foundationdb/fdbserver/workloads/BulkSetup.actor.h

342 lines
12 KiB
C
Raw Normal View History

2017-05-26 04:48:44 +08:00
/*
* BulkSetup.actor.h
*
* This source file is part of the FoundationDB open source project
*
* Copyright 2013-2018 Apple Inc. and the FoundationDB project authors
*
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
*
2017-05-26 04:48:44 +08:00
* http://www.apache.org/licenses/LICENSE-2.0
*
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.
*/
#pragma once
// When actually compiled (NO_INTELLISENSE), include the generated
// version of this file. In intellisense use the source version.
#if defined(NO_INTELLISENSE) && !defined(FDBSERVER_BULK_SETUP_ACTOR_G_H)
#define FDBSERVER_BULK_SETUP_ACTOR_G_H
#include "BulkSetup.actor.g.h"
#elif !defined(FDBSERVER_BULK_SETUP_ACTOR_H)
#define FDBSERVER_BULK_SETUP_ACTOR_H
#include "flow/actorcompiler.h"
#include "fdbclient/NativeAPI.h"
#include "workloads.h"
#include "fdbserver/ServerDBInfo.h"
#include "fdbserver/QuietDatabase.h"
#include "fdbrpc/simulator.h"
#include <string>
ACTOR template<class T>
Future<bool> checkRangeSimpleValueSize( Database cx, T* workload, uint64_t begin, uint64_t end) {
loop {
state Transaction tr(cx);
try {
state Standalone< KeyValueRef > firstKV = (*workload)( begin );
state Standalone< KeyValueRef > lastKV = (*workload)( end - 1 );
state Future< Optional< Value > > first = tr.get( firstKV.key );
state Future< Optional< Value > > last = tr.get( lastKV.key );
Void _ = wait( success( first ) && success( last ) );
return first.get().present() && last.get().present();
} catch (Error& e) {
TraceEvent("CheckRangeError").error(e).detail("Begin", begin).detail("End", end);
2017-05-26 04:48:44 +08:00
Void _ = wait( tr.onError(e) );
}
}
}
// Returns true if the range was added
ACTOR template<class T>
Future<uint64_t> setupRange( Database cx, T* workload, uint64_t begin, uint64_t end) {
state uint64_t bytesInserted = 0;
loop {
state Transaction tr(cx);
try {
//if( g_random->random01() < 0.001 )
// tr.debugTransaction( g_random->randomUniqueID() );
state Standalone<KeyValueRef> sampleKV = (*workload)( begin );
Optional<Value> f = wait( tr.get( sampleKV.key ) );
if (f.present()) {
// if( sampleKV.value.size() == f.get().size() ) {
//TraceEvent("BulkSetupRangeAlreadyPresent")
// .detail("Begin", begin)
// .detail("End", end)
// .detail("Version", tr.getReadVersion().get())
// .detail("Key", printable((*workload)(begin).key))
// .detailf("From", "%016llx", debug_lastLoadBalanceResultEndpointToken);
return bytesInserted; // The transaction already completed!
//}
//TraceEvent(SevError, "BulkSetupConflict")
// .detail("Begin", begin)
// .detail("End", end)
// .detail("ExpectedLength", sampleKV.value.size())
// .detail("FoundLength", f.get().size());
//throw operation_failed();
}
// Predefine a single large write conflict range over the whole key space
tr.addWriteConflictRange( KeyRangeRef(
workload->keyForIndex( begin ),
keyAfter( workload->keyForIndex( end ) ) ) );
bytesInserted = 0;
for(uint64_t n=begin; n<end; n++) {
Standalone<KeyValueRef> kv = (*workload)( n );
tr.set( kv.key, kv.value );
bytesInserted += kv.key.size() + kv.value.size();
}
Void _ = wait( tr.commit() );
return bytesInserted;
} catch (Error& e) {
Void _ = wait( tr.onError(e) );
}
}
}
ACTOR template<class T>
Future<uint64_t> setupRangeWorker( Database cx, T* workload, vector<std::pair<uint64_t,uint64_t>>* jobs, double maxKeyInsertRate, int keySaveIncrement, int actorId) {
state double nextStart;
state uint64_t loadedRanges = 0;
state int lastStoredKeysLoaded = 0;
state uint64_t keysLoaded = 0;
state uint64_t bytesStored = 0;
while (jobs->size()){
state pair<uint64_t, uint64_t> job = jobs->back();
jobs->pop_back();
nextStart = now() + (job.second-job.first)/maxKeyInsertRate;
uint64_t numBytes = wait( setupRange(cx, workload, job.first, job.second) );
if( numBytes > 0 )
loadedRanges++;
if(keySaveIncrement > 0)
{
keysLoaded += job.second - job.first;
bytesStored += numBytes;
if(keysLoaded - lastStoredKeysLoaded >= keySaveIncrement || jobs->size() == 0)
{
state Transaction tr(cx);
try
{
std::string countKey = format("keycount|%d|%d", workload->clientId, actorId);
std::string bytesKey = format("bytesstored|%d|%d", workload->clientId, actorId);
tr.set(StringRef(countKey), StringRef((uint8_t*)&keysLoaded, sizeof(uint64_t)));
tr.set(StringRef(bytesKey), StringRef((uint8_t*)&bytesStored, sizeof(uint64_t)));
Void _ = wait(tr.commit());
lastStoredKeysLoaded = keysLoaded;
}
catch(Error& e)
{
Void _ = wait(tr.onError(e));
}
}
}
if (now() < nextStart)
Void _ = wait( delayUntil(nextStart));
}
return loadedRanges;
}
//Periodically determines how many keys have been inserted. If the count has just exceeded a count of interest, computes the time taken
//to reach that mark. Returns a vector of times (in seconds) corresponding to the counts in the countsOfInterest vector.
//Expects countsOfInterest to be sorted in ascending order
ACTOR static Future<vector<pair<uint64_t, double> > > trackInsertionCount(Database cx, vector<uint64_t> countsOfInterest, double checkInterval)
{
state KeyRange keyPrefix = KeyRangeRef(std::string("keycount"), std::string("keycount") + char(255));
state KeyRange bytesPrefix = KeyRangeRef(std::string("bytesstored"), std::string("bytesstored") + char(255));
state Transaction tr(cx);
state uint64_t lastInsertionCount = 0;
state int currentCountIndex = 0;
state vector<pair<uint64_t, double> > countInsertionRates;
state double startTime = now();
while(currentCountIndex < countsOfInterest.size())
{
try
{
state Future<Standalone<RangeResultRef>> countFuture = tr.getRange(keyPrefix, 1000000000);
state Future<Standalone<RangeResultRef>> bytesFuture = tr.getRange(bytesPrefix, 1000000000);
Void __ = wait(success(countFuture) && success(bytesFuture));
Standalone<RangeResultRef> counts = countFuture.get();
Standalone<RangeResultRef> bytes = bytesFuture.get();
uint64_t numInserted = 0;
for(int i = 0; i < counts.size(); i++)
numInserted += *(uint64_t*)counts[i].value.begin();
uint64_t bytesInserted = 0;
for(int i = 0; i < bytes.size(); i++)
bytesInserted += *(uint64_t*)bytes[i].value.begin();
while(currentCountIndex < countsOfInterest.size() && countsOfInterest[currentCountIndex] > lastInsertionCount && countsOfInterest[currentCountIndex] <= numInserted)
countInsertionRates.push_back(pair<uint64_t, double>(countsOfInterest[currentCountIndex++], bytesInserted / (now() - startTime)));
lastInsertionCount = numInserted;
Void _ = wait(delay(checkInterval));
}
catch(Error& e)
{
Void _ = wait(tr.onError(e));
}
}
return countInsertionRates;
}
ACTOR template<class T>
Future<Void> bulkSetup( Database cx, T* workload, uint64_t nodeCount, Promise<double> setupTime,
bool valuesInconsequential = false, double postSetupWarming = 0.0, double maxKeyInsertRate = 1e12,
vector<uint64_t> insertionCountsToMeasure = vector<uint64_t>(), Promise<vector<pair<uint64_t, double> > > ratesAtKeyCounts = Promise<vector<pair<uint64_t, double> > >(),
int keySaveIncrement = 0, double keyCheckInterval = 0.1 ) {
state vector<pair<uint64_t,uint64_t>> jobs;
state uint64_t startNode = (nodeCount * workload->clientId) / workload->clientCount;
state uint64_t endNode = (nodeCount * (workload->clientId+1)) / workload->clientCount;
state double start = now();
TraceEvent("BulkSetupStart")
.detail("NodeCount", nodeCount)
.detail("ValuesInconsequential", valuesInconsequential)
.detail("PostSetupWarming", postSetupWarming)
.detail("MaxKeyInsertRate", maxKeyInsertRate);
// For bulk data schemes where the value of the key is not critical to operation, check to
// see if the database has already been set up.
if( valuesInconsequential ) {
bool present = wait( checkRangeSimpleValueSize( cx, workload, startNode, endNode ) );
if( present ) {
TraceEvent("BulkSetupRangeAlreadyPresent")
.detail("Begin", startNode)
.detail("End", endNode)
.detail("CheckMethod", "SimpleValueSize");
setupTime.send(0.0);
ratesAtKeyCounts.send(vector<pair<uint64_t, double> >());
return Void();
} else {
TraceEvent("BulkRangeNotFound")
.detail("Begin", startNode)
.detail("End", endNode)
.detail("CheckMethod", "SimpleValueSize");
}
}
Void _ = wait( delay( g_random->random01() / 4 ) ); // smear over .25 seconds
state int BULK_SETUP_WORKERS = 40;
// See that each chunk inserted is about 10KB
int size_total = 0;
for( int i = 0; i < 100; i++ ) {
Standalone<KeyValueRef> sampleKV = (*workload)( startNode + (uint64_t)(g_random->random01()*(endNode - startNode)) );
size_total += sampleKV.key.size() + sampleKV.value.size();
}
state int BULK_SETUP_RANGE_SIZE = size_total == 0 ? 50 : std::max(1, 10000 / (size_total / 100));
TraceEvent((workload->description() + "SetupStart").c_str())
.detail("ClientIdx", workload->clientId)
.detail("ClientCount", workload->clientCount)
.detail("StartingNode", startNode)
.detail("NodesAssigned", endNode-startNode)
.detail("NodesPerInsertion", BULK_SETUP_RANGE_SIZE)
.detail("SetupActors", BULK_SETUP_WORKERS);
// create a random vector of range-create jobs
for(uint64_t n=startNode; n<endNode; n+=BULK_SETUP_RANGE_SIZE)
jobs.push_back( std::make_pair( n, std::min(endNode, n+BULK_SETUP_RANGE_SIZE) ) );
g_random->randomShuffle(jobs);
// fire up the workers and wait for them to eat all the jobs
double maxWorkerInsertRate = maxKeyInsertRate / BULK_SETUP_WORKERS / workload->clientCount;
state vector<Future<uint64_t>> fs;
state Future<vector<pair<uint64_t, double> > > insertionTimes = vector<pair<uint64_t, double> >();
if(insertionCountsToMeasure.size() > 0)
{
std::sort(insertionCountsToMeasure.begin(), insertionCountsToMeasure.end());
for(int i = 0; i < insertionCountsToMeasure.size(); i++)
if(insertionCountsToMeasure[i] > nodeCount)
insertionCountsToMeasure.erase(insertionCountsToMeasure.begin() + i, insertionCountsToMeasure.end());
if(workload->clientId == 0)
insertionTimes = trackInsertionCount(cx, insertionCountsToMeasure, keyCheckInterval);
if(keySaveIncrement <= 0)
keySaveIncrement = BULK_SETUP_RANGE_SIZE;
}
else
keySaveIncrement = 0;
for(int j=0; j<BULK_SETUP_WORKERS; j++)
fs.push_back( setupRangeWorker(cx, workload, &jobs, maxWorkerInsertRate, keySaveIncrement, j) );
try {
Void _ = wait( success(insertionTimes) && waitForAll(fs) );
} catch(Error& e) {
if( e.code() == error_code_operation_failed ) {
TraceEvent(SevError, "BulkSetupFailed").error(e);
}
throw;
}
ratesAtKeyCounts.send(insertionTimes.get());
uint64_t rangesInserted = 0;
for(int i = 0; i < fs.size(); i++)
rangesInserted += fs[i].get();
state double elapsed = now() - start;
setupTime.send( elapsed );
TraceEvent("SetupLoadComplete")
.detail("LoadedRanges", rangesInserted)
.detail("Duration", elapsed)
.detail("Nodes", nodeCount);
// Here we wait for data in flight to go to 0 (this will not work on a database with other users)
if( postSetupWarming != 0 ) {
try {
Void _ = wait( delay( 5.0 ) ); // Wait for the data distribution in a small test to start
loop {
int64_t inFlight = wait( getDataInFlight( cx, workload->dbInfo ) );
TraceEvent("DynamicWarming").detail("InFlight", inFlight);
2017-05-26 04:48:44 +08:00
if( inFlight > 1e6 ) { // Wait for just 1 MB to be in flight
Void _ = wait( delay( 1.0 ) );
} else {
Void _ = wait( delay( 1.0 ) );
TraceEvent("DynamicWarmingDone");
break;
}
}
} catch (Error& e) {
if( e.code() == error_code_actor_cancelled )
throw;
TraceEvent("DynamicWarmingError").error(e);
if( postSetupWarming > 0 )
Void _ = wait( timeout( databaseWarmer( cx ), postSetupWarming, Void() ) );
}
}
TraceEvent((workload->description() + "SetupOK").c_str())
.detail("ClientIdx", workload->clientId)
.detail("WarmingDelay", postSetupWarming)
.detail("KeyLoadElapsedTime", elapsed);
return Void();
}
#endif