Many resources are needed to download a project. Please understand that we have to compensate our server costs. Thank you in advance. Project price only 1 $
You can buy this project and download/modify it how often you want.
/*
* Copyright (c) 2014-2016 VMware, Inc. All Rights Reserved.
*
* Licensed under the Apache License, Version 2.0 (the "License"); you may not
* use this file except in compliance with the License. You may obtain a copy of
* the License at http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software distributed
* under the License is distributed on an "AS IS" BASIS, without warranties or
* conditions of any kind, EITHER EXPRESS OR IMPLIED. See the License for the
* specific language governing permissions and limitations under the License.
*/
package com.vmware.xenon.common;
import java.net.URI;
import java.util.HashSet;
import java.util.Iterator;
import java.util.Map.Entry;
import java.util.Set;
import java.util.concurrent.CancellationException;
import java.util.concurrent.ConcurrentSkipListMap;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.logging.Level;
import com.vmware.xenon.common.NodeSelectorService.SelectOwnerResponse;
import com.vmware.xenon.common.Operation.CompletionHandler;
import com.vmware.xenon.common.Service.Action;
import com.vmware.xenon.common.Service.ProcessingStage;
import com.vmware.xenon.common.Service.ServiceOption;
import com.vmware.xenon.common.ServiceHost.MaintenanceStage;
import com.vmware.xenon.common.ServiceHost.ServiceHostState;
import com.vmware.xenon.common.ServiceMaintenanceRequest.MaintenanceReason;
import com.vmware.xenon.common.ServiceStats.ServiceStat;
import com.vmware.xenon.services.common.NodeGroupService;
import com.vmware.xenon.services.common.NodeGroupService.NodeGroupState;
import com.vmware.xenon.services.common.NodeSelectorSynchronizationService.SynchronizePeersRequest;
import com.vmware.xenon.services.common.ServiceUriPaths;
/**
* Sequences service periodic maintenance
*/
class ServiceSynchronizationTracker {
public static ServiceSynchronizationTracker create(ServiceHost host) {
ServiceSynchronizationTracker sst = new ServiceSynchronizationTracker();
sst.host = host;
return sst;
}
private ServiceHost host;
private final ConcurrentSkipListMap synchronizationTimes = new ConcurrentSkipListMap<>();
private final ConcurrentSkipListMap synchronizationRequiredServices = new ConcurrentSkipListMap<>();
private final ConcurrentSkipListMap synchronizationActiveServices = new ConcurrentSkipListMap<>();
private final ConcurrentSkipListMap pendingNodeSelectorsForFactorySynch = new ConcurrentSkipListMap<>();
public void addService(String servicePath, long timeMicros) {
this.synchronizationRequiredServices.put(servicePath, timeMicros);
}
public void removeService(String path) {
this.synchronizationActiveServices.remove(path);
this.synchronizationRequiredServices.remove(path);
}
private void scheduleNodeGroupChangeMaintenance(String nodeSelectorPath, Operation op) {
OperationContext.setAuthorizationContext(this.host.getSystemAuthorizationContext());
if (nodeSelectorPath == null) {
throw new IllegalArgumentException("nodeGroupPath is required");
}
NodeSelectorService nss = this.host.findNodeSelectorService(nodeSelectorPath, null);
if (nss == null) {
throw new IllegalArgumentException("Node selector not found: " + nodeSelectorPath);
}
String ngPath = nss.getNodeGroupPath();
Operation get = Operation
.createGet(UriUtils.buildUri(this.host, ngPath))
.setReferer(this.host.getUri())
.setCompletion(
(o, e) -> {
if (e != null) {
this.host.log(Level.WARNING,
"Failure getting node group state: %s", e.toString());
if (op != null) {
op.fail(e);
}
return;
}
NodeGroupState ngs = o.getBody(NodeGroupState.class);
this.pendingNodeSelectorsForFactorySynch.put(nodeSelectorPath, ngs);
if (op != null) {
op.complete();
}
});
this.host.sendRequest(get);
}
void failStartServiceOrSynchronize(
Service service, Operation start, Operation startRsp, Throwable startEx) {
boolean isMarkedDeleted = false;
if (startRsp.getStatusCode() == Operation.STATUS_CODE_CONFLICT && startRsp.hasBody()) {
ServiceErrorResponse rsp = startRsp.getBody(ServiceErrorResponse.class);
isMarkedDeleted = rsp.getErrorCode() == ServiceErrorResponse.ERROR_CODE_STATE_MARKED_DELETED;
}
// We check if if this was a failure because of
// a 409 error from a replica node. If it was,
// then this is mostly likely a new owner who does
// not have the service. Remember before reaching here
// we do check if the service is started locally in
// checkIfServiceExistsAndAttach. So, in this scenario,
// we will kick-off on-demand synchronization by kicking
// off a synch-post request (like the synch-task). This will
// start the service locally.
boolean isReplicaConflict = !isMarkedDeleted &&
ServiceHost.isServiceCreate(start) &&
service.hasOption(ServiceOption.REPLICATION) &&
start.getAction() == Action.POST &&
!start.isFromReplication() &&
startRsp.getStatusCode() == Operation.STATUS_CODE_CONFLICT;
if (isReplicaConflict) {
this.host.log(Level.INFO,
"%s not available on owner node, on-demand synchronizing ...",
service.getSelfLink());
URI factoryUri = startRsp.getUri();
String selfLink = startRsp.getLinkedState().documentSelfLink;
sendSynchRequest(factoryUri, selfLink, (synchOp, t) -> {
if (t != null) {
this.host.log(Level.SEVERE, "Synch failed for %s. Exception: %s",
service.getSelfLink(), t.toString());
}
// It's important that we fail the original POST request with the same
// failure, statusCode and body.
start.fail(startRsp.getStatusCode(), startEx, startRsp.getBodyRaw());
this.host.processPendingServiceAvailableOperations(
service, startEx, !start.isFailureLoggingDisabled());
});
return;
}
start.fail(startRsp.getStatusCode(), startEx, startRsp.getBodyRaw());
this.host.processPendingServiceAvailableOperations(
service, startEx, !start.isFailureLoggingDisabled());
}
void failWithNotFoundOrSynchronize(Service parent, String path, Operation op) {
// Because the service uses 1X replication, we don't need to synchronize it on-demand.
if (parent.getPeerNodeSelectorPath().equals(ServiceUriPaths.DEFAULT_1X_NODE_SELECTOR)) {
Operation.failServiceNotFound(op);
return;
}
this.host.log(Level.INFO,
"Service %s not found on owner. On-demand synchronizing.", op.getUri());
String documentSelfLink;
if (ServiceHost.isHelperServicePath(op.getUri().getPath())) {
documentSelfLink = UriUtils.getLastPathSegment(
UriUtils.getParentPath(op.getUri().getPath()));
} else {
documentSelfLink = UriUtils.getLastPathSegment(op.getUri().getPath());
}
sendSynchRequest(parent.getUri(), documentSelfLink, (o, e) -> {
if (e == null) {
// Service was found on a remote peer and has been
// synchronized successfully. We go ahead and retry
// the original request now.
this.host.handleRequest(null, op);
return;
}
boolean markedDeleted = false;
boolean notFound = o.getStatusCode() == Operation.STATUS_CODE_NOT_FOUND;
if (o.getStatusCode() == Operation.STATUS_CODE_CONFLICT) {
if (o.hasBody()) {
ServiceErrorResponse error = o.getBody(ServiceErrorResponse.class);
markedDeleted = error.getErrorCode() ==
ServiceErrorResponse.ERROR_CODE_STATE_MARKED_DELETED;
}
}
if (notFound || markedDeleted) {
if (op.getAction() == Action.DELETE) {
// do not queue DELETE actions for services not present, complete with success
op.complete();
return;
}
Operation.failServiceNotFound(op);
return;
}
this.host.log(Level.SEVERE, "Failed to synch service not found on owner. Failure: %s", e);
op.fail(e);
});
}
private void sendSynchRequest(URI parentUri, String documentSelfLink, CompletionHandler ch) {
ServiceDocument synchState = new ServiceDocument();
synchState.documentSelfLink = documentSelfLink;
Operation synchOp = Operation
.createPost(this.host, parentUri.getPath())
.setBody(synchState)
.addPragmaDirective(Operation.PRAGMA_DIRECTIVE_SYNCH_OWNER)
.setReferer(this.host.getUri())
.setExpiration(Utils.fromNowMicrosUtc(NodeGroupService.PEER_REQUEST_TIMEOUT_MICROS))
.setCompletion((o, e) -> ch.handle(o, e));
this.host.handleRequest(null, synchOp);
}
/**
* Infrastructure use only.
*
* Determines the owner for the given service and if the local node is owner, proceeds
* with synchronization.
*
* This method is called in the following cases:
*
* 1) Synchronization of a factory service, due to node group change. This includes
* synchronization after host restart. Since factory synchronization uses SYNCH_OWNER
* request that is ONLY sent to the document owner, this function will not be executed
* on non-owner nodes (during factory synchronization).
*
* 2) Synchronization due to conflict on epoch, version or owner, on a specific stateful
* service instance. The service instance will call this method to synchronize peers.
*
* 3) When an On-demand load service is re-started due to an incoming request.
*
* Note that case 1) actually causes SYNCH_PEER requests to be sent out to peers, that can
* implicitly invoke case 2). That's because a SYNCH_PEER POST request is converted to a PUT
* on the peer node and is forwarded to the StatefulService. Validating ownership in this
* method is critical to avoid recursive loops between OWNER and PEERs trying to synchronize
* each other endlessly.
*
* Also note that the SYNCH_OWNER request since it is only sent to owner nodes, it is
* not really required to again validate document ownership in this method. So, a future
* optimization could be to skip ownership validation for case 1) ONLY.
*/
void selectServiceOwnerAndSynchState(Service s, Operation op) {
CompletionHandler c = (o, e) -> {
if (e != null) {
this.host.log(Level.WARNING, "Failure partitioning %s: %s", op.getUri(),
e.toString());
op.fail(e);
return;
}
SelectOwnerResponse rsp = o.getBody(SelectOwnerResponse.class);
if (op.isFromReplication()) {
// replicated requests should not synchronize, that is done on the owner node
if (op.isCommit()) {
// remote node is telling us to commit the owner changes
s.toggleOption(ServiceOption.DOCUMENT_OWNER, rsp.isLocalHostOwner);
}
op.complete();
return;
}
s.toggleOption(ServiceOption.DOCUMENT_OWNER, rsp.isLocalHostOwner);
if (ServiceHost.isServiceCreate(op) || !rsp.isLocalHostOwner) {
// if this is from a client, do not synchronize. an conflict can be resolved
// when we attempt to replicate the POST.
// if this is synchronization attempt and we are not the owner, do nothing
op.complete();
return;
}
// we are on owner node, proceed with synchronization logic that will discover
// and push, latest, best state, to all peers
synchronizeWithPeers(s, op);
};
Operation selectOwnerOp = Operation.createPost(null)
.setExpiration(op.getExpirationMicrosUtc())
.setCompletion(c);
this.host.selectOwner(s.getPeerNodeSelectorPath(), s.getSelfLink(), selectOwnerOp);
}
private void synchronizeWithPeers(Service s, Operation op) {
// service is durable and replicated. We need to ask our peers if they
// have more recent state version than we do, then pick the latest one
// (or the most valid one, depending on peer consensus)
SynchronizePeersRequest t = SynchronizePeersRequest.create();
t.indexLink = s.getDocumentIndexPath();
t.stateDescription = this.host.buildDocumentDescription(s);
t.options = s.getOptions();
t.state = op.hasBody() ? op.getBody(s.getStateType()) : null;
t.factoryLink = UriUtils.getParentPath(s.getSelfLink());
if (t.factoryLink == null || t.factoryLink.isEmpty()) {
String error = String.format("Factory not found for %s."
+ "If the service is not created through a factory it should not set %s",
s.getSelfLink(), ServiceOption.OWNER_SELECTION);
op.fail(new IllegalStateException(error));
return;
}
if (t.state == null) {
// we have no initial state or state from storage. Create an empty state so we can
// compare with peers
ServiceDocument template = null;
try {
template = s.getStateType().newInstance();
} catch (Exception e) {
this.host.log(Level.SEVERE, "Could not create instance state type: %s",
e.toString());
op.fail(e);
return;
}
template.documentSelfLink = s.getSelfLink();
template.documentEpoch = 0L;
// set version to negative so we do not select this over peer state
template.documentVersion = -1;
t.state = template;
}
// We remove the SYNCH_OWNER pragma from the operation here.
// This allows the best state computed through
// NodeSelectorSynchronizationService get persisted locally.
op.removePragmaDirective(Operation.PRAGMA_DIRECTIVE_SYNCH_OWNER);
CompletionHandler c = (o, e) -> {
if (this.host.isStopping()) {
op.fail(new CancellationException("Host is stopping"));
return;
}
if (e != null) {
op.setStatusCode(o.getStatusCode());
op.fail(e);
return;
}
if (!o.hasBody()) {
// peers did not have a better state to offer
if (ServiceDocument.isDeleted(t.state)) {
Operation.failServiceMarkedDeleted(t.state.documentSelfLink, op);
return;
}
// avoid duplicate document version on owner
op.addPragmaDirective(Operation.PRAGMA_DIRECTIVE_NO_INDEX_UPDATE);
op.complete();
return;
}
ServiceDocument selectedState = o.getBody(s.getStateType());
boolean isVersionSame = ServiceDocument
.compare(selectedState, t.state, t.stateDescription, Utils.getTimeComparisonEpsilonMicros())
.contains(ServiceDocument.DocumentRelationship.EQUAL_VERSION);
if (ServiceDocument.isDeleted(selectedState)) {
Operation.failServiceMarkedDeleted(t.state.documentSelfLink, op);
// Only save the document, if the selected state is a newer version of the document
// than the local copy.
if (!isVersionSame ) {
selectedState.documentSelfLink = s.getSelfLink();
selectedState.documentUpdateAction = Action.DELETE.toString();
this.host.saveServiceState(s, Operation.createDelete(UriUtils.buildUri(this.host,
s.getSelfLink())).setReferer(s.getUri()),
selectedState);
}
return;
}
// indicate that synchronization occurred, we got an updated state from peers
op.addPragmaDirective(Operation.PRAGMA_DIRECTIVE_SYNCH_PEER);
if (isVersionSame) {
// avoid duplicate document version on owner
op.addPragmaDirective(Operation.PRAGMA_DIRECTIVE_NO_INDEX_UPDATE);
}
// The remote peers have a more recent state than the one we loaded from the store.
// Use the peer service state as the initial state. Also update the linked state,
// so that the correct documentVersion is indexed during startService.
op.linkState(selectedState);
op.setBodyNoCloning(selectedState).complete();
};
URI synchServiceForGroup = UriUtils.extendUri(
UriUtils.buildUri(this.host, s.getPeerNodeSelectorPath()),
ServiceUriPaths.SERVICE_URI_SUFFIX_SYNCHRONIZATION);
Operation synchPost = Operation
.createPost(synchServiceForGroup)
.setBodyNoCloning(t)
.setExpiration(op.getExpirationMicrosUtc())
.setRetryCount(0)
.setReferer(s.getUri())
.setCompletion(c);
this.host.sendRequest(synchPost);
}
public void scheduleNodeGroupChangeMaintenance(String nodeSelectorPath) {
long now = Utils.getNowMicrosUtc();
this.host.log(Level.INFO, "%s %d", nodeSelectorPath, now);
this.synchronizationTimes.put(nodeSelectorPath, now);
scheduleNodeGroupChangeMaintenance(nodeSelectorPath, null);
}
public void performNodeSelectorChangeMaintenance(Operation post, long now,
MaintenanceStage nextStage, boolean isCheckRequired, long deadline) {
if (isCheckRequired && checkAndScheduleNodeSelectorSynch(post, nextStage, deadline)) {
return;
}
try {
Iterator> it = this.pendingNodeSelectorsForFactorySynch
.entrySet()
.iterator();
while (it.hasNext()) {
Entry e = it.next();
it.remove();
performNodeSelectorChangeMaintenance(e);
}
} finally {
this.host.performMaintenanceStage(post, nextStage, deadline);
}
}
private boolean checkAndScheduleNodeSelectorSynch(Operation post, MaintenanceStage nextStage,
long deadline) {
boolean hasSynchOccuredAtLeastOnce = false;
for (Long synchTime : this.synchronizationTimes.values()) {
if (synchTime != null && synchTime > 0) {
hasSynchOccuredAtLeastOnce = true;
}
}
if (!hasSynchOccuredAtLeastOnce) {
return false;
}
Set selectorPathsToSynch = new HashSet<>();
// we have done at least once synchronization. Check if any services that require synch
// started after the last node group change, and if so, schedule them
for (Entry en : this.synchronizationRequiredServices.entrySet()) {
Long lastSynchTime = en.getValue();
String link = en.getKey();
Service s = this.host.findService(link, true);
if (s == null || s.getProcessingStage() != ProcessingStage.AVAILABLE) {
continue;
}
String selectorPath = s.getPeerNodeSelectorPath();
Long selectorSynchTime = this.synchronizationTimes.get(selectorPath);
if (selectorSynchTime == null) {
continue;
}
if (lastSynchTime < selectorSynchTime) {
this.host.log(Level.FINE, "Service %s started at %d, last synch at %d", link,
lastSynchTime, selectorSynchTime);
selectorPathsToSynch.add(s.getPeerNodeSelectorPath());
}
}
if (selectorPathsToSynch.isEmpty()) {
return false;
}
AtomicInteger pending = new AtomicInteger(selectorPathsToSynch.size());
CompletionHandler c = (o, e) -> {
if (e != null) {
if (!this.host.isStopping()) {
this.host.log(Level.WARNING, "skipping synchronization, error: %s",
Utils.toString(e));
}
this.host.performMaintenanceStage(post, nextStage, deadline);
return;
}
int r = pending.decrementAndGet();
if (r != 0) {
return;
}
// we refreshed the pending selector list, now ready to do kick of synchronization
performNodeSelectorChangeMaintenance(post, Utils.getSystemNowMicrosUtc(), nextStage,
false,
deadline);
};
for (String path : selectorPathsToSynch) {
Operation synch = Operation.createPost(this.host.getUri()).setCompletion(c);
scheduleNodeGroupChangeMaintenance(path, synch);
}
return true;
}
private void performNodeSelectorChangeMaintenance(Entry entry) {
String nodeSelectorPath = entry.getKey();
Long selectorSynchTime = this.synchronizationTimes.get(nodeSelectorPath);
NodeGroupState ngs = entry.getValue();
long now = Utils.getSystemNowMicrosUtc();
for (Entry en : this.synchronizationActiveServices.entrySet()) {
String link = en.getKey();
Service s = this.host.findService(link, true);
if (s == null) {
continue;
}
ServiceHostState hostState = this.host.getStateNoCloning();
long delta = now - en.getValue();
boolean shouldLog = false;
if (delta > hostState.operationTimeoutMicros) {
s.toggleOption(ServiceOption.INSTRUMENTATION, true);
s.adjustStat(Service.STAT_NAME_NODE_GROUP_SYNCH_DELAYED_COUNT, 1);
ServiceStat st = s.getStat(Service.STAT_NAME_NODE_GROUP_SYNCH_DELAYED_COUNT);
if (st != null && st.latestValue % 10 == 0) {
shouldLog = true;
}
}
long deltaSeconds = TimeUnit.MICROSECONDS.toSeconds(delta);
if (shouldLog) {
this.host.log(Level.WARNING, "Service %s has been synchronizing for %d seconds",
link, deltaSeconds);
}
if (hostState.peerSynchronizationTimeLimitSeconds < deltaSeconds) {
this.host.log(Level.WARNING, "Service %s has exceeded synchronization limit of %d",
link, hostState.peerSynchronizationTimeLimitSeconds);
this.synchronizationActiveServices.remove(link);
}
}
for (Entry en : this.synchronizationRequiredServices
.entrySet()) {
now = Utils.getSystemNowMicrosUtc();
if (this.host.isStopping()) {
return;
}
String link = en.getKey();
Long lastSynchTime = en.getValue();
if (lastSynchTime >= selectorSynchTime) {
continue;
}
Service s = this.host.findService(link, true);
if (s == null) {
continue;
}
if (s.getProcessingStage() != ProcessingStage.AVAILABLE) {
continue;
}
if (!s.hasOption(ServiceOption.FACTORY)) {
continue;
}
if (!s.hasOption(ServiceOption.REPLICATION)) {
continue;
}
String serviceSelectorPath = s.getPeerNodeSelectorPath();
if (!nodeSelectorPath.equals(serviceSelectorPath)) {
continue;
}
Operation maintOp = Operation.createPost(s.getUri()).setCompletion((o, e) -> {
this.synchronizationActiveServices.remove(link);
if (e != null) {
this.host.log(Level.WARNING, "Node group change maintenance failed for %s: %s",
s.getSelfLink(),
e.getMessage());
}
this.host.log(Level.FINE, "Synch done for selector %s, service %s",
nodeSelectorPath, s.getSelfLink());
});
// update service entry so we do not reschedule it
this.synchronizationRequiredServices.put(link, now);
this.synchronizationActiveServices.put(link, now);
ServiceMaintenanceRequest body = ServiceMaintenanceRequest.create();
body.reasons.add(MaintenanceReason.NODE_GROUP_CHANGE);
body.nodeGroupState = ngs;
maintOp.setBodyNoCloning(body);
long n = now;
// allow overlapping node group change maintenance requests
this.host
.run(() -> {
OperationContext.setAuthorizationContext(this.host
.getSystemAuthorizationContext());
this.host.log(Level.FINE, " Synchronizing %s (last:%d, sl: %d now:%d)",
link,
lastSynchTime, selectorSynchTime, n);
s.adjustStat(Service.STAT_NAME_NODE_GROUP_CHANGE_MAINTENANCE_COUNT, 1);
s.handleMaintenance(maintOp);
});
}
}
public void close() {
this.synchronizationTimes.clear();
this.synchronizationRequiredServices.clear();
this.synchronizationActiveServices.clear();
this.pendingNodeSelectorsForFactorySynch.clear();
}
}
