示例#1
0
        public ITargetBlock <string> Build(DataflowBuilderConfiguration config)
        {
            var sinkCompletions = new List <Task>();

            var fileParser =
                new TransformManyBlock <string, UnparsedRecord>(fileLocation => config.FileParser.Parse(fileLocation));

            var invalidParsedRecordExtractor = new TransformBlock <Validity <ParsedRecord>, InvalidRecord>(validity =>
                                                                                                           new InvalidRecord(validity.Data.Id, validity.InvalidReason));

            ITargetBlock <InvalidRecord> invalidRecordSink =
                DataflowExtensions.CreateBatchedActionBlock <InvalidRecord>(BatchSize, config.InvalidRecordSink.Persist);

            invalidParsedRecordExtractor.LinkTo(invalidRecordSink);
            sinkCompletions.Add(invalidRecordSink.Completion);

            foreach (RecordPipeline recordPipeline in config.RecordPipelines)
            {
                ITargetBlock <Validity <ParsedRecord> > sink = DataflowExtensions.ToTargetPipeline(
                    new TransformBlock <Validity <ParsedRecord>, ParsedRecord>(v => v.Data),
                    DataflowExtensions.CreateBatchedActionBlock <ParsedRecord>(BatchSize,
                                                                               recordPipeline.ValidRecordSink.Persist));

                fileParser.LinkTo(recordPipeline.Pipeline, DataflowExtensions.PropagateCompletion,
                                  r => recordPipeline.RoutingPredicate(r));

                recordPipeline.Pipeline.LinkTo(sink, DataflowExtensions.PropagateCompletion, validity => validity.IsValid);
                recordPipeline.Pipeline.LinkTo(invalidParsedRecordExtractor, validity => !validity.IsValid);

                sinkCompletions.Add(sink.Completion);
            }

            // We need to add unroutable last since it will greedily accept all messages.  This won't work if any of
            // the record pipeline have a limited capacity since they could overflow into this link
            var unroutable = new TransformBlock <UnparsedRecord, InvalidRecord>(unparsed =>
                                                                                new InvalidRecord(unparsed.Id, "Record did not match any of the pipelines' predicates."));

            fileParser.LinkTo(unroutable, DataflowExtensions.PropagateCompletion);

            invalidParsedRecordExtractor.CompleteWhenAllOrAnyFaulted(config.RecordPipelines.Select(p => p.Pipeline.Completion).ToList());
            invalidRecordSink.CompleteWhenAllOrAnyFaulted(new List <Task> {
                invalidParsedRecordExtractor.Completion, unroutable.Completion
            });

            return(DataflowExtensions.EncapsulateTarget(fileParser, Task.WhenAll(sinkCompletions)));
        }