Skip to content

Commit be22bfe

Browse files
sureshanapartisureshanaparti
authored
Management Server - Prepare for Maintenance and Cancel Maintenance improvements (#10995)
* Management Server - Prepare for Maintenance and Cancel Maintenance improvements: - Added new setting 'management.server.maintenance.ignore.maintenance.hosts' to ignore hosts in maintenance states while preparing management server for maintenance. This skips agent transfer and agents count check for hosts in maintenance. - Rebalance indirect agents after cancel maintenance, using rebalance parameter in cancelMaintenance API - Force maintenance after maintenance window timeout, using forced parameter in prepareForMaintenance API. - Propagate 'indirect.agent.lb.check.interval' setting change to the host agents. * rebases fixes * code improvements, cleanup * [UI] Set rebalance true by default in cancel maintenance dialog * Update MS state after executing cluster cmd in the target MS, and some code improvements * code improvements * Ensure the host lb algorithm 'shuffle' is applied once before disabling the indirect agent lb check background task
1 parent 16c60c7 commit be22bfe

File tree

26 files changed

+411
-145
lines changed

26 files changed

+411
-145
lines changed

‎agent/src/main/java/com/cloud/agent/Agent.java‎

Lines changed: 21 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -453,22 +453,30 @@ private void scheduleCertificateRenewalTask() {
453453
certExecutor.schedule(new PostCertificateRenewalTask(this), 5, TimeUnit.SECONDS);
454454
}
455455

456-
private void scheduleHostLBCheckerTask(final long checkInterval) {
456+
private void scheduleHostLBCheckerTask(final String lbAlgorithm, final long checkInterval) {
457457
String name = "HostLBCheckerTask";
458458
if (hostLbCheckExecutor != null && !hostLbCheckExecutor.isShutdown()) {
459+
logger.info("Shutting down the preferred host checker task {}", name);
459460
hostLbCheckExecutor.shutdown();
460461
try {
461462
if (!hostLbCheckExecutor.awaitTermination(1, TimeUnit.SECONDS)) {
462463
hostLbCheckExecutor.shutdownNow();
463464
}
464465
} catch (InterruptedException e) {
465-
logger.debug("Forcing {} shutdown as it did not shutdown in the desired time due to: {}",
466+
logger.debug("Forcing the preferred host checker task {} shutdown as it did not shutdown in the desired time due to: {}",
466467
name, e.getMessage());
467468
hostLbCheckExecutor.shutdownNow();
468469
}
469470
}
470471
if (checkInterval > 0L) {
471-
logger.info("Scheduling preferred host task with host.lb.interval={}ms", checkInterval);
472+
if ("shuffle".equalsIgnoreCase(lbAlgorithm)) {
473+
logger.info("Scheduling the preferred host checker task to trigger once (to apply lb algorithm '{}') after host.lb.interval={} ms", lbAlgorithm, checkInterval);
474+
hostLbCheckExecutor = Executors.newSingleThreadScheduledExecutor((new NamedThreadFactory(name)));
475+
hostLbCheckExecutor.schedule(new PreferredHostCheckerTask(), checkInterval, TimeUnit.MILLISECONDS);
476+
return;
477+
}
478+
479+
logger.info("Scheduling a recurring preferred host checker task with lb algorithm '{}' and host.lb.interval={} ms", lbAlgorithm, checkInterval);
472480
hostLbCheckExecutor = Executors.newSingleThreadScheduledExecutor((new NamedThreadFactory(name)));
473481
hostLbCheckExecutor.scheduleAtFixedRate(new PreferredHostCheckerTask(), checkInterval, checkInterval,
474482
TimeUnit.MILLISECONDS);
@@ -928,7 +936,7 @@ private Answer setupAgentCertificate(final SetupCertificateCommand cmd) {
928936
return new SetupCertificateAnswer(true);
929937
}
930938

931-
private void processManagementServerList(final List<String> msList, final List<String> avoidMsList, final String lbAlgorithm, final Long lbCheckInterval) {
939+
private void processManagementServerList(final List<String> msList, final List<String> avoidMsList, final String lbAlgorithm, final Long lbCheckInterval, final boolean triggerHostLB) {
932940
if (CollectionUtils.isNotEmpty(msList) && StringUtils.isNotEmpty(lbAlgorithm)) {
933941
try {
934942
final String newMSHosts = String.format("%s%s%s", com.cloud.utils.StringUtils.toCSVList(msList), IAgentShell.hostLbAlgorithmSeparator, lbAlgorithm);
@@ -941,22 +949,24 @@ private void processManagementServerList(final List<String> msList, final List<S
941949
}
942950
}
943951
shell.setAvoidHosts(avoidMsList);
944-
if ("shuffle".equals(lbAlgorithm)) {
945-
scheduleHostLBCheckerTask(0);
946-
} else {
947-
scheduleHostLBCheckerTask(shell.getLbCheckerInterval(lbCheckInterval));
952+
if (triggerHostLB) {
953+
logger.info("Triggering the preferred host checker task now");
954+
ScheduledExecutorService hostLbExecutor = Executors.newSingleThreadScheduledExecutor(new NamedThreadFactory("HostLB-Executor"));
955+
hostLbExecutor.schedule(new PreferredHostCheckerTask(), 0, TimeUnit.MILLISECONDS);
956+
hostLbExecutor.shutdown();
948957
}
958+
scheduleHostLBCheckerTask(lbAlgorithm, shell.getLbCheckerInterval(lbCheckInterval));
949959
}
950960

951961
private Answer setupManagementServerList(final SetupMSListCommand cmd) {
952-
processManagementServerList(cmd.getMsList(), cmd.getAvoidMsList(), cmd.getLbAlgorithm(), cmd.getLbCheckInterval());
962+
processManagementServerList(cmd.getMsList(), cmd.getAvoidMsList(), cmd.getLbAlgorithm(), cmd.getLbCheckInterval(), cmd.getTriggerHostLb());
953963
return new SetupMSListAnswer(true);
954964
}
955965

956966
private Answer migrateAgentToOtherMS(final MigrateAgentConnectionCommand cmd) {
957967
try {
958968
if (CollectionUtils.isNotEmpty(cmd.getMsList())) {
959-
processManagementServerList(cmd.getMsList(), cmd.getAvoidMsList(), cmd.getLbAlgorithm(), cmd.getLbCheckInterval());
969+
processManagementServerList(cmd.getMsList(), cmd.getAvoidMsList(), cmd.getLbAlgorithm(), cmd.getLbCheckInterval(), false);
960970
}
961971
Executors.newSingleThreadScheduledExecutor(new NamedThreadFactory("MigrateAgentConnection-Job")).schedule(() -> {
962972
migrateAgentConnection(cmd.getAvoidMsList());
@@ -1046,7 +1056,7 @@ public void processReadyCommand(final Command cmd) {
10461056
}
10471057

10481058
verifyAgentArch(ready.getArch());
1049-
processManagementServerList(ready.getMsHostList(), ready.getAvoidMsHostList(), ready.getLbAlgorithm(), ready.getLbCheckInterval());
1059+
processManagementServerList(ready.getMsHostList(), ready.getAvoidMsHostList(), ready.getLbAlgorithm(), ready.getLbCheckInterval(), false);
10501060

10511061
logger.info("Ready command is processed for agent [id: {}, uuid: {}, name: {}]", getId(), getUuid(), getName());
10521062
}

‎api/src/main/java/com/cloud/exception/OperationTimedoutException.java‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -40,7 +40,7 @@ public class OperationTimedoutException extends CloudException {
4040
boolean _isActive;
4141

4242
public OperationTimedoutException(Command[] cmds, long agentId, long seqId, int time, boolean isActive) {
43-
super("Commands " + seqId + " to Host " + agentId + " timed out after " + time);
43+
super("Commands " + seqId + " to Host " + agentId + " timed out after " + time + " secs");
4444
_agentId = agentId;
4545
_seqId = seqId;
4646
_time = time;

‎api/src/main/java/com/cloud/resource/ResourceState.java‎

Lines changed: 5 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -76,6 +76,10 @@ public static Event toEvent(String e) {
7676
}
7777
}
7878

79+
public static List<ResourceState> s_maintenanceStates = List.of(ResourceState.Maintenance,
80+
ResourceState.ErrorInMaintenance, ResourceState.PrepareForMaintenance,
81+
ResourceState.ErrorInPrepareForMaintenance);
82+
7983
public ResourceState getNextState(Event a) {
8084
return s_fsm.getNextState(this, a);
8185
}
@@ -98,8 +102,7 @@ public static String[] toString(ResourceState... states) {
98102
}
99103

100104
public static boolean isMaintenanceState(ResourceState state) {
101-
return Arrays.asList(ResourceState.Maintenance, ResourceState.ErrorInMaintenance,
102-
ResourceState.PrepareForMaintenance, ResourceState.ErrorInPrepareForMaintenance).contains(state);
105+
return s_maintenanceStates.contains(state);
103106
}
104107

105108
public static boolean canAttemptMaintenance(ResourceState state) {

‎api/src/main/java/org/apache/cloudstack/api/ApiConstants.java‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -441,6 +441,7 @@ public class ApiConstants {
441441
public static final String PUBLIC_END_PORT = "publicendport";
442442
public static final String PUBLIC_ZONE = "publiczone";
443443
public static final String PURGE_RESOURCES = "purgeresources";
444+
public static final String REBALANCE = "rebalance";
444445
public static final String RECEIVED_BYTES = "receivedbytes";
445446
public static final String RECONNECT = "reconnect";
446447
public static final String RECOVER = "recover";

‎api/src/main/java/org/apache/cloudstack/api/command/admin/systemvm/PatchSystemVMCmd.java‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -46,7 +46,7 @@ public class PatchSystemVMCmd extends BaseAsyncCmd {
4646
@Parameter(name = ApiConstants.FORCED, type = CommandType.BOOLEAN,
4747
description = "If true, initiates copy of scripts and restart of the agent, even if the scripts version matches." +
4848
"To be used with ID parameter only")
49-
private Boolean force;
49+
private Boolean forced;
5050

5151
/////////////////////////////////////////////////////
5252
/////////////////// Accessors ///////////////////////
@@ -58,7 +58,7 @@ public Long getId() {
5858
}
5959

6060
public boolean isForced() {
61-
return force != null && force;
61+
return forced != null && forced;
6262
}
6363

6464
/////////////////////////////////////////////////////

‎core/src/main/java/org/apache/cloudstack/agent/lb/SetupMSListCommand.java‎

Lines changed: 7 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -29,13 +29,15 @@ public class SetupMSListCommand extends Command {
2929
private List<String> avoidMsList;
3030
private String lbAlgorithm;
3131
private Long lbCheckInterval;
32+
private Boolean triggerHostLb;
3233

33-
public SetupMSListCommand(final List<String> msList, final List<String> avoidMsList, final String lbAlgorithm, final Long lbCheckInterval) {
34+
public SetupMSListCommand(final List<String> msList, final List<String> avoidMsList, final String lbAlgorithm, final Long lbCheckInterval, final Boolean triggerHostLb) {
3435
super();
3536
this.msList = msList;
3637
this.avoidMsList = avoidMsList;
3738
this.lbAlgorithm = lbAlgorithm;
3839
this.lbCheckInterval = lbCheckInterval;
40+
this.triggerHostLb = triggerHostLb;
3941
}
4042

4143
public List<String> getMsList() {
@@ -54,9 +56,12 @@ public Long getLbCheckInterval() {
5456
return lbCheckInterval;
5557
}
5658

59+
public boolean getTriggerHostLb() {
60+
return triggerHostLb;
61+
}
62+
5763
@Override
5864
public boolean executeInSequence() {
5965
return false;
6066
}
61-
6267
}

‎engine/components-api/src/main/java/com/cloud/agent/AgentManager.java‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -171,5 +171,5 @@ enum TapAgentsAction {
171171

172172
void propagateChangeToAgents(Map<String, String> params);
173173

174-
boolean transferDirectAgentsFromMS(String fromMsUuid, long fromMsId, long timeoutDurationInMs);
174+
boolean transferDirectAgentsFromMS(String fromMsUuid, long fromMsId, long timeoutDurationInMs, boolean excludeHostsInMaintenance);
175175
}

‎engine/orchestration/src/main/java/com/cloud/agent/manager/AgentManagerImpl.java‎

Lines changed: 2 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -273,8 +273,6 @@ public boolean configure(final String name, final Map<String, Object> params) th
273273

274274
_executor = new ThreadPoolExecutor(agentTaskThreads, agentTaskThreads, 60L, TimeUnit.SECONDS, new LinkedBlockingQueue<>(), new NamedThreadFactory("AgentTaskPool"));
275275

276-
initConnectExecutor();
277-
278276
maxConcurrentNewAgentConnections = RemoteAgentMaxConcurrentNewConnections.value();
279277

280278
_connection = new NioServer("AgentManager", Port.value(), Workers.value() + 10,
@@ -828,6 +826,7 @@ public boolean start() {
828826
return true;
829827
}
830828

829+
initConnectExecutor();
831830
startDirectlyConnectedHosts(false);
832831

833832
if (_connection != null) {
@@ -2193,7 +2192,7 @@ public void propagateChangeToAgents(Map<String, String> params) {
21932192
}
21942193

21952194
@Override
2196-
public boolean transferDirectAgentsFromMS(String fromMsUuid, long fromMsId, long timeoutDurationInMs) {
2195+
public boolean transferDirectAgentsFromMS(String fromMsUuid, long fromMsId, long timeoutDurationInMs, boolean excludeHostsInMaintenance) {
21972196
return true;
21982197
}
21992198

‎engine/orchestration/src/main/java/com/cloud/agent/manager/ClusteredAgentManagerImpl.java‎

Lines changed: 18 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -42,6 +42,7 @@
4242
import javax.net.ssl.SSLContext;
4343
import javax.net.ssl.SSLEngine;
4444

45+
import com.cloud.resource.ResourceState;
4546
import org.apache.cloudstack.ca.CAManager;
4647
import org.apache.cloudstack.framework.config.ConfigDepot;
4748
import org.apache.cloudstack.framework.config.ConfigKey;
@@ -431,10 +432,10 @@ public boolean routeToPeer(final String peer, final byte[] bytes) {
431432
ch = connectToPeer(peer, ch);
432433
if (ch == null) {
433434
try {
434-
logD(bytes, "Unable to route to peer: " + Request.parse(bytes));
435+
logD(bytes, "Unable to establish connection to route to peer: " + Request.parse(bytes));
435436
} catch (ClassNotFoundException | UnsupportedVersionException e) {
436437
// Request.parse thrown exception when we try to log it, log as much as we can
437-
logD(bytes, "Unable to route to peer, and Request.parse further caught exception" + e.getMessage());
438+
logD(bytes, "Unable to establish connection to route to peer, and Request.parse further caught exception" + e.getMessage());
438439
}
439440
return false;
440441
}
@@ -643,7 +644,6 @@ protected void doTask(final Task task) throws TaskExecutionException {
643644
final Link link = task.getLink();
644645

645646
if (Request.fromServer(data)) {
646-
647647
final AgentAttache agent = findAttache(hostId);
648648

649649
if (Request.isControl(data)) {
@@ -691,7 +691,6 @@ protected void doTask(final Task task) throws TaskExecutionException {
691691
cancel(Long.toString(Request.getManagementServerId(data)), hostId, Request.getSequence(data), e.getMessage());
692692
}
693693
} else {
694-
695694
final long mgmtId = Request.getManagementServerId(data);
696695
if (mgmtId != -1 && mgmtId != _nodeId) {
697696
routeToPeer(Long.toString(mgmtId), data);
@@ -1352,7 +1351,7 @@ private String handleShutdownManagementServerHostCommand(BaseShutdownManagementS
13521351
if (cmd instanceof PrepareForMaintenanceManagementServerHostCommand) {
13531352
logger.debug("Received PrepareForMaintenanceManagementServerHostCommand - preparing for maintenance");
13541353
try {
1355-
managementServerMaintenanceManager.prepareForMaintenance(((PrepareForMaintenanceManagementServerHostCommand) cmd).getLbAlgorithm());
1354+
managementServerMaintenanceManager.prepareForMaintenance(((PrepareForMaintenanceManagementServerHostCommand) cmd).getLbAlgorithm(), ((PrepareForMaintenanceManagementServerHostCommand) cmd).isForced());
13561355
return "Successfully prepared for maintenance";
13571356
} catch(CloudRuntimeException e) {
13581357
return e.getMessage();
@@ -1399,14 +1398,14 @@ private String handleShutdownManagementServerHostCommand(BaseShutdownManagementS
13991398
}
14001399

14011400
@Override
1402-
public boolean transferDirectAgentsFromMS(String fromMsUuid, long fromMsId, long timeoutDurationInMs) {
1401+
public boolean transferDirectAgentsFromMS(String fromMsUuid, long fromMsId, long timeoutDurationInMs, boolean excludeHostsInMaintenance) {
14031402
if (timeoutDurationInMs <= 0) {
14041403
logger.debug("Not transferring direct agents from management server node {} (id: {}) to other nodes, invalid timeout duration", fromMsId, fromMsUuid);
14051404
return false;
14061405
}
14071406

14081407
long transferStartTimeInMs = System.currentTimeMillis();
1409-
if (CollectionUtils.isEmpty(getDirectAgentHosts(fromMsId))) {
1408+
if (CollectionUtils.isEmpty(getDirectAgentHosts(fromMsId, excludeHostsInMaintenance))) {
14101409
logger.info("No direct agent hosts available on management server node {} (id: {}), to transfer", fromMsId, fromMsUuid);
14111410
return true;
14121411
}
@@ -1421,7 +1420,7 @@ public boolean transferDirectAgentsFromMS(String fromMsUuid, long fromMsId, long
14211420
int agentTransferFailedCount = 0;
14221421
List<DataCenterVO> dataCenterList = dcDao.listAll();
14231422
for (DataCenterVO dc : dataCenterList) {
1424-
List<HostVO> directAgentHostsInDc = getDirectAgentHostsInDc(fromMsId, dc.getId());
1423+
List<HostVO> directAgentHostsInDc = getDirectAgentHostsInDc(fromMsId, dc.getId(), excludeHostsInMaintenance);
14251424
if (CollectionUtils.isEmpty(directAgentHostsInDc)) {
14261425
continue;
14271426
}
@@ -1455,9 +1454,10 @@ public boolean transferDirectAgentsFromMS(String fromMsUuid, long fromMsId, long
14551454
return (agentTransferFailedCount == 0);
14561455
}
14571456

1458-
private List<HostVO> getDirectAgentHosts(long msId) {
1457+
private List<HostVO> getDirectAgentHosts(long msId, boolean excludeHostsInMaintenance) {
14591458
List<HostVO> directAgentHosts = new ArrayList<>();
1460-
List<HostVO> hosts = _hostDao.listHostsByMs(msId);
1459+
List<ResourceState> statesToExclude = excludeHostsInMaintenance ? ResourceState.s_maintenanceStates : List.of();
1460+
List<HostVO> hosts = _hostDao.listHostsByMsResourceState(msId, statesToExclude);
14611461
for (HostVO host : hosts) {
14621462
AgentAttache agent = findAttache(host.getId());
14631463
if (agent instanceof DirectAgentAttache) {
@@ -1468,9 +1468,11 @@ private List<HostVO> getDirectAgentHosts(long msId) {
14681468
return directAgentHosts;
14691469
}
14701470

1471-
private List<HostVO> getDirectAgentHostsInDc(long msId, long dcId) {
1471+
private List<HostVO> getDirectAgentHostsInDc(long msId, long dcId, boolean excludeHostsInMaintenance) {
14721472
List<HostVO> directAgentHosts = new ArrayList<>();
1473-
List<HostVO> hosts = _hostDao.listHostsByMsAndDc(msId, dcId);
1473+
// To exclude maintenance states use values from ResourceState as source of truth
1474+
List<ResourceState> statesToExclude = excludeHostsInMaintenance ? ResourceState.s_maintenanceStates : List.of();
1475+
List<HostVO> hosts = _hostDao.listHostsByMsDcResourceState(msId, dcId, statesToExclude);
14741476
for (HostVO host : hosts) {
14751477
AgentAttache agent = findAttache(host.getId());
14761478
if (agent instanceof DirectAgentAttache) {
@@ -1506,6 +1508,10 @@ public void onManagementServerPreparingForMaintenance() {
15061508
public void onManagementServerCancelPreparingForMaintenance() {
15071509
logger.debug("Management server cancel preparing for maintenance");
15081510
super.onManagementServerPreparingForMaintenance();
1511+
1512+
// needed for the case when Management Server in Preparing For Maintenance but didn't go to Maintenance state
1513+
// (where this variable will be reset)
1514+
_agentLbHappened = false;
15091515
}
15101516

15111517
@Override

‎engine/schema/src/main/java/com/cloud/host/dao/HostDao.java‎

Lines changed: 14 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -177,14 +177,24 @@ public interface HostDao extends GenericDao<HostVO, Long>, StateDao<Status, Stat
177177

178178
List<HostVO> listHostsByMsAndDc(long msId, long dcId);
179179

180+
List<HostVO> listHostsByMsDcResourceState(long msId, long dcId, List<ResourceState> excludedResourceStates);
181+
180182
List<HostVO> listHostsByMs(long msId);
181183

184+
List<HostVO> listHostsByMsResourceState(long msId, List<ResourceState> excludedResourceStates);
185+
182186
/**
183-
* Retrieves the number of hosts/agents this {@see ManagementServer} has responsibility over.
184-
* @param msId the id of the {@see ManagementServer}
185-
* @return the number of hosts/agents this {@see ManagementServer} has responsibility over
187+
* Count Hosts by given Management Server, Host and Hypervisor Types,
188+
* and exclude Hosts with given Resource States.
189+
*
190+
* @param msId Management Server Id
191+
* @param excludedResourceStates Resource States to be excluded
192+
* @param hostTypes Host Types
193+
* @param hypervisorTypes Hypervisor Types
194+
* @return Hosts count
186195
*/
187-
int countByMs(long msId);
196+
int countHostsByMsResourceStateTypeAndHypervisorType(long msId, List<ResourceState> excludedResourceStates,
197+
List<Type> hostTypes, List<HypervisorType> hypervisorTypes);
188198

189199
/**
190200
* Retrieves the host ids/agents this {@see ManagementServer} has responsibility over.

0 commit comments

Comments
 (0)