com.yahoo.vespa.hosted.provision.autoscale.AllocationOptimizer 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.
The newest version!
// 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.ClusterResources;
import com.yahoo.config.provision.Flavor;
import com.yahoo.config.provision.IntRange;
import com.yahoo.config.provision.NodeResources;
import com.yahoo.vespa.hosted.provision.NodeRepository;
import java.time.Instant;
import java.util.ArrayList;
import java.util.List;
import java.util.Optional;
import java.util.logging.Logger;
import static com.yahoo.vespa.hosted.provision.autoscale.Autoscaler.headroomRequiredToScaleDown;
/**
* A searcher of the space of possible allocation
*
* @author bratseth
*/
public class AllocationOptimizer {
private static final Logger log = Logger.getLogger(AllocationOptimizer.class.getName());
// The min and max nodes to consider when not using application supplied limits
private static final int minimumNodes = 2; // Since this number includes redundancy it cannot be lower than 2
private static final int maximumNodes = 150;
// A lower bound on the number of vCPUs to suggest. Flavors with fewer vCPUs than this are rarely good for
// anything but testing.
private static final int minimumSuggestedVcpus = 2;
private final NodeRepository nodeRepository;
public AllocationOptimizer(NodeRepository nodeRepository) {
this.nodeRepository = nodeRepository;
}
/**
* Searches the space of possible allocations given a target relative load
* and (optionally) cluster limits and returns the best alternative.
*
* @return the best allocation, if there are any possible legal allocations, fulfilling the target
* fully or partially, within the limits
*/
public Optional findBestAllocation(Load loadAdjustment, ClusterModel model, Limits limits, boolean logDetails) {
return findBestAllocations(loadAdjustment, model, limits, logDetails).stream().findFirst();
}
/**
* Searches the space of possible allocations given a target relative load
* and (optionally) cluster limits and returns the best alternative.
*
* @return the best allocations, if there are any possible legal allocations, fulfilling the target
* fully or partially, within the limits. The list contains the three best allocations, sorted from most to least preferred.
*/
public List findBestAllocations(Load loadAdjustment, ClusterModel model, Limits limits, boolean logDetails) {
if (limits.isEmpty())
limits = Limits.of(new ClusterResources(minimumNodes, 1, NodeResources.unspecified()),
new ClusterResources(maximumNodes, maximumNodes, NodeResources.unspecified()),
IntRange.empty());
else
limits = atLeast(minimumNodes, limits).fullySpecified(model.current().clusterSpec(), nodeRepository, model.application().id());
List bestAllocations = new ArrayList<>();
List availableRealHostResources = availableRealHostResources(model);
for (int groups = limits.min().groups(); groups <= limits.max().groups(); groups++) {
for (int nodes = limits.min().nodes(); nodes <= limits.max().nodes(); nodes++) {
if (nodes % groups != 0) continue;
if ( ! limits.groupSize().includes(nodes / groups)) continue;
var resources = new ClusterResources(nodes,
groups,
nodeResourcesWith(nodes, groups,
limits, loadAdjustment, model));
var allocatableResources = AllocatableResources.from(resources,
model.application().id(),
model.current().clusterSpec(),
limits,
availableRealHostResources,
model,
nodeRepository);
if (allocatableResources.isEmpty()) continue;
bestAllocations.add(allocatableResources.get());
if (logDetails) {
log.info("Adding allocatableResources to list for " + model.application().id() + " in " + model.current().clusterSpec().id() + ": "
+ "\n\t" + allocatableResources.get());
}
}
}
if (logDetails) {
log.info("Found " + bestAllocations.size() + " legal allocations for " + model.application().id() + " in " + model.current().clusterSpec().id());
}
return bestAllocations.stream()
.sorted((one, other) -> {
if (one.preferableTo(other, model))
return -1;
else if (other.preferableTo(one, model)) {
return 1;
}
return 0;
})
.limit(3)
.toList();
}
private List availableRealHostResources(ClusterModel model) {
if (nodeRepository.zone().cloud().dynamicProvisioning()) {
return nodeRepository.flavors().getFlavors().stream().map(Flavor::resources)
.filter(r -> r.vcpu() >= minimumSuggestedVcpus)
.toList();
}
return nodeRepository.nodes().list().hosts().stream().map(host -> host.flavor().resources())
.map(hostResources -> maxResourcesOf(hostResources, model))
.toList();
}
/** Returns the max resources of a host one node may allocate. */
private NodeResources maxResourcesOf(NodeResources hostResources, ClusterModel model) {
if (nodeRepository.exclusivity().allocation(model.clusterSpec())) return hostResources;
// static, shared hosts: Allocate at most half of the host cpu to simplify management
return hostResources.withVcpu(hostResources.vcpu() / 2);
}
/**
* For the observed load this instance is initialized with, returns the resources needed per node to be at
* the target relative load, given a target node and group count.
*/
private NodeResources nodeResourcesWith(int nodes,
int groups,
Limits limits,
Load loadAdjustment,
ClusterModel model) {
Instant now = nodeRepository.clock().instant();
var loadWithTarget = model.loadAdjustmentWith(nodes, groups, loadAdjustment, now);
// Leave some headroom above the ideal allocation to avoid immediately needing to scale back up
if (loadAdjustment.cpu() < 1 && (1.0 - loadWithTarget.cpu()) < headroomRequiredToScaleDown)
loadAdjustment = loadAdjustment.withCpu(Math.min(1.0, loadAdjustment.cpu() * (1.0 + headroomRequiredToScaleDown)));
if (loadAdjustment.memory() < 1 && (1.0 - loadWithTarget.memory()) < headroomRequiredToScaleDown)
loadAdjustment = loadAdjustment.withMemory(Math.min(1.0, loadAdjustment.memory() * (1.0 + headroomRequiredToScaleDown)));
if (loadAdjustment.disk() < 1 && (1.0 - loadWithTarget.disk()) < headroomRequiredToScaleDown)
loadAdjustment = loadAdjustment.withDisk(Math.min(1.0, loadAdjustment.disk() * (1.0 + headroomRequiredToScaleDown)));
loadWithTarget = model.loadAdjustmentWith(nodes, groups, loadAdjustment, now);
var scaled = loadWithTarget.scaled(model.current().realResources().nodeResources());
var nonScaled = limits.isEmpty() || limits.min().nodeResources().isUnspecified()
? model.current().advertisedResources().nodeResources()
: limits.min().nodeResources(); // min=max for non-scaled
return nonScaled.withVcpu(scaled.vcpu()).withMemoryGiB(scaled.memoryGiB()).withDiskGb(scaled.diskGb());
}
/** Returns a copy of the given limits where the minimum nodes are at least the given value when allowed */
private Limits atLeast(int min, Limits limits) {
if (limits.max().nodes() < min) return limits; // not allowed
return limits.withMin(limits.min().withNodes(Math.max(min, limits.min().nodes())));
}
}