/**
* Provides functionality to test errors
*
* @param errorType the error type. (start.non-transient, end.non-transient)
* @param externalStatus the external status to set.
* @param signalValue the signal value to set.
* @throws Exception
*/
private void _testError(String errorType, String externalStatus, String signalValue)
throws Exception {
String workflowPath = getTestCaseFileUri("workflow.xml");
Reader reader = IOUtils.getResourceAsReader("wf-ext-schema-valid.xml", -1);
Writer writer = new FileWriter(new File(getTestCaseDir(), "workflow.xml"));
IOUtils.copyCharStream(reader, writer);
final DagEngine engine = new DagEngine("u");
Configuration conf = new XConfiguration();
conf.set(OozieClient.APP_PATH, workflowPath);
conf.set(OozieClient.USER_NAME, getTestUser());
conf.set(OozieClient.LOG_TOKEN, "t");
conf.set("error", errorType);
conf.set("external-status", externalStatus);
conf.set("signal-value", signalValue);
final String jobId = engine.submitJob(conf, true);
final WorkflowStore store = Services.get().get(WorkflowStoreService.class).create();
store.beginTrx();
waitFor(
5000,
new Predicate() {
public boolean evaluate() throws Exception {
WorkflowJobBean bean = store.getWorkflow(jobId, false);
return (bean.getWorkflowInstance().getStatus() == WorkflowInstance.Status.KILLED);
}
});
assertEquals(WorkflowJob.Status.KILLED, engine.getJob(jobId).getStatus());
store.commitTrx();
store.closeTrx();
}
/**
* Provides functionality to test for set*Data calls not being made by the Action Handler.
*
* @param avoidParam set*Data function call to avoid.
* @param expActionErrorCode the expected action error code.
* @throws Exception
*/
private void _testDataNotSet(String avoidParam, String expActionErrorCode) throws Exception {
String workflowPath = getTestCaseFileUri("workflow.xml");
Reader reader = IOUtils.getResourceAsReader("wf-ext-schema-valid.xml", -1);
Writer writer = new FileWriter(new File(getTestCaseDir(), "workflow.xml"));
IOUtils.copyCharStream(reader, writer);
final DagEngine engine = new DagEngine("u");
Configuration conf = new XConfiguration();
conf.set(OozieClient.APP_PATH, workflowPath);
conf.set(OozieClient.USER_NAME, getTestUser());
conf.set(OozieClient.LOG_TOKEN, "t");
conf.set("external-status", "ok");
conf.set("signal-value", "based_on_action_status");
conf.set(avoidParam, "true");
final String jobId = engine.submitJob(conf, true);
final WorkflowStore store = Services.get().get(WorkflowStoreService.class).create();
store.beginTrx();
Thread.sleep(2000);
waitFor(
5000,
new Predicate() {
public boolean evaluate() throws Exception {
WorkflowJobBean bean = store.getWorkflow(jobId, false);
return (bean.getWorkflowInstance().getStatus() == WorkflowInstance.Status.FAILED);
}
});
store.commitTrx();
store.closeTrx();
final WorkflowStore store2 = Services.get().get(WorkflowStoreService.class).create();
store2.beginTrx();
assertEquals(
WorkflowInstance.Status.FAILED,
store2.getWorkflow(jobId, false).getWorkflowInstance().getStatus());
assertEquals(WorkflowJob.Status.FAILED, engine.getJob(jobId).getStatus());
List<WorkflowActionBean> actions = store2.getActionsForWorkflow(jobId, false);
WorkflowActionBean action = null;
for (WorkflowActionBean bean : actions) {
if (bean.getType().equals("test")) {
action = bean;
break;
}
}
assertNotNull(action);
assertEquals(expActionErrorCode, action.getErrorCode());
store2.commitTrx();
store2.closeTrx();
}
protected Void call(WorkflowStore store) throws StoreException, CommandException {
WorkflowJobBean workflow = store.getWorkflow(jobId, false);
setLogInfo(workflow);
WorkflowActionBean action = store.getAction(id, false);
setLogInfo(action);
if (action.isPending()
&& (action.getStatus() == WorkflowActionBean.Status.DONE
|| action.getStatus() == WorkflowActionBean.Status.END_RETRY
|| action.getStatus() == WorkflowActionBean.Status.END_MANUAL)) {
if (workflow.getStatus() == WorkflowJob.Status.RUNNING) {
ActionExecutor executor =
Services.get().get(ActionService.class).getExecutor(action.getType());
Configuration conf = workflow.getWorkflowInstance().getConf();
int maxRetries = conf.getInt(OozieClient.ACTION_MAX_RETRIES, executor.getMaxRetries());
long retryInterval =
conf.getLong(OozieClient.ACTION_RETRY_INTERVAL, executor.getRetryInterval());
executor.setMaxRetries(maxRetries);
executor.setRetryInterval(retryInterval);
if (executor != null) {
boolean isRetry = false;
if (action.getStatus() == WorkflowActionBean.Status.END_RETRY
|| action.getStatus() == WorkflowActionBean.Status.END_MANUAL) {
isRetry = true;
}
ActionExecutorContext context =
new ActionCommand.ActionExecutorContext(workflow, action, isRetry);
try {
XLog.getLog(getClass())
.debug(
"End, name [{0}] type [{1}] status[{2}] external status [{3}] signal value [{4}]",
action.getName(),
action.getType(),
action.getStatus(),
action.getExternalStatus(),
action.getSignalValue());
WorkflowInstance wfInstance = workflow.getWorkflowInstance();
DagELFunctions.setActionInfo(wfInstance, action);
workflow.setWorkflowInstance(wfInstance);
incrActionCounter(action.getType(), 1);
Instrumentation.Cron cron = new Instrumentation.Cron();
cron.start();
executor.end(context, action);
cron.stop();
addActionCron(action.getType(), cron);
if (!context.isEnded()) {
XLog.getLog(getClass())
.warn(
XLog.OPS,
"Action Ended, ActionExecutor [{0}] must call setEndData()",
executor.getType());
action.setErrorInfo(
END_DATA_MISSING, "Execution Ended, but End Data Missing from Action");
failJob(context);
store.updateAction(action);
store.updateWorkflow(workflow);
return null;
}
action.setRetries(0);
action.setEndTime(new Date());
store.updateAction(action);
store.updateWorkflow(workflow);
Status slaStatus = null;
switch (action.getStatus()) {
case OK:
slaStatus = Status.SUCCEEDED;
break;
case KILLED:
slaStatus = Status.KILLED;
break;
case FAILED:
slaStatus = Status.FAILED;
break;
case ERROR:
XLog.getLog(getClass()).info("ERROR is considered as FAILED for SLA");
slaStatus = Status.KILLED;
break;
default: // TODO: What will happen for other Action
// status
slaStatus = Status.FAILED;
break;
}
SLADbOperations.writeStausEvent(
action.getSlaXml(), action.getId(), store, slaStatus, SlaAppType.WORKFLOW_ACTION);
queueCallable(new NotificationCommand(workflow, action));
XLog.getLog(getClass())
.debug(
"Queuing commands for action "
+ id
+ " status "
+ action.getStatus()
+ ", Set pending="
+ action.getPending());
queueCallable(new SignalCommand(workflow.getId(), id));
} catch (ActionExecutorException ex) {
XLog.getLog(getClass())
.warn(
"Error ending action [{0}]. ErrorType [{1}], ErrorCode [{2}], Message [{3}]",
action.getName(), ex.getErrorType(), ex.getErrorCode(), ex.getMessage());
action.setErrorInfo(ex.getErrorCode(), ex.getMessage());
action.setEndTime(null);
switch (ex.getErrorType()) {
case TRANSIENT:
if (!handleTransient(context, executor, WorkflowAction.Status.END_RETRY)) {
handleNonTransient(context, executor, WorkflowAction.Status.END_MANUAL);
action.setPendingAge(new Date());
action.setRetries(0);
}
action.setEndTime(null);
break;
case NON_TRANSIENT:
handleNonTransient(context, executor, WorkflowAction.Status.END_MANUAL);
action.setEndTime(null);
break;
case ERROR:
handleError(context, executor, COULD_NOT_END, false, WorkflowAction.Status.ERROR);
queueCallable(new SignalCommand(workflow.getId(), id));
break;
case FAILED:
failJob(context);
break;
}
store.updateAction(action);
store.updateWorkflow(workflow);
}
} else {
throw new CommandException(ErrorCode.E0802, action.getType());
}
} else {
XLog.getLog(getClass())
.warn(
"Job state is not {0}. Skipping ActionEnd Execution",
WorkflowJob.Status.RUNNING.toString());
}
} else {
XLog.getLog(getClass())
.debug(
"Action pending={0}, status={1}. Skipping ActionEnd Execution",
action.getPending(), action.getStatusStr());
}
return null;
}
public void testFsFailover() throws Exception {
Reader reader = IOUtils.getResourceAsReader("failover-fs-wf.xml", -1);
Writer writer = new FileWriter(getTestCaseDir() + "/workflow.xml");
IOUtils.copyCharStream(reader, writer);
final OozieClient wfClient = LocalOozie.getClient();
Properties conf = wfClient.createConfiguration();
conf.setProperty(OozieClient.APP_PATH, getTestCaseDir());
conf.setProperty(OozieClient.USER_NAME, getTestUser());
conf.setProperty(OozieClient.GROUP_NAME, getTestGroup());
injectKerberosInfo(conf);
final Path source = new Path(getFsTestCaseDir(), "fsfailover-source");
getFileSystem().mkdirs(source);
final Path target = new Path(getFsTestCaseDir().toString(), "fsfailover-target");
conf.setProperty("source", source.toString());
conf.setProperty("target", target.toUri().getPath());
setSystemProperty(FaultInjection.FAULT_INJECTION, "true");
setSystemProperty(SkipCommitFaultInjection.ACTION_FAILOVER_FAULT_INJECTION, "true");
final String jobId1 = wfClient.submit(conf);
wfClient.start(jobId1);
waitFor(
10 * 1000,
new Predicate() {
public boolean evaluate() throws Exception {
return getFileSystem().exists(target);
}
});
assertTrue(getFileSystem().exists(target));
waitFor(
10 * 1000,
new Predicate() {
public boolean evaluate() throws Exception {
return FaultInjection.isActive("org.apache.oozie.command.SkipCommitFaultInjection");
}
});
assertFalse(FaultInjection.isActive("org.apache.oozie.command.SkipCommitFaultInjection"));
assertEquals(WorkflowJob.Status.RUNNING, wfClient.getJobInfo(jobId1).getStatus());
WorkflowStore store = Services.get().get(WorkflowStoreService.class).create();
List<WorkflowActionBean> actions = store.getActionsForWorkflow(jobId1, false);
WorkflowActionBean action = actions.get(0);
assertEquals(WorkflowAction.Status.PREP, action.getStatus());
setSystemProperty(FaultInjection.FAULT_INJECTION, "false");
setSystemProperty(SkipCommitFaultInjection.ACTION_FAILOVER_FAULT_INJECTION, "false");
ActionStartCommand actionStartCommand =
new ActionStartCommand(action.getId(), action.getType());
actionStartCommand.call();
store = Services.get().get(WorkflowStoreService.class).create();
actions = store.getActionsForWorkflow(jobId1, false);
action = actions.get(0);
assertEquals(WorkflowAction.Status.DONE, action.getStatus());
waitFor(
5 * 1000,
new Predicate() {
public boolean evaluate() throws Exception {
return wfClient.getJobInfo(jobId1).getStatus() == WorkflowJob.Status.SUCCEEDED;
}
});
assertEquals(WorkflowJob.Status.SUCCEEDED, wfClient.getJobInfo(jobId1).getStatus());
final String jobId2 = wfClient.submit(conf);
wfClient.start(jobId2);
waitFor(
10 * 1000,
new Predicate() {
public boolean evaluate() throws Exception {
return wfClient.getJobInfo(jobId2).getStatus() == WorkflowJob.Status.KILLED;
}
});
assertEquals(WorkflowJob.Status.KILLED, wfClient.getJobInfo(jobId2).getStatus());
}
/**
* Provides functionality to test transient failures.
*
* @param errorType the error type. (start.transient, end.transient)
* @param expStatus1 expected status after the first step (START_RETRY, END_RETRY)
* @param expStatus2 expected status after the second step (START_MANUAL, END_MANUAL)
* @param expErrorMsg the expected error message.
* @throws Exception
*/
private void _testTransient(
String errorType,
WorkflowActionBean.Status expStatus1,
final WorkflowActionBean.Status expStatus2,
String expErrorMsg)
throws Exception {
String workflowPath = getTestCaseFileUri("workflow.xml");
Reader reader = IOUtils.getResourceAsReader("wf-ext-schema-valid.xml", -1);
Writer writer = new FileWriter(new File(getTestCaseDir(), "workflow.xml"));
IOUtils.copyCharStream(reader, writer);
final int maxRetries = 2;
final int retryInterval = 10;
final DagEngine engine = new DagEngine("u");
Configuration conf = new XConfiguration();
conf.set(OozieClient.APP_PATH, workflowPath);
conf.set(OozieClient.USER_NAME, getTestUser());
conf.set(OozieClient.LOG_TOKEN, "t");
conf.set("signal-value", "OK");
conf.set("external-status", "ok");
conf.set("error", errorType);
conf.setInt(OozieClient.ACTION_MAX_RETRIES, maxRetries);
conf.setInt(OozieClient.ACTION_RETRY_INTERVAL, retryInterval);
final String jobId = engine.submitJob(conf, true);
int retryCount = 1;
WorkflowActionBean.Status expectedStatus = expStatus1;
int expectedRetryCount = 2;
Thread.sleep(20000);
String aId = null;
final WorkflowStore store = Services.get().get(WorkflowStoreService.class).create();
store.beginTrx();
while (retryCount <= maxRetries) {
List<WorkflowActionBean> actions = store.getActionsForWorkflow(jobId, false);
WorkflowActionBean action = null;
for (WorkflowActionBean bean : actions) {
if (bean.getType().equals("test")) {
action = bean;
break;
}
}
assertNotNull(action);
aId = action.getId();
assertEquals(expectedStatus, action.getStatus());
assertEquals(expectedRetryCount, action.getRetries());
assertEquals("TEST_ERROR", action.getErrorCode());
assertEquals(expErrorMsg, action.getErrorMessage());
if (action.getRetries() == maxRetries) {
expectedRetryCount = 0;
expectedStatus = expStatus2;
break;
} else {
expectedRetryCount++;
}
Thread.sleep(retryInterval * 1000);
retryCount++;
}
store.commitTrx();
store.closeTrx();
Thread.sleep(5000);
final String actionId = aId;
waitFor(
5000,
new Predicate() {
public boolean evaluate() throws Exception {
return (engine.getWorkflowAction(actionId).getStatus() == expStatus2);
}
});
final WorkflowStore store2 = Services.get().get(WorkflowStoreService.class).create();
store2.beginTrx();
WorkflowActionBean action = engine.getWorkflowAction(actionId);
assertEquals("TEST_ERROR", action.getErrorCode());
assertEquals(expErrorMsg, action.getErrorMessage());
assertEquals(expStatus2, action.getStatus());
assertTrue(action.isPending() == false);
assertEquals(WorkflowJob.Status.SUSPENDED, engine.getJob(jobId).getStatus());
store2.commitTrx();
store2.closeTrx();
}
/**
* Provides functionality to test non transient failures.
*
* @param errorType the error type. (start.non-transient, end.non-transient)
* @param expStatus1 expected status. (START_MANUAL, END_MANUAL)
* @param expErrorMsg expected error message.
* @throws Exception
*/
private void _testNonTransient(
String errorType, WorkflowActionBean.Status expStatus1, String expErrorMsg) throws Exception {
String workflowPath = getTestCaseFileUri("workflow.xml");
Reader reader = IOUtils.getResourceAsReader("wf-ext-schema-valid.xml", -1);
Writer writer = new FileWriter(new File(getTestCaseDir(), "workflow.xml"));
IOUtils.copyCharStream(reader, writer);
final DagEngine engine = new DagEngine("u");
Configuration conf = new XConfiguration();
conf.set(OozieClient.APP_PATH, workflowPath);
conf.set(OozieClient.USER_NAME, getTestUser());
conf.set(OozieClient.LOG_TOKEN, "t");
conf.set("signal-value", "OK");
conf.set("external-status", "ok");
conf.set("error", errorType);
final String jobId = engine.submitJob(conf, true);
waitFor(
5000,
new Predicate() {
public boolean evaluate() throws Exception {
return (engine.getJob(jobId).getStatus() == WorkflowJob.Status.SUSPENDED);
}
});
final WorkflowStore store = Services.get().get(WorkflowStoreService.class).create();
store.beginTrx();
List<WorkflowActionBean> actions = store.getActionsForWorkflow(jobId, true);
int n = actions.size();
WorkflowActionBean action = actions.get(n - 1);
assertEquals("TEST_ERROR", action.getErrorCode());
assertEquals(expErrorMsg, action.getErrorMessage());
assertEquals(expStatus1, action.getStatus());
assertTrue(action.isPending() == false);
assertTrue(engine.getJob(jobId).getStatus() == WorkflowJob.Status.SUSPENDED);
String actionConf = action.getConf();
String fixedActionConf = actionConf.replaceAll(errorType, "none");
action.setConf(fixedActionConf);
store.updateAction(action);
store.commitTrx();
store.closeTrx();
engine.resume(jobId);
waitFor(
5000,
new Predicate() {
public boolean evaluate() throws Exception {
return (engine.getJob(jobId).getStatus() == WorkflowJob.Status.SUCCEEDED);
}
});
assertEquals(WorkflowJob.Status.SUCCEEDED, engine.getJob(jobId).getStatus());
final WorkflowStore store2 = Services.get().get(WorkflowStoreService.class).create();
store2.beginTrx();
actions = store2.getActionsForWorkflow(jobId, false);
action = actions.get(0);
assertEquals(null, action.getErrorCode());
assertEquals(null, action.getErrorMessage());
assertEquals(WorkflowActionBean.Status.OK, action.getStatus());
store2.commitTrx();
store2.closeTrx();
}