1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
|
// Copyright 2017 Yahoo Holdings. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
package com.yahoo.vespa.hosted.provision.maintenance;
import com.yahoo.config.provision.ApplicationId;
import com.yahoo.config.provision.ClusterSpec;
import com.yahoo.config.provision.Deployer;
import com.yahoo.config.provision.Deployment;
import com.yahoo.config.provision.Flavor;
import com.yahoo.config.provision.NodeType;
import com.yahoo.config.provision.Zone;
import com.yahoo.transaction.Mutex;
import com.yahoo.vespa.hosted.provision.Node;
import com.yahoo.vespa.hosted.provision.NodeRepository;
import com.yahoo.vespa.hosted.provision.maintenance.retire.RetirementPolicy;
import com.yahoo.vespa.hosted.provision.node.Agent;
import com.yahoo.vespa.hosted.provision.provisioning.FlavorSpareChecker;
import java.time.Duration;
import java.util.Arrays;
import java.util.Collection;
import java.util.HashMap;
import java.util.Iterator;
import java.util.List;
import java.util.Map;
import java.util.Optional;
import java.util.Set;
import java.util.logging.Logger;
import java.util.stream.Collectors;
import java.util.stream.Stream;
/**
* Automatically retires ready and active nodes if they meet a certain criteria given by the {@link RetirementPolicy}
* and if there are enough remaining nodes to both replace the retiring node as well as to keep enough in spare.
*
* @author freva
*/
public class NodeRetirer extends Maintainer {
public static final FlavorSpareChecker.SpareNodesPolicy SPARE_NODES_POLICY = flavorSpareCount ->
flavorSpareCount.getNumReadyAmongReplacees() > 2;
private static final long MAX_SIMULTANEOUS_RETIRES_PER_CLUSTER = 1;
private static final Logger log = Logger.getLogger(NodeRetirer.class.getName());
private final Deployer deployer;
private final FlavorSpareChecker flavorSpareChecker;
private final RetirementPolicy retirementPolicy;
public NodeRetirer(NodeRepository nodeRepository, Zone zone, FlavorSpareChecker flavorSpareChecker, Duration interval,
Deployer deployer, JobControl jobControl, RetirementPolicy retirementPolicy, Zone... applies) {
super(nodeRepository, interval, jobControl);
if (! Arrays.asList(applies).contains(zone)) {
String targetZones = Arrays.stream(applies).map(Zone::toString).collect(Collectors.joining(", "));
log.info("NodeRetirer should only run in " + targetZones + " and not in " + zone + ", stopping.");
deconstruct();
}
this.deployer = deployer;
this.retirementPolicy = retirementPolicy;
this.flavorSpareChecker = flavorSpareChecker;
}
@Override
protected void maintain() {
if (retireUnallocated()) {
retireAllocated();
}
}
/**
* Retires unallocated nodes by moving them directly to parked.
* Returns true iff all there are no unallocated nodes that match the retirement policy
*/
boolean retireUnallocated() {
try (Mutex lock = nodeRepository().lockUnallocated()) {
List<Node> allNodes = nodeRepository().getNodes(NodeType.tenant);
Map<Flavor, Map<Node.State, Long>> numSpareNodesByFlavorByState = getNumberOfNodesByFlavorByNodeState(allNodes);
flavorSpareChecker.updateReadyAndActiveCountsByFlavor(numSpareNodesByFlavorByState);
long numFlavorsWithUnsuccessfullyRetiredNodes = allNodes.stream()
.filter(node -> node.state() == Node.State.ready)
.filter(retirementPolicy::shouldRetire)
.collect(Collectors.groupingBy(
Node::flavor,
Collectors.toSet()))
.entrySet().stream()
.filter(entry -> {
Set<Node> nodesThatShouldBeRetiredForFlavor = entry.getValue();
for (Iterator<Node> iter = nodesThatShouldBeRetiredForFlavor.iterator(); iter.hasNext(); ) {
Node nodeToRetire = iter.next();
if (! flavorSpareChecker.canRetireUnallocatedNodeWithFlavor(nodeToRetire.flavor())) break;
nodeRepository().write(nodeToRetire.with(nodeToRetire.status().withWantToDeprovision(true)));
nodeRepository().park(nodeToRetire.hostname(), Agent.NodeRetirer,
"Policy: " + retirementPolicy.getClass().getSimpleName());
iter.remove();
}
if (! nodesThatShouldBeRetiredForFlavor.isEmpty()) {
String commaSeparatedHostnames = nodesThatShouldBeRetiredForFlavor.stream().map(Node::hostname)
.collect(Collectors.joining(", "));
log.info(String.format("Failed to retire %s, wanted to retire %d nodes (%s), but there are no spare nodes left.",
entry.getKey(), nodesThatShouldBeRetiredForFlavor.size(), commaSeparatedHostnames));
}
return ! nodesThatShouldBeRetiredForFlavor.isEmpty();
}).count();
return numFlavorsWithUnsuccessfullyRetiredNodes == 0;
}
}
void retireAllocated() {
List<Node> allNodes = nodeRepository().getNodes(NodeType.tenant);
List<ApplicationId> activeApplications = getActiveApplicationIds(allNodes);
Map<Flavor, Map<Node.State, Long>> numSpareNodesByFlavorByState = getNumberOfNodesByFlavorByNodeState(allNodes);
flavorSpareChecker.updateReadyAndActiveCountsByFlavor(numSpareNodesByFlavorByState);
// Get all the nodes that we could retire along with their deployments
Map<Deployment, Set<Node>> nodesToRetireByDeployment = new HashMap<>();
for (ApplicationId applicationId : activeApplications) {
Map<ClusterSpec, Set<Node>> nodesByCluster = getNodesBelongingToApplication(allNodes, applicationId).stream()
.collect(Collectors.groupingBy(
node -> node.allocation().get().membership().cluster(),
Collectors.toSet()));
Map<ClusterSpec, Set<Node>> retireableNodesByCluster = nodesByCluster.entrySet().stream()
.collect(Collectors.toMap(
Map.Entry::getKey,
entry -> filterRetireableNodes(entry.getValue())));
if (retireableNodesByCluster.values().stream().mapToInt(Set::size).sum() == 0) continue;
Optional<Deployment> deployment = deployer.deployFromLocalActive(applicationId, Duration.ofMinutes(30));
if ( ! deployment.isPresent()) continue; // this will be done at another config server
Set<Node> replaceableNodes = retireableNodesByCluster.entrySet().stream()
.flatMap(entry -> entry.getValue().stream()
.filter(node -> flavorSpareChecker.canRetireAllocatedNodeWithFlavor(node.flavor()))
.limit(getNumberNodesAllowToRetireForCluster(nodesByCluster.get(entry.getKey()), MAX_SIMULTANEOUS_RETIRES_PER_CLUSTER)))
.collect(Collectors.toSet());
if (! replaceableNodes.isEmpty()) nodesToRetireByDeployment.put(deployment.get(), replaceableNodes);
}
nodesToRetireByDeployment.forEach(((deployment, nodes) -> {
ApplicationId app = nodes.iterator().next().allocation().get().owner();
Set<Node> nodesToRetire;
// While under application lock, get up-to-date node, and make sure that the state and the owner of the
// node has not changed in the meantime, mutate the up-to-date node (so to not overwrite other fields
// that may have changed) with wantToRetire and wantToDeprovision.
try (Mutex lock = nodeRepository().lock(app)) {
nodesToRetire = nodes.stream()
.map(node ->
nodeRepository().getNode(node.hostname())
.filter(upToDateNode -> node.state() == Node.State.active)
.filter(upToDateNode -> node.allocation().get().owner().equals(upToDateNode.allocation().get().owner())))
.flatMap(node -> node.map(Stream::of).orElseGet(Stream::empty))
.collect(Collectors.toSet());
nodesToRetire.forEach(node -> {
log.info("Setting wantToRetire and wantToDeprovision for host " + node.hostname() +
" with flavor " + node.flavor().name() +
" allocated to " + node.allocation().get().owner() + ". Policy: " +
retirementPolicy.getClass().getSimpleName());
Node updatedNode = node.with(node.status()
.withWantToRetire(true)
.withWantToDeprovision(true));
nodeRepository().write(updatedNode);
});
}
// This takes a while, so do it outside of the application lock
if (! nodesToRetire.isEmpty()) deployment.activate();
}));
}
private List<Node> getNodesBelongingToApplication(Collection<Node> allNodes, ApplicationId applicationId) {
return allNodes.stream()
.filter(node -> node.allocation().isPresent())
.filter(node -> node.allocation().get().owner().equals(applicationId))
.collect(Collectors.toList());
}
/**
* Returns a list of ApplicationIds sorted by number of active nodes the application has allocated to it
*/
List<ApplicationId> getActiveApplicationIds(Collection<Node> nodes) {
return nodes.stream()
.filter(node -> node.state() == Node.State.active)
.collect(Collectors.groupingBy(
node -> node.allocation().get().owner(),
Collectors.counting()))
.entrySet().stream()
.sorted((c1, c2) -> c2.getValue().compareTo(c1.getValue()))
.map(Map.Entry::getKey)
.collect(Collectors.toList());
}
/**
* @param nodes Collection of nodes that are considered for retirement
* @return Set of nodes that all should eventually be retired
*/
Set<Node> filterRetireableNodes(Collection<Node> nodes) {
return nodes.stream()
.filter(node -> node.state() == Node.State.active)
.filter(node -> !node.status().wantToRetire())
.filter(retirementPolicy::shouldRetire)
.collect(Collectors.toSet());
}
/**
* @param clusterNodes All the nodes allocated to an application belonging to a single cluster
* @return number of nodes we can safely start retiring
*/
long getNumberNodesAllowToRetireForCluster(Collection<Node> clusterNodes, long maxSimultaneousRetires) {
long numNodesInWantToRetire = clusterNodes.stream()
.filter(node -> node.status().wantToRetire())
.filter(node -> node.state() != Node.State.parked)
.count();
return Math.max(0, maxSimultaneousRetires - numNodesInWantToRetire);
}
private Map<Flavor, Map<Node.State, Long>> getNumberOfNodesByFlavorByNodeState(Collection<Node> allNodes) {
return allNodes.stream()
.collect(Collectors.groupingBy(
Node::flavor,
Collectors.groupingBy(Node::state, Collectors.counting())));
}
}
|
