use of io.cdap.cdap.master.spi.twill.SecretDisk in project cdap by caskdata.
the class DistributedPreviewManager method run.
@Override
public void run() {
TwillController activeController = null;
for (TwillController controller : twillRunner.lookup(PreviewRunnerTwillApplication.NAME)) {
// If detected more than one controller, terminate those extra controllers.
if (activeController != null) {
controller.terminate();
} else {
activeController = controller;
}
}
// If there is no preview runner running, create one
if (activeController == null) {
try {
Path tmpDir = new File(cConf.get(Constants.CFG_LOCAL_DATA_DIR), cConf.get(Constants.AppFabric.TEMP_DIR)).toPath();
Files.createDirectories(tmpDir);
Path runDir = Files.createTempDirectory(tmpDir, "preview");
try {
CConfiguration cConfCopy = CConfiguration.copy(cConf);
Path cConfPath = runDir.resolve("cConf.xml");
if (!cConf.getBoolean(Constants.Twill.Security.WORKER_MOUNT_SECRET)) {
// Unset the internal certificate path since certificate is stored cdap-security which
// is not going to be exposed to preview runner.
// TODO: CDAP-18768 this will break preview when certificate checking is enabled.
cConfCopy.unset(Constants.Security.SSL.INTERNAL_CERT_PATH);
}
try (Writer writer = Files.newBufferedWriter(cConfPath, StandardCharsets.UTF_8)) {
cConfCopy.writeXml(writer);
}
Path hConfPath = runDir.resolve("hConf.xml");
try (Writer writer = Files.newBufferedWriter(hConfPath, StandardCharsets.UTF_8)) {
hConf.writeXml(writer);
}
Boolean artifactLocalizerEnabled = cConf.getBoolean(Constants.Preview.ARTIFACT_LOCALIZER_ENABLED);
ResourceSpecification runnerResourceSpec = ResourceSpecification.Builder.with().setVirtualCores(cConf.getInt(Constants.Preview.CONTAINER_CORES)).setMemory(cConf.getInt(Constants.Preview.CONTAINER_MEMORY_MB), ResourceSpecification.SizeUnit.MEGA).setInstances(cConf.getInt(Constants.Preview.CONTAINER_COUNT)).build();
Optional<ResourceSpecification> artifactLocalizerResourceSpec = Optional.empty();
if (artifactLocalizerEnabled) {
artifactLocalizerResourceSpec = Optional.of(ResourceSpecification.Builder.with().setVirtualCores(cConf.getInt(Constants.ArtifactLocalizer.CONTAINER_CORES)).setMemory(cConf.getInt(Constants.ArtifactLocalizer.CONTAINER_MEMORY_MB), ResourceSpecification.SizeUnit.MEGA).setInstances(cConf.getInt(Constants.TaskWorker.CONTAINER_COUNT)).build());
}
LOG.info("Starting preview runners with {} instances and artifactLocalizer {}", runnerResourceSpec.getInstances(), artifactLocalizerEnabled ? "enabled" : "disabled");
TwillPreparer twillPreparer = twillRunner.prepare(new PreviewRunnerTwillApplication(cConfPath.toUri(), hConfPath.toUri(), runnerResourceSpec, artifactLocalizerResourceSpec));
String priorityClass = cConf.get(Constants.Preview.CONTAINER_PRIORITY_CLASS_NAME);
if (priorityClass != null) {
twillPreparer = twillPreparer.setSchedulerQueue(priorityClass);
}
if (twillPreparer instanceof DependentTwillPreparer) {
if (artifactLocalizerEnabled) {
twillPreparer = ((DependentTwillPreparer) twillPreparer).dependentRunnableNames(PreviewRunnerTwillRunnable.class.getSimpleName(), ArtifactLocalizerTwillRunnable.class.getSimpleName());
}
}
if (twillPreparer instanceof StatefulTwillPreparer) {
int diskSize = cConf.getInt(Constants.Preview.CONTAINER_DISK_SIZE_GB);
twillPreparer = ((StatefulTwillPreparer) twillPreparer).withStatefulRunnable(PreviewRunnerTwillRunnable.class.getSimpleName(), false, new StatefulDisk("preview-runner-data", diskSize, cConf.get(Constants.CFG_LOCAL_DATA_DIR)));
}
if (twillPreparer instanceof SecureTwillPreparer) {
String twillUserIdentity = cConf.get(Constants.Twill.Security.IDENTITY_USER);
if (twillUserIdentity != null) {
SecurityContext securityContext = new SecurityContext.Builder().withIdentity(twillUserIdentity).build();
twillPreparer = ((SecureTwillPreparer) twillPreparer).withSecurityContext(PreviewRunnerTwillRunnable.class.getSimpleName(), securityContext);
}
if (artifactLocalizerEnabled) {
// Mount secret in ArtifactLocalizer sidecar which only runs trusted code,
// so requests originated by ArtifactLocalizer can run with system identity when internal auth
// is enabled.
String secretName = cConf.get(Constants.Twill.Security.MASTER_SECRET_DISK_NAME);
String secretPath = cConf.get(Constants.Twill.Security.MASTER_SECRET_DISK_PATH);
twillPreparer = ((SecureTwillPreparer) twillPreparer).withSecretDisk(ArtifactLocalizerTwillRunnable.class.getSimpleName(), new SecretDisk(secretName, secretPath));
}
if (cConf.getBoolean(Constants.Twill.Security.WORKER_MOUNT_SECRET)) {
String secretName = cConf.get(Constants.Twill.Security.WORKER_SECRET_DISK_NAME);
String secretPath = cConf.get(Constants.Twill.Security.WORKER_SECRET_DISK_PATH);
twillPreparer = ((SecureTwillPreparer) twillPreparer).withSecretDisk(PreviewRunnerTwillRunnable.class.getSimpleName(), new SecretDisk(secretName, secretPath));
}
}
activeController = twillPreparer.start(5, TimeUnit.MINUTES);
activeController.onRunning(() -> deleteDir(runDir), Threads.SAME_THREAD_EXECUTOR);
activeController.onTerminated(() -> deleteDir(runDir), Threads.SAME_THREAD_EXECUTOR);
} catch (Exception e) {
deleteDir(runDir);
throw e;
}
} catch (Exception e) {
LOG.warn("Failed to launch preview runner. It will be retried", e);
}
}
controller = activeController;
}
use of io.cdap.cdap.master.spi.twill.SecretDisk in project cdap by caskdata.
the class TaskWorkerServiceLauncher method run.
public void run() {
TwillController activeController = null;
for (TwillController controller : twillRunner.lookup(TaskWorkerTwillApplication.NAME)) {
// If detected more than one controller, terminate those extra controllers.
if (activeController != null) {
controller.terminate();
} else {
activeController = controller;
}
}
// If there is no task worker runner running, create one
if (activeController == null) {
try {
Path tmpDir = new File(cConf.get(Constants.CFG_LOCAL_DATA_DIR), cConf.get(Constants.AppFabric.TEMP_DIR)).toPath();
Files.createDirectories(tmpDir);
Path runDir = Files.createTempDirectory(tmpDir, "task.worker.launcher");
try {
// Unset the internal certificate path since certificate is stored cdap-security which
// is not exposed (i.e. mounted in k8s) to TaskWorkerService.
CConfiguration cConfCopy = CConfiguration.copy(cConf);
cConfCopy.unset(Constants.Security.SSL.INTERNAL_CERT_PATH);
Path cConfPath = runDir.resolve("cConf.xml");
try (Writer writer = Files.newBufferedWriter(cConfPath, StandardCharsets.UTF_8)) {
cConfCopy.writeXml(writer);
}
Path hConfPath = runDir.resolve("hConf.xml");
try (Writer writer = Files.newBufferedWriter(hConfPath, StandardCharsets.UTF_8)) {
hConf.writeXml(writer);
}
ResourceSpecification taskworkerResourceSpec = ResourceSpecification.Builder.with().setVirtualCores(cConf.getInt(Constants.TaskWorker.CONTAINER_CORES)).setMemory(cConf.getInt(Constants.TaskWorker.CONTAINER_MEMORY_MB), ResourceSpecification.SizeUnit.MEGA).setInstances(cConf.getInt(Constants.TaskWorker.CONTAINER_COUNT)).build();
ResourceSpecification artifactLocalizerResourceSpec = ResourceSpecification.Builder.with().setVirtualCores(cConf.getInt(Constants.ArtifactLocalizer.CONTAINER_CORES)).setMemory(cConf.getInt(Constants.ArtifactLocalizer.CONTAINER_MEMORY_MB), ResourceSpecification.SizeUnit.MEGA).setInstances(cConf.getInt(Constants.TaskWorker.CONTAINER_COUNT)).build();
LOG.info("Starting TaskWorker pool with {} instances", taskworkerResourceSpec.getInstances());
TwillPreparer twillPreparer = twillRunner.prepare(new TaskWorkerTwillApplication(cConfPath.toUri(), hConfPath.toUri(), taskworkerResourceSpec, artifactLocalizerResourceSpec));
String priorityClass = cConf.get(Constants.TaskWorker.CONTAINER_PRIORITY_CLASS_NAME);
if (priorityClass != null) {
twillPreparer = twillPreparer.setSchedulerQueue(priorityClass);
}
if (twillPreparer instanceof DependentTwillPreparer) {
twillPreparer = ((DependentTwillPreparer) twillPreparer).dependentRunnableNames(TaskWorkerTwillRunnable.class.getSimpleName(), ArtifactLocalizerTwillRunnable.class.getSimpleName());
}
if (twillPreparer instanceof StatefulTwillPreparer) {
int diskSize = cConf.getInt(Constants.TaskWorker.CONTAINER_DISK_SIZE_GB);
twillPreparer = ((StatefulTwillPreparer) twillPreparer).withStatefulRunnable(TaskWorkerTwillRunnable.class.getSimpleName(), false, new StatefulDisk(STATEFUL_DISK_NAME, diskSize, cConf.get(Constants.CFG_LOCAL_DATA_DIR)));
if (cConf.getBoolean(Constants.TaskWorker.CONTAINER_DISK_READONLY)) {
twillPreparer = ((StatefulTwillPreparer) twillPreparer).withReadonlyDisk(TaskWorkerTwillRunnable.class.getSimpleName(), STATEFUL_DISK_NAME);
}
}
if (twillPreparer instanceof SecureTwillPreparer) {
SecurityContext securityContext = createSecurityContext();
twillPreparer = ((SecureTwillPreparer) twillPreparer).withSecurityContext(TaskWorkerTwillRunnable.class.getSimpleName(), securityContext);
// Mount secret in ArtifactLocalizer sidecar which only run trusted code,
// so requests originated by ArtifactLocalizer can run with system identity when internal auth
// is enabled.
twillPreparer = ((SecureTwillPreparer) twillPreparer).withSecretDisk(ArtifactLocalizerTwillRunnable.class.getSimpleName(), new SecretDisk(cConf.get(Constants.Twill.Security.MASTER_SECRET_DISK_NAME), cConf.get(Constants.Twill.Security.MASTER_SECRET_DISK_PATH)));
if (cConf.getBoolean(Constants.Twill.Security.WORKER_MOUNT_SECRET)) {
String secretName = cConf.get(Constants.Twill.Security.WORKER_SECRET_DISK_NAME);
String secretPath = cConf.get(Constants.Twill.Security.WORKER_SECRET_DISK_PATH);
twillPreparer = ((SecureTwillPreparer) twillPreparer).withSecretDisk(TaskWorkerTwillRunnable.class.getSimpleName(), new SecretDisk(secretName, secretPath));
}
}
activeController = twillPreparer.start(5, TimeUnit.MINUTES);
activeController.onRunning(() -> deleteDir(runDir), Threads.SAME_THREAD_EXECUTOR);
activeController.onTerminated(() -> deleteDir(runDir), Threads.SAME_THREAD_EXECUTOR);
} catch (Exception e) {
deleteDir(runDir);
throw e;
}
} catch (Exception e) {
LOG.warn(String.format("Failed to launch TaskWorker pool, retry in %d", cConf.getInt(Constants.TaskWorker.POOL_CHECK_INTERVAL)), e);
}
}
this.twillController = activeController;
}
use of io.cdap.cdap.master.spi.twill.SecretDisk in project cdap by caskdata.
the class KubeTwillPreparer method addSecreteVolMountIfNeeded.
private List<V1VolumeMount> addSecreteVolMountIfNeeded(RuntimeSpecification runtimeSpec, List<V1VolumeMount> mounts) {
List<V1VolumeMount> updatedMounts = new ArrayList<>(mounts);
// Add secret disks as secret volume mounts
if (secretDiskRunnables.containsKey(runtimeSpec.getName())) {
for (SecretDisk secretDisk : secretDiskRunnables.get(runtimeSpec.getName()).getSecretDisks()) {
String secretName = secretDisk.getName();
String mountPath = secretDisk.getPath();
updatedMounts.add(new V1VolumeMount().name(secretName).mountPath(mountPath).readOnly(true));
}
}
return updatedMounts;
}
use of io.cdap.cdap.master.spi.twill.SecretDisk in project cdap by caskdata.
the class DistributedProgramRunner method run.
@Override
public final ProgramController run(final Program program, ProgramOptions oldOptions) {
validateOptions(program, oldOptions);
CConfiguration cConf = CConfiguration.copy(this.cConf);
// Reload config for log extension jar update (CDAP-15091)
cConf.reloadConfiguration();
File tempDir = DirUtils.createTempDir(new File(cConf.get(Constants.CFG_LOCAL_DATA_DIR), cConf.get(Constants.AppFabric.TEMP_DIR)).getAbsoluteFile());
try {
ProgramLaunchConfig launchConfig = new ProgramLaunchConfig();
if (clusterMode == ClusterMode.ISOLATED) {
// For isolated mode, the hadoop classes comes from the hadoop classpath in the target cluster directly
launchConfig.addExtraClasspath(Collections.singletonList("$HADOOP_CLASSPATH"));
}
setupLaunchConfig(launchConfig, program, oldOptions, cConf, hConf, tempDir);
// Add extra localize resources needed by the program runner
final Map<String, LocalizeResource> localizeResources = new HashMap<>(launchConfig.getExtraResources());
final List<String> additionalClassPaths = new ArrayList<>();
addContainerJars(cConf, localizeResources, additionalClassPaths);
addAdditionalLogAppenderJars(cConf, tempDir, localizeResources, SystemArguments.getProfileProvisioner(oldOptions.getArguments().asMap()));
prepareHBaseDDLExecutorResources(tempDir, cConf, localizeResources);
List<URI> configResources = localizeConfigs(createContainerCConf(cConf), createContainerHConf(this.hConf), tempDir, localizeResources);
// Localize the program jar
Location programJarLocation = program.getJarLocation();
final String programJarName = programJarLocation.getName();
localizeResources.put(programJarName, new LocalizeResource(programJarLocation.toURI(), false));
// Localize the app spec
localizeResources.put(APP_SPEC_FILE_NAME, new LocalizeResource(saveJsonFile(program.getApplicationSpecification(), ApplicationSpecification.class, File.createTempFile("appSpec", ".json", tempDir))));
URI logbackURI = getLogBackURI(program);
if (logbackURI != null) {
// Localize the logback xml
localizeResources.put(LOGBACK_FILE_NAME, new LocalizeResource(logbackURI, false));
}
// Update the ProgramOptions to carry program and runtime information necessary to reconstruct the program
// and runs it in the remote container
Map<String, String> extraSystemArgs = new HashMap<>(launchConfig.getExtraSystemArguments());
extraSystemArgs.put(ProgramOptionConstants.PROGRAM_JAR, programJarName);
extraSystemArgs.put(ProgramOptionConstants.HADOOP_CONF_FILE, HADOOP_CONF_FILE_NAME);
extraSystemArgs.put(ProgramOptionConstants.CDAP_CONF_FILE, CDAP_CONF_FILE_NAME);
extraSystemArgs.put(ProgramOptionConstants.APP_SPEC_FILE, APP_SPEC_FILE_NAME);
ProgramOptions options = updateProgramOptions(oldOptions, localizeResources, DirUtils.createTempDir(tempDir), extraSystemArgs);
ProgramRunId programRunId = program.getId().run(ProgramRunners.getRunId(options));
// Localize the serialized program options
localizeResources.put(PROGRAM_OPTIONS_FILE_NAME, new LocalizeResource(saveJsonFile(options, ProgramOptions.class, File.createTempFile("program.options", ".json", tempDir))));
Callable<ProgramController> callable = () -> {
ProgramTwillApplication twillApplication = new ProgramTwillApplication(programRunId, options, launchConfig.getRunnables(), launchConfig.getLaunchOrder(), localizeResources, createEventHandler(cConf, programRunId, options));
TwillPreparer twillPreparer = twillRunner.prepare(twillApplication);
// Also add the configuration files to container classpath so that the
// TwillAppLifecycleEventHandler can get it. This can be removed when TWILL-246 is fixed.
// Only ON_PREMISE mode will be using EventHandler
twillPreparer.withResources(configResources);
Map<String, String> userArgs = options.getUserArguments().asMap();
// Setup log level
twillPreparer.setLogLevels(transformLogLevels(SystemArguments.getLogLevels(userArgs)));
// Set the configuration for the twill application
Map<String, String> twillConfigs = new HashMap<>();
if (DistributedProgramRunner.this instanceof LongRunningDistributedProgramRunner) {
twillConfigs.put(Configs.Keys.YARN_ATTEMPT_FAILURES_VALIDITY_INTERVAL, cConf.get(Constants.AppFabric.YARN_ATTEMPT_FAILURES_VALIDITY_INTERVAL));
} else {
// For non long running program type, set the max attempts to 1 to avoid YARN retry.
// If the AM container dies, the program execution will be marked as failure.
// Note that this setting is only applicable to the Twill YARN application
// (e.g. workflow, Spark client, MR client, etc), but not to the actual Spark / MR job.
twillConfigs.put(Configs.Keys.YARN_MAX_APP_ATTEMPTS, Integer.toString(1));
}
// Add twill configurations coming from the runtime arguments
twillConfigs.putAll(SystemArguments.getTwillApplicationConfigs(userArgs));
twillPreparer.withConfiguration(twillConfigs);
// Setup per runnable configurations
for (Map.Entry<String, RunnableDefinition> entry : launchConfig.getRunnables().entrySet()) {
String runnable = entry.getKey();
RunnableDefinition runnableDefinition = entry.getValue();
if (runnableDefinition.getMaxRetries() != null) {
twillPreparer.withMaxRetries(runnable, runnableDefinition.getMaxRetries());
}
twillPreparer.setLogLevels(runnable, transformLogLevels(runnableDefinition.getLogLevels()));
twillPreparer.withConfiguration(runnable, runnableDefinition.getTwillRunnableConfigs());
// Add cdap-security.xml if using secrets, and set the runnable identity.
if (twillPreparer instanceof SecureTwillPreparer) {
String twillSystemIdentity = cConf.get(Constants.Twill.Security.IDENTITY_SYSTEM);
if (twillSystemIdentity != null) {
SecurityContext securityContext = new SecurityContext.Builder().withIdentity(twillSystemIdentity).build();
twillPreparer = ((SecureTwillPreparer) twillPreparer).withSecurityContext(runnable, securityContext);
}
String securityName = cConf.get(Constants.Twill.Security.MASTER_SECRET_DISK_NAME);
String securityPath = cConf.get(Constants.Twill.Security.MASTER_SECRET_DISK_PATH);
twillPreparer = ((SecureTwillPreparer) twillPreparer).withSecretDisk(runnable, new SecretDisk(securityName, securityPath));
}
}
if (options.isDebug()) {
twillPreparer.enableDebugging();
}
logProgramStart(program, options);
// Add scheduler queue name if defined
String schedulerQueueName = options.getArguments().getOption(Constants.AppFabric.APP_SCHEDULER_QUEUE);
if (schedulerQueueName != null && !schedulerQueueName.isEmpty()) {
LOG.info("Setting scheduler queue for app {} as {}", program.getId(), schedulerQueueName);
twillPreparer.setSchedulerQueue(schedulerQueueName);
}
// Set JVM options based on configuration
String jvmOpts = cConf.get(Constants.AppFabric.PROGRAM_JVM_OPTS);
if (!Strings.isNullOrEmpty(jvmOpts)) {
twillPreparer.addJVMOptions(jvmOpts);
}
if (logbackURI != null) {
twillPreparer.addJVMOptions("-Dlogback.configurationFile=" + LOGBACK_FILE_NAME);
}
addLogHandler(twillPreparer, cConf);
// Setup the environment for the container logback.xml
twillPreparer.withEnv(Collections.singletonMap("CDAP_LOG_DIR", ApplicationConstants.LOG_DIR_EXPANSION_VAR));
// Add dependencies
Set<Class<?>> extraDependencies = addExtraDependencies(cConf, new HashSet<>(launchConfig.getExtraDependencies()));
twillPreparer.withDependencies(extraDependencies);
// Add the additional classes to the classpath that comes from the container jar setting
twillPreparer.withClassPaths(additionalClassPaths);
twillPreparer.withClassPaths(launchConfig.getExtraClasspath());
twillPreparer.withEnv(launchConfig.getExtraEnv());
// Add the YARN_APPLICATION_CLASSPATH so that yarn classpath are included in the twill container.
// The Yarn app classpath goes last
List<String> yarnAppClassPath = Arrays.asList(hConf.getTrimmedStrings(YarnConfiguration.YARN_APPLICATION_CLASSPATH, YarnConfiguration.DEFAULT_YARN_APPLICATION_CLASSPATH));
twillPreparer.withApplicationClassPaths(yarnAppClassPath).withClassPaths(yarnAppClassPath);
twillPreparer.withBundlerClassAcceptor(launchConfig.getClassAcceptor()).withApplicationArguments(PROGRAM_OPTIONS_FILE_NAME).setClassLoader(MainClassLoader.class.getName());
TwillController twillController;
// Change the context classloader to the combine classloader of this ProgramRunner and
// all the classloaders of the dependencies classes so that Twill can trace classes.
ClassLoader oldClassLoader = ClassLoaders.setContextClassLoader(new CombineClassLoader(DistributedProgramRunner.this.getClass().getClassLoader(), extraDependencies.stream().map(Class::getClassLoader)::iterator));
try {
twillController = twillPreparer.start(cConf.getLong(Constants.AppFabric.PROGRAM_MAX_START_SECONDS), TimeUnit.SECONDS);
// Block on the twill controller until it is in running state or terminated (due to failure)
CountDownLatch latch = new CountDownLatch(1);
twillController.onRunning(latch::countDown, Threads.SAME_THREAD_EXECUTOR);
twillController.onTerminated(latch::countDown, Threads.SAME_THREAD_EXECUTOR);
latch.await(cConf.getLong(Constants.AppFabric.PROGRAM_MAX_START_SECONDS), TimeUnit.SECONDS);
} finally {
ClassLoaders.setContextClassLoader(oldClassLoader);
}
return createProgramController(programRunId, addCleanupListener(twillController, program, tempDir));
};
return impersonator.doAs(programRunId, callable);
} catch (Exception e) {
deleteDirectory(tempDir);
throw Throwables.propagate(e);
}
}
Aggregations