So I've got 9000+ original images, each with 5-7 generated sizes, already stored on S3. Moving production to Heroku and one of the issues we have is timeout on large master images.
Started trying Carrierwave Direct gem to go straight to the S3 store and then process async, but ran into an issue with existing images disappearing as soon as I added the third include line to the base uploader:
class BaseUploader < CarrierWave::Uploader::Base
include CarrierWave::RMagick
include CarrierWave::ImageOptimizer
include CarrierWaveDirect::Uploader
...
end
CarrierwaveDirect changes the filename format (here is the method: https://github.com/dwilkie/carrierwave_direct/blob/29caaa8d399968258299b57c84eac2e5e6fc4cf7/lib/carrierwave_direct/uploader.rb#L153)
The filenames are changed in this manner:
Orig carrierwave gen:
https://images.patchvault.org/uploads/issue/imageback/629/full_003b_a1_back.png
CW Direct url gen:
https://images.patchvault.org/uploads/issue/imageback/629/003b_a1_back_full.png
So the version name moves from the begining to the end of the uploaded filename. That actually makes sense, but I'm dealing with 2 years and 9k images :)
When I enable include CarrierWaveDirect::Uploader in the BaseUploader it changes the manner in which urls are generated in the view, and therefore all images 404 (since they no longer exist)
- What is the best practice here? I'm saving to Fog, but since it's uploading to Heroku's
/tmp, should I just move the processing to async? - Can I revert the behavior by overriding that method or is that a no-go?
- Is there a way to easily rewrite the images filenames on S3 to the new format or is that a no-go?
Open to other solutions I'm not thinking about as well :)
The path is generated in base_uploader.rb:
def store_dir
"uploads/#{model.class.to_s.underscore}/#{mounted_as}/#{model.id}"
end
Rails 5.2.1, Ruby 2.5.2, Heroku-18 Standard 2x dyno (web) 1x dyno (worker). images.patchvault.org is the CNAME for a cloudfront CDN distro on the S3 bucket.
