Ejemplo n.º 1
0
 /**
  * Counts the number of instances in EC2 currently running that are using the specified image and
  * a template.
  *
  * @param ami If AMI is left null, then all instances are counted and template description is
  *     ignored.
  *     <p>This includes those instances that may be started outside Jenkins.
  * @param templateDesc
  */
 public int countCurrentEC2Slaves(String ami, String templateDesc) throws AmazonClientException {
   int n = 0;
   for (Reservation r : connect().describeInstances().getReservations()) {
     for (Instance i : r.getInstances()) {
       if (isEc2ProvisionedAmiSlave(i, ami, templateDesc)) {
         InstanceStateName stateName = InstanceStateName.fromValue(i.getState().getName());
         if (stateName == InstanceStateName.Pending || stateName == InstanceStateName.Running) {
           EC2AbstractSlave foundSlave = null;
           for (EC2AbstractSlave ec2Node : NodeIterator.nodes(EC2AbstractSlave.class)) {
             if (ec2Node.getInstanceId().equals(i.getInstanceId())) {
               foundSlave = ec2Node;
               break;
             }
           }
           // Don't count disconnected slaves as being used, we will connected them later is
           // required
           if (foundSlave != null && foundSlave.toComputer().isOffline()) continue;
           n++;
         }
       }
     }
   }
   // Count pending spot requests too
   for (SpotInstanceRequest sir :
       connect().describeSpotInstanceRequests().getSpotInstanceRequests()) {
     // Count Spot requests that are open and still have a
     // chance to be active.
     if (sir.getState().equals("open")) {
       n++;
     }
   }
   return n;
 }
Ejemplo n.º 2
0
  @Override
  public Collection<PlannedNode> provision(Label label, int excessWorkload) {
    try {
      // Count number of pending executors from spot requests
      for (EC2SpotSlave n : NodeIterator.nodes(EC2SpotSlave.class)) {
        // If the slave is online then it is already counted by Jenkins
        // We only want to count potential additional Spot instance
        // slaves
        if (n.getComputer().isOffline() && label.matches(n.getAssignedLabels())) {
          DescribeSpotInstanceRequestsRequest dsir =
              new DescribeSpotInstanceRequestsRequest()
                  .withSpotInstanceRequestIds(n.getSpotInstanceRequestId());

          for (SpotInstanceRequest sir :
              connect().describeSpotInstanceRequests(dsir).getSpotInstanceRequests()) {
            // Count Spot requests that are open and still have a
            // chance to be active
            // A request can be active and not yet registered as a
            // slave. We check above
            // to ensure only unregistered slaves get counted
            if (sir.getState().equals("open") || sir.getState().equals("active")) {
              excessWorkload -= n.getNumExecutors();
            }
          }
        }
      }
      LOGGER.log(Level.INFO, "Excess workload after pending Spot instances: " + excessWorkload);

      List<PlannedNode> r = new ArrayList<PlannedNode>();

      final SlaveTemplate t = getTemplate(label);
      int amiCap = t.getInstanceCap();

      while (excessWorkload > 0) {

        if (!addProvisionedSlave(t.ami, amiCap, t.description)) {
          break;
        }

        r.add(
            new PlannedNode(
                t.getDisplayName(),
                Computer.threadPoolForRemoting.submit(
                    new Callable<Node>() {
                      public Node call() throws Exception {
                        // TODO: record the output somewhere
                        try {
                          EC2AbstractSlave s = t.provision(StreamTaskListener.fromStdout());
                          Hudson.getInstance().addNode(s);
                          // EC2 instances may have a long init script. If we
                          // declare
                          // the provisioning complete by returning without
                          // the connect
                          // operation, NodeProvisioner may decide that it
                          // still wants
                          // one more instance, because it sees that (1) all
                          // the slaves
                          // are offline (because it's still being launched)
                          // and
                          // (2) there's no capacity provisioned yet.
                          //
                          // deferring the completion of provisioning until
                          // the launch
                          // goes successful prevents this problem.
                          s.toComputer().connect(false).get();
                          return s;
                        } finally {
                          decrementAmiSlaveProvision(t.ami);
                        }
                      }
                    }),
                t.getNumExecutors()));

        excessWorkload -= t.getNumExecutors();
      }
      return r;
    } catch (AmazonClientException e) {
      LOGGER.log(Level.WARNING, "Failed to count the # of live instances on EC2", e);
      return Collections.emptyList();
    }
  }