_t> getDataInFlight(Database cx, WorkerInterface distributorWorker) {
|
2017-05-26 04:48:44 +08:00
|
|
|
try {
|
2018-12-14 05:31:37 +08:00
|
|
|
TraceEvent("DataInFlight").detail("Stage", "ContactingDataDistributor");
|
2021-03-11 02:06:03 +08:00
|
|
|
TraceEventFields md = wait(timeoutError(
|
|
|
|
distributorWorker.eventLogRequest.getReply(EventLogRequest(LiteralStringRef("TotalDataInFlight"))), 1.0));
|
2019-05-22 01:53:36 +08:00
|
|
|
int64_t dataInFlight = boost::lexical_cast<int64_t>(md.getValue("TotalBytes"));
|
2017-05-26 04:48:44 +08:00
|
|
|
return dataInFlight;
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
|
|
|
TraceEvent("QuietDatabaseFailure", distributorWorker.id())
|
|
|
|
.error(e)
|
|
|
|
.detail("Reason", "Failed to extract DataInFlight");
|
2017-05-26 04:48:44 +08:00
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-01-23 13:19:31 +08:00
|
|
|
// Gets the number of bytes in flight from the data distributor.
|
2021-07-12 13:04:38 +08:00
|
|
|
ACTOR Future<int64_t> getDataInFlight(Database cx, Reference<AsyncVar<ServerDBInfo> const> dbInfo) {
|
2021-03-11 02:06:03 +08:00
|
|
|
WorkerInterface distributorInterf = wait(getDataDistributorWorker(cx, dbInfo));
|
2019-01-31 11:53:15 +08:00
|
|
|
int64_t dataInFlight = wait(getDataInFlight(cx, distributorInterf));
|
2017-05-26 04:48:44 +08:00
|
|
|
return dataInFlight;
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
// Computes the queue size for storage servers and tlogs using the bytesInput and bytesDurable attributes
|
|
|
|
int64_t getQueueSize(const TraceEventFields& md) {
|
2017-11-15 04:59:42 +08:00
|
|
|
double inputRate, durableRate;
|
|
|
|
double inputRoughness, durableRoughness;
|
|
|
|
int64_t inputBytes, durableBytes;
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2019-05-05 01:52:02 +08:00
|
|
|
sscanf(md.getValue("BytesInput").c_str(), "%lf %lf %" SCNd64, &inputRate, &inputRoughness, &inputBytes);
|
|
|
|
sscanf(md.getValue("BytesDurable").c_str(), "%lf %lf %" SCNd64, &durableRate, &durableRoughness, &durableBytes);
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2017-11-15 04:59:42 +08:00
|
|
|
return inputBytes - durableBytes;
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
// Computes the popped version lag for tlogs
|
|
|
|
int64_t getPoppedVersionLag(const TraceEventFields& md) {
|
2019-05-22 01:53:36 +08:00
|
|
|
int64_t persistentDataDurableVersion = boost::lexical_cast<int64_t>(md.getValue("PersistentDataDurableVersion"));
|
|
|
|
int64_t queuePoppedVersion = boost::lexical_cast<int64_t>(md.getValue("QueuePoppedVersion"));
|
2019-05-21 14:58:45 +08:00
|
|
|
|
|
|
|
return persistentDataDurableVersion - queuePoppedVersion;
|
|
|
|
}
|
|
|
|
|
2021-09-17 08:42:34 +08:00
|
|
|
ACTOR Future<std::vector<WorkerInterface>> getCoordWorkers(Database cx,
|
|
|
|
Reference<AsyncVar<ServerDBInfo> const> dbInfo) {
|
2019-07-23 05:10:35 +08:00
|
|
|
state std::vector<WorkerDetails> workers = wait(getWorkers(dbInfo));
|
2019-06-20 02:20:44 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
Optional<Value> coordinators =
|
|
|
|
wait(runRYWTransaction(cx, [=](Reference<ReadYourWritesTransaction> tr) -> Future<Optional<Value>> {
|
|
|
|
tr->setOption(FDBTransactionOptions::ACCESS_SYSTEM_KEYS);
|
|
|
|
tr->setOption(FDBTransactionOptions::LOCK_AWARE);
|
|
|
|
return tr->get(coordinatorsKey);
|
|
|
|
}));
|
2019-06-20 02:20:44 +08:00
|
|
|
if (!coordinators.present()) {
|
|
|
|
throw operation_failed();
|
|
|
|
}
|
2022-01-21 16:42:03 +08:00
|
|
|
state ClusterConnectionString ccs(coordinators.get().toString());
|
2022-02-22 03:53:54 +08:00
|
|
|
wait(ccs.resolveHostnames());
|
2022-01-21 16:42:03 +08:00
|
|
|
std::vector<NetworkAddress> coordinatorsAddr = ccs.coordinators();
|
2019-07-23 05:10:35 +08:00
|
|
|
std::set<NetworkAddress> coordinatorsAddrSet;
|
2021-03-11 02:06:03 +08:00
|
|
|
for (const auto& addr : coordinatorsAddr) {
|
2019-06-20 02:20:44 +08:00
|
|
|
TraceEvent(SevDebug, "CoordinatorAddress").detail("Addr", addr);
|
|
|
|
coordinatorsAddrSet.insert(addr);
|
|
|
|
}
|
|
|
|
|
2021-09-17 08:42:34 +08:00
|
|
|
std::vector<WorkerInterface> result;
|
2021-03-11 02:06:03 +08:00
|
|
|
for (const auto& worker : workers) {
|
2019-06-20 02:20:44 +08:00
|
|
|
NetworkAddress primary = worker.interf.address();
|
|
|
|
Optional<NetworkAddress> secondary = worker.interf.tLog.getEndpoint().addresses.secondaryAddress;
|
2021-03-11 02:06:03 +08:00
|
|
|
if (coordinatorsAddrSet.find(primary) != coordinatorsAddrSet.end() ||
|
|
|
|
(secondary.present() && (coordinatorsAddrSet.find(secondary.get()) != coordinatorsAddrSet.end()))) {
|
2019-06-20 02:20:44 +08:00
|
|
|
result.push_back(worker.interf);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
2017-05-26 04:48:44 +08:00
|
|
|
// This is not robust in the face of a TLog failure
|
2021-07-12 13:04:38 +08:00
|
|
|
ACTOR Future<std::pair<int64_t, int64_t>> getTLogQueueInfo(Database cx,
|
|
|
|
Reference<AsyncVar<ServerDBInfo> const> dbInfo) {
|
2018-08-17 01:24:12 +08:00
|
|
|
TraceEvent("MaxTLogQueueSize").detail("Stage", "ContactingLogs");
|
2017-11-15 04:59:42 +08:00
|
|
|
|
2019-03-09 00:25:07 +08:00
|
|
|
state std::vector<WorkerDetails> workers = wait(getWorkers(dbInfo));
|
2017-11-15 04:59:42 +08:00
|
|
|
std::map<NetworkAddress, WorkerInterface> workersMap;
|
2021-03-11 02:06:03 +08:00
|
|
|
for (auto worker : workers) {
|
2019-03-09 00:25:07 +08:00
|
|
|
workersMap[worker.interf.address()] = worker.interf;
|
2017-11-15 04:59:42 +08:00
|
|
|
}
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2018-05-03 01:44:38 +08:00
|
|
|
state std::vector<Future<TraceEventFields>> messages;
|
2017-05-26 04:48:44 +08:00
|
|
|
state std::vector<TLogInterface> tlogs = dbInfo->get().logSystemConfig.allPresentLogs();
|
2021-03-11 02:06:03 +08:00
|
|
|
for (int i = 0; i < tlogs.size(); i++) {
|
2017-11-15 04:59:42 +08:00
|
|
|
auto itr = workersMap.find(tlogs[i].address());
|
2021-03-11 02:06:03 +08:00
|
|
|
if (itr == workersMap.end()) {
|
|
|
|
TraceEvent("QuietDatabaseFailure")
|
|
|
|
.detail("Reason", "Could not find worker for log server")
|
|
|
|
.detail("Tlog", tlogs[i].id());
|
2017-11-15 04:59:42 +08:00
|
|
|
throw attribute_not_found();
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
messages.push_back(timeoutError(
|
|
|
|
itr->second.eventLogRequest.getReply(EventLogRequest(StringRef(tlogs[i].id().toString() + "/TLogMetrics"))),
|
|
|
|
1.0));
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(waitForAll(messages));
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2018-08-17 01:24:12 +08:00
|
|
|
TraceEvent("MaxTLogQueueSize").detail("Stage", "ComputingMax").detail("MessageCount", messages.size());
|
2017-05-26 04:48:44 +08:00
|
|
|
|
|
|
|
state int64_t maxQueueSize = 0;
|
2019-05-21 14:58:45 +08:00
|
|
|
state int64_t maxPoppedVersionLag = 0;
|
2017-05-26 04:48:44 +08:00
|
|
|
state int i = 0;
|
2021-03-11 02:06:03 +08:00
|
|
|
for (; i < messages.size(); i++) {
|
2017-05-26 04:48:44 +08:00
|
|
|
try {
|
2021-03-11 02:06:03 +08:00
|
|
|
maxQueueSize = std::max(maxQueueSize, getQueueSize(messages[i].get()));
|
|
|
|
maxPoppedVersionLag = std::max(maxPoppedVersionLag, getPoppedVersionLag(messages[i].get()));
|
|
|
|
} catch (Error& e) {
|
|
|
|
TraceEvent("QuietDatabaseFailure")
|
|
|
|
.detail("Reason", "Failed to extract MaxTLogQueue")
|
|
|
|
.detail("Tlog", tlogs[i].id());
|
2017-05-26 04:48:44 +08:00
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
return std::make_pair(maxQueueSize, maxPoppedVersionLag);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
2021-09-29 07:15:32 +08:00
|
|
|
// Returns a vector of blob worker interfaces which have been persisted under the system key space
|
2021-10-14 04:37:56 +08:00
|
|
|
ACTOR Future<std::vector<BlobWorkerInterface>> getBlobWorkers(Database cx, bool use_system_priority = false) {
|
2021-09-21 03:42:20 +08:00
|
|
|
state Transaction tr(cx);
|
|
|
|
loop {
|
|
|
|
if (use_system_priority) {
|
|
|
|
tr.setOption(FDBTransactionOptions::PRIORITY_SYSTEM_IMMEDIATE);
|
|
|
|
}
|
2022-02-26 03:43:20 +08:00
|
|
|
tr.setOption(FDBTransactionOptions::READ_SYSTEM_KEYS);
|
2021-09-21 03:42:20 +08:00
|
|
|
tr.setOption(FDBTransactionOptions::LOCK_AWARE);
|
|
|
|
try {
|
|
|
|
RangeResult blobWorkersList = wait(tr.getRange(blobWorkerListKeys, CLIENT_KNOBS->TOO_MANY));
|
|
|
|
ASSERT(!blobWorkersList.more && blobWorkersList.size() < CLIENT_KNOBS->TOO_MANY);
|
|
|
|
|
2021-10-14 04:37:56 +08:00
|
|
|
std::vector<BlobWorkerInterface> blobWorkers;
|
2021-09-21 03:42:20 +08:00
|
|
|
blobWorkers.reserve(blobWorkersList.size());
|
|
|
|
for (int i = 0; i < blobWorkersList.size(); i++) {
|
|
|
|
blobWorkers.push_back(decodeBlobWorkerListValue(blobWorkersList[i].value));
|
|
|
|
}
|
|
|
|
return blobWorkers;
|
|
|
|
} catch (Error& e) {
|
|
|
|
wait(tr.onError(e));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-09-17 08:42:34 +08:00
|
|
|
ACTOR Future<std::vector<StorageServerInterface>> getStorageServers(Database cx, bool use_system_priority = false) {
|
2021-03-11 02:06:03 +08:00
|
|
|
state Transaction tr(cx);
|
2017-05-26 04:48:44 +08:00
|
|
|
loop {
|
2019-08-17 06:13:53 +08:00
|
|
|
if (use_system_priority) {
|
|
|
|
tr.setOption(FDBTransactionOptions::PRIORITY_SYSTEM_IMMEDIATE);
|
|
|
|
}
|
2022-02-26 03:43:20 +08:00
|
|
|
tr.setOption(FDBTransactionOptions::READ_SYSTEM_KEYS);
|
2019-08-17 06:13:53 +08:00
|
|
|
tr.setOption(FDBTransactionOptions::LOCK_AWARE);
|
2017-05-26 04:48:44 +08:00
|
|
|
try {
|
2021-05-04 04:14:16 +08:00
|
|
|
RangeResult serverList = wait(tr.getRange(serverListKeys, CLIENT_KNOBS->TOO_MANY));
|
2021-03-11 02:06:03 +08:00
|
|
|
ASSERT(!serverList.more && serverList.size() < CLIENT_KNOBS->TOO_MANY);
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-09-17 08:42:34 +08:00
|
|
|
std::vector<StorageServerInterface> servers;
|
2021-03-04 11:36:21 +08:00
|
|
|
servers.reserve(serverList.size());
|
2021-03-11 02:06:03 +08:00
|
|
|
for (int i = 0; i < serverList.size(); i++)
|
|
|
|
servers.push_back(decodeServerListValue(serverList[i].value));
|
2017-05-26 04:48:44 +08:00
|
|
|
return servers;
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
|
|
|
wait(tr.onError(e));
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-09-17 08:42:34 +08:00
|
|
|
ACTOR Future<std::vector<WorkerInterface>> getStorageWorkers(Database cx,
|
|
|
|
Reference<AsyncVar<ServerDBInfo> const> dbInfo,
|
|
|
|
bool localOnly) {
|
2019-07-21 16:00:29 +08:00
|
|
|
state std::vector<StorageServerInterface> servers = wait(getStorageServers(cx));
|
2019-06-20 05:02:15 +08:00
|
|
|
state std::map<NetworkAddress, WorkerInterface> workersMap;
|
2019-07-21 16:00:29 +08:00
|
|
|
std::vector<WorkerDetails> workers = wait(getWorkers(dbInfo));
|
2019-07-13 01:56:27 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
for (const auto& worker : workers) {
|
2019-06-20 02:20:44 +08:00
|
|
|
workersMap[worker.interf.address()] = worker.interf;
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
Optional<Value> regionsValue =
|
|
|
|
wait(runRYWTransaction(cx, [=](Reference<ReadYourWritesTransaction> tr) -> Future<Optional<Value>> {
|
|
|
|
tr->setOption(FDBTransactionOptions::ACCESS_SYSTEM_KEYS);
|
|
|
|
tr->setOption(FDBTransactionOptions::LOCK_AWARE);
|
|
|
|
return tr->get(LiteralStringRef("usable_regions").withPrefix(configKeysPrefix));
|
|
|
|
}));
|
2019-07-27 06:01:05 +08:00
|
|
|
int usableRegions = 1;
|
|
|
|
if (regionsValue.present()) {
|
|
|
|
usableRegions = atoi(regionsValue.get().toString().c_str());
|
|
|
|
}
|
2019-06-20 05:02:15 +08:00
|
|
|
auto masterDcId = dbInfo->get().master.locality.dcId();
|
|
|
|
|
2021-09-17 08:42:34 +08:00
|
|
|
std::vector<WorkerInterface> result;
|
2021-03-11 02:06:03 +08:00
|
|
|
for (const auto& server : servers) {
|
2019-06-20 05:02:15 +08:00
|
|
|
TraceEvent(SevDebug, "DcIdInfo")
|
2021-03-11 02:06:03 +08:00
|
|
|
.detail("ServerLocalityID", server.locality.dcId())
|
|
|
|
.detail("MasterDcID", masterDcId);
|
2019-06-20 05:02:15 +08:00
|
|
|
if (!localOnly || (usableRegions == 1 || server.locality.dcId() == masterDcId)) {
|
|
|
|
auto itr = workersMap.find(server.address());
|
2021-03-11 02:06:03 +08:00
|
|
|
if (itr == workersMap.end()) {
|
|
|
|
TraceEvent(SevWarn, "GetStorageWorkers")
|
|
|
|
.detail("Reason", "Could not find worker for storage server")
|
|
|
|
.detail("SS", server.id());
|
2019-07-04 02:02:35 +08:00
|
|
|
throw operation_failed();
|
2019-06-20 05:02:15 +08:00
|
|
|
}
|
|
|
|
result.push_back(itr->second);
|
2019-06-20 02:20:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
2021-06-16 06:49:27 +08:00
|
|
|
// Helper function to extract he maximum SQ size of all provided messages. All futures in the
|
|
|
|
// messages vector have to be ready.
|
2021-06-16 06:00:28 +08:00
|
|
|
int64_t extractMaxQueueSize(const std::vector<Future<TraceEventFields>>& messages,
|
|
|
|
const std::vector<StorageServerInterface>& servers) {
|
|
|
|
|
|
|
|
int64_t maxQueueSize = 0;
|
|
|
|
UID maxQueueServer;
|
|
|
|
|
|
|
|
for (int i = 0; i < messages.size(); i++) {
|
|
|
|
try {
|
|
|
|
auto queueSize = getQueueSize(messages[i].get());
|
|
|
|
if (queueSize > maxQueueSize) {
|
|
|
|
maxQueueSize = queueSize;
|
|
|
|
maxQueueServer = servers[i].id();
|
|
|
|
}
|
|
|
|
} catch (Error& e) {
|
|
|
|
TraceEvent("QuietDatabaseFailure")
|
|
|
|
.detail("Reason", "Failed to extract MaxStorageServerQueue")
|
|
|
|
.detail("SS", servers[i].id());
|
|
|
|
for (auto& m : messages) {
|
|
|
|
TraceEvent("Messages").detail("Info", m.get().toString());
|
|
|
|
}
|
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TraceEvent("QuietDatabaseGotMaxStorageServerQueueSize")
|
|
|
|
.detail("Stage", "MaxComputed")
|
|
|
|
.detail("Max", maxQueueSize)
|
|
|
|
.detail("MaxQueueServer", format("%016" PRIx64, maxQueueServer.first()));
|
|
|
|
return maxQueueSize;
|
|
|
|
}
|
|
|
|
|
2021-06-16 06:49:27 +08:00
|
|
|
// Timeout wrapper when getting the storage metrics. This will do some additional tracing
|
2021-06-16 06:00:28 +08:00
|
|
|
ACTOR Future<TraceEventFields> getStorageMetricsTimeout(UID storage, WorkerInterface wi) {
|
|
|
|
state Future<TraceEventFields> result =
|
|
|
|
wi.eventLogRequest.getReply(EventLogRequest(StringRef(storage.toString() + "/StorageMetrics")));
|
|
|
|
state Future<Void> timeout = delay(1.0);
|
|
|
|
choose {
|
|
|
|
when(TraceEventFields res = wait(result)) { return res; }
|
|
|
|
when(wait(timeout)) {
|
|
|
|
TraceEvent("QuietDatabaseFailure")
|
|
|
|
.detail("Reason", "Could not fetch StorageMetrics")
|
|
|
|
.detail("Storage", format("%016" PRIx64, storage.first()));
|
|
|
|
throw timed_out();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
// Gets the maximum size of all the storage server queues
|
2021-07-12 13:04:38 +08:00
|
|
|
ACTOR Future<int64_t> getMaxStorageServerQueueSize(Database cx, Reference<AsyncVar<ServerDBInfo> const> dbInfo) {
|
2018-08-17 01:24:12 +08:00
|
|
|
TraceEvent("MaxStorageServerQueueSize").detail("Stage", "ContactingStorageServers");
|
2017-11-15 04:59:42 +08:00
|
|
|
|
|
|
|
Future<std::vector<StorageServerInterface>> serversFuture = getStorageServers(cx);
|
2019-03-09 00:25:07 +08:00
|
|
|
state Future<std::vector<WorkerDetails>> workersFuture = getWorkers(dbInfo);
|
2017-11-15 04:59:42 +08:00
|
|
|
|
|
|
|
state std::vector<StorageServerInterface> servers = wait(serversFuture);
|
2019-03-09 00:25:07 +08:00
|
|
|
state std::vector<WorkerDetails> workers = wait(workersFuture);
|
2017-11-15 04:59:42 +08:00
|
|
|
|
|
|
|
std::map<NetworkAddress, WorkerInterface> workersMap;
|
2021-03-11 02:06:03 +08:00
|
|
|
for (auto worker : workers) {
|
2019-03-09 00:25:07 +08:00
|
|
|
workersMap[worker.interf.address()] = worker.interf;
|
2017-11-15 04:59:42 +08:00
|
|
|
}
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2018-05-03 01:44:38 +08:00
|
|
|
state std::vector<Future<TraceEventFields>> messages;
|
2021-03-11 02:06:03 +08:00
|
|
|
for (int i = 0; i < servers.size(); i++) {
|
2017-11-15 04:59:42 +08:00
|
|
|
auto itr = workersMap.find(servers[i].address());
|
2021-03-11 02:06:03 +08:00
|
|
|
if (itr == workersMap.end()) {
|
|
|
|
TraceEvent("QuietDatabaseFailure")
|
|
|
|
.detail("Reason", "Could not find worker for storage server")
|
|
|
|
.detail("SS", servers[i].id());
|
2017-11-15 04:59:42 +08:00
|
|
|
throw attribute_not_found();
|
|
|
|
}
|
2021-06-19 01:10:50 +08:00
|
|
|
messages.push_back(getStorageMetricsTimeout(servers[i].id(), itr->second));
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(waitForAll(messages));
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2018-08-17 01:24:12 +08:00
|
|
|
TraceEvent("MaxStorageServerQueueSize").detail("Stage", "ComputingMax").detail("MessageCount", messages.size());
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-06-16 06:00:28 +08:00
|
|
|
return extractMaxQueueSize(messages, servers);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
// Gets the size of the data distribution queue. If reportInFlight is true, then data in flight is considered part of
|
|
|
|
// the queue
|
|
|
|
ACTOR Future<int64_t> getDataDistributionQueueSize(Database cx,
|
|
|
|
WorkerInterface distributorWorker,
|
|
|
|
bool reportInFlight) {
|
2017-05-26 04:48:44 +08:00
|
|
|
try {
|
2018-12-14 05:31:37 +08:00
|
|
|
TraceEvent("DataDistributionQueueSize").detail("Stage", "ContactingDataDistributor");
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
TraceEventFields movingDataMessage = wait(timeoutError(
|
|
|
|
distributorWorker.eventLogRequest.getReply(EventLogRequest(LiteralStringRef("MovingData"))), 1.0));
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2018-08-17 01:24:12 +08:00
|
|
|
TraceEvent("DataDistributionQueueSize").detail("Stage", "GotString");
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2019-05-22 01:53:36 +08:00
|
|
|
int64_t inQueue = boost::lexical_cast<int64_t>(movingDataMessage.getValue("InQueue"));
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (reportInFlight) {
|
2019-05-22 01:53:36 +08:00
|
|
|
int64_t inFlight = boost::lexical_cast<int64_t>(movingDataMessage.getValue("InFlight"));
|
2017-05-26 04:48:44 +08:00
|
|
|
inQueue += inFlight;
|
|
|
|
}
|
|
|
|
|
|
|
|
return inQueue;
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
|
|
|
TraceEvent("QuietDatabaseFailure", distributorWorker.id())
|
|
|
|
.detail("Reason", "Failed to extract DataDistributionQueueSize");
|
2017-05-26 04:48:44 +08:00
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
// Gets the size of the data distribution queue. If reportInFlight is true, then data in flight is considered part of
|
|
|
|
// the queue Convenience method that first finds the master worker from a zookeeper interface
|
|
|
|
ACTOR Future<int64_t> getDataDistributionQueueSize(Database cx,
|
2021-07-12 13:04:38 +08:00
|
|
|
Reference<AsyncVar<ServerDBInfo> const> dbInfo,
|
2021-03-11 02:06:03 +08:00
|
|
|
bool reportInFlight) {
|
|
|
|
WorkerInterface distributorInterf = wait(getDataDistributorWorker(cx, dbInfo));
|
|
|
|
int64_t inQueue = wait(getDataDistributionQueueSize(cx, distributorInterf, reportInFlight));
|
2017-05-26 04:48:44 +08:00
|
|
|
return inQueue;
|
|
|
|
}
|
|
|
|
|
2019-02-13 05:41:18 +08:00
|
|
|
// Gets if the number of process and machine teams does not exceed the maximum allowed number of teams
|
2019-02-16 09:03:40 +08:00
|
|
|
ACTOR Future<bool> getTeamCollectionValid(Database cx, WorkerInterface dataDistributorWorker) {
|
2019-02-06 06:18:25 +08:00
|
|
|
state int attempts = 0;
|
2019-07-16 08:32:49 +08:00
|
|
|
state bool ret = false;
|
2019-02-06 06:18:25 +08:00
|
|
|
loop {
|
|
|
|
try {
|
2019-09-06 01:34:57 +08:00
|
|
|
if (!g_network->isSimulated()) {
|
2019-07-16 09:17:46 +08:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2019-02-06 06:18:25 +08:00
|
|
|
TraceEvent("GetTeamCollectionValid").detail("Stage", "ContactingMaster");
|
|
|
|
|
2019-02-13 05:41:18 +08:00
|
|
|
TraceEventFields teamCollectionInfoMessage = wait(timeoutError(
|
2021-03-11 02:06:03 +08:00
|
|
|
dataDistributorWorker.eventLogRequest.getReply(EventLogRequest(LiteralStringRef("TeamCollectionInfo"))),
|
|
|
|
1.0));
|
2019-02-06 06:18:25 +08:00
|
|
|
|
|
|
|
TraceEvent("GetTeamCollectionValid").detail("Stage", "GotString");
|
|
|
|
|
2021-03-09 07:20:50 +08:00
|
|
|
state int64_t currentTeams =
|
|
|
|
boost::lexical_cast<int64_t>(teamCollectionInfoMessage.getValue("CurrentServerTeams"));
|
2021-03-11 02:06:03 +08:00
|
|
|
state int64_t desiredTeams =
|
|
|
|
boost::lexical_cast<int64_t>(teamCollectionInfoMessage.getValue("DesiredTeams"));
|
2019-07-16 10:29:06 +08:00
|
|
|
state int64_t maxTeams = boost::lexical_cast<int64_t>(teamCollectionInfoMessage.getValue("MaxTeams"));
|
2021-03-11 02:06:03 +08:00
|
|
|
state int64_t currentMachineTeams =
|
|
|
|
boost::lexical_cast<int64_t>(teamCollectionInfoMessage.getValue("CurrentMachineTeams"));
|
|
|
|
state int64_t healthyMachineTeams =
|
|
|
|
boost::lexical_cast<int64_t>(teamCollectionInfoMessage.getValue("CurrentHealthyMachineTeams"));
|
|
|
|
state int64_t desiredMachineTeams =
|
|
|
|
boost::lexical_cast<int64_t>(teamCollectionInfoMessage.getValue("DesiredMachineTeams"));
|
|
|
|
state int64_t maxMachineTeams =
|
|
|
|
boost::lexical_cast<int64_t>(teamCollectionInfoMessage.getValue("MaxMachineTeams"));
|
2019-02-06 06:18:25 +08:00
|
|
|
|
2019-07-16 10:29:06 +08:00
|
|
|
state int64_t minServerTeamsOnServer =
|
2019-07-09 06:45:28 +08:00
|
|
|
boost::lexical_cast<int64_t>(teamCollectionInfoMessage.getValue("MinTeamsOnServer"));
|
2019-07-16 10:29:06 +08:00
|
|
|
state int64_t maxServerTeamsOnServer =
|
2019-07-09 06:45:28 +08:00
|
|
|
boost::lexical_cast<int64_t>(teamCollectionInfoMessage.getValue("MaxTeamsOnServer"));
|
2019-07-16 10:29:06 +08:00
|
|
|
state int64_t minMachineTeamsOnMachine =
|
2019-07-09 06:45:28 +08:00
|
|
|
boost::lexical_cast<int64_t>(teamCollectionInfoMessage.getValue("MinMachineTeamsOnMachine"));
|
2019-07-16 10:29:06 +08:00
|
|
|
state int64_t maxMachineTeamsOnMachine =
|
2019-07-09 06:45:28 +08:00
|
|
|
boost::lexical_cast<int64_t>(teamCollectionInfoMessage.getValue("MaxMachineTeamsOnMachine"));
|
2019-02-06 06:18:25 +08:00
|
|
|
|
2019-07-12 13:05:20 +08:00
|
|
|
// The if condition should be consistent with the condition in serverTeamRemover() and
|
2019-07-10 02:45:17 +08:00
|
|
|
// machineTeamRemover() that decides if redundant teams exist.
|
2019-07-12 13:05:20 +08:00
|
|
|
// Team number is always valid when we disable teamRemover, which avoids false positive in simulation test.
|
|
|
|
// The minimun team number per server (and per machine) should be no less than 0 so that newly added machine
|
|
|
|
// can host data on it.
|
2019-07-19 07:47:42 +08:00
|
|
|
//
|
|
|
|
// If the machineTeamRemover does not remove the machine team with the most machine teams,
|
|
|
|
// we may oscillate between building more server teams by teamBuilder() and removing those teams by
|
|
|
|
// teamRemover To avoid false positive in simulation, we skip the consistency check in this case.
|
|
|
|
// This is a corner case. This is a work-around if case the team number requirements cannot be satisfied.
|
|
|
|
//
|
2021-03-11 02:06:03 +08:00
|
|
|
// The checking for too many teams is disabled because teamRemover may not remove a team if it leads to 0
|
|
|
|
// team on a server
|
2019-07-17 03:38:52 +08:00
|
|
|
//(!SERVER_KNOBS->TR_FLAG_DISABLE_MACHINE_TEAM_REMOVER &&
|
|
|
|
// healthyMachineTeams > desiredMachineTeams) ||
|
|
|
|
// (!SERVER_KNOBS->TR_FLAG_DISABLE_SERVER_TEAM_REMOVER && currentTeams > desiredTeams) ||
|
|
|
|
if ((minMachineTeamsOnMachine <= 0 || minServerTeamsOnServer <= 0) &&
|
2021-03-11 02:06:03 +08:00
|
|
|
SERVER_KNOBS->TR_FLAG_REMOVE_MT_WITH_MOST_TEAMS) {
|
2019-07-16 08:32:49 +08:00
|
|
|
ret = false;
|
|
|
|
|
2019-07-10 02:45:17 +08:00
|
|
|
if (attempts++ < 10) {
|
|
|
|
wait(delay(60));
|
|
|
|
continue; // We may not receive the most recent TeamCollectionInfo
|
|
|
|
}
|
2019-07-16 08:32:49 +08:00
|
|
|
|
2019-06-29 07:01:05 +08:00
|
|
|
// When DESIRED_TEAMS_PER_SERVER == 1, we see minMachineTeamOnMachine can be 0 in one out of 30k test
|
|
|
|
// cases. Only check DESIRED_TEAMS_PER_SERVER == 3 for now since it is mostly used configuration.
|
2019-07-12 13:05:20 +08:00
|
|
|
// TODO: Remove the constraint SERVER_KNOBS->DESIRED_TEAMS_PER_SERVER == 3 to ensure that
|
|
|
|
// the minimun team number per server (and per machine) is always > 0 for any number of replicas
|
2019-02-08 03:25:01 +08:00
|
|
|
TraceEvent("GetTeamCollectionValid")
|
2021-03-09 07:20:50 +08:00
|
|
|
.detail("CurrentServerTeams", currentTeams)
|
2019-07-05 23:54:21 +08:00
|
|
|
.detail("DesiredTeams", desiredTeams)
|
|
|
|
.detail("MaxTeams", maxTeams)
|
|
|
|
.detail("CurrentHealthyMachineTeams", healthyMachineTeams)
|
|
|
|
.detail("DesiredMachineTeams", desiredMachineTeams)
|
|
|
|
.detail("CurrentMachineTeams", currentMachineTeams)
|
2019-07-09 06:37:00 +08:00
|
|
|
.detail("MaxMachineTeams", maxMachineTeams)
|
2019-07-09 06:45:28 +08:00
|
|
|
.detail("MinTeamsOnServer", minServerTeamsOnServer)
|
|
|
|
.detail("MaxTeamsOnServer", maxServerTeamsOnServer)
|
|
|
|
.detail("MinMachineTeamsOnMachine", minMachineTeamsOnMachine)
|
|
|
|
.detail("MaxMachineTeamsOnMachine", maxMachineTeamsOnMachine)
|
2019-06-27 10:38:12 +08:00
|
|
|
.detail("DesiredTeamsPerServer", SERVER_KNOBS->DESIRED_TEAMS_PER_SERVER)
|
2019-07-16 08:32:49 +08:00
|
|
|
.detail("MaxTeamsPerServer", SERVER_KNOBS->MAX_TEAMS_PER_SERVER)
|
2019-07-16 09:40:14 +08:00
|
|
|
.detail("RemoveMTWithMostTeams", SERVER_KNOBS->TR_FLAG_REMOVE_MT_WITH_MOST_TEAMS);
|
2019-07-16 08:32:49 +08:00
|
|
|
return ret;
|
2019-02-06 06:18:25 +08:00
|
|
|
} else {
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2019-02-13 05:41:18 +08:00
|
|
|
} catch (Error& e) {
|
2021-03-11 02:06:03 +08:00
|
|
|
if (e.code() == error_code_actor_cancelled) {
|
2019-07-26 07:15:28 +08:00
|
|
|
throw;
|
|
|
|
}
|
2019-02-16 09:03:40 +08:00
|
|
|
TraceEvent("QuietDatabaseFailure", dataDistributorWorker.id())
|
2019-02-13 05:41:18 +08:00
|
|
|
.detail("Reason", "Failed to extract GetTeamCollectionValid information");
|
2019-02-06 06:18:25 +08:00
|
|
|
attempts++;
|
2019-02-13 05:41:18 +08:00
|
|
|
if (attempts > 10) {
|
2019-02-16 09:03:40 +08:00
|
|
|
TraceEvent("QuietDatabaseNoTeamCollectionInfo", dataDistributorWorker.id())
|
2019-02-13 05:41:18 +08:00
|
|
|
.detail("Reason", "Had never called build team to build any team");
|
2019-02-06 06:18:25 +08:00
|
|
|
return true;
|
|
|
|
}
|
2019-02-13 05:41:18 +08:00
|
|
|
// throw;
|
|
|
|
wait(delay(10.0));
|
2019-02-05 09:35:07 +08:00
|
|
|
}
|
2019-02-06 06:18:25 +08:00
|
|
|
};
|
2019-02-05 09:35:07 +08:00
|
|
|
}
|
|
|
|
|
2019-02-13 05:41:18 +08:00
|
|
|
// Gets if the number of process and machine teams does not exceed the maximum allowed number of teams
|
|
|
|
// Convenience method that first finds the master worker from a zookeeper interface
|
2021-07-12 13:04:38 +08:00
|
|
|
ACTOR Future<bool> getTeamCollectionValid(Database cx, Reference<AsyncVar<ServerDBInfo> const> dbInfo) {
|
2019-02-16 09:03:40 +08:00
|
|
|
WorkerInterface dataDistributorWorker = wait(getDataDistributorWorker(cx, dbInfo));
|
|
|
|
bool valid = wait(getTeamCollectionValid(cx, dataDistributorWorker));
|
2019-02-05 09:35:07 +08:00
|
|
|
return valid;
|
|
|
|
}
|
|
|
|
|
2018-12-14 05:31:37 +08:00
|
|
|
// Checks that data distribution is active
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<bool> getDataDistributionActive(Database cx, WorkerInterface distributorWorker) {
|
2017-05-26 04:48:44 +08:00
|
|
|
try {
|
2018-12-14 05:31:37 +08:00
|
|
|
TraceEvent("DataDistributionActive").detail("Stage", "ContactingDataDistributor");
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
TraceEventFields activeMessage = wait(timeoutError(
|
|
|
|
distributorWorker.eventLogRequest.getReply(EventLogRequest(LiteralStringRef("DDTrackerStarting"))), 1.0));
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2018-05-03 01:44:38 +08:00
|
|
|
return activeMessage.getValue("State") == "Active";
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
|
|
|
TraceEvent("QuietDatabaseFailure", distributorWorker.id())
|
|
|
|
.detail("Reason", "Failed to extract DataDistributionActive");
|
2017-05-26 04:48:44 +08:00
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-12-14 05:31:37 +08:00
|
|
|
// Checks to see if any storage servers are being recruited
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<bool> getStorageServersRecruiting(Database cx, WorkerInterface distributorWorker, UID distributorUID) {
|
2017-05-26 04:48:44 +08:00
|
|
|
try {
|
2018-12-14 05:31:37 +08:00
|
|
|
TraceEvent("StorageServersRecruiting").detail("Stage", "ContactingDataDistributor");
|
2021-03-11 02:06:03 +08:00
|
|
|
TraceEventFields recruitingMessage =
|
|
|
|
wait(timeoutError(distributorWorker.eventLogRequest.getReply(
|
|
|
|
EventLogRequest(StringRef("StorageServerRecruitment_" + distributorUID.toString()))),
|
|
|
|
1.0));
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2018-12-14 05:31:37 +08:00
|
|
|
TraceEvent("StorageServersRecruiting").detail("Message", recruitingMessage.toString());
|
2021-03-06 03:28:15 +08:00
|
|
|
|
|
|
|
if (recruitingMessage.getValue("State") == "Recruiting") {
|
|
|
|
std::string tssValue;
|
|
|
|
// if we're tss recruiting, that's fine because that can block indefinitely if only 1 free storage process
|
|
|
|
if (!recruitingMessage.tryGetValue("IsTSS", tssValue) || tssValue == "False") {
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return false;
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
2018-12-14 05:31:37 +08:00
|
|
|
TraceEvent("QuietDatabaseFailure", distributorWorker.id())
|
2021-03-11 02:06:03 +08:00
|
|
|
.detail("Reason", "Failed to extract StorageServersRecruiting")
|
|
|
|
.detail("DataDistributorID", distributorUID);
|
2017-05-26 04:48:44 +08:00
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-07-12 13:04:38 +08:00
|
|
|
ACTOR Future<Void> repairDeadDatacenter(Database cx,
|
|
|
|
Reference<AsyncVar<ServerDBInfo> const> dbInfo,
|
|
|
|
std::string context) {
|
2021-03-11 02:06:03 +08:00
|
|
|
if (g_network->isSimulated() && g_simulator.usableRegions > 1) {
|
2018-07-17 01:06:57 +08:00
|
|
|
bool primaryDead = g_simulator.datacenterDead(g_simulator.primaryDcId);
|
|
|
|
bool remoteDead = g_simulator.datacenterDead(g_simulator.remoteDcId);
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
// FIXME: the primary and remote can both be considered dead because excludes are not handled properly by the
|
|
|
|
// datacenterDead function
|
|
|
|
if (primaryDead && remoteDead) {
|
2021-07-27 10:55:10 +08:00
|
|
|
TraceEvent(SevWarnAlways, "CannotDisableFearlessConfiguration").log();
|
2019-07-31 07:35:52 +08:00
|
|
|
return Void();
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
if (primaryDead || remoteDead) {
|
|
|
|
TraceEvent(SevWarnAlways, "DisablingFearlessConfiguration")
|
|
|
|
.detail("Location", context)
|
|
|
|
.detail("Stage", "Repopulate")
|
|
|
|
.detail("RemoteDead", remoteDead)
|
|
|
|
.detail("PrimaryDead", primaryDead);
|
2018-07-17 01:06:57 +08:00
|
|
|
g_simulator.usableRegions = 1;
|
2021-09-21 05:41:20 +08:00
|
|
|
wait(success(ManagementAPI::changeConfig(
|
|
|
|
cx.getReference(),
|
|
|
|
(primaryDead ? g_simulator.disablePrimary : g_simulator.disableRemote) + " repopulate_anti_quorum=1",
|
|
|
|
true)));
|
2021-03-11 02:06:03 +08:00
|
|
|
while (dbInfo->get().recoveryState < RecoveryState::STORAGE_RECOVERED) {
|
|
|
|
wait(dbInfo->onChange());
|
2018-07-17 01:06:57 +08:00
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
TraceEvent(SevWarnAlways, "DisablingFearlessConfiguration")
|
|
|
|
.detail("Location", context)
|
|
|
|
.detail("Stage", "Usable_Regions");
|
2021-09-21 05:41:20 +08:00
|
|
|
wait(success(ManagementAPI::changeConfig(cx.getReference(), "usable_regions=1", true)));
|
2018-05-02 06:31:49 +08:00
|
|
|
}
|
|
|
|
}
|
2018-07-17 01:06:57 +08:00
|
|
|
return Void();
|
|
|
|
}
|
2018-05-02 06:31:49 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> reconfigureAfter(Database cx,
|
|
|
|
double time,
|
2021-07-12 13:04:38 +08:00
|
|
|
Reference<AsyncVar<ServerDBInfo> const> dbInfo,
|
2021-03-11 02:06:03 +08:00
|
|
|
std::string context) {
|
|
|
|
wait(delay(time));
|
|
|
|
wait(repairDeadDatacenter(cx, dbInfo, context));
|
2018-05-02 06:31:49 +08:00
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
2021-06-16 06:49:27 +08:00
|
|
|
// Waits until a database quiets down (no data in flight, small tlog queue, low SQ, no active data distribution). This
|
|
|
|
// requires the database to be available and healthy in order to succeed.
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> waitForQuietDatabase(Database cx,
|
2021-07-12 13:04:38 +08:00
|
|
|
Reference<AsyncVar<ServerDBInfo> const> dbInfo,
|
2021-03-11 02:06:03 +08:00
|
|
|
std::string phase,
|
|
|
|
int64_t dataInFlightGate = 2e6,
|
|
|
|
int64_t maxTLogQueueGate = 5e6,
|
|
|
|
int64_t maxStorageServerQueueGate = 5e6,
|
|
|
|
int64_t maxDataDistributionQueueSize = 0,
|
|
|
|
int64_t maxPoppedVersionLag = 30e6) {
|
|
|
|
state Future<Void> reconfig =
|
|
|
|
reconfigureAfter(cx, 100 + (deterministicRandom()->random01() * 100), dbInfo, "QuietDatabase");
|
2021-06-16 06:00:28 +08:00
|
|
|
state Future<int64_t> dataInFlight;
|
|
|
|
state Future<std::pair<int64_t, int64_t>> tLogQueueInfo;
|
|
|
|
state Future<int64_t> dataDistributionQueueSize;
|
|
|
|
state Future<bool> teamCollectionValid;
|
|
|
|
state Future<int64_t> storageQueueSize;
|
|
|
|
state Future<bool> dataDistributionActive;
|
|
|
|
state Future<bool> storageServersRecruiting;
|
2020-04-23 05:14:56 +08:00
|
|
|
auto traceMessage = "QuietDatabase" + phase + "Begin";
|
2021-09-22 07:49:55 +08:00
|
|
|
TraceEvent(traceMessage.c_str()).log();
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
// In a simulated environment, wait 5 seconds so that workers can move to their optimal locations
|
|
|
|
if (g_network->isSimulated())
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(delay(5.0));
|
2021-09-22 07:49:55 +08:00
|
|
|
|
2021-10-22 00:01:52 +08:00
|
|
|
TraceEvent("QuietDatabaseWaitingOnFullRecovery").log();
|
|
|
|
while (dbInfo->get().recoveryState != RecoveryState::FULLY_RECOVERED) {
|
|
|
|
wait(dbInfo->onChange());
|
|
|
|
}
|
|
|
|
|
2021-06-04 03:58:28 +08:00
|
|
|
// The quiet database check (which runs at the end of every test) will always time out due to active data movement.
|
|
|
|
// To get around this, quiet Database will disable the perpetual wiggle in the setup phase.
|
2021-09-22 07:49:55 +08:00
|
|
|
|
2021-07-29 05:11:25 +08:00
|
|
|
printf("Set perpetual_storage_wiggle=0 ...\n");
|
2021-07-17 15:11:40 +08:00
|
|
|
wait(setPerpetualStorageWiggle(cx, false, LockAware::True));
|
2021-07-29 05:11:25 +08:00
|
|
|
printf("Set perpetual_storage_wiggle=0 Done.\n");
|
2021-06-04 03:58:28 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
// Require 3 consecutive successful quiet database checks spaced 2 second apart
|
2017-05-26 04:48:44 +08:00
|
|
|
state int numSuccesses = 0;
|
|
|
|
loop {
|
|
|
|
try {
|
2021-07-27 10:55:10 +08:00
|
|
|
TraceEvent("QuietDatabaseWaitingOnDataDistributor").log();
|
2021-03-11 02:06:03 +08:00
|
|
|
WorkerInterface distributorWorker = wait(getDataDistributorWorker(cx, dbInfo));
|
2020-01-17 13:21:25 +08:00
|
|
|
UID distributorUID = dbInfo->get().distributor.get().id();
|
2021-03-11 02:06:03 +08:00
|
|
|
TraceEvent("QuietDatabaseGotDataDistributor", distributorUID)
|
|
|
|
.detail("Locality", distributorWorker.locality.toString());
|
2018-12-14 05:31:37 +08:00
|
|
|
|
2021-06-16 06:00:28 +08:00
|
|
|
dataInFlight = getDataInFlight(cx, distributorWorker);
|
|
|
|
tLogQueueInfo = getTLogQueueInfo(cx, dbInfo);
|
|
|
|
dataDistributionQueueSize = getDataDistributionQueueSize(cx, distributorWorker, dataInFlightGate == 0);
|
|
|
|
teamCollectionValid = getTeamCollectionValid(cx, distributorWorker);
|
|
|
|
storageQueueSize = getMaxStorageServerQueueSize(cx, dbInfo);
|
|
|
|
dataDistributionActive = getDataDistributionActive(cx, distributorWorker);
|
|
|
|
storageServersRecruiting = getStorageServersRecruiting(cx, distributorWorker, distributorUID);
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2020-01-17 13:21:25 +08:00
|
|
|
wait(success(dataInFlight) && success(tLogQueueInfo) && success(dataDistributionQueueSize) &&
|
|
|
|
success(teamCollectionValid) && success(storageQueueSize) && success(dataDistributionActive) &&
|
|
|
|
success(storageServersRecruiting));
|
2019-05-22 02:31:34 +08:00
|
|
|
|
2017-05-26 04:48:44 +08:00
|
|
|
TraceEvent(("QuietDatabase" + phase).c_str())
|
2019-08-13 01:08:12 +08:00
|
|
|
.detail("DataInFlight", dataInFlight.get())
|
|
|
|
.detail("DataInFlightGate", dataInFlightGate)
|
|
|
|
.detail("MaxTLogQueueSize", tLogQueueInfo.get().first)
|
|
|
|
.detail("MaxTLogQueueGate", maxTLogQueueGate)
|
|
|
|
.detail("MaxTLogPoppedVersionLag", tLogQueueInfo.get().second)
|
|
|
|
.detail("MaxTLogPoppedVersionLagGate", maxPoppedVersionLag)
|
|
|
|
.detail("DataDistributionQueueSize", dataDistributionQueueSize.get())
|
|
|
|
.detail("DataDistributionQueueSizeGate", maxDataDistributionQueueSize)
|
|
|
|
.detail("TeamCollectionValid", teamCollectionValid.get())
|
|
|
|
.detail("MaxStorageQueueSize", storageQueueSize.get())
|
|
|
|
.detail("MaxStorageServerQueueGate", maxStorageServerQueueGate)
|
|
|
|
.detail("DataDistributionActive", dataDistributionActive.get())
|
|
|
|
.detail("StorageServersRecruiting", storageServersRecruiting.get())
|
|
|
|
.detail("NumSuccesses", numSuccesses);
|
2019-02-13 05:41:18 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (dataInFlight.get() > dataInFlightGate || tLogQueueInfo.get().first > maxTLogQueueGate ||
|
|
|
|
tLogQueueInfo.get().second > maxPoppedVersionLag ||
|
2019-02-13 05:41:18 +08:00
|
|
|
dataDistributionQueueSize.get() > maxDataDistributionQueueSize ||
|
2021-09-22 07:49:55 +08:00
|
|
|
storageQueueSize.get() > maxStorageServerQueueGate || !dataDistributionActive.get() ||
|
|
|
|
storageServersRecruiting.get() || !teamCollectionValid.get()) {
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(delay(1.0));
|
2017-05-26 04:48:44 +08:00
|
|
|
numSuccesses = 0;
|
|
|
|
} else {
|
2021-03-11 02:06:03 +08:00
|
|
|
if (++numSuccesses == 3) {
|
2020-04-23 05:14:56 +08:00
|
|
|
auto msg = "QuietDatabase" + phase + "Done";
|
2021-09-22 07:49:55 +08:00
|
|
|
TraceEvent(msg.c_str()).log();
|
2017-05-26 04:48:44 +08:00
|
|
|
break;
|
2019-08-15 05:19:50 +08:00
|
|
|
} else {
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(delay(g_network->isSimulated() ? 2.0 : 30.0));
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
} catch (Error& e) {
|
2022-02-25 04:25:52 +08:00
|
|
|
TraceEvent(("QuietDatabase" + phase + "Error").c_str()).errorUnsuppressed(e);
|
2019-08-15 05:19:50 +08:00
|
|
|
if (e.code() != error_code_actor_cancelled && e.code() != error_code_attribute_not_found &&
|
|
|
|
e.code() != error_code_timed_out)
|
2017-05-26 04:48:44 +08:00
|
|
|
TraceEvent(("QuietDatabase" + phase + "Error").c_str()).error(e);
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
// Client invalid operation occurs if we don't get back a message from one of the servers, often corrected
|
|
|
|
// by retrying
|
|
|
|
if (e.code() != error_code_attribute_not_found && e.code() != error_code_timed_out)
|
2017-05-26 04:48:44 +08:00
|
|
|
throw;
|
|
|
|
|
2021-06-16 06:00:28 +08:00
|
|
|
auto evtType = "QuietDatabase" + phase + "Retry";
|
|
|
|
TraceEvent evt(evtType.c_str());
|
|
|
|
evt.error(e);
|
|
|
|
int notReadyCount = 0;
|
|
|
|
if (dataInFlight.isReady() && dataInFlight.isError()) {
|
|
|
|
auto key = "NotReady" + std::to_string(notReadyCount++);
|
|
|
|
evt.detail(key.c_str(), "dataInFlight");
|
|
|
|
}
|
|
|
|
if (tLogQueueInfo.isReady() && tLogQueueInfo.isError()) {
|
|
|
|
auto key = "NotReady" + std::to_string(notReadyCount++);
|
|
|
|
evt.detail(key.c_str(), "tLogQueueInfo");
|
|
|
|
}
|
|
|
|
if (dataDistributionQueueSize.isReady() && dataDistributionQueueSize.isError()) {
|
|
|
|
auto key = "NotReady" + std::to_string(notReadyCount++);
|
|
|
|
evt.detail(key.c_str(), "dataDistributionQueueSize");
|
|
|
|
}
|
|
|
|
if (teamCollectionValid.isReady() && teamCollectionValid.isError()) {
|
|
|
|
auto key = "NotReady" + std::to_string(notReadyCount++);
|
|
|
|
evt.detail(key.c_str(), "teamCollectionValid");
|
|
|
|
}
|
|
|
|
if (storageQueueSize.isReady() && storageQueueSize.isError()) {
|
|
|
|
auto key = "NotReady" + std::to_string(notReadyCount++);
|
|
|
|
evt.detail(key.c_str(), "storageQueueSize");
|
|
|
|
}
|
|
|
|
if (dataDistributionActive.isReady() && dataDistributionActive.isError()) {
|
|
|
|
auto key = "NotReady" + std::to_string(notReadyCount++);
|
|
|
|
evt.detail(key.c_str(), "dataDistributionActive");
|
|
|
|
}
|
|
|
|
if (storageServersRecruiting.isReady() && storageServersRecruiting.isError()) {
|
|
|
|
auto key = "NotReady" + std::to_string(notReadyCount++);
|
|
|
|
evt.detail(key.c_str(), "storageServersRecruiting");
|
|
|
|
}
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(delay(1.0));
|
2017-05-26 04:48:44 +08:00
|
|
|
numSuccesses = 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
Future<Void> quietDatabase(Database const& cx,
|
2021-07-12 13:04:38 +08:00
|
|
|
Reference<AsyncVar<ServerDBInfo> const> const& dbInfo,
|
2021-03-11 02:06:03 +08:00
|
|
|
std::string phase,
|
|
|
|
int64_t dataInFlightGate,
|
|
|
|
int64_t maxTLogQueueGate,
|
|
|
|
int64_t maxStorageServerQueueGate,
|
|
|
|
int64_t maxDataDistributionQueueSize,
|
|
|
|
int64_t maxPoppedVersionLag) {
|
|
|
|
return waitForQuietDatabase(cx,
|
|
|
|
dbInfo,
|
|
|
|
phase,
|
|
|
|
dataInFlightGate,
|
|
|
|
maxTLogQueueGate,
|
|
|
|
maxStorageServerQueueGate,
|
|
|
|
maxDataDistributionQueueSize,
|
|
|
|
maxPoppedVersionLag);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|