private void testRestartApp(StorageAgent agent, String appPath1) throws Exception {
FileUtils.deleteDirectory(new File(testMeta.dir)); // clean any state from previous run
String appId1 = "app1";
String appId2 = "app2";
String appPath2 = testMeta.dir + "/" + appId2;
LogicalPlan dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_ID, appId1);
dag.setAttribute(LogicalPlan.APPLICATION_PATH, appPath1);
dag.setAttribute(LogicalPlan.APPLICATION_ATTEMPT_ID, 1);
dag.setAttribute(OperatorContext.STORAGE_AGENT, agent);
dag.addOperator("o1", StatsListeningOperator.class);
FSRecoveryHandler recoveryHandler =
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false));
StreamingContainerManager.getInstance(recoveryHandler, dag, false);
// test restore initial snapshot + log
dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_PATH, appPath1);
StreamingContainerManager scm =
StreamingContainerManager.getInstance(
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false)), dag, false);
PhysicalPlan plan = scm.getPhysicalPlan();
dag = plan.getLogicalPlan(); // original plan
Assert.assertNotNull("operator", dag.getOperatorMeta("o1"));
PTOperator o1p1 = plan.getOperators(dag.getOperatorMeta("o1")).get(0);
long[] ids =
new FSStorageAgent(appPath1 + "/" + LogicalPlan.SUBDIR_CHECKPOINTS, new Configuration())
.getWindowIds(o1p1.getId());
Assert.assertArrayEquals(new long[] {o1p1.getRecoveryCheckpoint().getWindowId()}, ids);
Assert.assertNull(o1p1.getContainer().getExternalId());
// trigger journal write
o1p1.getContainer().setExternalId("cid1");
scm.writeJournal(o1p1.getContainer().getSetContainerState());
dag = new LogicalPlan();
dag.setAttribute(LogicalPlan.APPLICATION_PATH, appPath2);
dag.setAttribute(LogicalPlan.APPLICATION_ID, appId2);
StramClient sc = new StramClient(new Configuration(), dag);
try {
sc.start();
sc.copyInitialState(new Path(appPath1));
} finally {
sc.stop();
}
scm =
StreamingContainerManager.getInstance(
new FSRecoveryHandler(dag.assertAppPath(), new Configuration(false)), dag, false);
plan = scm.getPhysicalPlan();
dag = plan.getLogicalPlan();
assertEquals("modified appId", appId2, dag.getValue(LogicalPlan.APPLICATION_ID));
assertEquals("modified appPath", appPath2, dag.getValue(LogicalPlan.APPLICATION_PATH));
Assert.assertNotNull("operator", dag.getOperatorMeta("o1"));
o1p1 = plan.getOperators(dag.getOperatorMeta("o1")).get(0);
assertEquals("journal copied", "cid1", o1p1.getContainer().getExternalId());
ids =
new FSStorageAgent(appPath2 + "/" + LogicalPlan.SUBDIR_CHECKPOINTS, new Configuration())
.getWindowIds(o1p1.getId());
Assert.assertArrayEquals(
"checkpoints copied", new long[] {o1p1.getRecoveryCheckpoint().getWindowId()}, ids);
}
/**
* Create deploy info for operator.
*
* <p>
*
* @return {@link com.datatorrent.stram.api.OperatorDeployInfo}
*/
private OperatorDeployInfo createOperatorDeployInfo(PTOperator oper) {
OperatorDeployInfo ndi;
if (oper.isUnifier()) {
UnifierDeployInfo udi = new UnifierDeployInfo(); /* the constructor auto sets the type */
try {
udi.operatorAttributes = oper.getUnifiedOperatorMeta().getAttributes().clone();
} catch (CloneNotSupportedException ex) {
throw new RuntimeException("Cannot clone unifier attributes", ex);
}
ndi = udi;
} else {
ndi = new OperatorDeployInfo();
Operator operator = oper.getOperatorMeta().getOperator();
if (operator instanceof InputOperator) {
ndi.type = OperatorType.INPUT;
if (!oper.getInputs().isEmpty()) {
// If there are no input ports then it has to be an input operator. But if there are input
// ports then
// we check if any input port is connected which would make it a Generic operator.
for (PTOperator.PTInput ptInput : oper.getInputs()) {
if (ptInput.logicalStream != null && ptInput.logicalStream.getSource() != null) {
ndi.type = OperatorType.GENERIC;
break;
}
}
}
} else {
ndi.type = OperatorType.GENERIC;
}
}
Checkpoint checkpoint = oper.getRecoveryCheckpoint();
ProcessingMode pm = oper.getOperatorMeta().getValue(OperatorContext.PROCESSING_MODE);
if (pm == ProcessingMode.AT_MOST_ONCE || pm == ProcessingMode.EXACTLY_ONCE) {
// TODO: following should be handled in the container at deploy time
// for exactly once container should also purge previous checkpoint
// whenever new checkpoint is written.
StorageAgent agent =
oper.getOperatorMeta().getAttributes().get(OperatorContext.STORAGE_AGENT);
if (agent == null) {
agent = initCtx.getValue(OperatorContext.STORAGE_AGENT);
}
// pick checkpoint most recently written
try {
long[] windowIds = agent.getWindowIds(oper.getId());
long checkpointId = Stateless.WINDOW_ID;
for (long windowId : windowIds) {
if (windowId > checkpointId) {
checkpointId = windowId;
}
}
if (checkpoint == null || checkpoint.windowId != checkpointId) {
checkpoint = new Checkpoint(checkpointId, 0, 0);
}
} catch (Exception e) {
throw new RuntimeException("Failed to determine checkpoint window id " + oper, e);
}
}
LOG.debug("{} recovery checkpoint {}", oper, checkpoint);
ndi.checkpoint = checkpoint;
ndi.name = oper.getOperatorMeta().getName();
ndi.id = oper.getId();
try {
// clone map before modifying it
ndi.contextAttributes = oper.getOperatorMeta().getAttributes().clone();
} catch (CloneNotSupportedException ex) {
throw new RuntimeException("Cannot clone operator attributes", ex);
}
if (oper.isOperatorStateLess()) {
ndi.contextAttributes.put(OperatorContext.STATELESS, true);
}
return ndi;
}
