com.yahoo.vespa.hosted.provision.autoscale.ClusterModel Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of node-repository Show documentation
Show all versions of node-repository Show documentation
Keeps track of node assignment in a multi-application setup.
// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
package com.yahoo.vespa.hosted.provision.autoscale;
import com.yahoo.config.provision.CloudAccount;
import com.yahoo.config.provision.ClusterSpec;
import com.yahoo.vespa.hosted.provision.Node;
import com.yahoo.vespa.hosted.provision.NodeList;
import com.yahoo.vespa.hosted.provision.NodeRepository;
import com.yahoo.vespa.hosted.provision.applications.Application;
import com.yahoo.vespa.hosted.provision.applications.Cluster;
import com.yahoo.config.provision.CapacityPolicies;
import java.time.Clock;
import java.time.Duration;
import java.time.Instant;
import java.util.Objects;
import java.util.OptionalDouble;
/**
* A cluster with its associated metrics which allows prediction about its future behavior.
* For single-threaded, short-term usage.
*
* @author bratseth
*/
public final class ClusterModel {
/** Containers typically use more cpu right after generation change, so discard those metrics */
public static final Duration warmupDuration = Duration.ofMinutes(7);
/** If we have less than this query rate, we cannot be fully confident in our load data, which influences some decisions. */
public static final double queryRateGivingFullConfidence = 100.0;
static final double idealQueryCpuLoad = 0.75;
static final double idealWriteCpuLoad = 0.95;
static final double idealContainerMemoryLoad = 0.9;
static final double idealContentMemoryLoad = 0.65;
static final double idealContainerDiskLoad = 0.95;
static final double idealContentDiskLoad = 0.6;
// Memory for other processes running on the node (config-proxy, metrics-proxy).
// Keep in sync with config-model/NodeResourcesTuning.
static final double nodeMemoryOverheadGb = 0.7;
// When a query is issued on a node the cost is the sum of a fixed cost component and a cost component
// proportional to document count. We must account for this when comparing configurations with more or fewer nodes.
// TODO: Measure this, and only take it into account with queries
private static final double fixedCpuCostFraction = 0.1;
private final NodeRepository nodeRepository;
private final CapacityPolicies capacityPolicies;
private final Application application;
private final ClusterSpec clusterSpec;
private final Cluster cluster;
private final AllocatableResources current;
private final MemoryModel memory = new MemoryModel();
private final DiskModel disk = new DiskModel();
/**
* The current active nodes of this cluster, including retired,
* or empty if this models a new cluster not yet deployed.
*/
private final NodeList nodes;
private final Clock clock;
private final Duration scalingDuration;
private final Duration allocationDuration;
private final ClusterTimeseries clusterTimeseries;
private final ClusterNodesTimeseries nodeTimeseries;
private final Instant at;
// Lazily initialized members
private Double queryFractionOfMax = null;
private Double maxQueryGrowthRate = null;
private OptionalDouble averageQueryRate = OptionalDouble.empty();
public ClusterModel(NodeRepository nodeRepository,
Application application,
ClusterSpec clusterSpec,
Cluster cluster,
NodeList clusterNodes,
AllocatableResources current,
MetricsDb metricsDb,
Clock clock) {
this.nodeRepository = nodeRepository;
this.capacityPolicies = nodeRepository.capacityPoliciesFor(application.id());
this.application = application;
this.clusterSpec = clusterSpec;
this.cluster = cluster;
this.nodes = clusterNodes;
this.current = current;
this.clock = clock;
this.scalingDuration = cluster.scalingDuration();
this.allocationDuration = cluster.allocationDuration(clusterSpec);
this.clusterTimeseries = metricsDb.getClusterTimeseries(application.id(), cluster.id());
this.nodeTimeseries = new ClusterNodesTimeseries(scalingDuration(), cluster, nodes, metricsDb);
this.at = clock.instant();
}
ClusterModel(NodeRepository nodeRepository,
Application application,
ClusterSpec clusterSpec,
Cluster cluster,
AllocatableResources current,
Clock clock,
Duration scalingDuration,
Duration allocationDuration,
ClusterTimeseries clusterTimeseries,
ClusterNodesTimeseries nodeTimeseries) {
this.nodeRepository = nodeRepository;
this.capacityPolicies = nodeRepository.capacityPoliciesFor(application.id());
this.application = application;
this.clusterSpec = clusterSpec;
this.cluster = cluster;
this.nodes = NodeList.of();
this.current = current;
this.clock = clock;
this.scalingDuration = scalingDuration;
this.allocationDuration = allocationDuration;
this.clusterTimeseries = clusterTimeseries;
this.nodeTimeseries = nodeTimeseries;
this.at = clock.instant();
}
/**
* The central decision made in autoscaling.
*
* @return the relative load adjustment that should be made to this cluster given available measurements.
* For example, a load adjustment of 2 means we should allocate twice the amount of that resources.
*/
public Load loadAdjustment() {
if (nodeTimeseries().measurementsPerNode() < 0.5) return Load.one(); // Don't change based on very little data
Load adjustment = peakLoad().divide(idealLoad());
if (! safeToScaleDown())
adjustment = adjustment.map(v -> v < 1 ? 1 : v);
return adjustment;
}
public Application application() { return application; }
public ClusterSpec clusterSpec() { return clusterSpec; }
public CloudAccount cloudAccount() { return cluster.cloudAccount().orElse(CloudAccount.empty); }
public AllocatableResources current() { return current; }
private ClusterNodesTimeseries nodeTimeseries() { return nodeTimeseries; }
private ClusterTimeseries clusterTimeseries() { return clusterTimeseries; }
/** Returns the instant this model was created. */
public Instant at() { return at; }
/** Returns the predicted duration of a rescaling of this cluster */
public Duration scalingDuration() { return scalingDuration; }
/**
* Returns the predicted duration of a resource change in this cluster,
* until we, or the application , will change it again.
*/
public Duration allocationDuration() { return allocationDuration; }
public boolean isEmpty() { return nodeTimeseries().isEmpty(); }
public boolean isContent() { return clusterSpec.type().isContent(); }
/** Returns the predicted duration of data redistribution in this cluster. */
public Duration redistributionDuration() {
if (! isContent()) return Duration.ofMinutes(0);
return scalingDuration(); // TODO: Estimate separately
}
/** Returns the predicted duration of replacing all the nodes in this cluster. */
public Duration nodeReplacementDuration() {
return Duration.ofMinutes(5); // TODO: Estimate?
}
/** Returns the average of the peak load measurement in each dimension, from each node. */
public Load peakLoad() {
return nodeTimeseries().peakLoad();
}
/** Returns the relative load adjustment accounting for redundancy in this. */
private Load redundancyAdjustment(Instant now) {
return loadWith(nodeCount(), groupCount(), now);
}
public boolean isExclusive() {
return nodeRepository.exclusivity().allocation(clusterSpec);
}
public boolean isStable(NodeRepository nodeRepository) {
// The cluster is processing recent changes
if (nodes.stream().anyMatch(node -> node.status().wantToRetire() ||
node.allocation().get().membership().retired() ||
node.allocation().get().removable()))
return false;
// A deployment is ongoing
if ( ! nodeRepository.nodes().list(Node.State.reserved, Node.State.provisioned).owner(application.id()).isEmpty())
return false;
return true;
}
/** Are we in a position to make decisions to scale down at this point? */
public boolean safeToScaleDown() {
if (hasScaledIn(scalingDuration().multipliedBy(3))) return false;
if (nodeTimeseries().nodesMeasured() != nodeCount()) return false;
return true;
}
public static Duration minScalingDuration() {
return Duration.ofMinutes(5);
}
/** Transforms the given load adjustment to an equivalent adjustment given a target number of nodes and groups. */
public Load loadAdjustmentWith(int nodes, int groups, Load loadAdjustment, Instant now) {
return loadAdjustment // redundancy adjusted target relative to current load
.multiply(loadWith(nodes, groups, now)) // redundancy aware adjustment with these counts
.divide(redundancyAdjustment(now)); // correct for double redundancy adjustment
}
/**
* Returns the relative load adjustment accounting for redundancy given these nodes+groups
* relative to node nodes+groups in this.
*/
Load loadWith(int givenNodes, int givenGroups, Instant now) {
int nodes = nodesAdjustedForRedundancy(givenNodes, givenGroups);
int groups = groupsAdjustedForRedundancy(givenNodes, givenGroups);
if (clusterSpec().type() == ClusterSpec.Type.content) { // load scales with node share of content
int groupSize = nodes / groups;
// Cpu: Query cpu scales with cluster size, write cpu scales with group size
// Memory and disk: Scales with group size
// The fixed cost portion of cpu does not scale with changes to the node count
double queryCpuPerGroup = fixedCpuCostFraction + (1 - fixedCpuCostFraction) * groupSize() / groupSize;
double queryCpu = queryCpuPerGroup * groupCount() / groups;
double writeCpu = (double)groupSize() / groupSize;
CpuModel cpu = cpu(now);
return new Load(cpu.queryFraction() * queryCpu + (1 - cpu.queryFraction()) * writeCpu,
(1 - memory.fixedFraction()) * (double) groupSize() / groupSize + memory.fixedFraction() * 1,
(double)groupSize() / groupSize,
1,
1);
}
else {
return new Load((double) nodeCount() / nodes, 1, 1, 1, 1);
}
}
/**
* Returns the ideal load across the nodes of this such that each node will be at ideal load
* if one of the nodes go down.
*/
public Load idealLoad() {
CpuModel cpu = cpu(clock.instant());
var ideal = new Load(cpu.idealLoad(), memory.idealLoad(), disk.idealLoad(), cpu.idealLoad(), memory.idealLoad()).divide(redundancyAdjustment(cpu.at()));
if ( !cluster.bcpGroupInfo().isEmpty() && cluster.bcpGroupInfo().queryRate() > 0) {
// Since we have little local information, use information about query cost in other groups
Load bcpGroupIdeal = adjustQueryDependentIdealLoadByBcpGroupInfo(ideal, cpu);
// Do a weighted sum of the ideal "vote" based on local and bcp group info.
// This avoids any discontinuities with a near-zero local query rate.
double localInformationWeight = Math.min(1, averageQueryRate(cpu.at()).orElse(0) /
Math.min(queryRateGivingFullConfidence, cluster.bcpGroupInfo().queryRate()));
ideal = ideal.multiply(localInformationWeight).add(bcpGroupIdeal.multiply(1 - localInformationWeight));
}
return ideal;
}
public CpuModel cpu(Instant now) {
return new CpuModel(now);
}
private boolean canRescaleWithinBcpDeadline() {
return scalingDuration().minus(cluster.clusterInfo().bcpDeadline()).isNegative();
}
public Autoscaling.Metrics metrics() {
Instant now = clock.instant();
return new Autoscaling.Metrics(averageQueryRate(now).orElse(0),
growthRateHeadroom(now),
cpu(now).costPerQuery().orElse(0));
}
private Load adjustQueryDependentIdealLoadByBcpGroupInfo(Load ideal, CpuModel cpu) {
double currentClusterTotalVcpuPerGroup = nodes.not().retired().first().get().resources().vcpu() * groupSize();
double targetQueryRateToHandle = ( canRescaleWithinBcpDeadline() ? averageQueryRate(cpu.at()).orElse(0)
: cluster.bcpGroupInfo().queryRate() )
* cluster.bcpGroupInfo().growthRateHeadroom() * trafficShiftHeadroom(cpu.at());
double neededTotalVcpuPerGroup = cluster.bcpGroupInfo().cpuCostPerQuery() * targetQueryRateToHandle / groupCount() +
( 1 - cpu.queryFraction()) * cpu.idealLoad() *
(clusterSpec.type().isContainer() ? 1 : groupSize());
// Max 1: Only use bcp group info if it indicates that we need to scale *up*
double cpuAdjustment = Math.max(1.0, neededTotalVcpuPerGroup / currentClusterTotalVcpuPerGroup);
return ideal.withCpu(ideal.cpu() / cpuAdjustment);
}
private boolean hasScaledIn(Duration period) {
if (cluster.lastScalingEvent().isEmpty()) return false;
var lastCompletion = cluster.lastScalingEvent().get().completion();
if (lastCompletion.isEmpty()) return true; // Ongoing
return lastCompletion.get().isAfter(clock.instant().minus(period));
}
/**
* Returns the predicted max query growth rate per minute as a fraction of the average traffic
* in the scaling window.
*/
private double maxQueryGrowthRate(Instant now) {
if (maxQueryGrowthRate != null) return maxQueryGrowthRate;
return maxQueryGrowthRate = clusterTimeseries().maxQueryGrowthRate(scalingDuration(), now);
}
/** Returns the average query rate in the scaling window as a fraction of the max observed query rate. */
private double queryFractionOfMax(Instant now) {
if (queryFractionOfMax != null) return queryFractionOfMax;
return queryFractionOfMax = clusterTimeseries().queryFractionOfMax(scalingDuration(), now);
}
/** Returns the average query rate in the scaling window. */
private OptionalDouble averageQueryRate(Instant now) {
if (averageQueryRate.isPresent()) return averageQueryRate;
return averageQueryRate = clusterTimeseries().queryRate(scalingDuration(), now);
}
/** The number of nodes this cluster has, or will have if not deployed yet. */
// TODO: Make this the deployed, not current count
public int nodeCount() {
if ( ! nodes.isEmpty()) return (int)nodes.not().retired().stream().count();
return cluster.minResources().nodes();
}
/** The number of groups this cluster has, or will have if not deployed yet. */
// TODO: Make this the deployed, not current count
private int groupCount() {
if ( ! nodes.isEmpty()) return (int)nodes.not().retired().stream().mapToInt(node -> node.allocation().get().membership().cluster().group().get().index()).distinct().count();
return cluster.minResources().groups();
}
private int groupSize() {
// ceil: If the division does not produce a whole number we assume some node is missing
return (int)Math.ceil((double)nodeCount() / groupCount());
}
private static int nodesAdjustedForRedundancy(int nodes, int groups) {
int groupSize = (int)Math.ceil((double)nodes / groups);
return nodes > 1 ? (groups == 1 ? nodes - 1 : nodes - groupSize) : nodes;
}
private static int groupsAdjustedForRedundancy(int nodes, int groups) {
return nodes > 1 ? (groups == 1 ? 1 : groups - 1) : groups;
}
/** Returns the headroom for growth during organic traffic growth as a multiple of current resources. */
private double growthRateHeadroom(Instant now) {
if ( ! nodeRepository.zone().environment().isProduction()) return 1;
double growthRateHeadroom = 1 + maxQueryGrowthRate(now) * scalingDuration().toMinutes();
// Cap headroom at 10% above the historical observed peak
if (queryFractionOfMax(now) != 0)
growthRateHeadroom = Math.min(growthRateHeadroom, 1 / queryFractionOfMax(now) + 0.1);
return adjustByConfidence(growthRateHeadroom, now);
}
/**
* Returns the headroom is needed to handle sudden arrival of additional traffic due to another zone going down
* as a multiple of current resources.
*/
private double trafficShiftHeadroom(Instant now) {
if ( ! nodeRepository.zone().environment().isProduction()) return 1;
if (canRescaleWithinBcpDeadline()) return 1;
double trafficShiftHeadroom;
if (application.status().maxReadShare() == 0) // No traffic fraction data
trafficShiftHeadroom = 2.0; // assume we currently get half of the max possible share of traffic
else if (application.status().currentReadShare() == 0)
trafficShiftHeadroom = 1/application.status().maxReadShare();
else
trafficShiftHeadroom = application.status().maxReadShare() / application.status().currentReadShare();
return adjustByConfidence(Math.min(trafficShiftHeadroom, 1/application.status().maxReadShare()), now);
}
/**
* Headroom values are a multiplier of the current query rate.
* Adjust this value closer to 1 if the query rate is too low to derive statistical conclusions
* with high confidence to avoid large adjustments caused by random noise due to low traffic numbers.
*/
private double adjustByConfidence(double headroom, Instant now) {
return ( (headroom -1 ) * Math.min(1, averageQueryRate(now).orElse(0) / queryRateGivingFullConfidence) ) + 1;
}
public class CpuModel {
private final Instant at;
public CpuModel(Instant at) {
this.at = Objects.requireNonNull(at);
}
Instant at() {
return at;
}
/** Ideal cpu load must take the application traffic fraction into account. */
double idealLoad() {
double queryCpuFraction = queryFraction();
// Assumptions: 1) Write load is not organic so we should not increase to handle potential future growth.
// (TODO: But allow applications to set their target write rate and size for that)
// 2) Write load does not change in BCP scenarios.
return queryCpuFraction * 1/growthRateHeadroom(at) * 1 / trafficShiftHeadroom(at) * idealQueryCpuLoad +
(1 - queryCpuFraction) * idealWriteCpuLoad;
}
OptionalDouble costPerQuery() {
if (averageQueryRate(at()).isEmpty() || averageQueryRate(at()).getAsDouble() == 0.0) return OptionalDouble.empty();
// TODO: Query rate should generally be sampled at the time where we see the peak resource usage
int fanOut = clusterSpec.type().isContainer() ? 1 : groupSize();
return OptionalDouble.of(peakLoad().cpu() * queryFraction() * fanOut * nodes.not().retired().first().get().resources().vcpu()
/ averageQueryRate(at).getAsDouble() / groupCount());
}
/** The estimated fraction of cpu usage which goes to processing queries vs. writes */
double queryFraction() {
OptionalDouble writeRate = clusterTimeseries().writeRate(scalingDuration(), at);
if (averageQueryRate(at).orElse(0) == 0 && writeRate.orElse(0) == 0) return queryFraction(0.5);
return queryFraction(averageQueryRate(at).orElse(0) / (averageQueryRate(at).orElse(0) + writeRate.orElse(0)));
}
double queryFraction(double queryRateFraction) {
double relativeQueryCost = 9; // How much more expensive are queries than writes? TODO: Measure
double writeFraction = 1 - queryRateFraction;
return queryRateFraction * relativeQueryCost / (queryRateFraction * relativeQueryCost + writeFraction);
}
public String toString() {
return "cpu model idealLoad: " + idealLoad() + ", queryFraction: " + queryFraction() +
", growthRateHeadroom: " + growthRateHeadroom(at) + ", trafficShiftHeadroom: " + trafficShiftHeadroom(at);
}
}
private class MemoryModel {
double idealLoad() {
if (clusterSpec.type().isContainer()) return idealContainerMemoryLoad;
if (clusterSpec.type() == ClusterSpec.Type.admin) return idealContainerMemoryLoad; // Not autoscaled, but ideal shown in console
return idealContentMemoryLoad;
}
/**
* Returns the fraction of memory of the current allocation which is currently consumed by
* fixed data structures which take the same amount of space regardless of document volume.
*/
double fixedFraction() {
if (clusterSpec().type().isContainer()) return 1.0;
double fixedMemory = nodeMemoryOverheadGb +
(averageReal() - nodeMemoryOverheadGb) * 0.05; // TODO: Measure actual content node usage
return fixedMemory / averageReal();
}
double averageReal() {
if (nodes.isEmpty()) { // we're estimating
var initialResources = capacityPolicies.specifyFully(cluster.minResources().nodeResources(), clusterSpec);
return nodeRepository.resourcesCalculator().requestToReal(initialResources,
cloudAccount(),
nodeRepository.exclusivity().allocation(clusterSpec),
false).memoryGiB();
}
else {
return nodes.stream()
.mapToDouble(node -> nodeRepository.resourcesCalculator().realResourcesOf(node, nodeRepository).memoryGiB())
.average()
.getAsDouble();
}
}
}
private class DiskModel {
double idealLoad() {
// Stateless clusters are not expected to consume more disk over time -
// if they do it is due to logs which will be rotated away right before the disk is full
return clusterSpec.isStateful() ? idealContentDiskLoad : idealContainerDiskLoad;
}
}
}
© 2015 - 2024 Weber Informatics LLC | Privacy Policy