Search in sources :

Example 31 with Deployment

use of com.spotify.helios.common.descriptors.Deployment in project helios by spotify.

the class ZooKeeperMasterModel method getUndeployOperations.

private List<ZooKeeperOperation> getUndeployOperations(final ZooKeeperClient client, final String host, final JobId jobId, final String token) throws HostNotFoundException, JobNotDeployedException, TokenVerificationException {
    assertHostExists(client, host);
    final Deployment deployment = getDeployment(host, jobId);
    if (deployment == null) {
        throw new JobNotDeployedException(host, jobId);
    }
    final Job job = getJob(client, jobId);
    verifyToken(token, job);
    final String configHostJobPath = Paths.configHostJob(host, jobId);
    try {
        // use listRecursive to remove both job node and its child creation node
        final List<String> nodes = newArrayList(reverse(client.listRecursive(configHostJobPath)));
        nodes.add(Paths.configJobHost(jobId, host));
        final List<Integer> staticPorts = staticPorts(job);
        for (final int port : staticPorts) {
            nodes.add(Paths.configHostPort(host, port));
        }
        return ImmutableList.of(delete(nodes));
    } catch (NoNodeException e) {
        // throw an exception and handle it the same as if we discovered this earlier.
        throw new JobNotDeployedException(host, jobId);
    } catch (KeeperException e) {
        throw new HeliosRuntimeException("calculating undeploy operations failed", e);
    }
}
Also used : NoNodeException(org.apache.zookeeper.KeeperException.NoNodeException) HeliosRuntimeException(com.spotify.helios.common.HeliosRuntimeException) Deployment(com.spotify.helios.common.descriptors.Deployment) Job(com.spotify.helios.common.descriptors.Job) KeeperException(org.apache.zookeeper.KeeperException)

Example 32 with Deployment

use of com.spotify.helios.common.descriptors.Deployment in project helios by spotify.

the class TerminationTest method testTermOnExit.

@Test
public void testTermOnExit() throws Exception {
    startDefaultMaster();
    final String host = testHost();
    startDefaultAgent(host);
    final HeliosClient client = defaultClient();
    awaitHostStatus(client, host, UP, LONG_WAIT_SECONDS, SECONDS);
    // Note: signal 15 is SIGTERM
    final Job jobToInterrupt = Job.newBuilder().setName(testJobName).setVersion(testJobVersion).setImage(BUSYBOX).setCommand(asList("/bin/sh", "-c", "trap handle 15; handle() { echo term; exit 0; }; " + "while true; do sleep 1; done")).build();
    final JobId jobId = createJob(jobToInterrupt);
    deployJob(jobId, host);
    awaitTaskState(jobId, host, RUNNING);
    client.setGoal(new Deployment(jobId, Goal.STOP, Deployment.EMTPY_DEPLOYER_USER, Deployment.EMPTY_DEPLOYER_MASTER, Deployment.EMPTY_DEPLOYMENT_GROUP_NAME), host);
    final TaskStatus taskStatus = awaitTaskState(jobId, host, STOPPED);
    final String log;
    try (final DockerClient dockerClient = getNewDockerClient();
        LogStream logs = dockerClient.logs(taskStatus.getContainerId(), stdout())) {
        log = logs.readFully();
    }
    // Message expected, because the SIGTERM handler in the script should have run
    assertEquals("term\n", log);
}
Also used : DockerClient(com.spotify.docker.client.DockerClient) Deployment(com.spotify.helios.common.descriptors.Deployment) LogStream(com.spotify.docker.client.LogStream) HeliosClient(com.spotify.helios.client.HeliosClient) Job(com.spotify.helios.common.descriptors.Job) TaskStatus(com.spotify.helios.common.descriptors.TaskStatus) JobId(com.spotify.helios.common.descriptors.JobId) Test(org.junit.Test)

Example 33 with Deployment

use of com.spotify.helios.common.descriptors.Deployment in project helios by spotify.

the class DeploymentTest method testLotsOfConcurrentJobs.

@Test
public void testLotsOfConcurrentJobs() throws Exception {
    startDefaultMaster();
    final HeliosClient client = defaultClient();
    startDefaultAgent(testHost());
    awaitHostRegistered(client, testHost(), LONG_WAIT_SECONDS, SECONDS);
    awaitHostStatus(client, testHost(), UP, LONG_WAIT_SECONDS, SECONDS);
    final int numberOfJobs = 40;
    final List<JobId> jobIds = Lists.newArrayListWithCapacity(numberOfJobs);
    final String jobName = testJobName + "_" + toHexString(ThreadLocalRandom.current().nextInt());
    // create and deploy a bunch of jobs
    for (Integer i = 0; i < numberOfJobs; i++) {
        final Job job = Job.newBuilder().setName(jobName).setVersion(i.toString()).setImage(BUSYBOX).setCommand(IDLE_COMMAND).setCreatingUser(TEST_USER).build();
        final JobId jobId = job.getId();
        final CreateJobResponse created = client.createJob(job).get();
        assertEquals(CreateJobResponse.Status.OK, created.getStatus());
        final Deployment deployment = Deployment.of(jobId, START, TEST_USER);
        final JobDeployResponse deployed = client.deploy(deployment, testHost()).get();
        assertEquals(JobDeployResponse.Status.OK, deployed.getStatus());
        jobIds.add(jobId);
    }
    // get the container ID's for the jobs
    final Set<String> containerIds = Sets.newHashSetWithExpectedSize(numberOfJobs);
    for (final JobId jobId : jobIds) {
        final TaskStatus taskStatus = awaitJobState(client, testHost(), jobId, RUNNING, LONG_WAIT_SECONDS, SECONDS);
        containerIds.add(taskStatus.getContainerId());
    }
    try (final DockerClient dockerClient = getNewDockerClient()) {
        // kill all the containers for the jobs
        for (final String containerId : containerIds) {
            dockerClient.killContainer(containerId);
        }
        // make sure all the containers come back up
        final int restartedContainers = Polling.await(LONG_WAIT_SECONDS, SECONDS, new Callable<Integer>() {

            @Override
            public Integer call() throws Exception {
                int matchingContainerCount = 0;
                for (final Container c : dockerClient.listContainers()) {
                    for (final String name : c.names()) {
                        if (name.contains(jobName)) {
                            matchingContainerCount++;
                        }
                    }
                }
                if (matchingContainerCount < containerIds.size()) {
                    return null;
                } else {
                    return matchingContainerCount;
                }
            }
        });
        assertEquals(numberOfJobs, restartedContainers);
    }
}
Also used : DockerClient(com.spotify.docker.client.DockerClient) Deployment(com.spotify.helios.common.descriptors.Deployment) Integer.toHexString(java.lang.Integer.toHexString) HeliosClient(com.spotify.helios.client.HeliosClient) TaskStatus(com.spotify.helios.common.descriptors.TaskStatus) JobDeployResponse(com.spotify.helios.common.protocol.JobDeployResponse) Container(com.spotify.docker.client.messages.Container) CreateJobResponse(com.spotify.helios.common.protocol.CreateJobResponse) Job(com.spotify.helios.common.descriptors.Job) JobId(com.spotify.helios.common.descriptors.JobId) Test(org.junit.Test)

Example 34 with Deployment

use of com.spotify.helios.common.descriptors.Deployment in project helios by spotify.

the class DeregisterTest method testDeregisterJobDeployedWithoutStatus.

// Verify that we can deregister a host there are jobs deployed to it, for which there's no
// corresponding status information. For example, if a job was deployed to the host after is went
// down.
@Test
public void testDeregisterJobDeployedWithoutStatus() throws Exception {
    startDefaultMaster();
    final String host = testHost();
    final HeliosClient client = defaultClient();
    final DefaultZooKeeperClient zkClient = new DefaultZooKeeperClient(zk().curatorWithSuperAuth());
    final String idPath = Paths.configHostId(host);
    ZooKeeperRegistrarUtil.registerHost(zkClient, idPath, host, UUID.randomUUID().toString());
    // Create a job
    final Job job = Job.newBuilder().setName(testJobName).setVersion(testJobVersion).setImage(BUSYBOX).setCommand(IDLE_COMMAND).setPorts(ImmutableMap.of("foo", PortMapping.of(4711), "bar", PortMapping.of(4712, ports.localPort("bar")))).build();
    final JobId jobId = job.getId();
    final CreateJobResponse created = client.createJob(job).get();
    assertEquals(CreateJobResponse.Status.OK, created.getStatus());
    // Deploy the job on the agent
    final Deployment deployment = Deployment.of(jobId, START);
    final JobDeployResponse deployed = client.deploy(deployment, host).get();
    assertEquals(JobDeployResponse.Status.OK, deployed.getStatus());
    // Deregister agent
    final HostDeregisterResponse deregisterResponse = client.deregisterHost(host).get();
    assertEquals(HostDeregisterResponse.Status.OK, deregisterResponse.getStatus());
    // Verify that it's possible to remove the job
    final JobDeleteResponse deleteResponse = client.deleteJob(jobId).get();
    assertEquals(JobDeleteResponse.Status.OK, deleteResponse.getStatus());
}
Also used : HostDeregisterResponse(com.spotify.helios.common.protocol.HostDeregisterResponse) CreateJobResponse(com.spotify.helios.common.protocol.CreateJobResponse) Deployment(com.spotify.helios.common.descriptors.Deployment) HeliosClient(com.spotify.helios.client.HeliosClient) DefaultZooKeeperClient(com.spotify.helios.servicescommon.coordination.DefaultZooKeeperClient) Job(com.spotify.helios.common.descriptors.Job) JobDeployResponse(com.spotify.helios.common.protocol.JobDeployResponse) JobId(com.spotify.helios.common.descriptors.JobId) JobDeleteResponse(com.spotify.helios.common.protocol.JobDeleteResponse) Test(org.junit.Test)

Example 35 with Deployment

use of com.spotify.helios.common.descriptors.Deployment in project helios by spotify.

the class DeploymentGroupTest method testRollingUpdateMigrateNothingToUndeploy.

@Test
public void testRollingUpdateMigrateNothingToUndeploy() throws Exception {
    final String host = testHost();
    startDefaultAgent(host, "--labels", TEST_LABEL);
    // Wait for agent to come up
    final HeliosClient client = defaultClient();
    awaitHostStatus(client, testHost(), UP, LONG_WAIT_SECONDS, SECONDS);
    // Manually deploy a job on the host
    final String manualJobVersion = "foo-" + testJobVersion;
    final JobId manualJobId = createJob(testJobName, manualJobVersion, BUSYBOX, IDLE_COMMAND);
    deployJob(manualJobId, host);
    awaitTaskState(manualJobId, host, TaskStatus.State.RUNNING);
    // create a deployment group and trigger a migration rolling-update -- with a different
    // job that the one deployed manually! The manually deployed job should remain running on the
    // host.
    final JobId jobId = createJob(testJobName, testJobVersion, BUSYBOX, IDLE_COMMAND);
    cli("create-deployment-group", "--json", TEST_GROUP, TEST_LABEL);
    cli("rolling-update", "--async", "--migrate", testJobNameAndVersion, TEST_GROUP);
    // rolling-update should succeed & job should be running
    awaitDeploymentGroupStatus(defaultClient(), TEST_GROUP, DeploymentGroupStatus.State.DONE);
    awaitTaskState(jobId, host, TaskStatus.State.RUNNING);
    final String jobDeploymentGroup = Polling.await(LONG_WAIT_SECONDS, SECONDS, new Callable<String>() {

        @Override
        public String call() throws Exception {
            final Deployment deployment = defaultClient().hostStatus(host).get().getJobs().get(jobId);
            if (deployment != null && !isNullOrEmpty(deployment.getDeploymentGroupName())) {
                return deployment.getDeploymentGroupName();
            } else {
                return null;
            }
        }
    });
    assertEquals(TEST_GROUP, jobDeploymentGroup);
    // Ensure that the manually deployed job is still there & running
    final Deployment manualDeployment = defaultClient().hostStatus(host).get().getJobs().get(manualJobId);
    assertNotNull(manualDeployment);
    assertEquals(Goal.START, manualDeployment.getGoal());
}
Also used : Deployment(com.spotify.helios.common.descriptors.Deployment) Matchers.containsString(org.hamcrest.Matchers.containsString) HeliosClient(com.spotify.helios.client.HeliosClient) JobId(com.spotify.helios.common.descriptors.JobId) Test(org.junit.Test)

Aggregations

Deployment (com.spotify.helios.common.descriptors.Deployment)44 JobId (com.spotify.helios.common.descriptors.JobId)30 Test (org.junit.Test)24 Job (com.spotify.helios.common.descriptors.Job)21 HeliosClient (com.spotify.helios.client.HeliosClient)19 JobDeployResponse (com.spotify.helios.common.protocol.JobDeployResponse)18 TaskStatus (com.spotify.helios.common.descriptors.TaskStatus)16 CreateJobResponse (com.spotify.helios.common.protocol.CreateJobResponse)14 DockerClient (com.spotify.docker.client.DockerClient)8 HostStatus (com.spotify.helios.common.descriptors.HostStatus)6 Matchers.containsString (org.hamcrest.Matchers.containsString)6 AgentMain (com.spotify.helios.agent.AgentMain)5 HeliosRuntimeException (com.spotify.helios.common.HeliosRuntimeException)5 JobStatus (com.spotify.helios.common.descriptors.JobStatus)5 ZooKeeperClient (com.spotify.helios.servicescommon.coordination.ZooKeeperClient)5 KeeperException (org.apache.zookeeper.KeeperException)5 NoNodeException (org.apache.zookeeper.KeeperException.NoNodeException)5 RolloutTask (com.spotify.helios.common.descriptors.RolloutTask)3 Task (com.spotify.helios.common.descriptors.Task)3 JobUndeployResponse (com.spotify.helios.common.protocol.JobUndeployResponse)3