the advantage that divide dockerfile into some stages

[alicera]

Description

what is the advantage for the dockerfile with different stages ?

example

FROM nvcr.io/nvidia/tensorrt:20.06-py3 AS base
# do something
FROM base AS trtlab
# do something

https://github.com/NVIDIA/tensorrt-laboratory/blob/v2/Dockerfile

# stage 1 - development container
# holds the core nvidia libraries but does not container the project source code
# use this container for development by mapping our source into the image which
# persists your source code outside of the container lifecycle

FROM nvcr.io/nvidia/tensorrt:20.06-py3 AS base

RUN apt update
RUN apt install -y clang-format libssl-dev openssl libz-dev software-properties-common

# remove base cmake
RUN apt remove --purge -y cmake
RUN apt autoremove -y
RUN apt autoclean -y

# install cmake ppa from kitware - https://apt.kitware.com/
RUN apt install -y apt-transport-https ca-certificates gnupg software-properties-common wget
RUN wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | apt-key add -
RUN apt-add-repository 'deb https://apt.kitware.com/ubuntu/ bionic main'
RUN apt update && apt install -y cmake

# then remove FindGTest.cmake installed by cmake
RUN find / -name "FindGTest.cmake" -exec rm -f {} \;

# add cufft and nvml to the container image
RUN apt install -y libcufft-dev-11-0 cuda-nvml-dev-11-0

# override some envs
ENV LD_LIBRARY_PATH=/externals/myelin/x86_64/cuda-11.0/lib:/externals/cudnn/x86_64/8.0/cuda-11.0/lib64:/usr/local/cuda-11.0/targets/x86_64-linux/lib
ENV CCACHE_DIR=/tmp/.ccache
RUN cd /usr/lib/x86_64-linux-gnu && ln -s libnvidia-ml.so.1 libnvidia-ml.so


# stage 2: build the project inside the dev container

FROM base AS trtlab

WORKDIR /work

COPY . .

RUN mkdir build && cd build && cmake .. && make -j

[jedevc]

The main advantages of having multiple stages are:

1. Multiple stages that don't have dependencies between each other can be run in parallel, speeding up the build.
2. Multiple stages can be used to avoid packaging dependencies in the final image result.

Hope that answers your question 🎉

[alicera]

Thanks for your answer.
But the 1 point that you mention "each other can be run in parallel"
I dont understand.

In the dockerfile, the LINE3 command is based on the LINE1.
Can LINE1 and LINE2 be run in parallel ?
LINE2 should wait for LINE1 to finish building ?

FROM nvcr.io/nvidia/tensorrt:20.06-py3 AS base   #  LINE1
# do something                                                       #  LINE2
FROM base AS trtlab                                              #  LINE3
# do something                                                       #  LINE4

If the dockerfile like this
I can understand the LINE1 and LINE3 can be run in parallel.

FROM nvcr.io/nvidia/tensorrt:20.06-py3 AS base   #  LINE1
# do something                                                       #  LINE2
FROM ubuntu20.04 AS trtlab                                  #  LINE3
# do something                                                       #  LINE4