Pivotal Cloud Foundry - "access denied" for custom buildpacks written on Windows

Writing custom buildpacks on Windows can lead to some unnecessary frustrations. Today I would like to cover one of those frustrations: buildpack scripts not being executable on PCF. Creating buildpack scripts on windows can lead to cryptic errors during buildpack execution like "access denied" or "permission denied".

The problem is caused by the fact that any new file created on Windows is automatically set to mode 644 aka rw-r--r-- (linux standards), but at the same type is automatically executable (-rwxr-xr-x) by windows standards.

Result? Listing your buildpack scripts using git bash on windows you see:

drwxr-xr-x 1 Piotr_Justyna 1049089    0 Feb 13 20:46 ./
drwxr-xr-x 1 Piotr_Justyna 1049089    0 Feb 13 20:44 ../
-rwxr-xr-x 1 Piotr_Justyna 1049089  137 Feb 13 20:44 detect*
-rwxr-xr-x 1 Piotr_Justyna 1049089   40 Feb 13 20:46 finalize*
-rwxr-xr-x 1 Piotr_Justyna 1049089  102 Feb 13 20:44 release*
-rwxr-xr-x 1 Piotr_Justyna 1049089  135 Feb 13 20:46 supply*

While in fact all of them are by default mode 644 (rw-r--r--). Trying to execute those scripts on PCF will result in "access denied" errors. It is pretty annoying, so I prepared a little cheat sheet for those of you who work with custom buildpacks, windows and git bash:

Executable files:

• DON'T:

  chmod +x file
  

• DO:

  git update-index --chmod=+x file
  

List files with their access modes:

• DON'T:

  ls -la
  

• DO:

  git ls-files -s
  

Copying files:

• DON'T:

  cp file file_copy (that defaults the access mode of the copy to 644 even if the file is e.g. 755)
  

• DO:

  cp file file_copy *and then* git update-index --chmod=+x file_copy
  

chmod-calculator.com is a very handy tool, especially for those who don't work with linux too often.

Haskell, Stack, Docker, AWS and a notepad of your choice

I've been playing with Haskell on and off for the last few years and at this stage, while I still consider myself a newbie, I feel quite confident I could write a small application (*something*) and have it do relatively simple tasks for me. It's great to have your application published and running somewhere in the cloud, but the whole process of getting your code to the cloud is much less so. At least it has been in my case. What I like to do is more programming and less configuration and manual tasks!

Check now inactive fontbot for example. The way it is written is pretty basic and its iterations are as basic as one can wish for:

* write code
* build it with Stack
* copy the compiled code manually to a prepared EC2 machine

Basic.

But, as I'm here to hack fridges, I want more. I want the cycle to be more automatic and less manual. I want to be able to publish my built code anytime I'm happy with the build. Preferably to a Docker container hosted by AWS (not necessarily a single EC2 machine). My daily work does not require any interaction with Docker (so your feedback is much appreciated!), but since it looks like a very useful thing, I plan to explore it. I'm pretty new to Docker and the idea of ECS, so it should be good fun.

Let's sketch a plan of this exercise (now partially updated with the parts I've already finished):

• Prepare a basic Docker image to host static HTML files locally.
• Configure ECS to use that image and host the static files globally.
• Make another image, this time able to host some basic Haskell server code.
• Adjust ECS if needed for that new image.
• Configure a bitbucket pipeline to publish your code to ECS whenever a new version is pushed to a certain branch.

At the end of this exercise I should have a setup where every time I push my code to a chosen branch, it gets deployed to ECS by the pipeline.

Stay tuned and keep hacking!
Piotr