Pipelines is too slow

[ZenobiusJ]

Every step costs at minimum 55secs due to the git clone.

We need a way to declare when we want the clone and artifacts operations occur.

I recommended in a ticket that the following be converted to bash scripts which would be injected into the container:

- git_clone(depth)

- artifact_save([]globs)

- artifact_restore(step_name)

[ZenobiusJ]

Yeah being able to decide what impact the step execution time cost is going to be a huge positive.

We've been able to mitigate the huge risk of npm installing assets to use gulp, browserify and sass by baking all those npm modules into a docker image we use, but when the team saw the unavoidable 55sec cost of each step, there was dismay.

I know there'll be an avoidable delay of container warmup for each step. but having these bitbucket* commands mounted at a configurable path would help speed everyone's build times up.

[mwatson]

So as I understand it - you have subsequent steps of a multiple step pipeline that only need the artifacts from the previous step, not the checkout, or sometimes don't even need the artifacts, they just execute commands?

Thus the close and artefact retrieval is time wasted for you?

[ZenobiusJ]

Really though, I invite you to see how circleci does this: https://circleci.com/docs/2.0/configuration-reference/#checkout

[ZenobiusJ]

As to your comment about needing to perform a git clone in order to know what to do:

I'd imagine that the whole pipeline yaml file is provided out of band separate from the git clone.

[ZenobiusJ]

if you wanted to be even more cautious about what you're putting into peoples containers you could provide optional key like so:

bitbucket_api_location: /opt/bitbucket

Then we'd change the above calls to bitbucket_checkout, bitbucket_artifact_save bitbucket_artifact_restore would change to: /opt/bitbucket/bitbucket_checkout /opt/bitbucket/bitbucket_artifact_save /opt/bitbucket/bitbucket_artifact_restore

[ZenobiusJ]

@mwatson 

How would I use it?

Like this:

with opt-in yaml key:

`git: [manual|auto (default)]`

using proposed bash scripts: 

- `bitbucket_checkout depth`

  - `git checkout at commit with depth configured`

- `bitbucket_artifact_save glob glob glob`

  - saves all files found matching glob(s)

- `bitbucket_artifact_restore step_name`

  - restores all globbed files from step_name

pipelines:
  default:
    - step:
        image: our-image-based-on-node:8-alpine
        name: setup
        git: manual
        script:
          - |
             bitbucket_checkout full
             npm run prod
             bitbucket_artifact_save ./client/build/**/* ./cilent/build/docs/**/*

    - step:
        image: atlassians-awscli-image
        name: publish
        git: manual
        script:
          - |
             ls -al ./
             bitbucket_artifact_restore setup
             aws s3 sync --delete ./client/build/ s3://our-bucket/releases/$BITBUCKET_BUILD_NO/

    - step:
        image: cloudfoundries-concourse-ci-slack-resource-image
        name: notify
        git: manual
        script:
          - |
             echo "complex json object"  | envsubst | /opt/resource/out  

If a step required files from the commit, my options here are :

- use artifacts to persist it through the pipeline

- run a git checkout

I estimate this would save roughly 40-50 secs for each step.

[mwatson]

Hi ZenobiusJ,

It'd be useful to better understand your use case.

If we made these operations able to be manually applied by the user, how would you use them to speed things up? What would you do differently to the current order of doing things? Would you not always need the git clone to determine what to do?

Thanks,

Matt Watson (Bitbucket Pipelines Development Manager)