building_a_pipeline.md

Building a Pipeline

Prerequisites The following examples assume that the example dataset has been fetched. From the root of the Morpheus repo, run:

./scripts/fetch_data.py fetch examples

To build a pipeline via the CLI, users must first specify the type of pipeline, a source object, followed by a sequential list of stages. For each stage, options can be specified to configure the particular stage. Since stages are listed sequentially the output of one stage becomes the input to the next. Unless heavily customized, pipelines typically start with morpheus run followed by the pipeline mode such as pipeline-nlp or pipeline-fil. For example, to run the NLP pipeline, use:

morpheus run pipeline-nlp ...

While each stage has configuration options, there are options that apply to the pipeline as a whole as well. Check morpheus run --help, morpheus run pipeline-nlp --help and morpheus run pipeline-fil --help for these global Pipeline options.

Source Stages

All pipelines configured with the CLI need to start with a source object. Two commonly used source stages included with Morpheus are:

• from-file
  • Reads from a local file into the Pipeline
  • Supports CSV, JSON, JSON lines and Parquet formats
  • All lines are read at the start and queued into the pipeline at one time. Useful for performance testing.
  • Refer to morpheus.stages.input.file_source_stage.FileSourceStage for more information
• from-kafka
  • Pulls messages from a Kafka cluster into the Pipeline
  • Kafka cluster can be running on the localhost or remotely
  • Refer to morpheus.stages.input.kafka_source_stage.KafkaSourceStage for more information

Stages

From this point on, any number of stages can be sequentially added to the command line from start to finish. For example, we could build a trivial pipeline that reads from a file, deserializes messages, serializes them, and then writes to a file use the following:

morpheus --log_level=DEBUG run pipeline-other --viz_file=.tmp/simple_identity.png \
  from-file --filename=examples/data/pcap_dump.jsonlines \
  deserialize \
  serialize \
  to-file --overwrite --filename .tmp/temp_out.json

The output should be similar to:

Configuring Pipeline via CLI
Starting pipeline via CLI... Ctrl+C to Quit
Config:
{
  "_model_max_batch_size": 8,
  "_pipeline_batch_size": 256,
  "ae": null,
  "class_labels": [],
  "debug": false,
  "edge_buffer_size": 128,
  "feature_length": 1,
  "fil": {
    "feature_columns": null
  },
  "log_config_file": null,
  "log_level": 10,
  "mode": "OTHER",
  "num_threads": 64,
  "plugins": []
}
CPP Enabled: True
====Starting Pipeline====
====Pipeline Started====
====Building Segment: linear_segment_0====
Added source: <from-file-0; FileSourceStage(filename=examples/data/pcap_dump.jsonlines, iterative=False, file_type=FileTypes.Auto, repeat=1, filter_null=True, filter_null_columns=(), parser_kwargs={})>
  └─> morpheus.MessageMeta
Added stage: <deserialize-1; DeserializeStage(ensure_sliceable_index=True, task_type=None, task_payload=None)>
  └─ morpheus.MessageMeta -> morpheus.ControlMessage
Added stage: <serialize-2; SerializeStage(include=(), exclude=(), fixed_columns=True)>
  └─ morpheus.ControlMessage -> morpheus.MessageMeta
Added stage: <to-file-3; WriteToFileStage(filename=.tmp/temp_out.json, overwrite=True, file_type=FileTypes.Auto, include_index_col=True, flush=False)>
  └─ morpheus.MessageMeta -> morpheus.MessageMeta
====Building Segment Complete!====
====Pipeline Complete====

Pipeline Build Checks

After the ====Building Pipeline==== message, if logging is INFO or greater, the CLI prints a list of all stages and the type transformations of each stage. To be a valid Pipeline, the output type of one stage must match the input type of the next. Many stages are flexible and determine their type at runtime but some stages require a specific input type. If your Pipeline is configured incorrectly, Morpheus reports the error. For example, if we run the same command as above but forget the serialize stage:

morpheus --log_level=DEBUG run pipeline-other \
  from-file --filename=examples/data/pcap_dump.jsonlines \
  deserialize \
  to-file --overwrite --filename .tmp/temp_out.json

Then the following error displays:

RuntimeError: The to-file stage cannot handle input of <class 'morpheus.messages.control_message.ControlMessage'>. Accepted input types: (<class 'morpheus.messages.message_meta.MessageMeta'>,)

This indicates that the to-file stage cannot accept the input type of morpheus.messages.ControlMessage. This is because the to-file stage has no idea how to write that class to a file; it only knows how to write instances of morpheus.messages.message_meta.MessageMeta. To ensure you have a valid pipeline, examine the Accepted input types: (<class 'morpheus.messages.message_meta.MessageMeta'>,) portion of the message. This indicates you need a stage that converts from the output type of the deserialize stage, ControlMessage, to MessageMeta, which is exactly what the serialize stage does.

Kafka Source Example

The above example essentially just copies a file. However, it is an important to note that most Morpheus pipelines are similar in structure, in that they begin with a source stage (from-file) followed by a deserialize stage, end with a serialize stage followed by a sink stage (to-file), with the actual training or inference logic occurring in between.

We could also easily swap out the source or sink stages in the above example without any impact to the pipeline as a whole. For example, to read from a Kafka topic, simply replace the from-file stage with from-kafka:

Note: This assumes a Kafka broker running on the localhost listening to port 9092. For testing Morpheus with Kafka follow steps 1-8 in Quick Launch Kafka Cluster section of contributing.md, creating a topic named test_pcap then replace port 9092 with the port your Kafka instance is listening on.

morpheus --log_level=DEBUG run pipeline-other \
  from-kafka --input_topic test_pcap --bootstrap_servers localhost:9092 \
  deserialize \
  serialize \
  to-file --filename .tmp/temp_out.json

Available Stages

For a complete list of available stages for a particular pipeline mode, use the CLI help commands. First morpheus run --help can be used to list the available pipeline modes. Then morpheus run <mode> --help can be used to list the available stages for that mode. For example, to list the available stages for the pipeline-nlp mode:

morpheus run pipeline-nlp --help

Basic Usage Examples

Remove Fields from JSON Objects

This example only copies the fields timestamp, src_ip and dest_ip from examples/data/pcap_dump.jsonlines to out.jsonlines.

morpheus run pipeline-other --viz_file=.tmp/remove_fields_from_json_objects.png \
   from-file --filename examples/data/pcap_dump.jsonlines \
   deserialize \
   serialize --include 'timestamp' --include 'src_ip' --include 'dest_ip' \
   to-file --overwrite --filename out.jsonlines

Monitor Throughput

This example reports the throughput on the command line.

morpheus --log_level=INFO run pipeline-other --viz_file=.tmp/monitor_throughput.png  \
   from-file --filename examples/data/pcap_dump.jsonlines \
   deserialize \
   monitor --description "Lines Throughput" --smoothing 0.1 --unit "lines" \
   serialize \
   to-file --overwrite --filename out.jsonlines

Output:

Configuring Pipeline via CLI
Starting pipeline via CLI... Ctrl+C to Quit
Lines Throughput[Complete]: 93085 lines [00:03, 29446.18 lines/s]
Pipeline visualization saved to .tmp/monitor_throughput.png

Note: By default the monitor stage will omit itself from the pipeline if the log_level is set to WARNING or below.

Multi-Monitor Throughput

This example reports the throughput for each stage independently.

morpheus --log_level=INFO run pipeline-nlp --viz_file=.tmp/multi_monitor_throughput.png  \
   from-file --filename examples/data/pcap_dump.jsonlines \
   monitor --description "From File Throughput" \
   deserialize \
   monitor --description "Deserialize Throughput" \
   serialize \
   monitor --description "Serialize Throughput" \
   to-file --filename out.jsonlines --overwrite

Output:

Configuring Pipeline via CLI
Starting pipeline via CLI... Ctrl+C to Quit
From File Throughput[Complete]: 93085 messages [00:00, 168118.35 messages/s]
Deserialize Throughput[Complete]: 93085 messages [00:04, 22584.37 messages/s]
Serialize Throughput[Complete]: 93085 messages [00:06, 14095.36 messages/s]
Pipeline visualization saved to .tmp/multi_monitor_throughput.png

NLP Kitchen Sink

This example shows an NLP Pipeline which uses several stages available in Morpheus. This example utilizes the Triton Inference Server to perform inference, and writes the output to a Kafka topic named inference_output. Both of which need to be started prior to launching Morpheus.

Launching Triton

Run the following to launch Triton and load the sid-minibert model:

docker run --rm -ti --gpus=all -p8000:8000 -p8001:8001 -p8002:8002 nvcr.io/nvidia/morpheus/morpheus-tritonserver-models:25.02 tritonserver --model-repository=/models/triton-model-repo --exit-on-error=false --model-control-mode=explicit --load-model sid-minibert-onnx

Launching Kafka

Follow steps 1-8 in Quick Launch Kafka Cluster section of contributing.md, creating a topic named inference_output then replace port 9092 with the port your Kafka instance is listening on.

morpheus  --log_level=INFO run  --pipeline_batch_size=1024 --model_max_batch_size=32 \
   pipeline-nlp --viz_file=.tmp/nlp_kitchen_sink.png  \
   from-file --filename examples/data/pcap_dump.jsonlines \
   deserialize \
   preprocess \
   inf-triton --model_name=sid-minibert-onnx --server_url=localhost:8000 \
   monitor --description "Inference Rate" --smoothing=0.001 --unit "inf" \
   add-class \
   filter --filter_source=TENSOR --threshold=0.8 \
   serialize --include 'timestamp' --exclude '^_ts_' \
   to-kafka --bootstrap_servers localhost:9092 --output_topic "inference_output" \
   monitor --description "ToKafka Rate" --smoothing=0.001 --unit "msg"

Output:

Configuring Pipeline via CLI
Starting pipeline via CLI... Ctrl+C to Quit
Inference Rate[Complete]: 93085 inf [00:07, 12334.49 inf/s]
ToKafka Rate[Complete]: 93085 msg [00:07, 13297.85 msg/s]
Pipeline visualization saved to .tmp/nlp_kitchen_sink.png