纵有疾风起
人生不言弃

Mesos源码分析(11): Mesos-Master接收到launchTasks消息

根据Mesos源码分析(6): Mesos Master的初始化中的代码分析,当Mesos-Master接收到launchTask消息的时候,会调用Master::launchTasks函数。

 

  1. void Master::launchTasks(
  2.     const UPID& from,
  3.     const FrameworkID& frameworkId,
  4.     const vector<TaskInfo>& tasks,
  5.     const Filters& filters,
  6.     const vector<OfferID>& offerIds)
  7. {
  8.   Framework* framework = getFramework(frameworkId);
  9.  
  10.   if (framework == NULL) {
  11.     LOG(WARNING)
  12.       << “Ignoring launch tasks message for offers ” << stringify(offerIds)
  13.       << “ of framework ” << frameworkId
  14.       << “ because the framework cannot be found“;
  15.  
  16.     return;
  17.   }
  18.  
  19.   if (framework->pid != from) {
  20.     LOG(WARNING)
  21.       << “Ignoring launch tasks message for offers ” << stringify(offerIds)
  22.       << “ from ‘” << from << “‘ because it is not from the
  23.       << “ registered framework ” << *framework;
  24.  
  25.     return;
  26.   }
  27.  
  28.   // Currently when no tasks are specified in the launchTasks message
  29.   // it is implicitly considered a decline of the offers.
  30.   if (!tasks.empty()) {
  31.     scheduler::Call::Accept message;
  32.     message.mutable_filters()->CopyFrom(filters);
  33.  
  34.     Offer::Operation* operation = message.add_operations();
  35.     operation->set_type(Offer::Operation::LAUNCH);
  36.  
  37.     foreach (const TaskInfo& task, tasks) {
  38.       operation->mutable_launch()->add_task_infos()->CopyFrom(task);
  39.     }
  40.  
  41.     foreach (const OfferID& offerId, offerIds) {
  42.       message.add_offer_ids()->CopyFrom(offerId);
  43.     }
  44.  
  45.     accept(framework, message);
  46.   } else {
  47.     scheduler::Call::Decline message;
  48.     message.mutable_filters()->CopyFrom(filters);
  49.  
  50.     foreach (const OfferID& offerId, offerIds) {
  51.       message.add_offer_ids()->CopyFrom(offerId);
  52.     }
  53.  
  54.     decline(framework, message);
  55.   }
  56. }

 

它会进一步调用accept函数

  1. void Master::accept(
  2.     Framework* framework,
  3.     const scheduler::Call::Accept& accept)
  4. {
  5.   CHECK_NOTNULL(framework);
  6.  
  7.   foreach (const Offer::Operation& operation, accept.operations()) {
  8.     if (operation.type() == Offer::Operation::LAUNCH) {
  9.       if (operation.launch().task_infos().size() > 0) {
  10.         ++metrics->messages_launch_tasks;
  11.       } else {
  12.         ++metrics->messages_decline_offers;
  13.       }
  14.     }
  15.  
  16.     // TODO(jieyu): Add metrics for non launch operations.
  17.   }
  18.  
  19.   // TODO(bmahler): We currently only support using multiple offers
  20.   // for a single slave.
  21.   Resources offeredResources;
  22.   Option<SlaveID> slaveId = None();
  23.   Option<Error> error = None();
  24.  
  25.   if (accept.offer_ids().size() == 0) {
  26.     error = Error(“No offers specified“);
  27.   } else {
  28.     // Validate the offers.
  29.     error = validation::offer::validate(accept.offer_ids(), this, framework);
  30.  
  31.     // Compute offered resources and remove the offers. If the
  32.     // validation failed, return resources to the allocator.
  33.     foreach (const OfferID& offerId, accept.offer_ids()) {
  34.       Offer* offer = getOffer(offerId);
  35.  
  36.       // Since we re-use `OfferID`s, it is possible to arrive here with either
  37.       // a resource offer, or an inverse offer. We first try as a resource offer
  38.       // and if that fails, then we assume it is an inverse offer. This is
  39.       // correct as those are currently the only 2 ways to get an `OfferID`.
  40.       if (offer != NULL) {
  41.         slaveId = offer->slave_id();
  42.         offeredResources += offer->resources();
  43.  
  44.         if (error.isSome()) {
  45.           allocator->recoverResources(
  46.               offer->framework_id(),
  47.               offer->slave_id(),
  48.               offer->resources(),
  49.               None());
  50.         }
  51.         removeOffer(offer);
  52.         continue;
  53.       }
  54.  
  55.       // Try it as an inverse offer. If this fails then the offer is no longer
  56.       // valid.
  57.       InverseOffer* inverseOffer = getInverseOffer(offerId);
  58.       if (inverseOffer != NULL) {
  59.         mesos::master::InverseOfferStatus status;
  60.         status.set_status(mesos::master::InverseOfferStatus::ACCEPT);
  61.         status.mutable_framework_id()->CopyFrom(inverseOffer->framework_id());
  62.         status.mutable_timestamp()->CopyFrom(protobuf::getCurrentTime());
  63.  
  64.         allocator->updateInverseOffer(
  65.             inverseOffer->slave_id(),
  66.             inverseOffer->framework_id(),
  67.             UnavailableResources{
  68.                 inverseOffer->resources(),
  69.                 inverseOffer->unavailability()},
  70.             status,
  71.             accept.filters());
  72.  
  73.         removeInverseOffer(inverseOffer);
  74.         continue;
  75.       }
  76.  
  77.       // If the offer was neither in our offer or inverse offer sets, then this
  78.       // offer is no longer valid.
  79.       LOG(WARNING) << “Ignoring accept of offer ” << offerId
  80.                    << “ since it is no longer valid“;
  81.     }
  82.   }
  83.  
  84.   // If invalid, send TASK_LOST for the launch attempts.
  85.   // TODO(jieyu): Consider adding a ‘drop’ overload for ACCEPT call to
  86.   // consistently handle message dropping. It would be ideal if the
  87.   // ‘drop’ overload can handle both resource recovery and lost task
  88.   // notifications.
  89.   if (error.isSome()) {
  90.     LOG(WARNING) << “ACCEPT call used invalid offers ‘” << accept.offer_ids()
  91.                  << “‘: ” << error.get().message;
  92.  
  93.     foreach (const Offer::Operation& operation, accept.operations()) {
  94.       if (operation.type() != Offer::Operation::LAUNCH) {
  95.         continue;
  96.       }
  97.  
  98.       foreach (const TaskInfo& task, operation.launch().task_infos()) {
  99.         const StatusUpdate& update = protobuf::createStatusUpdate(
  100.             framework->id(),
  101.             task.slave_id(),
  102.             task.task_id(),
  103.             TASK_LOST,
  104.             TaskStatus::SOURCE_MASTER,
  105.             None(),
  106.             “Task launched with invalid offers: ” + error.get().message,
  107.             TaskStatus::REASON_INVALID_OFFERS);
  108.  
  109.         metrics->tasks_lost++;
  110.  
  111.         metrics->incrementTasksStates(
  112.             TASK_LOST,
  113.             TaskStatus::SOURCE_MASTER,
  114.             TaskStatus::REASON_INVALID_OFFERS);
  115.  
  116.         forward(update, UPID(), framework);
  117.       }
  118.     }
  119.  
  120.     return;
  121.   }
  122.  
  123.   CHECK_SOME(slaveId);
  124.   Slave* slave = slaves.registered.get(slaveId.get());
  125.   CHECK_NOTNULL(slave);
  126.  
  127.   LOG(INFO) << “Processing ACCEPT call for offers: ” << accept.offer_ids()
  128.             << “ on slave ” << *slave << “ for framework ” << *framework;
  129.  
  130.   list<Future<bool>> futures;
  131.   foreach (const Offer::Operation& operation, accept.operations()) {
  132.     switch (operation.type()) {
  133.       case Offer::Operation::LAUNCH: {
  134.         // Authorize the tasks. A task is in ‘framework->pendingTasks’
  135.         // before it is authorized.
  136.         foreach (const TaskInfo& task, operation.launch().task_infos()) {
  137.           futures.push_back(authorizeTask(task, framework));
  138.  
  139.           // Add to pending tasks.
  140.           //
  141.           // NOTE: The task ID here hasn’t been validated yet, but it
  142.           // doesn’t matter. If the task ID is not valid, the task won’t
  143.           // be launched anyway. If two tasks have the same ID, the second
  144.           // one will not be put into ‘framework->pendingTasks’, therefore
  145.           // will not be launched.
  146.           if (!framework->pendingTasks.contains(task.task_id())) {
  147.             framework->pendingTasks[task.task_id()] = task;
  148.           }
  149.         }
  150.         break;
  151.       }
  152.  
  153.       // NOTE: When handling RESERVE and UNRESERVE operations, authorization
  154.       // will proceed even if no principal is specified, although currently
  155.       // resources cannot be reserved or unreserved unless a principal is
  156.       // provided. Any RESERVE/UNRESERVE operation with no associated principal
  157.       // will be found invalid when `validate()` is called in `_accept()` below.
  158.  
  159.       // The RESERVE operation allows a principal to reserve resources.
  160.       case Offer::Operation::RESERVE: {
  161.         Option<string> principal = framework->info.has_principal()
  162.           ? framework->info.principal()
  163.           : Option<string>::none();
  164.  
  165.         futures.push_back(
  166.             authorizeReserveResources(
  167.                 operation.reserve(), principal));
  168.  
  169.         break;
  170.       }
  171.  
  172.       // The UNRESERVE operation allows a principal to unreserve resources.
  173.       case Offer::Operation::UNRESERVE: {
  174.         Option<string> principal = framework->info.has_principal()
  175.           ? framework->info.principal()
  176.           : Option<string>::none();
  177.  
  178.         futures.push_back(
  179.             authorizeUnreserveResources(
  180.                 operation.unreserve(), principal));
  181.  
  182.         break;
  183.       }
  184.  
  185.       // The CREATE operation allows the creation of a persistent volume.
  186.       case Offer::Operation::CREATE: {
  187.         Option<string> principal = framework->info.has_principal()
  188.           ? framework->info.principal()
  189.           : Option<string>::none();
  190.  
  191.         futures.push_back(
  192.             authorizeCreateVolume(
  193.                 operation.create(), principal));
  194.  
  195.         break;
  196.       }
  197.  
  198.       // The DESTROY operation allows the destruction of a persistent volume.
  199.       case Offer::Operation::DESTROY: {
  200.         Option<string> principal = framework->info.has_principal()
  201.           ? framework->info.principal()
  202.           : Option<string>::none();
  203.  
  204.         futures.push_back(
  205.             authorizeDestroyVolume(
  206.                 operation.destroy(), principal));
  207.  
  208.         break;
  209.       }
  210.     }
  211.   }
  212.  
  213.   // Wait for all the tasks to be authorized.
  214.   await(futures)
  215.     .onAny(defer(self(),
  216.                  &Master::_accept,
  217.                  framework->id(),
  218.                  slaveId.get(),
  219.                  offeredResources,
  220.                  accept,
  221.                  lambda::_1));
  222. }

 

如果鉴权通过,则调用Master::_accept

  1. void Master::_accept(
  2.     const FrameworkID& frameworkId,
  3.     const SlaveID& slaveId,
  4.     const Resources& offeredResources,
  5.     const scheduler::Call::Accept& accept,
  6.     const Future<list<Future<bool>>>& _authorizations)
  7. {
  8.   Framework* framework = getFramework(frameworkId);
  9. ……
  10.  
  11.   Slave* slave = slaves.registered.get(slaveId);
  12.  
  13.   if (slave == NULL || !slave->connected) {
  14.     foreach (const Offer::Operation& operation, accept.operations()) {
  15.       if (operation.type() != Offer::Operation::LAUNCH) {
  16.         continue;
  17.       }
  18.  
  19.       foreach (const TaskInfo& task, operation.launch().task_infos()) {
  20.         const TaskStatus::Reason reason =
  21.             slave == NULL ? TaskStatus::REASON_SLAVE_REMOVED
  22.                           : TaskStatus::REASON_SLAVE_DISCONNECTED;
  23.         const StatusUpdate& update = protobuf::createStatusUpdate(
  24.             framework->id(),
  25.             task.slave_id(),
  26.             task.task_id(),
  27.             TASK_LOST,
  28.             TaskStatus::SOURCE_MASTER,
  29.             None(),
  30.             slave == NULL ? “Slave removed” : “Slave disconnected“,
  31.             reason);
  32.  
  33.         metrics->tasks_lost++;
  34.  
  35.         metrics->incrementTasksStates(
  36.             TASK_LOST,
  37.             TaskStatus::SOURCE_MASTER,
  38.             reason);
  39.  
  40.         forward(update, UPID(), framework);
  41.       }
  42.     }
  43.  
  44.     // Tell the allocator about the recovered resources.
  45.     allocator->recoverResources(
  46.         frameworkId,
  47.         slaveId,
  48.         offeredResources,
  49.         None());
  50.  
  51.     return;
  52.   }
  53.  
  54.   // Some offer operations update the offered resources. We keep
  55.   // updated offered resources here. When a task is successfully
  56.   // launched, we remove its resource from offered resources.
  57.   Resources _offeredResources = offeredResources;
  58.  
  59.   // The order of `authorizations` must match the order of the operations in
  60.   // `accept.operations()`, as they are iterated through simultaneously.
  61.   CHECK_READY(_authorizations);
  62.   list<Future<bool>> authorizations = _authorizations.get();
  63.  
  64.   foreach (const Offer::Operation& operation, accept.operations()) {
  65.     switch (operation.type()) {
  66.       // The RESERVE operation allows a principal to reserve resources.
  67.       case Offer::Operation::RESERVE: {
  68.         Future<bool> authorization = authorizations.front();
  69.         authorizations.pop_front();
  70.  
  71.         CHECK(!authorization.isDiscarded());
  72.  
  73.         if (authorization.isFailed()) {
  74.           // TODO(greggomann): We may want to retry this failed authorization
  75.           // request rather than dropping it immediately.
  76.           drop(framework,
  77.                operation,
  78.                “Authorization of principal ‘” + framework->info.principal() +
  79.                “‘ to reserve resources failed: ” +
  80.                authorization.failure());
  81.  
  82.           continue;
  83.         } elseif (!authorization.get()) {
  84.           drop(framework,
  85.                operation,
  86.                “Not authorized to reserve resources as ‘” +
  87.                  framework->info.principal() + ““);
  88.  
  89.           continue;
  90.         }
  91.  
  92.         Option<string> principal = framework->info.has_principal()
  93.           ? framework->info.principal()
  94.           : Option<string>::none();
  95.  
  96.         // Make sure this reserve operation is valid.
  97.         Option<Error> error = validation::operation::validate(
  98.             operation.reserve(), principal);
  99.  
  100.         if (error.isSome()) {
  101.           drop(framework, operation, error.get().message);
  102.           continue;
  103.         }
  104.  
  105.         // Test the given operation on the included resources.
  106.         Try<Resources> resources = _offeredResources.apply(operation);
  107.         if (resources.isError()) {
  108.           drop(framework, operation, resources.error());
  109.           continue;
  110.         }
  111.  
  112.         _offeredResources = resources.get();
  113.  
  114.         LOG(INFO) << “Applying RESERVE operation for resources
  115.                   << operation.reserve().resources() << “ from framework
  116.                   << *framework << “ to slave ” << *slave;
  117.  
  118.         apply(framework, slave, operation);
  119.         break;
  120.       }
  121.  
  122.       // The UNRESERVE operation allows a principal to unreserve resources.
  123.       case Offer::Operation::UNRESERVE: {
  124.         Future<bool> authorization = authorizations.front();
  125.         authorizations.pop_front();
  126.  
  127.         CHECK(!authorization.isDiscarded());
  128.  
  129.         if (authorization.isFailed()) {
  130.           // TODO(greggomann): We may want to retry this failed authorization
  131.           // request rather than dropping it immediately.
  132.           drop(framework,
  133.                operation,
  134.                “Authorization of principal ‘” + framework->info.principal() +
  135.                “‘ to unreserve resources failed: ” +
  136.                authorization.failure());
  137.  
  138.           continue;
  139.         } elseif (!authorization.get()) {
  140.           drop(framework,
  141.                operation,
  142.                “Not authorized to unreserve resources as ‘” +
  143.                  framework->info.principal() + ““);
  144.  
  145.           continue;
  146.         }
  147.  
  148.         // Make sure this unreserve operation is valid.
  149.         Option<Error> error = validation::operation::validate(
  150.             operation.unreserve());
  151.  
  152.         if (error.isSome()) {
  153.           drop(framework, operation, error.get().message);
  154.           continue;
  155.         }
  156.  
  157.         // Test the given operation on the included resources.
  158.         Try<Resources> resources = _offeredResources.apply(operation);
  159.         if (resources.isError()) {
  160.           drop(framework, operation, resources.error());
  161.           continue;
  162.         }
  163.  
  164.         _offeredResources = resources.get();
  165.  
  166.         LOG(INFO) << “Applying UNRESERVE operation for resources
  167.                   << operation.unreserve().resources() << “ from framework
  168.                   << *framework << “ to slave ” << *slave;
  169.  
  170.         apply(framework, slave, operation);
  171.         break;
  172.       }
  173.  
  174.       case Offer::Operation::CREATE: {
  175.         Future<bool> authorization = authorizations.front();
  176.         authorizations.pop_front();
  177.  
  178.         CHECK(!authorization.isDiscarded());
  179.  
  180.         if (authorization.isFailed()) {
  181.           // TODO(greggomann): We may want to retry this failed authorization
  182.           // request rather than dropping it immediately.
  183.           drop(framework,
  184.                operation,
  185.                “Authorization of principal ‘” + framework->info.principal() +
  186.                “‘ to create persistent volumes failed: ” +
  187.                authorization.failure());
  188.  
  189.           continue;
  190.         } elseif (!authorization.get()) {
  191.           drop(framework,
  192.                operation,
  193.                “Not authorized to create persistent volumes as ‘” +
  194.                  framework->info.principal() + ““);
  195.  
  196.           continue;
  197.         }
  198.  
  199.         // Make sure this create operation is valid.
  200.         Option<Error> error = validation::operation::validate(
  201.             operation.create(), slave->checkpointedResources);
  202.  
  203.         if (error.isSome()) {
  204.           drop(framework, operation, error.get().message);
  205.           continue;
  206.         }
  207.  
  208.         Try<Resources> resources = _offeredResources.apply(operation);
  209.         if (resources.isError()) {
  210.           drop(framework, operation, resources.error());
  211.           continue;
  212.         }
  213.  
  214.         _offeredResources = resources.get();
  215.  
  216.         LOG(INFO) << “Applying CREATE operation for volumes
  217.                   << operation.create().volumes() << “ from framework
  218.                   << *framework << “ to slave ” << *slave;
  219.  
  220.         apply(framework, slave, operation);
  221.         break;
  222.       }
  223.  
  224.       case Offer::Operation::DESTROY: {
  225.         Future<bool> authorization = authorizations.front();
  226.         authorizations.pop_front();
  227.  
  228.         CHECK(!authorization.isDiscarded());
  229.  
  230.         if (authorization.isFailed()) {
  231.           // TODO(greggomann): We may want to retry this failed authorization
  232.           // request rather than dropping it immediately.
  233.           drop(framework,
  234.                operation,
  235.                “Authorization of principal ‘” + framework->info.principal() +
  236.                “‘ to destroy persistent volumes failed: ” +
  237.                authorization.failure());
  238.  
  239.           continue;
  240.         } elseif (!authorization.get()) {
  241.           drop(framework,
  242.                operation,
  243.                “Not authorized to destroy persistent volumes as ‘” +
  244.                  framework->info.principal() + ““);
  245.  
  246.           continue;
  247.         }
  248.  
  249.         // Make sure this destroy operation is valid.
  250.         Option<Error> error = validation::operation::validate(
  251.             operation.destroy(), slave->checkpointedResources);
  252.  
  253.         if (error.isSome()) {
  254.           drop(framework, operation, error.get().message);
  255.           continue;
  256.         }
  257.  
  258.         Try<Resources> resources = _offeredResources.apply(operation);
  259.         if (resources.isError()) {
  260.           drop(framework, operation, resources.error());
  261.           continue;
  262.         }
  263.  
  264.         _offeredResources = resources.get();
  265.  
  266.         LOG(INFO) << “Applying DESTROY operation for volumes
  267.                   << operation.create().volumes() << “ from framework
  268.                   << *framework << “ to slave ” << *slave;
  269.  
  270.         apply(framework, slave, operation);
  271.         break;
  272.       }
  273.  
  274.       case Offer::Operation::LAUNCH: {
  275.         foreach (const TaskInfo& task, operation.launch().task_infos()) {
  276.           Future<bool> authorization = authorizations.front();
  277.           authorizations.pop_front();
  278.  
  279.           // NOTE: The task will not be in ‘pendingTasks’ if
  280.           // ‘killTask()’ for the task was called before we are here.
  281.           // No need to launch the task if it’s no longer pending.
  282.           // However, we still need to check the authorization result
  283.           // and do the validation so that we can send status update
  284.           // in case the task has duplicated ID.
  285.           bool pending = framework->pendingTasks.contains(task.task_id());
  286.  
  287.           // Remove from pending tasks.
  288.           framework->pendingTasks.erase(task.task_id());
  289.  
  290.           CHECK(!authorization.isDiscarded());
  291.  
  292.           if (authorization.isFailed() || !authorization.get()) {
  293.             string user = framework->info.user(); // Default user.
  294.             if (task.has_command() && task.command().has_user()) {
  295.               user = task.command().user();
  296.             } elseif (task.has_executor() &&
  297.                        task.executor().command().has_user()) {
  298.               user = task.executor().command().user();
  299.             }
  300.  
  301.             const StatusUpdate& update = protobuf::createStatusUpdate(
  302.                 framework->id(),
  303.                 task.slave_id(),
  304.                 task.task_id(),
  305.                 TASK_ERROR,
  306.                 TaskStatus::SOURCE_MASTER,
  307.                 None(),
  308.                 authorization.isFailed() ?
  309.                     “Authorization failure: ” + authorization.failure() :
  310.                     “Not authorized to launch as user ‘” + user + ““,
  311.                 TaskStatus::REASON_TASK_UNAUTHORIZED);
  312.  
  313.             metrics->tasks_error++;
  314.  
  315.             metrics->incrementTasksStates(
  316.                 TASK_ERROR,
  317.                 TaskStatus::SOURCE_MASTER,
  318.                 TaskStatus::REASON_TASK_UNAUTHORIZED);
  319.  
  320.             forward(update, UPID(), framework);
  321.  
  322.             continue;
  323.           }
  324.  
  325.           // Validate the task.
  326.  
  327.           // Make a copy of the original task so that we can
  328.           // fill the missing `framework_id` in ExecutorInfo
  329.           // if needed. This field was added to the API later
  330.           // and thus was made optional.
  331.           TaskInfo task_(task);
  332.           if (task.has_executor() && !task.executor().has_framework_id()) {
  333.             task_.mutable_executor()
  334.                 ->mutable_framework_id()->CopyFrom(framework->id());
  335.           }
  336.  
  337.           const Option<Error>& validationError = validation::task::validate(
  338.               task_,
  339.               framework,
  340.               slave,
  341.               _offeredResources);
  342.  
  343.           if (validationError.isSome()) {
  344.             const StatusUpdate& update = protobuf::createStatusUpdate(
  345.                 framework->id(),
  346.                 task_.slave_id(),
  347.                 task_.task_id(),
  348.                 TASK_ERROR,
  349.                 TaskStatus::SOURCE_MASTER,
  350.                 None(),
  351.                 validationError.get().message,
  352.                 TaskStatus::REASON_TASK_INVALID);
  353.  
  354.             metrics->tasks_error++;
  355.  
  356.             metrics->incrementTasksStates(
  357.                 TASK_ERROR,
  358.                 TaskStatus::SOURCE_MASTER,
  359.                 TaskStatus::REASON_TASK_INVALID);
  360.  
  361.             forward(update, UPID(), framework);
  362.  
  363.             continue;
  364.           }
  365.  
  366.           // Add task.
  367.           if (pending) {
  368.             _offeredResources -= addTask(task_, framework, slave);
  369.  
  370.             // TODO(bmahler): Consider updating this log message to
  371.             // indicate when the executor is also being launched.
  372.             LOG(INFO) << “Launching task ” << task_.task_id()
  373.                       << “ of framework ” << *framework
  374.                       << “ with resources ” << task_.resources()
  375.                       << “ on slave ” << *slave;
  376.  
  377.             RunTaskMessage message;
  378.             message.mutable_framework()->MergeFrom(framework->info);
  379.  
  380.             // TODO(anand): We set ‘pid’ to UPID() for http frameworks
  381.             // as ‘pid’ was made optional in 0.24.0. In 0.25.0, we
  382.             // no longer have to set pid here for http frameworks.
  383.             message.set_pid(framework->pid.getOrElse(UPID()));
  384.             message.mutable_task()->MergeFrom(task_);
  385.  
  386.             if (HookManager::hooksAvailable()) {
  387.               // Set labels retrieved from label-decorator hooks.
  388.               message.mutable_task()->mutable_labels()->CopyFrom(
  389.                   HookManager::masterLaunchTaskLabelDecorator(
  390.                       task_,
  391.                       framework->info,
  392.                       slave->info));
  393.             }
  394.  
  395.             send(slave->pid, message);
  396.           }
  397.         }
  398.         break;
  399.       }
  400.  
  401.       default:
  402.         LOG(ERROR) << “Unsupported offer operation ” << operation.type();
  403.         break;
  404.     }
  405.   }
  406.  
  407.   if (!_offeredResources.empty()) {
  408.     // Tell the allocator about the unused (e.g., refused) resources.
  409.     allocator->recoverResources(
  410.         frameworkId,
  411.         slaveId,
  412.         _offeredResources,
  413.         accept.filters());
  414.   }
  415. }

 

Mesos-Master将RunTaskMessage消息发送给Mesos-Slave

文章转载于:https://www.cnblogs.com/popsuper1982/p/5724166.html

原著是一个有趣的人,若有侵权,请通知删除

未经允许不得转载:起风网 » Mesos源码分析(11): Mesos-Master接收到launchTasks消息
分享到: 生成海报

评论 抢沙发

评论前必须登录!

立即登录