Commands

login

docker login -u _json_key -p "$(cat gcr-json-key-file.json)" https://gcr.io

list

docker ps
docker ps -a  # including stopped
docker ps -a -q # all processes(containers)' id
docker container ls
docker container ls -a # including stopped

rm

1. remove container + image

   docker container stop name_or_id
   docker container rm name_or_id
   docker image rm name_or_id
   

2. remove all containers

   docker rm $(docker ps -a -q)
   

3. rm vs rmi images size listed in docker images is cumulated with parent size. If parent is pre-built separately at local machine, remove the image might not release the total size listed. To check actual released size, calculate the difference between the two commands docker system df. To check each layer size, docker history <img>

   rm is for container
   rmi is for image

4. Remove all dangling(not referring to any tagged images) images. If -a is specified, will also remove all images not referenced by any container.

   docker image prune

build

docker build -t tag -f Dockerfile .

. project context

tag

docker tag image:v1 image:v2
docker push image:v2

run

1. run simple command

   docker run some_image ls
   

   • ls: the command to run in the container. After it exits, the container stops.

   interactive command like /bin/bash or python

   docker run -ti some_image /bin/bash
   

   • -ti
     • -t into the container. Interactive commands must have it.
     • -i with STDIN pass to the container. Only effective with -t
     • If only -t, a bash shell is created but user input is not passed to it.
     • If only -i, no effect
   • /bin/bash: the command to run in the container. After it exits, the container stops.
   • ctrl+p ctl+q: if -ti, detach the running docker to background. docker attach name_of_the_container will resume it.
   • ctrl+d: if -ti, will exist the bash in container, stop the container and come back to the terminal.

   port-forwarding

   docker run -p 8000:8080 --rm -it --name apple image_tag
   

   • -p/--publish 8000:8080: asks Docker to forward traffic incoming on the host’s port 8000, to the container’s port 8080
   • --rm: once stopped, removed the containers
   • -t: into the container
   • -i: with STDIN
   • --name: give it a name

   Containers are merely an instance of the image you use to run them; docker run will create a temporary container and execute the command. After command finishes, the container is also stopped

2. exec

   docker exec apple
   docker exec id
   docker exec -t -i mycontainer /bin/bash
   

   run commands on the running container

3. top

   docker top apple
   docker top id
   

   see processes

docker-compose

docker-compose start name: start a service

docker-compose up -d name: start this service and any binding ones

docker-compose run --service-ports name [NEW COMMAND]: start service with new command and with port-forwarding

tf serving

check serving status: localhost:8501/v1/models/model_name

check serving schema: localhost:8501/v1/models/model_name/metadata

1. way 1

   gRPC

   docker run -p 8500:8500 \
   --mount type=bind,source=/tmp/mnist,target=/models/mnist \
   -e MODEL_NAME=mnist -t tensorflow/serving &
   

   rest api

   docker run -p 8501:8501 \
   --mount type=bind,source=/tmp/mnist,target=/models/mnist \
   -e MODEL_NAME=mnist -t tensorflow/serving &
   

   bash parameters:

   • & as long as the container does not output logs, it will run in the background. Otherwise, the terminal is occupied.
     • jobs will list background jobs
     • fg will take a background job in the front
     • use docker parameter -d -ti is better.

   docker parameters:

   • --mount type=bind,source=x,target=y: creates a folder y in the container, and x,y must be absolute paths
   • models/mnist contains version folder 1
   • -e MODEL_NAME must equal target, if no -e MODEL_BASE_PATH
   • -e MODEL_NAME, equals --model_name, --MODEL_NAME in tf serving parameters
   • -e TF_CPP_MIN_VLOG_LEVEL=4 enables verbose logging of tf serving, and it logs continuously thus occupies the terminal even if &
     • TF_CPP_MIN_LOG_LEVEL: controls LOG with format [module] msg; 0 logs everything, 4 logs nothing
     • TF_CPP_MIN_VLOG_LEVEL: controls VLOB with format [time level file] msg; extra info per allowed LOG level; 0 logs nothing, 4 logs everything
     • TF_CPP_MIN_VLOG_LEVEL has been renamed to TF_CPP_MAX_VLOG_LEVEL after 2.5.0rc0
   • -e TF_FORCE_GPU_ALLOW_GROWTH=true: occupies gpu mem step by step
   • --gpus '"device=2"' (requires nvidia-docker and docker >= 19.03, for lower version of docker, use --runtime=nvidia)

   tf serving parameters, must adds after image name:

   • --enable_batching: adds after image name to batch the requests

2. way 2

   export TESTDATA="$(pwd)/models/model_name/"
   
   docker run -t --rm -p 8501:8501 \
   -v "$TESTDATA:/models/keras_dpr_model" \
   -e MODEL_NAME=keras_dpr_model \
   tensorflow/serving &
   

   • -v: creates a folder in the container

3. way 3

   docker run -d --name serving_base tensorflow/serving
   
   docker cp /tmp/resnet serving_base:/models/resnet
   
   docker commit --change "ENV MODEL_NAME resnet" serving_base $USER/new_image
   
   docker kill serving_base
   docker rm serving_base
   
   docker run -p 8500:8500 -t $USER/new_image &
   

   • -d: detach

tg bot

In google cloud:

• Update environment variables in .env and run.sh
• If session expired: python main.py
• sudo sysctl -w vm.max_map_count=262144 for elastic search to work
• sudo docker-compose up -d to start
• To index history messages: /download_history
• sudo docker-compose down to stop