Sidecar container for tensorflow serving. Polls GCS bucket for new versions to serve.
164
model_poller is a sidecar container to use with tensorflow/serving to push new model versions without service interruption.
It polls a google cloud storage model bucket for new versions to serve. And then download and unzip the models in the folder used by tensorflow/serving as the source of models.
The source Dockerfile and utilization are detailed in the associated GitHub repository: https://github.com/fpaupier/sidecar-pattern_tf-serving
Usage example
# Update env variables accordingly
PROJECT_ID=tensorflow-serving-sidecar
SUBSCRIPTION=tf_sub
SAVE_DIR=/shared_dir/models
GOOGLE_APPLICATION_CREDENTIALS="/path/to/gcp-key.json"
docker run -t \
-v "/Users/fpaupier/projects/model_poller:/shared_dir" \
-e PROJECT_ID=$PROJECT_ID \
-e SUBSCRIPTION=$SUBSCRIPTION \
-e SAVE_DIR=$SAVE_DIR \
-e GOOGLE_APPLICATION_CREDENTIALS=$GOOGLE_APPLICATION_CREDENTIALS \
popszer/model_poller:v0.1.6 &
Content type
Image
Digest
Size
284.8 MB
Last updated
over 7 years ago
docker pull popszer/model_poller:v0.1.6