Stages Documentation

Stages are the building blocks of Morpheus pipelines. Below is a list of the most commonly used stages. For a full list of stages, refer to the stages API morpheus.stages. In addition to this there are several custom stages contained in the Examples and Developer Guides.

• Doca

• General

• Inference

• Input

• LLM

• Output

• Post-process

• Pre-process

• Doca Source Stage DocaSourceStage A source stage used to receive raw packet data in GPU memory from a ConnectX NIC using DOCA GPUNetIO function within a CUDA kernel to actually receive and process Ethernet network packets. Receive packets information is passed to next pipeline stage in the form of RawPacketMessage. This stage is not compiled by default refer to the Doca Example for details on building this stage.

• Doca Convert Stage DocaConvertStage Convert the RawPacketMessage format received by the DOCA Source Stage into a more complex message format MetaMessage. Packets’ info never leave the GPU memory. This stage is not compiled by default refer to the Doca Example for details on building this stage.

• Linear Modules Stage LinearModulesStage Loads an existing, registered, module and wraps it as a Morpheus stage. Refer to Morpheus Modules for details on modules.

• Monitor Stage MonitorStage Display throughput numbers at a specific point in the pipeline.

• Multi Port Module Stage MultiPortModulesStage Loads an existing, registered, multi-port module and wraps it as a multi-port Morpheus stage. Refer to Morpheus Modules for details on modules.

• Trigger Stage TriggerStage Buffer data until the previous stage has completed, useful for testing performance of one stage at a time.

• Auto Encoder Inference Stage AutoEncoderInferenceStage PyTorch inference stage used for Auto Encoder pipeline mode.

• PyTorch Inference Stage PyTorchInferenceStage PyTorch inference stage used for most pipeline modes with the exception of Auto Encoder.

• Triton Inference Stage TritonInferenceStage Inference stage which utilizes a Triton Inference Server.

• AppShield Source Stage AppShieldSourceStage Load Appshield messages from one or more plugins into a dataframe.

• Azure Source Stage AzureSourceStage Load Azure Active Directory messages.

• Cloud Trail Source Stage CloudTrailSourceStage Load messages from a Cloudtrail directory.

• Control Message File Source Stage ControlMessageFileSourceStage Recieves control messages from different sources specified by a list of (fsspec)[https://filesystem-spec.readthedocs.io/en/latest/api.html?highlight=open_files#fsspec.open_files] strings.

• Control Message Kafka Source Stage ControlMessageKafkaSourceStage Load control messages from a Kafka cluster.

• Databricks Delta Lake Source Stage DataBricksDeltaLakeSourceStage Source stage used to load messages from a DeltaLake table.

• Duo Source Stage DuoSourceStage Load Duo Authentication messages.

• File Source Stage FileSourceStage Load messages from a file.

• HTTP Client Source Stage HttpClientSourceStage Poll a remote HTTP server for incoming data.

• HTTP Server Source Stage HttpServerSourceStage Start an HTTP server and listens for incoming requests on a specified endpoint.

• In Memory Source Stage InMemorySourceStage Input source that emits a pre-defined list of dataframes.

• Kafka Source Stage KafkaSourceStage Load messages from a Kafka cluster.

• RSS Source Stage RSSSourceStage Load RSS feed items into a pandas DataFrame.

• LLM Engine Stage LLMEngineStage Execute an LLM engine within a Morpheus pipeline.

• HTTP Client Sink Stage HttpClientSinkStage Write all messages to an HTTP endpoint.

• HTTP Server Sink Stage HttpServerSinkStage Start an HTTP server and listens for incoming requests on a specified endpoint.

• In Memory Sink Stage InMemorySinkStage Collect incoming messages into a list that can be accessed after the pipeline is complete.

• Databricks Delta Lake Sink Stage DataBricksDeltaLakeSinkStage Write messages to a DeltaLake table.

• Write To Elastic Search Stage WriteToElasticsearchStage Write the messages as documents to Elasticsearch.

• Write To File Stage WriteToFileStage Write all messages to a file.

• Write To Kafka Stage WriteToKafkaStage Write all messages to a Kafka cluster.

• Write To Vector DB Stage WriteToVectorDBStage Write all messages to a Vector Database.

• Add Classifications Stage AddClassificationsStage Add detected classifications to each message.

• Add Scores Stage AddScoresStage Add probability scores to each message.

• Filter Detections Stage FilterDetectionsStage Filter message by a classification threshold.

• Generate Viz Frames Stage GenerateVizFramesStage Write out visualization DataFrames.

• MLflow Drift Stage MLFlowDriftStage Report model drift statistics to MLflow.

• Serialize Stage SerializeStage Include & exclude columns from messages.

• Timeseries Stage TimeSeriesStage Perform time series anomaly detection and add prediction.

• Deserialize Stage DeserializeStage Partition messages based on the pipeline config’s pipeline_batch_size parameter.

• Drop Null Stage DropNullStage Drop null data entries from a DataFrame.

• Preprocess AE Stage PreprocessAEStage Prepare Autoencoder input DataFrames for inference.

• Preprocess FIL Stage PreprocessFILStage Prepare FIL input DataFrames for inference.

• Preprocess NLP Stage PreprocessNLPStage Prepare NLP input DataFrames for inference.

• Train AE Stage TrainAEStage Train an Autoencoder model on incoming data.