use of org.apache.beam.sdk.transforms.reflect.DoFnSignature.StateDeclaration in project beam by apache.
the class DoFnSignatures method analyzeExtraParameter.
private static Parameter analyzeExtraParameter(ErrorReporter methodErrors, FnAnalysisContext fnContext, MethodAnalysisContext methodContext, ParameterDescription param, TypeDescriptor<?> inputT, TypeDescriptor<?> outputT) {
TypeDescriptor<?> expectedProcessContextT = doFnProcessContextTypeOf(inputT, outputT);
TypeDescriptor<?> expectedStartBundleContextT = doFnStartBundleContextTypeOf(inputT, outputT);
TypeDescriptor<?> expectedFinishBundleContextT = doFnFinishBundleContextTypeOf(inputT, outputT);
TypeDescriptor<?> expectedOnTimerContextT = doFnOnTimerContextTypeOf(inputT, outputT);
TypeDescriptor<?> expectedOnWindowExpirationContextT = doFnOnWindowExpirationContextTypeOf(inputT, outputT);
TypeDescriptor<?> paramT = param.getType();
Class<?> rawType = paramT.getRawType();
ErrorReporter paramErrors = methodErrors.forParameter(param);
String fieldAccessString = getFieldAccessId(param.getAnnotations());
if (fieldAccessString != null) {
return Parameter.schemaElementParameter(paramT, fieldAccessString, param.getIndex());
} else if (hasAnnotation(DoFn.Element.class, param.getAnnotations())) {
return paramT.equals(inputT) ? Parameter.elementParameter(paramT) : Parameter.schemaElementParameter(paramT, null, param.getIndex());
} else if (hasAnnotation(DoFn.Restriction.class, param.getAnnotations())) {
return Parameter.restrictionParameter(paramT);
} else if (hasAnnotation(DoFn.WatermarkEstimatorState.class, param.getAnnotations())) {
return Parameter.watermarkEstimatorState(paramT);
} else if (hasAnnotation(DoFn.Timestamp.class, param.getAnnotations())) {
methodErrors.checkArgument(rawType.equals(Instant.class), "@Timestamp argument must have type org.joda.time.Instant.");
return Parameter.timestampParameter();
} else if (hasAnnotation(DoFn.Key.class, param.getAnnotations())) {
methodErrors.checkArgument(KV.class.equals(inputT.getRawType()), "@Key argument is expected to be use with input element of type KV.");
Type keyType = ((ParameterizedType) inputT.getType()).getActualTypeArguments()[0];
methodErrors.checkArgument(TypeDescriptor.of(keyType).equals(paramT), "@Key argument is expected to be type of %s, but found %s.", keyType, rawType);
return Parameter.keyT(paramT);
} else if (rawType.equals(TimeDomain.class)) {
return Parameter.timeDomainParameter();
} else if (hasAnnotation(DoFn.SideInput.class, param.getAnnotations())) {
String sideInputId = getSideInputId(param.getAnnotations());
paramErrors.checkArgument(sideInputId != null, "%s missing %s annotation", sideInputId, format(SideInput.class));
return Parameter.sideInputParameter(paramT, sideInputId);
} else if (rawType.equals(PaneInfo.class)) {
return Parameter.paneInfoParameter();
} else if (rawType.equals(DoFn.BundleFinalizer.class)) {
return Parameter.bundleFinalizer();
} else if (rawType.equals(DoFn.ProcessContext.class)) {
paramErrors.checkArgument(paramT.equals(expectedProcessContextT), "ProcessContext argument must have type %s", format(expectedProcessContextT));
return Parameter.processContext();
} else if (rawType.equals(DoFn.StartBundleContext.class)) {
paramErrors.checkArgument(paramT.equals(expectedStartBundleContextT), "StartBundleContext argument must have type %s", format(expectedProcessContextT));
return Parameter.startBundleContext();
} else if (rawType.equals(DoFn.FinishBundleContext.class)) {
paramErrors.checkArgument(paramT.equals(expectedFinishBundleContextT), "FinishBundleContext argument must have type %s", format(expectedProcessContextT));
return Parameter.finishBundleContext();
} else if (rawType.equals(DoFn.OnTimerContext.class)) {
paramErrors.checkArgument(paramT.equals(expectedOnTimerContextT), "OnTimerContext argument must have type %s", format(expectedOnTimerContextT));
return Parameter.onTimerContext();
} else if (rawType.equals(DoFn.OnWindowExpirationContext.class)) {
paramErrors.checkArgument(paramT.equals(expectedOnWindowExpirationContextT), "OnWindowExpirationContext argument must have type %s", format(expectedOnWindowExpirationContextT));
return Parameter.onWindowExpirationContext();
} else if (BoundedWindow.class.isAssignableFrom(rawType)) {
methodErrors.checkArgument(!methodContext.hasParameter(WindowParameter.class), "Multiple %s parameters", format(BoundedWindow.class));
return Parameter.boundedWindow((TypeDescriptor<? extends BoundedWindow>) paramT);
} else if (rawType.equals(OutputReceiver.class)) {
// It's a schema row receiver if it's an OutputReceiver<Row> _and_ the output type is not
// already Row.
boolean schemaRowReceiver = paramT.equals(outputReceiverTypeOf(TypeDescriptor.of(Row.class))) && !outputT.equals(TypeDescriptor.of(Row.class));
if (!schemaRowReceiver) {
TypeDescriptor<?> expectedReceiverT = outputReceiverTypeOf(outputT);
paramErrors.checkArgument(paramT.equals(expectedReceiverT), "OutputReceiver should be parameterized by %s", outputT);
}
return Parameter.outputReceiverParameter(schemaRowReceiver);
} else if (rawType.equals(MultiOutputReceiver.class)) {
return Parameter.taggedOutputReceiverParameter();
} else if (PipelineOptions.class.equals(rawType)) {
methodErrors.checkArgument(!methodContext.hasParameter(PipelineOptionsParameter.class), "Multiple %s parameters", format(PipelineOptions.class));
return Parameter.pipelineOptions();
} else if (RestrictionTracker.class.isAssignableFrom(rawType)) {
methodErrors.checkArgument(!methodContext.hasParameter(RestrictionTrackerParameter.class), "Multiple %s parameters", format(RestrictionTracker.class));
return Parameter.restrictionTracker(paramT);
} else if (WatermarkEstimator.class.isAssignableFrom(rawType)) {
methodErrors.checkArgument(!methodContext.hasParameter(WatermarkEstimatorParameter.class), "Multiple %s parameters", format(WatermarkEstimator.class));
return Parameter.watermarkEstimator(paramT);
} else if (rawType.equals(Timer.class)) {
// m.getParameters() is not available until Java 8
String id = getTimerId(param.getAnnotations());
paramErrors.checkArgument(id != null, "%s missing %s annotation", format(Timer.class), format(TimerId.class));
paramErrors.checkArgument(!methodContext.getTimerParameters().containsKey(id), "duplicate %s: \"%s\"", format(TimerId.class), id);
TimerDeclaration timerDecl = fnContext.getTimerDeclarations().get(id);
paramErrors.checkArgument(timerDecl != null, "reference to undeclared %s: \"%s\"", format(TimerId.class), id);
paramErrors.checkArgument(timerDecl.field().getDeclaringClass().equals(getDeclaringClass(param.getMethod())), "%s %s declared in a different class %s." + " Timers may be referenced only in the lexical scope where they are declared.", format(TimerId.class), id, timerDecl.field().getDeclaringClass().getName());
return Parameter.timerParameter(timerDecl);
} else if (hasAnnotation(DoFn.TimerId.class, param.getAnnotations())) {
boolean isValidTimerIdForTimerFamily = fnContext.getTimerFamilyDeclarations().size() > 0 && rawType.equals(String.class);
paramErrors.checkArgument(isValidTimerIdForTimerFamily, "%s not allowed here", format(DoFn.TimerId.class));
return Parameter.timerIdParameter();
} else if (rawType.equals(TimerMap.class)) {
String id = getTimerFamilyId(param.getAnnotations());
paramErrors.checkArgument(id != null, "%s missing %s annotation", format(TimerMap.class), format(DoFn.TimerFamily.class));
paramErrors.checkArgument(!methodContext.getTimerFamilyParameters().containsKey(id), "duplicate %s: \"%s\"", format(DoFn.TimerFamily.class), id);
TimerFamilyDeclaration timerDecl = fnContext.getTimerFamilyDeclarations().get(id);
paramErrors.checkArgument(timerDecl != null, "reference to undeclared %s: \"%s\"", format(DoFn.TimerFamily.class), id);
paramErrors.checkArgument(timerDecl.field().getDeclaringClass().equals(getDeclaringClass(param.getMethod())), "%s %s declared in a different class %s." + " Timers may be referenced only in the lexical scope where they are declared.", format(DoFn.TimerFamily.class), id, timerDecl.field().getDeclaringClass().getName());
return Parameter.timerFamilyParameter(timerDecl);
} else if (State.class.isAssignableFrom(rawType)) {
// m.getParameters() is not available until Java 8
String id = getStateId(param.getAnnotations());
paramErrors.checkArgument(id != null, "missing %s annotation", format(DoFn.StateId.class));
paramErrors.checkArgument(!methodContext.getStateParameters().containsKey(id), "duplicate %s: \"%s\"", format(DoFn.StateId.class), id);
// By static typing this is already a well-formed State subclass
TypeDescriptor<? extends State> stateType = (TypeDescriptor<? extends State>) param.getType();
StateDeclaration stateDecl = fnContext.getStateDeclarations().get(id);
paramErrors.checkArgument(stateDecl != null, "reference to undeclared %s: \"%s\"", format(DoFn.StateId.class), id);
paramErrors.checkArgument(stateDecl.stateType().isSubtypeOf(stateType), "data type of reference to %s %s must be a supertype of %s", format(StateId.class), id, format(stateDecl.stateType()));
paramErrors.checkArgument(stateDecl.field().getDeclaringClass().equals(getDeclaringClass(param.getMethod())), "%s %s declared in a different class %s." + " State may be referenced only in the class where it is declared.", format(StateId.class), id, stateDecl.field().getDeclaringClass().getName());
boolean alwaysFetched = getStateAlwaysFetched(param.getAnnotations());
if (alwaysFetched) {
paramErrors.checkArgument(ReadableState.class.isAssignableFrom(rawType), "@AlwaysFetched can only be used on ReadableStates. It cannot be used on %s", format(stateDecl.stateType()));
}
return Parameter.stateParameter(stateDecl, alwaysFetched);
} else {
paramErrors.throwIllegalArgument("%s is not a valid context parameter.", format(paramT));
// Unreachable
return null;
}
}
use of org.apache.beam.sdk.transforms.reflect.DoFnSignature.StateDeclaration in project beam by apache.
the class ParDoTranslation method toProto.
public static ParDoPayload toProto(ParDo.MultiOutput<?, ?> parDo, SdkComponents components) throws IOException {
DoFn<?, ?> doFn = parDo.getFn();
DoFnSignature signature = DoFnSignatures.getSignature(doFn.getClass());
Map<String, StateDeclaration> states = signature.stateDeclarations();
Map<String, TimerDeclaration> timers = signature.timerDeclarations();
List<Parameter> parameters = signature.processElement().extraParameters();
ParDoPayload.Builder builder = ParDoPayload.newBuilder();
builder.setDoFn(toProto(parDo.getFn(), parDo.getMainOutputTag()));
for (PCollectionView<?> sideInput : parDo.getSideInputs()) {
builder.putSideInputs(sideInput.getTagInternal().getId(), toProto(sideInput));
}
for (Parameter parameter : parameters) {
Optional<RunnerApi.Parameter> protoParameter = toProto(parameter);
if (protoParameter.isPresent()) {
builder.addParameters(protoParameter.get());
}
}
for (Map.Entry<String, StateDeclaration> state : states.entrySet()) {
RunnerApi.StateSpec spec = toProto(getStateSpecOrCrash(state.getValue(), doFn), components);
builder.putStateSpecs(state.getKey(), spec);
}
for (Map.Entry<String, TimerDeclaration> timer : timers.entrySet()) {
RunnerApi.TimerSpec spec = toProto(getTimerSpecOrCrash(timer.getValue(), doFn));
builder.putTimerSpecs(timer.getKey(), spec);
}
return builder.build();
}
use of org.apache.beam.sdk.transforms.reflect.DoFnSignature.StateDeclaration in project beam by apache.
the class DoFnSignatures method analyzeExtraParameter.
private static Parameter analyzeExtraParameter(ErrorReporter methodErrors, FnAnalysisContext fnContext, MethodAnalysisContext methodContext, TypeDescriptor<? extends DoFn<?, ?>> fnClass, ParameterDescription param, TypeDescriptor<?> inputT, TypeDescriptor<?> outputT) {
TypeDescriptor<?> expectedProcessContextT = doFnProcessContextTypeOf(inputT, outputT);
TypeDescriptor<?> expectedOnTimerContextT = doFnOnTimerContextTypeOf(inputT, outputT);
TypeDescriptor<?> paramT = param.getType();
Class<?> rawType = paramT.getRawType();
ErrorReporter paramErrors = methodErrors.forParameter(param);
if (rawType.equals(DoFn.ProcessContext.class)) {
paramErrors.checkArgument(paramT.equals(expectedProcessContextT), "ProcessContext argument must have type %s", formatType(expectedProcessContextT));
return Parameter.processContext();
} else if (rawType.equals(DoFn.OnTimerContext.class)) {
paramErrors.checkArgument(paramT.equals(expectedOnTimerContextT), "OnTimerContext argument must have type %s", formatType(expectedOnTimerContextT));
return Parameter.onTimerContext();
} else if (BoundedWindow.class.isAssignableFrom(rawType)) {
methodErrors.checkArgument(!methodContext.hasWindowParameter(), "Multiple %s parameters", BoundedWindow.class.getSimpleName());
return Parameter.boundedWindow((TypeDescriptor<? extends BoundedWindow>) paramT);
} else if (RestrictionTracker.class.isAssignableFrom(rawType)) {
methodErrors.checkArgument(!methodContext.hasRestrictionTrackerParameter(), "Multiple %s parameters", RestrictionTracker.class.getSimpleName());
return Parameter.restrictionTracker(paramT);
} else if (rawType.equals(Timer.class)) {
// m.getParameters() is not available until Java 8
String id = getTimerId(param.getAnnotations());
paramErrors.checkArgument(id != null, "%s missing %s annotation", Timer.class.getSimpleName(), TimerId.class.getSimpleName());
paramErrors.checkArgument(!methodContext.getTimerParameters().containsKey(id), "duplicate %s: \"%s\"", TimerId.class.getSimpleName(), id);
TimerDeclaration timerDecl = fnContext.getTimerDeclarations().get(id);
paramErrors.checkArgument(timerDecl != null, "reference to undeclared %s: \"%s\"", TimerId.class.getSimpleName(), id);
paramErrors.checkArgument(timerDecl.field().getDeclaringClass().equals(param.getMethod().getDeclaringClass()), "%s %s declared in a different class %s." + " Timers may be referenced only in the lexical scope where they are declared.", TimerId.class.getSimpleName(), id, timerDecl.field().getDeclaringClass().getName());
return Parameter.timerParameter(timerDecl);
} else if (State.class.isAssignableFrom(rawType)) {
// m.getParameters() is not available until Java 8
String id = getStateId(param.getAnnotations());
paramErrors.checkArgument(id != null, "missing %s annotation", DoFn.StateId.class.getSimpleName());
paramErrors.checkArgument(!methodContext.getStateParameters().containsKey(id), "duplicate %s: \"%s\"", DoFn.StateId.class.getSimpleName(), id);
// By static typing this is already a well-formed State subclass
TypeDescriptor<? extends State> stateType = (TypeDescriptor<? extends State>) param.getType();
StateDeclaration stateDecl = fnContext.getStateDeclarations().get(id);
paramErrors.checkArgument(stateDecl != null, "reference to undeclared %s: \"%s\"", DoFn.StateId.class.getSimpleName(), id);
paramErrors.checkArgument(stateDecl.stateType().equals(stateType), "reference to %s %s with different type %s", StateId.class.getSimpleName(), id, formatType(stateDecl.stateType()));
paramErrors.checkArgument(stateDecl.field().getDeclaringClass().equals(param.getMethod().getDeclaringClass()), "%s %s declared in a different class %s." + " State may be referenced only in the class where it is declared.", StateId.class.getSimpleName(), id, stateDecl.field().getDeclaringClass().getName());
return Parameter.stateParameter(stateDecl);
} else {
List<String> allowedParamTypes = Arrays.asList(formatType(new TypeDescriptor<BoundedWindow>() {
}), formatType(new TypeDescriptor<RestrictionTracker<?>>() {
}));
paramErrors.throwIllegalArgument("%s is not a valid context parameter. Should be one of %s", formatType(paramT), allowedParamTypes);
// Unreachable
return null;
}
}
use of org.apache.beam.sdk.transforms.reflect.DoFnSignature.StateDeclaration in project beam by apache.
the class SimpleParDoFn method processSystemTimer.
private void processSystemTimer(TimerData timer) throws Exception {
// Timer owned by this class, for cleaning up state in expired windows
if (timer.getTimerId().equals(CLEANUP_TIMER_ID)) {
checkState(timer.getDomain().equals(TimeDomain.EVENT_TIME), "%s received cleanup timer with domain not EVENT_TIME: %s", this, timer);
checkState(timer.getNamespace() instanceof WindowNamespace, "%s received cleanup timer not for a %s: %s", this, WindowNamespace.class.getSimpleName(), timer);
BoundedWindow window = ((WindowNamespace) timer.getNamespace()).getWindow();
Instant targetTime = earliestAllowableCleanupTime(window, fnInfo.getWindowingStrategy());
checkState(!targetTime.isAfter(timer.getTimestamp()), "%s received state cleanup timer for window %s " + " that is before the appropriate cleanup time %s", this, window, targetTime);
fnRunner.onWindowExpiration(window, timer.getOutputTimestamp(), this.stepContext.stateInternals().getKey());
// This is for a timer for a window that is expired, so clean it up.
for (StateDeclaration stateDecl : fnSignature.stateDeclarations().values()) {
StateTag<?> tag;
try {
tag = StateTags.tagForSpec(stateDecl.id(), (StateSpec) stateDecl.field().get(fnInfo.getDoFn()));
} catch (IllegalAccessException e) {
throw new RuntimeException(String.format("Error accessing %s for %s", StateSpec.class.getName(), fnInfo.getDoFn().getClass().getName()), e);
}
StateInternals stateInternals = userStepContext.stateInternals();
org.apache.beam.sdk.state.State state = stateInternals.state(timer.getNamespace(), tag);
state.clear();
}
}
}
Aggregations