google-cloud-cli has a huge install size

[jonsneyers]

It's something like 1 gigabyte of install size when I follow the instructions on https://cloud.google.com/storage/docs/gsutil_install (and then I'm not even counting the dependencies, I'm only looking at /usr/lib/google-cloud-sdk)

This seems very excessive for what is supposed to be just a command line interface to a cloud platform.

Please consider being a bit more size-aware when packaging these things. For example, I see a huge unstripped binary in /usr/lib/google-cloud-sdk/bin/anthoscli (174 MB) and a big uncompressed JSON file in /usr/lib/google-cloud-sdk/data/cli/gcloud.json (83 MB). Also it looks like a bunch of test code is being shipped as part of the package:

$ ls -R /usr/lib/google-cloud-sdk |grep test |wc -l
1609

which makes me think this is not actually a packaged tool but more like a dump of a development repository.

It feels like there is a lot of low hanging fruit to reduce the install size of this tool.

[wiardvanrij]

root@9c0e8edf5bdd:/opt# du -sh *
1.7G    google-cloud-sdk

We only need 1 addon. I would like to stretch this issue by saying the size is even ridiculous. We really want minimal image sizes.

[Mcfloy]

On top of that I don't understand why Python should be installed when using a CLI that is available through multiple links, categorized by their core architecture. If I download the CLI for linux x86_64, why isn't it just a static binary ? On docker images it bloats the size because of dependencies that could be avoided.

[tmancill]

+1 on taking action to make this more efficient. It doesn't seem like it scales very well.

This package gets installed automatically on some GCP instance types (reference) and takes a long time (1-2 minutes) to update on smaller instances. Having nearly 40000 files, over a third of which are manpages, probably doesn't help...

$ dpkg -L google-cloud-cli | wc -l
38775

$ dpkg -L google-cloud-cli | grep /usr/share/man/man1/gcloud  | wc -l
13271

For a clean install on a Debian image:

The following NEW packages will be installed:
  google-cloud-cli
0 upgraded, 1 newly installed, 0 to remove and 0 not upgraded.
Need to get 154 MB of archives.
After this operation, 760 MB of additional disk space will be used.

It also releases on a strict weekly cadence, so if you're keeping up with security updates, etc. you're going to end up (re)installing it continually.

[binaryronin]

+1 on this. We have some very small endpoint instances in GCP and this single package takes over 10 minutes to unpack and update during which our alerting goes crazy!

[mihaiav]

Someone would expect more from Google engineers. Instead of an efficient Command Line Interface we are served with GBs of bloat. I'm using the CLI to deploy a Cloud Run docker image. The Google Cloud Cli competes with Docker itself in size...

[ViliusS]

I can confirm this is not only a problem for disk footprint but a problem during installation and update on very small GCP instances (e2-micro, etc). Every time DNF is trying to unpack Cloud SDK RPM image VM instance just goes down or segfaults with out of memory errors.

This happens on all instances with < 1GB of memory -> https://issuetracker.google.com/issues/239207289

[tonymet]

I entered this ticket and later found this convo. Please upvote this bug to help get it some attention.

https://issuetracker.google.com/issues/324114897

I agree the install size is abhorrent.

[tonymet]

@jonsneyers Here's a review of the installation and some of the wasted space. Fun fact: there is an entire python installation within the google-cloud-sdk

/google-cloud-sdk/...
* /platform/bundledpythonunix 270.1MB -- this contains a complete python installation
* /generatedclients 165.7MB
* /lib/third_party/botocore 94.8MB
* /lib/third_party/kubernetes 25.1MB 

[framegrace]

Installing it manually instead of using the distro package can save you precious space. This method reduced the docker image size 400Mb from installing with apt. Also removed bq, which is small, but I don't need and also do some config and cleanup from updating. Adapt to suit your needs:

ENV PATH /google-cloud-sdk/bin:$PATH
RUN ARCH=`arch` && CLOUD_SDK_VERSION="470.0.0" && curl -O https://dl.google.com/dl/cloudsdk/channels/rapid/downloads/google-cloud-cli-${CLOUD_SDK_VERSION}-linux-${ARCH}.tar.gz && \
    tar xzf google-cloud-cli-${CLOUD_SDK_VERSION}-linux-${ARCH}.tar.gz && \
    rm google-cloud-cli-${CLOUD_SDK_VERSION}-linux-${ARCH}.tar.gz && \
    gcloud config set core/disable_usage_reporting true && \
    gcloud config set component_manager/disable_update_check true && \
    gcloud config set metrics/environment github_docker_image && \
    gcloud components remove  bq && \
    gcloud components update && \
    rm -rf $(find google-cloud-sdk/ -regex ".*/__pycache__") && \
    rm -rf google-cloud-sdk/.install/.backup && \
    gcloud --version
RUN git config --system credential.'https://source.developers.google.com'.helper gcloud.sh

[framegrace]

For ubuntu (jammy) docker images, have a Dockerfile which completelly removes the included python and forces the use of the OS version.

Reduces the image size to around 800Mb uncompressed (200 compressed). THat's 2G less than the official cloud-sdk image.

FROM ubuntu:jammy as base

...
# Add your base stuff  if needed
...

# Minimalized Google cloud sdk
FROM base as gcloud-installer

# Download python3 module dependencies  (will also install python-minimal which is only around 25Mb)
RUN apt-get install -y \
        python3-crcmod \
        python3-openssl
ARG CLOUD_SDK_VERSION=452.0.1
ENV CLOUD_SDK_VERSION=$CLOUD_SDK_VERSION
ENV PATH /google-cloud-sdk/bin:$PATH
ENV CLOUDSDK_PYTHON=/usr/bin/python3
# Download and install cloud sdk. Review the components I install, you may not need them.
RUN ARCH=x86_64 && \
    curl -O https://dl.google.com/dl/cloudsdk/channels/rapid/downloads/google-cloud-cli-${CLOUD_SDK_VERSION}-linux-${ARCH}.tar.gz && \
    tar xzf google-cloud-cli-${CLOUD_SDK_VERSION}-linux-${ARCH}.tar.gz && \
    rm google-cloud-cli-${CLOUD_SDK_VERSION}-linux-${ARCH}.tar.gz && \
    rm -rf /google-cloud-sdk/platform/bundledpythonunix && \
    gcloud config set core/disable_usage_reporting true && \
    gcloud config set component_manager/disable_update_check true && \
    gcloud config set metrics/environment github_docker_image && \
    gcloud components remove -q bq && \
    gcloud components install -q beta && \
    gcloud components install -q gke-gcloud-auth-plugin && \
    rm -rf $(find google-cloud-sdk/ -regex ".*/__pycache__") && \
    rm -rf google-cloud-sdk/.install/.backup && \
    rm -rf google-cloud-sdk/bin/anthoscli && \
    gcloud --version

#...
#<Add more stages if you need>
#...

# On your final stage, (here simply from base, for example)
FROM base as final

# Add to the path
ENV PATH /google-cloud-sdk/bin:$PATH
# Ask gcloud to use local python3
ENV CLOUDSDK_PYTHON=/usr/bin/python3
# Copy just the installed files
copy --from=gcloud-installer /google-cloud-sdk /google-cloud-sdk
# This is to be able to update gcloud packages
RUN git config --system credential.'https://source.developers.google.com'.helper gcloud.sh

I really don't understand the little care Google engineers have put on this. 2GB of python is a lot of python and 400Mb of documentation is also a waste. SO, well, there's nothing really special there, I think this is easily ported to alpine,etc.. I may do it when I have some time.

Edit: Cleaned up some stuff and added comments to make it more readable. This was directly extracted from an internal dockerfile.

[tonymet]

For ubuntu (jammy) docker images, have a Dockerfile which completelly removes the included python and forces the use of the OS version.

• https://gist.github.com/tonymet/86021d88ca0a0585f575c23f61ebeb95
• https://hub.docker.com/r/tonymet/gcloud-lite

Here is an improved version

• image size 389 MB
• supports CLI arguments e.g. docker run -v$HOME:/root tonymet/gcloud-lite compute instances list
• based on ALPINE

docker pull tonymet/gcloud-lite

[tonymet]

I'm working on the CI to automate these artifacts for general use. Here are the two categories we'll publish

• "gcloud-lite" docker images like the above about 400MB
• a "gcloud-lite" tgz distribution based on google's official images at < 100MB. Published to github releases page

thumbs up👍 here if you would find this useful and ill ping you when it's ready to be tested.

[tonymet]

471 & 472 releases are ready to test out here. You can find instructions on how to obtain the docker images & .tgz artifacts

https://github.com/tonymet/gcloud-lite

[ad-m-ss]

Please be understanding, this is the industry standard: https://github.com/Azure/azure-cli/issues/22955 https://github.com/Azure/azure-cli/issues/7387 , unfortunately.

[tonymet]

In the second example aws is 3mb and azure is 700mb

On Tue, Apr 16, 2024 at 16:46 Adam Dobrawy @.***> wrote:

Please be understanding, this is the industry standard: Azure/azure-cli#22955 https://github.com/Azure/azure-cli/issues/22955 Azure/azure-cli#7387 https://github.com/Azure/azure-cli/issues/7387 .

— Reply to this email directly, view it on GitHub https://github.com/GoogleCloudPlatform/gsutil/issues/1732#issuecomment-2060080464, or unsubscribe https://github.com/notifications/unsubscribe-auth/AADBFK4DCJDQVGU2OYSK56TY5WZ4PAVCNFSM6AAAAAA3A2W3Z6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANRQGA4DANBWGQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[tonymet]

please upvote this GCP ticket https://issuetracker.google.com/issues/324114897?pli=1 -- maintainers also raised an internal ticket that's connected.

[ViliusS]

please upvote this GCP ticket https://issuetracker.google.com/issues/324114897?pli=1 -- maintainers also raised an internal ticket that's connected.

Voted. Please also upvote https://issuetracker.google.com/issues/239207289 to fix this in RPM packages too.

[tonymet]

automation is working now on a 24-hour latency. we have lite releases 471,472 & 473 in runnable docker images (86% faster) and tgz release . https://github.com/tonymet/gcloud-lite

give it a run

[max0x7ba]

I really don't understand the little care Google engineers have put on this. 2GB of python is a lot of python and 400Mb of documentation is also a waste.

Quick recap:

1. Google creates Go language which produces compact executables with no dependencies and uses that for its own applications.
2. Google creates Compute Engine and sells CPU time and storage space.
3. Google creates gcloud command line application for Compute Engine users which takes 2GB+ of storage and takes longer to install than Windows OS.

The above makes perfect business sense for Google, does it not?

[tonymet]

Quick recap:

1. Google creates Go language which produces compact executables with no dependencies and uses that for its own applications.

Let's be charitable to the team and make sure we are being reasonable. A Go rewrite is out of scope but a stripped gcloud CLI distribution is feasible. I don't think the package size is intentional, it's likely that the teams haven't assessed customer impact. I think our best strategy here is to make sure GCP is aware of the customer impact and give them resources to help prepare a stripped distribution.

[max0x7ba]

Quick recap:

1. Google creates Go language which produces compact executables with no dependencies and uses that for its own applications.

Let's be charitable to the team and make sure we are being reasonable. A Go rewrite is out of scope but a stripped gcloud CLI distribution is feasible. I don't think the package size is intentional, it's likely that the teams haven't assessed customer impact. I think our best strategy here is to make sure GCP is aware of the customer impact and give them resources to help prepare a stripped distribution.

I am a paying customer of Google Compute Engine. And I am nice and kind. Charitable donations or gifts are not involved here, hence, word charitable is not applicable and is totally out of place here.

I expect world-class service for my money from a company which takes pride in its technical excellence. Currently, gcloud is the opposite of technical excellence, is it not?

Is it too much to ask of a corporation to fulfill its promises https://about.google/intl/en-GB/philosophy/ ?

1. Focus on the user and all else will follow.
2. It’s best to do one thing really, really well.
3. Fast is better than slow. ...
4. Great just isn’t good enough.

[ViliusS]

Exactly. This is not free software, nor this is some small business which requires our nurture and forgiving attention. This is a multi-billion dollar company and we are paying customers. If their margins and profits cannot hire proper management or engineering which are capable of delivering stellar experience for their clients we might just turn off the lights and go home.

[max0x7ba]

Just to add context,

I had no issues invoking gcloud since 2021 until sometime April-May 2024, when invoking gcloud would freeze an instance forever and prevent it from accepting new ssh connections to diagnose the issue. Wasting a day debugging the issue revealed that snap auto-updating gcloud was the root cause of the problem.

In other words, gcloud automatic update broke my business processes and made me waste a day debugging the issue, all the while paying Google for using its resources. A mini-disaster caused solely by gcloud developers.

Next, 6+ months since the issue was reported and not resolved, cute dude @tonymet tells me to be charitable while cutting out one point of mine out of context. Which is a text-book example of trolling and virtue signalling.

[tonymet]

Next, 6+ months since the issue was reported and not resolved, cute dude @tonymet tells me to be charitable while cutting out one point of mine out of context. Which is a text-book example of trolling and virtue signalling.

I'm sorry i meant no offense. I meant that we will catch more flies with honey than vinegar.

And I'm certainly not virtue signaling. Look at all the effort I invested to create an open source distro and raise awareness with google. We both want the same thing here.

https://github.com/tonymet/gcloud-lite

[max0x7ba]

Next, 6+ months since the issue was reported and not resolved, cute dude @tonymet tells me to be charitable while cutting out one point of mine out of context. Which is a text-book example of trolling and virtue signalling.

I'm sorry i meant no offense. I meant that we will catch more flies with honey than vinegar.

And I'm certainly not virtue signaling. Look at all the effort I invested to create an open source distro and raise awareness with google. We both want the same thing here.

You are right. I am sorry for any offence or irritation caused. Sometimes, I cannot hold in the fire, and that's a limitation of mine.

[tonymet]

good to hear it and thanks. let's hope for the best