diff options
author | Jon Bratseth <bratseth@gmail.com> | 2020-06-12 12:51:22 +0200 |
---|---|---|
committer | Jon Bratseth <bratseth@gmail.com> | 2020-06-12 12:51:22 +0200 |
commit | 7f7b6777514bf05916e2edcbc3e27b1bfd28906c (patch) | |
tree | c530cbc56b80eb5128d2d9254b92c0486923f0d4 /node-repository/src/main | |
parent | 9fc05281d6a79c26efe04edeb7604300f0c05845 (diff) |
SpareCapacityMaintainer sketch
Diffstat (limited to 'node-repository/src/main')
19 files changed, 274 insertions, 96 deletions
diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/Node.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/Node.java index a9861497ca3..b6237886dc7 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/Node.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/Node.java @@ -128,6 +128,8 @@ public final class Node { return parentHostname.isPresent() && parentHostname.get().equals(hostname); } + public NodeResources resources() { return flavor.resources(); } + /** Returns the flavor of this node */ public Flavor flavor() { return flavor; } diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/NodeList.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/NodeList.java index 1b2f73a2f5f..1cc2abef734 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/NodeList.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/NodeList.java @@ -46,13 +46,21 @@ public class NodeList extends AbstractFilteringList<Node, NodeList> { } /** Returns the subset of nodes having exactly the given resources */ - public NodeList resources(NodeResources resources) { return matching(node -> node.flavor().resources().equals(resources)); } + public NodeList resources(NodeResources resources) { return matching(node -> node.resources().equals(resources)); } + + /** Returns the subset of nodes which satisfy the given resources */ + public NodeList satisfies(NodeResources resources) { return matching(node -> node.resources().satisfies(resources)); } /** Returns the subset of nodes of the given flavor */ public NodeList flavor(String flavor) { return matching(node -> node.flavor().name().equals(flavor)); } + /** Returns the subset of nodes not in the given collection */ + public NodeList except(Collection<Node> nodes) { + return matching(node -> ! nodes.contains(node)); + } + /** Returns the subset of nodes assigned to the given cluster type */ public NodeList type(ClusterSpec.Type type) { return matching(node -> node.allocation().isPresent() && node.allocation().get().membership().cluster().type().equals(type)); @@ -109,6 +117,11 @@ public class NodeList extends AbstractFilteringList<Node, NodeList> { return matching(node -> nodeTypes.contains(node.type())); } + /** Returns the subset of nodes of the host type */ + public NodeList hosts() { + return matching(node -> node.type() == NodeType.host); + } + /** Returns the subset of nodes that are parents */ public NodeList parents() { return matching(n -> n.parentHostname().isEmpty()); diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/AllocatableClusterResources.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/AllocatableClusterResources.java index e6cbddf96f2..267bfefa332 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/AllocatableClusterResources.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/autoscale/AllocatableClusterResources.java @@ -47,7 +47,7 @@ public class AllocatableClusterResources { this.nodes = nodes.size(); this.groups = (int)nodes.stream().map(node -> node.allocation().get().membership().cluster().group()).distinct().count(); this.realResources = averageRealResourcesOf(nodes, nodeRepository); // Average since we average metrics over nodes - this.advertisedResources = nodes.get(0).flavor().resources(); + this.advertisedResources = nodes.get(0).resources(); this.clusterType = nodes.get(0).allocation().get().membership().cluster().type(); this.fulfilment = 1; } @@ -125,11 +125,11 @@ public class AllocatableClusterResources { NodeResources sum = new NodeResources(0, 0, 0, 0); for (Node node : nodes) sum = sum.add(nodeRepository.resourcesCalculator().realResourcesOf(node, nodeRepository).justNumbers()); - return nodes.get(0).flavor().resources().justNonNumbers() - .withVcpu(sum.vcpu() / nodes.size()) - .withMemoryGb(sum.memoryGb() / nodes.size()) - .withDiskGb(sum.diskGb() / nodes.size()) - .withBandwidthGbps(sum.bandwidthGbps() / nodes.size()); + return nodes.get(0).resources().justNonNumbers() + .withVcpu(sum.vcpu() / nodes.size()) + .withMemoryGb(sum.memoryGb() / nodes.size()) + .withDiskGb(sum.diskGb() / nodes.size()) + .withBandwidthGbps(sum.bandwidthGbps() / nodes.size()); } public static Optional<AllocatableClusterResources> from(ClusterResources wantedResources, diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/AutoscalingMaintainer.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/AutoscalingMaintainer.java index fa8e8375e23..c32b7854d4e 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/AutoscalingMaintainer.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/AutoscalingMaintainer.java @@ -85,7 +85,7 @@ public class AutoscalingMaintainer extends NodeRepositoryMaintainer { int currentGroups = (int)clusterNodes.stream().map(node -> node.allocation().get().membership().cluster().group()).distinct().count(); ClusterSpec.Type clusterType = clusterNodes.get(0).allocation().get().membership().cluster().type(); log.info("Autoscaling " + application + " " + clusterType + " " + clusterId + ":" + - "\nfrom " + toString(clusterNodes.size(), currentGroups, clusterNodes.get(0).flavor().resources()) + + "\nfrom " + toString(clusterNodes.size(), currentGroups, clusterNodes.get(0).resources()) + "\nto " + toString(target.nodes(), target.groups(), target.nodeResources())); } diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/CapacityChecker.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/CapacityChecker.java index 97253df900b..97810c0b329 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/CapacityChecker.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/CapacityChecker.java @@ -136,7 +136,7 @@ public class CapacityChecker { int occupiedIps = 0; Set<String> ipPool = host.ipAddressPool().asSet(); for (var child : nodeChildren.get(host)) { - hostResources = hostResources.subtract(child.flavor().resources().justNumbers()); + hostResources = hostResources.subtract(child.resources().justNumbers()); occupiedIps += child.ipAddresses().stream().filter(ipPool::contains).count(); } availableResources.put(host, new AllocationResources(hostResources, host.ipAddressPool().asSet().size() - occupiedIps)); @@ -250,7 +250,7 @@ public class CapacityChecker { long eligibleParents = hosts.stream().filter(h -> !violatesParentHostPolicy(node, h, containedAllocations) - && availableResources.get(h).satisfies(AllocationResources.from(node.flavor().resources()))).count(); + && availableResources.get(h).satisfies(AllocationResources.from(node.resources()))).count(); allocationHistory.addEntry(node, newParent.get(), eligibleParents + 1); } } @@ -302,7 +302,7 @@ public class CapacityChecker { reason.violatesParentHostPolicy = violatesParentHostPolicy(node, host, containedAllocations); NodeResources l = availableHostResources.nodeResources; - NodeResources r = node.allocation().map(Allocation::requestedResources).orElse(node.flavor().resources()); + NodeResources r = node.allocation().map(Allocation::requestedResources).orElse(node.resources()); if (l.vcpu() < r.vcpu()) reason.insufficientVcpu = true; @@ -391,7 +391,7 @@ public class CapacityChecker { if (node.allocation().isPresent()) return from(node.allocation().get().requestedResources()); else - return from(node.flavor().resources()); + return from(node.resources()); } public static AllocationResources from(NodeResources nodeResources) { @@ -514,7 +514,7 @@ public class CapacityChecker { public String toString() { return String.format("%-20s %-65s -> %15s [%3d valid]", tenant.hostname().replaceFirst("\\..+", ""), - tenant.flavor().resources(), + tenant.resources(), newParent == null ? "x" : newParent.hostname().replaceFirst("\\..+", ""), this.eligibleParents ); diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/MaintenanceDeployment.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/MaintenanceDeployment.java index b006b2f964b..db331e88d64 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/MaintenanceDeployment.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/MaintenanceDeployment.java @@ -131,14 +131,19 @@ class MaintenanceDeployment implements Closeable { public static class Move { - final Node node; - final Node toHost; + private final Node node; + private final Node fromHost, toHost; - Move(Node node, Node toHost) { + Move(Node node, Node fromHost, Node toHost) { this.node = node; + this.fromHost = fromHost; this.toHost = toHost; } + public Node node() { return node; } + public Node fromHost() { return fromHost; } + public Node toHost() { return toHost; } + /** * Try to deploy to make this move. * @@ -197,10 +202,10 @@ class MaintenanceDeployment implements Closeable { @Override public String toString() { return "move " + - ( isEmpty() ? "none" : (node.hostname() + " to " + toHost)); + ( isEmpty() ? "none" : (node.hostname() + " from " + fromHost + " to " + toHost)); } - public static Move empty() { return new Move(null, null); } + public static Move empty() { return new Move(null, null, null); } } diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeRepositoryMaintenance.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeRepositoryMaintenance.java index caf845d36cb..c3f8afae4a4 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeRepositoryMaintenance.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/NodeRepositoryMaintenance.java @@ -88,7 +88,7 @@ public class NodeRepositoryMaintenance extends AbstractComponent { new LoadBalancerExpirer(nodeRepository, defaults.loadBalancerExpirerInterval, lbService)); dynamicProvisioningMaintainer = provisionServiceProvider.getHostProvisioner().map(hostProvisioner -> new DynamicProvisioningMaintainer(nodeRepository, defaults.dynamicProvisionerInterval, hostProvisioner, flagSource)); - spareCapacityMaintainer = new SpareCapacityMaintainer(deployer, nodeRepository, metric, clock, defaults.spareCapacityMaintenanceInterval); + spareCapacityMaintainer = new SpareCapacityMaintainer(deployer, nodeRepository, metric, defaults.spareCapacityMaintenanceInterval); osUpgradeActivator = new OsUpgradeActivator(nodeRepository, defaults.osUpgradeActivatorInterval); rebalancer = new Rebalancer(deployer, nodeRepository, metric, clock, defaults.rebalancerInterval); nodeMetricsDbMaintainer = new NodeMetricsDbMaintainer(nodeRepository, nodeMetrics, nodeMetricsDb, defaults.nodeMetricsCollectionInterval); diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/Rebalancer.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/Rebalancer.java index da3b9bd0e65..e10ac68fde0 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/Rebalancer.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/Rebalancer.java @@ -6,16 +6,14 @@ import com.yahoo.config.provision.Deployer; import com.yahoo.config.provision.NodeResources; import com.yahoo.config.provision.NodeType; import com.yahoo.jdisc.Metric; -import com.yahoo.transaction.Mutex; 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.provisioning.DockerHostCapacity; +import com.yahoo.vespa.hosted.provision.provisioning.HostCapacity; import java.time.Clock; import java.time.Duration; -import java.util.Optional; /** * @author bratseth @@ -53,7 +51,7 @@ public class Rebalancer extends NodeRepositoryMaintainer { /** We do this here rather than in MetricsReporter because it is expensive and frequent updates are unnecessary */ private void updateSkewMetric(NodeList allNodes) { - DockerHostCapacity capacity = new DockerHostCapacity(allNodes, nodeRepository().resourcesCalculator()); + HostCapacity capacity = new HostCapacity(allNodes, nodeRepository().resourcesCalculator()); double totalSkew = 0; int hostCount = 0; for (Node host : allNodes.nodeType((NodeType.host)).state(Node.State.active)) { @@ -75,7 +73,7 @@ public class Rebalancer extends NodeRepositoryMaintainer { * Returns Move.none if no moves can be made to reduce skew. */ private Move findBestMove(NodeList allNodes) { - DockerHostCapacity capacity = new DockerHostCapacity(allNodes, nodeRepository().resourcesCalculator()); + HostCapacity capacity = new HostCapacity(allNodes, nodeRepository().resourcesCalculator()); Move bestMove = Move.empty(); for (Node node : allNodes.nodeType(NodeType.tenant).state(Node.State.active)) { if (node.parentHostname().isEmpty()) continue; @@ -84,29 +82,29 @@ public class Rebalancer extends NodeRepositoryMaintainer { if (deployedRecently(applicationId)) continue; for (Node toHost : allNodes.matching(nodeRepository()::canAllocateTenantNodeTo)) { if (toHost.hostname().equals(node.parentHostname().get())) continue; - if ( ! capacity.freeCapacityOf(toHost).satisfies(node.flavor().resources())) continue; + if ( ! capacity.freeCapacityOf(toHost).satisfies(node.resources())) continue; double skewReductionAtFromHost = skewReductionByRemoving(node, allNodes.parentOf(node).get(), capacity); double skewReductionAtToHost = skewReductionByAdding(node, toHost, capacity); double netSkewReduction = skewReductionAtFromHost + skewReductionAtToHost; if (netSkewReduction > bestMove.netSkewReduction) - bestMove = new Move(node, toHost, netSkewReduction); + bestMove = new Move(node, nodeRepository().getNode(node.parentHostname().get()).get(), toHost, netSkewReduction); } } return bestMove; } - private double skewReductionByRemoving(Node node, Node fromHost, DockerHostCapacity capacity) { + private double skewReductionByRemoving(Node node, Node fromHost, HostCapacity capacity) { NodeResources freeHostCapacity = capacity.freeCapacityOf(fromHost); double skewBefore = Node.skew(fromHost.flavor().resources(), freeHostCapacity); double skewAfter = Node.skew(fromHost.flavor().resources(), freeHostCapacity.add(node.flavor().resources().justNumbers())); return skewBefore - skewAfter; } - private double skewReductionByAdding(Node node, Node toHost, DockerHostCapacity capacity) { + private double skewReductionByAdding(Node node, Node toHost, HostCapacity capacity) { NodeResources freeHostCapacity = capacity.freeCapacityOf(toHost); double skewBefore = Node.skew(toHost.flavor().resources(), freeHostCapacity); - double skewAfter = Node.skew(toHost.flavor().resources(), freeHostCapacity.subtract(node.flavor().resources().justNumbers())); + double skewAfter = Node.skew(toHost.flavor().resources(), freeHostCapacity.subtract(node.resources().justNumbers())); return skewBefore - skewAfter; } @@ -122,20 +120,19 @@ public class Rebalancer extends NodeRepositoryMaintainer { final double netSkewReduction; - Move(Node node, Node toHost, double netSkewReduction) { - super(node, toHost); + Move(Node node, Node fromHost, Node toHost, double netSkewReduction) { + super(node, fromHost, toHost); this.netSkewReduction = netSkewReduction; } @Override public String toString() { - return "move " + - ( node == null ? "none" : - (node.hostname() + " to " + toHost + " [skew reduction " + netSkewReduction + "]")); + if (isEmpty()) return "move none"; + return super.toString() + " [skew reduction " + netSkewReduction + "]"; } public static Move empty() { - return new Move(null, null, 0); + return new Move(null, null, null, 0); } } diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/SpareCapacityMaintainer.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/SpareCapacityMaintainer.java index c0e39b39e94..4179d6d3f83 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/SpareCapacityMaintainer.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/maintenance/SpareCapacityMaintainer.java @@ -2,15 +2,21 @@ package com.yahoo.vespa.hosted.provision.maintenance; import com.yahoo.config.provision.Deployer; +import com.yahoo.config.provision.NodeResources; import com.yahoo.jdisc.Metric; 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.maintenance.MaintenanceDeployment.Move; +import com.yahoo.vespa.hosted.provision.node.Agent; +import com.yahoo.vespa.hosted.provision.provisioning.HostCapacity; -import java.time.Clock; import java.time.Duration; +import java.util.ArrayList; +import java.util.Iterator; import java.util.List; import java.util.Optional; +import java.util.Set; import java.util.logging.Level; import java.util.stream.Collectors; @@ -29,19 +35,18 @@ import java.util.stream.Collectors; */ public class SpareCapacityMaintainer extends NodeRepositoryMaintainer { + private static final int maxMoves = 5; + private final Deployer deployer; private final Metric metric; - private final Clock clock; public SpareCapacityMaintainer(Deployer deployer, NodeRepository nodeRepository, Metric metric, - Clock clock, Duration interval) { super(nodeRepository, interval); this.deployer = deployer; this.metric = metric; - this.clock = clock; } @Override @@ -60,46 +65,194 @@ public class SpareCapacityMaintainer extends NodeRepositoryMaintainer { Optional<CapacityChecker.HostFailurePath> failurePath = capacityChecker.worstCaseHostLossLeadingToFailure(); if (failurePath.isPresent()) { - int worstCaseHostLoss = failurePath.get().hostsCausingFailure.size(); - metric.set("spareHostCapacity", worstCaseHostLoss - 1, null); - if (worstCaseHostLoss == 1) { // Try to get back to needing 2 hosts to fail in the worst case - Optional<Move> moveCandidate = identifyMoveCandidate(failurePath.get()); - if (moveCandidate.isPresent()) - move(moveCandidate.get()); + int spareHostCapacity = failurePath.get().hostsCausingFailure.size() - 1; + if (spareHostCapacity == 0) { + Move move = findMitigatingMove(failurePath.get()); + boolean success = move.execute(Agent.SpareCapacityMaintainer, deployer, metric, nodeRepository()); + if (success) { + // This may strictly be a (noble) lie: We succeeded in taking one step to mitigate, + // but not necessarily *all* steps needed to justify adding 1 to spare capacity. + // However, we alert on this being 0 and we don't want to do that as long as we're not stuck. + spareHostCapacity++; + } } + metric.set("spareHostCapacity", spareHostCapacity, null); } } - private Optional<Move> identifyMoveCandidate(CapacityChecker.HostFailurePath failurePath) { + private Move findMitigatingMove(CapacityChecker.HostFailurePath failurePath) { Optional<Node> nodeWhichCantMove = failurePath.failureReason.tenant; - if (nodeWhichCantMove.isEmpty()) return Optional.empty(); - return findMoveWhichMakesRoomFor(nodeWhichCantMove.get()); + if (nodeWhichCantMove.isEmpty()) return Move.empty(); + return moveTowardsSpareFor(nodeWhichCantMove.get()); } - private Optional<Move> findMoveWhichMakesRoomFor(Node node) { - return Optional.empty(); + private Move moveTowardsSpareFor(Node node) { + NodeList allNodes = nodeRepository().list(); + // Allocation will assign the two most empty nodes as "spares", which will not be allocated on + // unless needed for node failing. Our goal here is to make room on these spares for the given node + HostCapacity hostCapacity = new HostCapacity(allNodes, nodeRepository().resourcesCalculator()); + Set<Node> spareHosts = hostCapacity.findSpareHosts(allNodes.hosts().satisfies(node.resources()).asList(), 2); + List<Node> hosts = allNodes.hosts().except(spareHosts).asList(); + + CapacitySolver capacitySolver = new CapacitySolver(hostCapacity); + List<Move> shortestMitigation = null; + for (Node spareHost : spareHosts) { + List<Move> mitigation = capacitySolver.makeRoomFor(node, spareHost, hosts, List.of(), maxMoves); + if (mitigation == null) continue; + if (shortestMitigation == null || shortestMitigation.size() > mitigation.size()) + shortestMitigation = mitigation; + } + if (shortestMitigation == null || shortestMitigation.isEmpty()) return Move.empty(); + return shortestMitigation.get(0); } - private void move(Move move) { + private static class CapacitySolver { + + private final HostCapacity hostCapacity; + + CapacitySolver(HostCapacity hostCapacity) { + this.hostCapacity = hostCapacity; + } + + /** + * Finds the shortest sequence of moves which makes room for the given node on the given host, + * assuming the given moves already made over the given hosts' current allocation. + * + * @param node the node to make room for + * @param host the target host to make room on + * @param hosts the hosts onto which we can move nodes + * @param movesMade the moves already made in this scenario + * @return the list of movesMade with the moves needed for this appended, in the order they should be performed, + * or null if no sequence could be found + */ + List<Move> makeRoomFor(Node node, Node host, List<Node> hosts, List<Move> movesMade, int movesLeft) { + if ( ! host.resources().satisfies(node.resources())) return null; + NodeResources freeCapacity = freeCapacityWith(movesMade, host); + if (freeCapacity.satisfies(node.resources())) return List.of(); + if (movesLeft == 0) return null; + + List<Move> shortest = null; + for (var i = Subsets(hostCapacity.allNodes().childrenOf(host), movesLeft); i.hasNext(); ) { + List<Node> childrenToMove = i.next(); + if ( ! addResourcesOf(childrenToMove, freeCapacity).satisfies(node.resources())) continue; + List<Move> moves = move(childrenToMove, host, hosts, movesMade, movesLeft); + if (moves == null) continue; + if (shortest == null || moves.size() < shortest.size()) + shortest = moves; + } + if (shortest == null) return null; + List<Move> total = append(movesMade, shortest); + if (total.size() > movesLeft) return null; + return total; + } + + private List<Move> move(List<Node> nodes, Node host, List<Node> hosts, List<Move> movesMade, int movesLeft) { + List<Move> moves = new ArrayList<>(); + for (Node childToMove : nodes) { + List<Move> childMoves = move(childToMove, host, hosts, append(movesMade, moves), movesLeft - moves.size()); + if (childMoves == null) return null; + moves.addAll(childMoves); + if (moves.size() > movesLeft) return null; + } + return moves; + } + + private List<Move> move(Node node, Node host, List<Node> hosts, List<Move> movesMade, int movesLeft) { + List<Move> shortest = null; + for (Node target : hosts) { + List<Move> childMoves = makeRoomFor(node, target, hosts, movesMade, movesLeft - 1); + if (childMoves == null) continue; + if (shortest == null || shortest.size() > childMoves.size() + 1) { + shortest = new ArrayList<>(childMoves); + shortest.add(new Move(node, host, target)); + } + } + return shortest; + } + + private NodeResources addResourcesOf(List<Node> nodes, NodeResources resources) { + for (Node node : nodes) + resources = resources.add(node.resources()); + return resources; + } + + private Iterator<List<Node>> Subsets(NodeList nodes, int maxLength) { + return new SubsetIterator(nodes.asList(), maxLength); + } + + private List<Move> append(List<Move> a, List<Move> b) { + List<Move> list = new ArrayList<>(); + list.addAll(a); + list.addAll(b); + return list; + } + + private NodeResources freeCapacityWith(List<Move> moves, Node host) { + NodeResources resources = hostCapacity.freeCapacityOf(host); + for (Move move : moves) { + if ( ! move.toHost().equals(host)) continue; + resources = resources.subtract(move.node().resources()); + } + for (Move move : moves) { + if ( ! move.fromHost().equals(host)) continue; + resources = resources.add(move.fromHost().resources()); + } + return resources; + } } - private static class Move { + private static class SubsetIterator implements Iterator<List<Node>> { + + private final List<Node> nodes; + private final int maxLength; + + // A number whose binary representation determines which items of list we'll include + private int i = 0; // first "previous" = 0 -> skip the empty set + private List<Node> next = null; - static final Move none = new Move(null, null); + public SubsetIterator(List<Node> nodes, int maxLength) { + this.nodes = new ArrayList<>(nodes.subList(0, Math.min(nodes.size(), 31))); + this.maxLength = maxLength; + } + + @Override + public boolean hasNext() { + if (next != null) return true; - final Node node; - final Node toHost; + // find next + while (++i < 1<<nodes.size()) { + int ones = onesIn(i); + if (ones > maxLength) continue; - Move(Node node, Node toHost) { - this.node = node; - this.toHost = toHost; + next = new ArrayList<>(ones); + for (int position = 0; position < nodes.size(); position++) { + if (hasOneAtPosition(position, i)) + next.add(nodes.get(position)); + } + return true; + } + return false; } @Override - public String toString() { - return "move " + - ( node == null ? "none" : (node.hostname() + " to " + toHost)); + public List<Node> next() { + next = null; + if ( ! hasNext()) throw new IllegalStateException("No more elements"); + return next; + } + + private boolean hasOneAtPosition(int position, int number) { + return (number & (1 << position)) > 0; + } + + private int onesIn(int number) { + int ones = 0; + for (int position = 0; Math.pow(2, position) <= number; position++) { + if (hasOneAtPosition(position, number)) + ones++; + } + return ones; } } diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/node/Agent.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/node/Agent.java index 31b7181a58a..eba9e4a1ac9 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/node/Agent.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/node/Agent.java @@ -21,6 +21,7 @@ public enum Agent { ProvisionedExpirer, ReservationExpirer, DynamicProvisioningMaintainer, - RetiringUpgrader; + RetiringUpgrader, + SpareCapacityMaintainer } diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/Activator.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/Activator.java index ebe9327967e..7ad69d673e7 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/Activator.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/Activator.java @@ -184,12 +184,12 @@ class Activator { for (Node node : nodes) { HostSpec hostSpec = getHost(node.hostname(), hosts); node = hostSpec.membership().get().retired() ? node.retire(nodeRepository.clock().instant()) : node.unretire(); - if (! hostSpec.advertisedResources().equals(node.flavor().resources())) // A resized node + if (! hostSpec.advertisedResources().equals(node.resources())) // A resized node node = node.with(new Flavor(hostSpec.advertisedResources())); Allocation allocation = node.allocation().get() .with(hostSpec.membership().get()) .withRequestedResources(hostSpec.requestedResources() - .orElse(node.flavor().resources())); + .orElse(node.resources())); if (hostSpec.networkPorts().isPresent()) allocation = allocation.withNetworkPorts(hostSpec.networkPorts().get()); node = node.with(allocation); diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/EmptyProvisionServiceProvider.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/EmptyProvisionServiceProvider.java index e4b1e0fcbc0..004c74b5f70 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/EmptyProvisionServiceProvider.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/EmptyProvisionServiceProvider.java @@ -34,7 +34,7 @@ public class EmptyProvisionServiceProvider implements ProvisionServiceProvider { private static class IdentityHostResourcesCalculator implements HostResourcesCalculator { @Override - public NodeResources realResourcesOf(Node node, NodeRepository repository) { return node.flavor().resources(); } + public NodeResources realResourcesOf(Node node, NodeRepository repository) { return node.resources(); } @Override public NodeResources advertisedResourcesOf(Flavor flavor) { return flavor.resources(); } diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/DockerHostCapacity.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/HostCapacity.java index b508198db3a..4125345d492 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/DockerHostCapacity.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/HostCapacity.java @@ -3,10 +3,14 @@ package com.yahoo.vespa.hosted.provision.provisioning; import com.yahoo.config.provision.NodeResources; import com.yahoo.config.provision.NodeType; +import com.yahoo.vespa.hosted.provision.LockedNodeList; import com.yahoo.vespa.hosted.provision.Node; import com.yahoo.vespa.hosted.provision.NodeList; +import java.util.List; import java.util.Objects; +import java.util.Set; +import java.util.stream.Collectors; /** * Capacity calculation for docker hosts. @@ -16,17 +20,38 @@ import java.util.Objects; * * @author smorgrav */ -public class DockerHostCapacity { +public class HostCapacity { private final NodeList allNodes; private final HostResourcesCalculator hostResourcesCalculator; - public DockerHostCapacity(NodeList allNodes, HostResourcesCalculator hostResourcesCalculator) { + public HostCapacity(NodeList allNodes, HostResourcesCalculator hostResourcesCalculator) { this.allNodes = Objects.requireNonNull(allNodes, "allNodes must be non-null"); this.hostResourcesCalculator = Objects.requireNonNull(hostResourcesCalculator, "hostResourcesCalculator must be non-null"); } - int compareWithoutInactive(Node hostA, Node hostB) { + public NodeList allNodes() { return allNodes; } + + /** + * Spare hosts are the two hosts in the system with the most free capacity. + * + * We do not count retired or inactive nodes as used capacity (as they could have been + * moved to create space for the spare node in the first place). + * + * @param candidates the candidates to consider. This list may contain all kinds of nodes. + * @param count the max number of spare hosts to return + */ + public Set<Node> findSpareHosts(List<Node> candidates, int count) { + return candidates.stream() + .filter(node -> node.type() == NodeType.host) + .filter(dockerHost -> dockerHost.state() == Node.State.active) + .filter(dockerHost -> freeIPs(dockerHost) > 0) + .sorted(this::compareWithoutInactive) + .limit(count) + .collect(Collectors.toSet()); + } + + private int compareWithoutInactive(Node hostA, Node hostB) { int result = compare(freeCapacityOf(hostB, true), freeCapacityOf(hostA, true)); if (result != 0) return result; diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/NodeAllocation.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/NodeAllocation.java index 47d1b30a8e7..df8a7e45917 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/NodeAllocation.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/NodeAllocation.java @@ -148,7 +148,7 @@ class NodeAllocation { } node.node = offered.allocate(application, ClusterMembership.from(cluster, highestIndex.add(1)), - requestedNodes.resources().orElse(node.node.flavor().resources()), + requestedNodes.resources().orElse(node.node.resources()), nodeRepository.clock().instant()); accepted.add(acceptNode(node, false, false)); } @@ -242,7 +242,7 @@ class NodeAllocation { Node node = prioritizableNode.node; if (node.allocation().isPresent()) // Record the currently requested resources - node = node.with(node.allocation().get().withRequestedResources(requestedNodes.resources().orElse(node.flavor().resources()))); + node = node.with(node.allocation().get().withRequestedResources(requestedNodes.resources().orElse(node.resources()))); if (! wantToRetire) { accepted++; diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/NodePrioritizer.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/NodePrioritizer.java index 5e297900767..8560dd424e7 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/NodePrioritizer.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/NodePrioritizer.java @@ -13,7 +13,6 @@ 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.IP; -import com.yahoo.vespa.hosted.provision.persistence.NameResolver; import java.util.EnumSet; import java.util.HashMap; @@ -21,7 +20,6 @@ import java.util.List; import java.util.Map; import java.util.Optional; import java.util.Set; -import java.util.function.Predicate; import java.util.logging.Logger; import java.util.stream.Collectors; @@ -39,7 +37,7 @@ public class NodePrioritizer { private final Map<Node, PrioritizableNode> nodes = new HashMap<>(); private final LockedNodeList allNodes; - private final DockerHostCapacity capacity; + private final HostCapacity capacity; private final NodeSpec requestedNodes; private final ApplicationId application; private final ClusterSpec clusterSpec; @@ -55,11 +53,11 @@ public class NodePrioritizer { NodePrioritizer(LockedNodeList allNodes, ApplicationId application, ClusterSpec clusterSpec, NodeSpec nodeSpec, int spares, int wantedGroups, boolean allocateFully, NodeRepository nodeRepository) { this.allNodes = allNodes; - this.capacity = new DockerHostCapacity(allNodes, nodeRepository.resourcesCalculator()); + this.capacity = new HostCapacity(allNodes, nodeRepository.resourcesCalculator()); this.requestedNodes = nodeSpec; this.clusterSpec = clusterSpec; this.application = application; - this.spareHosts = findSpareHosts(allNodes, capacity, spares); + this.spareHosts = capacity.findSpareHosts(allNodes.asList(), spares); this.allocateFully = allocateFully; this.nodeRepository = nodeRepository; @@ -83,22 +81,6 @@ public class NodePrioritizer { this.isDocker = resources(requestedNodes) != null; } - /** - * Spare hosts are the two hosts in the system with the most free capacity. - * - * We do not count retired or inactive nodes as used capacity (as they could have been - * moved to create space for the spare node in the first place). - */ - private static Set<Node> findSpareHosts(LockedNodeList nodes, DockerHostCapacity capacity, int spares) { - return nodes.asList().stream() - .filter(node -> node.type() == NodeType.host) - .filter(dockerHost -> dockerHost.state() == Node.State.active) - .filter(dockerHost -> capacity.freeIPs(dockerHost) > 0) - .sorted(capacity::compareWithoutInactive) - .limit(spares) - .collect(Collectors.toSet()); - } - /** Returns the list of nodes sorted by PrioritizableNode::compare */ List<PrioritizableNode> prioritize() { return nodes.values().stream().sorted().collect(Collectors.toList()); @@ -207,7 +189,7 @@ public class NodePrioritizer { if (!isNewNode) builder.resizable(! allocateFully - && requestedNodes.canResize(node.flavor().resources(), parentCapacity, isTopologyChange, currentClusterSize)); + && requestedNodes.canResize(node.resources(), parentCapacity, isTopologyChange, currentClusterSize)); if (spareHosts.contains(parent)) builder.violatesSpares(true); diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/NodeSpec.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/NodeSpec.java index a8abdc3f38a..9971aae1714 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/NodeSpec.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/NodeSpec.java @@ -139,7 +139,7 @@ public interface NodeSpec { @Override public boolean needsResize(Node node) { - return ! node.flavor().resources().compatibleWith(requestedNodeResources); + return ! node.resources().compatibleWith(requestedNodeResources); } @Override diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/PrioritizableNode.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/PrioritizableNode.java index 3fc60c1192d..0c1b396c40c 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/PrioritizableNode.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/provisioning/PrioritizableNode.java @@ -126,7 +126,7 @@ class PrioritizableNode implements Comparable<PrioritizableNode> { double skewWithoutThis() { return skewWith(zeroResources); } /** Returns the allocation skew of the parent of this after adding this node to it */ - double skewWithThis() { return skewWith(node.flavor().resources()); } + double skewWithThis() { return skewWith(node.resources()); } private double skewWith(NodeResources resources) { if (parent.isEmpty()) return 0; diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/ApplicationSerializer.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/ApplicationSerializer.java index aa81aae84fe..a4161a318ab 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/ApplicationSerializer.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/ApplicationSerializer.java @@ -43,7 +43,7 @@ public class ApplicationSerializer { if (nodes.isEmpty()) return; int groups = (int)nodes.stream().map(node -> node.allocation().get().membership().cluster().group()).distinct().count(); - ClusterResources currentResources = new ClusterResources(nodes.size(), groups, nodes.get(0).flavor().resources()); + ClusterResources currentResources = new ClusterResources(nodes.size(), groups, nodes.get(0).resources()); toSlime(cluster.minResources(), clusterObject.setObject("min")); toSlime(cluster.maxResources(), clusterObject.setObject("max")); diff --git a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/HostCapacityResponse.java b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/HostCapacityResponse.java index 7e81a9cc002..e28b03d7517 100644 --- a/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/HostCapacityResponse.java +++ b/node-repository/src/main/java/com/yahoo/vespa/hosted/provision/restapi/HostCapacityResponse.java @@ -129,7 +129,7 @@ public class HostCapacityResponse extends HttpResponse { ); failurePath.failureReason.tenant.ifPresent(tenant -> { object.setString("failedTenant", tenant.hostname()); - object.setString("failedTenantResources", tenant.flavor().resources().toString()); + object.setString("failedTenantResources", tenant.resources().toString()); tenant.allocation().ifPresent(allocation -> object.setString("failedTenantAllocation", allocation.toString()) ); |