com.yahoo.vespa.hosted.provision.provisioning.NodeAllocation 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.provisioning;
import com.yahoo.config.provision.ApplicationId;
import com.yahoo.config.provision.ClusterMembership;
import com.yahoo.config.provision.ClusterSpec;
import com.yahoo.config.provision.Flavor;
import com.yahoo.config.provision.NodeResources;
import com.yahoo.config.provision.NodeType;
import com.yahoo.config.provision.SystemName;
import com.yahoo.net.HostName;
import com.yahoo.vespa.flags.Dimension;
import com.yahoo.vespa.flags.PermanentFlags;
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.node.Agent;
import com.yahoo.vespa.hosted.provision.node.Allocation;
import java.util.ArrayList;
import java.util.Collection;
import java.util.EnumSet;
import java.util.HashSet;
import java.util.LinkedHashMap;
import java.util.List;
import java.util.Map;
import java.util.Optional;
import java.util.Set;
import java.util.function.Predicate;
import java.util.function.Supplier;
import java.util.logging.Logger;
import java.util.stream.Collectors;
import java.util.stream.Stream;
/**
* Used to manage a list of nodes during the node reservation process to fulfill the nodespec.
*
* @author bratseth
*/
class NodeAllocation {
private static final Logger LOG = Logger.getLogger(NodeAllocation.class.getName());
/** List of all nodes in node-repository */
private final NodeList allNodes;
/** The application this list is for */
private final ApplicationId application;
/** The cluster this list is for */
private final ClusterSpec cluster;
/** The requested nodes of this list */
private final NodeSpec requested;
/** The node candidates this has accepted so far, keyed on hostname */
private final Map nodes = new LinkedHashMap<>();
/** The number of already allocated nodes of compatible size */
private int acceptedAndCompatible = 0;
/** The number of already allocated nodes which can be made compatible */
private int acceptedAndCompatibleOrResizable = 0;
/** The number of nodes rejected because of clashing parentHostname */
private int rejectedDueToClashingParentHost = 0;
/** The number of nodes rejected due to exclusivity constraints */
private int rejectedDueToExclusivity = 0;
private int rejectedDueToInsufficientRealResources = 0;
/** The number of nodes that just now was changed to retired */
private int wasRetiredJustNow = 0;
/** The number of nodes that just now was changed to retired to upgrade its host flavor */
private int wasRetiredDueToFlavorUpgrade = 0;
/** The node indexes to verify uniqueness of each member's index */
private final Set indexes = new HashSet<>();
/** The next membership index to assign to a new node */
private final Supplier nextIndex;
private final NodeRepository nodeRepository;
private final Optional requiredHostFlavor;
NodeAllocation(NodeList allNodes, ApplicationId application, ClusterSpec cluster, NodeSpec requested,
Supplier nextIndex, NodeRepository nodeRepository) {
this.allNodes = allNodes;
this.application = application;
this.cluster = cluster;
this.requested = requested;
this.nextIndex = nextIndex;
this.nodeRepository = nodeRepository;
this.requiredHostFlavor = Optional.of(PermanentFlags.HOST_FLAVOR.bindTo(nodeRepository.flagSource())
.with(Dimension.INSTANCE_ID, application.serializedForm())
.with(Dimension.CLUSTER_TYPE, cluster.type().name())
.with(Dimension.CLUSTER_ID, cluster.id().value())
.value())
.filter(s -> !s.isBlank());
}
/**
* Offer some nodes to this. The nodes may have an allocation to a different application or cluster,
* an allocation to this cluster, or no current allocation (in which case one is assigned).
*
* Note that if unallocated nodes are offered before allocated nodes, this will unnecessarily
* reject allocated nodes due to index duplicates.
*
* @param candidates the nodes which are potentially on offer. These may belong to a different application etc.
*/
void offer(List candidates) {
for (NodeCandidate candidate : candidates) {
if (candidate.allocation().isPresent()) {
Allocation allocation = candidate.allocation().get();
ClusterMembership membership = allocation.membership();
if ( ! allocation.owner().equals(application)) continue; // wrong application
if ( ! membership.cluster().satisfies(cluster)) continue; // wrong cluster id/type
if (candidate.state() == Node.State.active && allocation.removable()) continue; // don't accept; causes removal
if (candidate.state() == Node.State.active && candidate.wantToFail()) continue; // don't accept; causes failing
if (indexes.contains(membership.index())) continue; // duplicate index (just to be sure)
if (nodeRepository.zone().cloud().allowEnclave() && candidate.parent.isPresent() && ! candidate.parent.get().cloudAccount().equals(requested.cloudAccount())) continue; // wrong account
boolean resizeable = requested.considerRetiring() && candidate.isResizable;
if (( ! saturated() && hasCompatibleResources(candidate) && requested.acceptable(candidate)) || acceptIncompatible(candidate)) {
candidate = candidate.withNode();
if (candidate.isValid())
acceptNode(candidate, shouldRetire(candidate, candidates), resizeable);
}
}
else if ( ! saturated() && hasCompatibleResources(candidate)) {
if ( ! nodeRepository.nodeResourceLimits().isWithinRealLimits(candidate, cluster)) {
++rejectedDueToInsufficientRealResources;
continue;
}
if ( violatesParentHostPolicy(candidate)) {
++rejectedDueToClashingParentHost;
continue;
}
switch (violatesExclusivity(candidate)) {
case PARENT_HOST_NOT_EXCLUSIVE -> candidate = candidate.withExclusiveParent(true);
case NONE -> {}
case YES -> {
++rejectedDueToExclusivity;
continue;
}
}
if (candidate.wantToRetire()) {
continue;
}
if (requested.type() == NodeType.tenant && !requested.canFail()) {
continue; // Cannot allocate nodes when bootstrapping
}
candidate = candidate.allocate(application,
ClusterMembership.from(cluster, nextIndex.get()),
requested.resources().orElse(candidate.resources()),
nodeRepository.clock().instant());
if (candidate.isValid()) {
acceptNode(candidate, Retirement.none, false);
}
}
}
}
/** Returns the cause of retirement for given candidate */
private Retirement shouldRetire(NodeCandidate candidate, List candidates) {
if ( ! requested.considerRetiring()) {
boolean alreadyRetired = candidate.allocation().map(a -> a.membership().retired()).orElse(false);
return alreadyRetired ? Retirement.alreadyRetired : Retirement.none;
}
if ( ! nodeRepository.nodeResourceLimits().isWithinRealLimits(candidate, cluster)) return Retirement.outsideRealLimits;
if (violatesParentHostPolicy(candidate)) return Retirement.violatesParentHostPolicy;
if ( ! hasCompatibleResources(candidate)) return Retirement.incompatibleResources;
if (candidate.parent.map(node -> node.status().wantToUpgradeFlavor()).orElse(false)) return Retirement.violatesHostFlavorGeneration;
if (candidate.wantToRetire()) return Retirement.hardRequest;
if (candidate.preferToRetire() && candidate.replaceableBy(candidates)) return Retirement.softRequest;
if (violatesExclusivity(candidate) != NodeCandidate.ExclusivityViolation.NONE) return Retirement.violatesExclusivity;
if (requiredHostFlavor.isPresent() && ! candidate.parent.map(node -> node.flavor().name()).equals(requiredHostFlavor)) return Retirement.violatesHostFlavor;
if (candidate.violatesSpares) return Retirement.violatesSpares;
var group = candidate.allocation().get().membership().cluster().group();
if (cluster.isStateful() && group.isPresent() && requested.count().isPresent()) {
long nodesInGroup = nodes.values().stream().filter(n -> groupOf(n).equals(group) && ! isRetired(n)).count();
if (nodesInGroup >= requested.groupSize())
return Retirement.groupSurplus;
}
return Retirement.none;
}
private boolean violatesParentHostPolicy(NodeCandidate candidate) {
return checkForClashingParentHost() && offeredNodeHasParentHostnameAlreadyAccepted(candidate);
}
private boolean checkForClashingParentHost() {
return nodeRepository.zone().system() == SystemName.main &&
nodeRepository.zone().environment().isProduction() &&
! application.instance().isTester();
}
private boolean offeredNodeHasParentHostnameAlreadyAccepted(NodeCandidate candidate) {
if (candidate.parentHostname().isEmpty()) return false;
return nodes.values().stream().anyMatch(acceptedNode -> acceptedNode.parentHostname().equals(candidate.parentHostname()));
}
private NodeCandidate.ExclusivityViolation violatesExclusivity(NodeCandidate candidate) {
return candidate.violatesExclusivity(cluster, application,
nodeRepository.exclusivity().clusterType(cluster),
nodeRepository.exclusivity().allocation(cluster),
nodeRepository.exclusivity().provisioning(cluster),
nodeRepository.zone().cloud().allowHostSharing(), allNodes);
}
/**
* Returns whether this node should be accepted into the cluster even if it is not currently desired
* (already enough nodes, or wrong resources, etc.).
* Such nodes will be marked retired during finalization of the list of accepted nodes when allowed.
* The conditions for this are:
*
* - We are forced to accept since we cannot remove gracefully (bootstrap).
*
* - This is a stateful node. These must always be retired before being removed to allow the cluster to
* migrate away data.
*
* - This is a container node and it is not desired due to having the wrong flavor. In this case this
* will (normally) obtain for all the current nodes in the cluster and so retiring before removing must
* be used to avoid removing all the current nodes at once, before the newly allocated replacements are
* initialized. (In the other case, where a container node is not desired because we have enough nodes we
* do want to remove it immediately to get immediate feedback on how the size reduction works out.)
*/
private boolean acceptIncompatible(NodeCandidate candidate) {
if (candidate.state() != Node.State.active) return false;
if (candidate.allocation().get().membership().retired()) return true; // don't second-guess if already retired
if ( ! requested.considerRetiring()) // the node is active and we are not allowed to remove gracefully, so keep
return true;
return cluster.isStateful() ||
(cluster.type() == ClusterSpec.Type.container && !hasCompatibleResources(candidate));
}
private boolean hasCompatibleResources(NodeCandidate candidate) {
return requested.isCompatible(candidate.resources()) || candidate.isResizable;
}
private Node acceptNode(NodeCandidate candidate, Retirement retirement, boolean resizeable) {
Node node = candidate.toNode();
if (node.allocation().isPresent()) // Record the currently requested resources
node = node.with(node.allocation().get().withRequestedResources(requested.resources().orElse(node.resources())));
if (retirement == Retirement.none) {
// We want to allocate new nodes rather than unretiring with resize, so count without those
// for the purpose of deciding when to stop accepting nodes (saturation)
if (node.allocation().isEmpty()
|| (canBeUsedInGroupWithDeficiency(node) &&
! (requested.needsResize(node) && (node.allocation().get().membership().retired() || ! requested.considerRetiring())))) {
acceptedAndCompatible++;
}
if (hasCompatibleResources(candidate))
acceptedAndCompatibleOrResizable++;
if (resizeable && ! ( node.allocation().isPresent() && node.allocation().get().membership().retired()))
node = resize(node);
if (node.state() != Node.State.active) // reactivated node - wipe state that deactivated it
node = node.unretire().removable(false);
} else if (retirement != Retirement.alreadyRetired) {
LOG.info("Retiring " + node + " because " + retirement.description());
++wasRetiredJustNow;
if (retirement == Retirement.violatesHostFlavorGeneration) {
++wasRetiredDueToFlavorUpgrade;
}
node = node.retire(nodeRepository.clock().instant());
}
if ( ! node.allocation().get().membership().cluster().equals(cluster)) {
// Cluster has the updated settings but do not set a group
node = setCluster(cluster.with(node.allocation().get().membership().cluster().group()), node);
}
candidate = candidate.withNode(node, retirement != Retirement.none && retirement != Retirement.alreadyRetired );
indexes.add(node.allocation().get().membership().index());
nodes.put(node.hostname(), candidate);
return node;
}
private boolean canBeUsedInGroupWithDeficiency(Node node) {
if (requested.count().isEmpty()) return true;
if (node.allocation().isEmpty()) return true;
var group = node.allocation().get().membership().cluster().group();
if (group.isEmpty()) return true;
long nodesInGroup = nodes.values().stream().filter(n -> groupOf(n).equals(group)).count();
return nodesInGroup < requested.groupSize();
}
private Optional groupOf(NodeCandidate candidate) {
return candidate.allocation().flatMap(a -> a.membership().cluster().group());
}
private boolean isRetired(NodeCandidate candidate) {
return candidate.allocation().map(a -> a.membership().retired()).orElse(false);
}
private Node resize(Node node) {
NodeResources hostResources = allNodes.parentOf(node).get().flavor().resources();
return node.with(new Flavor(requested.resources().get()
.with(hostResources.diskSpeed())
.with(hostResources.storageType())
.with(hostResources.architecture())),
Agent.application, nodeRepository.clock().instant());
}
private Node setCluster(ClusterSpec cluster, Node node) {
ClusterMembership membership = node.allocation().get().membership().with(cluster);
return node.with(node.allocation().get().with(membership));
}
/** Returns true if no more nodes are needed in this list */
public boolean saturated() {
return requested.saturatedBy(acceptedAndCompatible);
}
/** Returns true if the content of this list is sufficient to meet the request */
boolean fulfilled() {
return requested.fulfilledBy(acceptedAndCompatibleOrResizable());
}
/** Returns true if this allocation was already fulfilled and resulted in no new changes */
boolean fulfilledWithoutChanges() {
return fulfilled() && !changes();
}
/** Returns true if this allocation changed any nodes */
boolean changes() {
return !reservableNodes().isEmpty() || !newNodes().isEmpty();
}
/** Returns true if this allocation has retired nodes */
boolean hasRetiredJustNow() {
return wasRetiredJustNow > 0;
}
/**
* Returns {@link HostDeficit} describing the host deficit for the given {@link NodeSpec}.
*
* @return empty if the requested spec is already fulfilled. Otherwise, returns {@link HostDeficit} containing the
* flavor and host count required to cover the deficit.
*/
Optional hostDeficit() {
if (nodeType().isHost()) {
return Optional.empty(); // Hosts are provisioned as required by the child application
}
int deficit = requested.fulfilledDeficitCount(acceptedAndCompatibleOrResizable());
// We can only require flavor upgrade if the entire deficit is caused by upgrades
boolean dueToFlavorUpgrade = deficit == wasRetiredDueToFlavorUpgrade;
return Optional.of(new HostDeficit(requested.resources().orElseGet(NodeResources::unspecified),
deficit,
dueToFlavorUpgrade))
.filter(hostDeficit -> hostDeficit.count() > 0);
}
/** Returns the indices to use when provisioning hosts for this */
List provisionIndices(int count) {
if (count < 1) throw new IllegalArgumentException("Count must be positive");
NodeType hostType = requested.type().hostType();
// Tenant hosts have a continuously increasing index
if (hostType == NodeType.host) return nodeRepository.database().readProvisionIndices(count);
// Infrastructure hosts have fixed indices, starting at 1
Set currentIndices = allNodes.nodeType(hostType)
.not().state(Node.State.deprovisioned)
.hostnames()
.stream()
// TODO(mpolden): Use cluster index instead of parsing hostname, once all
// config servers have been replaced once and have switched
// to compact indices
.map(NodeAllocation::parseIndex)
.collect(Collectors.toSet());
List indices = new ArrayList<>(count);
for (int i = 1; indices.size() < count; i++) {
if (!currentIndices.contains(i)) {
indices.add(i);
}
}
// Ignore our own index as we should never try to provision ourselves. This can happen in the following scenario:
// - cfg1 has been deprovisioned
// - cfg2 has triggered provisioning of a new cfg1
// - cfg1 is starting and redeploys its infrastructure application during bootstrap. A deficit is detected
// (itself, because cfg1 is only added to the repository *after* it is up)
// - cfg1 tries to provision a new host for itself
Integer myIndex = parseIndex(HostName.getLocalhost());
indices.remove(myIndex);
return indices;
}
/** The node type this is allocating */
NodeType nodeType() {
return requested.type();
}
List parentsRequiredToBeExclusive() {
return nodes.values()
.stream()
.filter(candidate -> candidate.exclusiveParent)
.map(candidate -> candidate.parent.orElseThrow())
.toList();
}
List finalNodes() {
GroupAssigner groupAssigner = new GroupAssigner(requested, allNodes, nodeRepository.clock());
Collection finalNodes = groupAssigner.assignTo(nodes.values());
nodes.clear();
finalNodes.forEach(candidate -> nodes.put(candidate.toNode().hostname(), candidate));
return finalNodes.stream().map(NodeCandidate::toNode).toList();
}
List reservableNodes() {
// Include already reserved nodes to extend reservation period and to potentially update their cluster spec.
EnumSet reservableStates = EnumSet.of(Node.State.inactive, Node.State.ready, Node.State.reserved);
return matching(n -> ! n.isNew && reservableStates.contains(n.state())).toList();
}
List newNodes() {
return matching(node -> node.isNew).toList();
}
private Stream matching(Predicate predicate) {
return nodes.values().stream().filter(predicate).map(NodeCandidate::toNode);
}
/** Returns the number of nodes accepted this far */
private int acceptedAndCompatibleOrResizable() {
if (nodeType() == NodeType.tenant) return acceptedAndCompatibleOrResizable;
// Infrastructure nodes are always allocated by type. Count all nodes as accepted so that we never exceed
// the wanted number of nodes for the type.
return allNodes.nodeType(nodeType()).size();
}
String allocationFailureDetails() {
List reasons = new ArrayList<>();
if (rejectedDueToExclusivity > 0)
reasons.add("host exclusivity constraints");
if (rejectedDueToClashingParentHost > 0)
reasons.add("insufficient nodes available on separate physical hosts");
if (wasRetiredJustNow > 0)
reasons.add("retirement of allocated nodes");
if (rejectedDueToInsufficientRealResources > 0)
reasons.add("insufficient real resources on hosts");
if (reasons.isEmpty()) return "";
return ": Not enough suitable nodes available due to " + String.join(", ", reasons);
}
private static Integer parseIndex(String hostname) {
// Node index is the first number appearing in the hostname, before the first dot
try {
return Integer.parseInt(hostname.replaceFirst("^\\D+(\\d+)\\..*", "$1"));
} catch (NumberFormatException e) {
throw new IllegalArgumentException("Could not parse index from hostname '" + hostname + "'", e);
}
}
/** Possible causes of node retirement */
private enum Retirement {
alreadyRetired("node is already retired"),
outsideRealLimits("node real resources is outside limits"),
violatesParentHostPolicy("node violates parent host policy"),
incompatibleResources("node resources are incompatible"),
hardRequest("node is requested and required to retire"),
softRequest("node is requested to retire"),
violatesExclusivity("node violates host exclusivity"),
violatesHostFlavor("node violates host flavor"),
violatesHostFlavorGeneration("node violates host flavor generation"),
violatesSpares("node is assigned to a host we want to use as a spare"),
groupSurplus("group has enough nodes"),
none("");
private final String description;
Retirement(String description) {
this.description = description;
}
/** Human-readable description of this cause */
public String description() {
return description;
}
}
/** A host deficit, the number of missing hosts, for a deployment */
record HostDeficit(NodeResources resources, int count, boolean dueToFlavorUpgrade) {
@Override
public String toString() {
return "deficit of " + count + " nodes with " + resources + (dueToFlavorUpgrade ? ", due to flavor upgrade" : "");
}
}
}