Command Line Scripts

Seldon provides several scripts to aid starting Seldon, provisioning services and stopping Seldon.



Create Seldon on a running kubernetes cluster.



To launch seldon with all components run


To start with GlusterFS run


To start without a Spark cluster

SELDON_WITH_SPARK=false seldon-up



Shutdown Seldon running on a Kubernetes cluster



See the detailed seldon-cli documentation.



Start one or more microservices for a particular client. The microservices can be REST or for predictions also gRPC based. The script allows you to start microservices of two types:

usage: start-microservice [-h] [-i name image microservice API ratio]
                          [-p name folder microservice API ratio] --client
                          CLIENT [--replicas REPLICAS] --type

optional arguments:
  -h, --help            show this help message and exit
  -i name image microservice API ratio
                        microservice image defn: <name> <image> <API type
                        (rest or rpc)> <ratio>
  -p name folder microservice API ratio
                        microservice from pipeline defn: <name> <folder> <API
                        type (rest or rpc) <ratio>
  --client CLIENT       client name
  --replicas REPLICAS   number of replicas
  --type {recommendation,prediction}
                        microservice type


Start a recommendation microservice from a built Docker image exposed as REST endpoint for the client “reuters”. See the worked Reuters content recommendation example.

start-microservice --type recommendation --client reuters -i reuters-example seldonio/reuters-example:2.0.7 rest 1.0

Start a prediction REST microservice from a saved pipeline previously saved to /seldon-data/seldon-models/finefoods/1 for client “test”. See the worked sentiment analysis demo.

start-microservice --type prediction --client test -p finefoods-xgboost /seldon-data/seldon-models/finefoods/1/ rest 1.0

Start a prediction microservice from an xgboost mode exposed as a REST service and packaged in a docker image. See worked example in the Iris prediction demo.

start-microservice --type prediction --client test -i iris-xgboost seldonio/iris_xgboost:2.1 rest 1.0

Start and AB test with two microservices.

start-microservice --type prediction --client test -i iris-xgboost seldonio/iris_xgboost:2.1 rest 0.5 -i iris-scikit seldonio/iris_scikit:2.1 rest 0.5

Start gRPC microservice for Iris demo

start-microservice --type prediction --client test -i xgboostrpc seldonio/iris_xgboost_rpc:2.1 rpc 1.0



Create a locust loadtest. Presently for prediction services only. Handles REST and gRPC

usage: launch-locust-load-test [-h] --seldon-client SELDON_CLIENT
                               [--locust-slaves LOCUST_SLAVES]
                               [--locust-hatch-rate LOCUST_HATCH_RATE]
                               [--locust-clients LOCUST_CLIENTS] 
			       --test-type {js-predict,grpc-predict}
                               [--seldon-grpc-endpoint SELDON_GRPC_ENDPOINT]
                               [--seldon-oauth-endpoint SELDON_OAUTH_ENDPOINT]
                               [--seldon-predict-default-data-size SELDON_PREDICT_DEFAULT_DATA_SIZE]

optional arguments:
  -h, --help            show this help message and exit
  --seldon-client SELDON_CLIENT
                        client name
  --locust-slaves LOCUST_SLAVES
                        number of slaves to run
  --locust-hatch-rate LOCUST_HATCH_RATE
                        locust hatch rate
  --locust-clients LOCUST_CLIENTS
                        number of locust clients
  --test-type {js-predict,grpc-predict}
                        type of test to run
  --seldon-grpc-endpoint SELDON_GRPC_ENDPOINT
                        seldon grpc endpoint
  --seldon-oauth-endpoint SELDON_OAUTH_ENDPOINT
                        seldon oauth endpoint
  --seldon-predict-default-data-size SELDON_PREDICT_DEFAULT_DATA_SIZE
                        the size of the default list of random floats to send
                        to predict endpoint


# launch grpc prediction load test
launch-locust-load-test --seldon-client deep_mnist_client --test-type grpc-predict