A Faster Collect Static (sometimes) for Django
# Collect Stupid - A Faster, Smarter, and Stupider Collect Static for Django
When using collectstatic with remote file systems like Amazon S3 it can often slow down because up to 3 remote operations are needed for each file.
- Determine file time stamp
- Delete file if old
- Copy the new file
collectstupid speeds up your static file deploy with two tricks
- The Stupid One: Never checks remote file timestamps and never deletes the old file. It just overwrites the file blindly. This saves two remote operations.
- The Smart One: Stores a list of deployed file md5sums on the remote file system. This keeps track of what files need to be deployed and md5sums are only computed at the source for speed. This can be faster than other collectstatic libraries which a lot of times use local cache and thus other developers don’t benefit from the last deploy. Also the deployment tracking is more permanent and not subject to cache clearing unless you explicitly need to clear it.
## Installation and Usage
pip install collectstupid
Add collectstupid to INSTALLED_APPS
Run: python manage.py collectstupid
## Storage Mixin
Some storage implementations execute an exists and delete on the storage class even when using collectstupid. You can add the collectstupid.storage.StupidStorageMixin to your storage class to skip those requests.
## Personal Example Where CollectStupid Improved Deployment Speed
For a project, whenever we deployed, a huge amount of memory (+13GB) was used and the whole process took more than 30 minutes on a fast Mac and over 2 hours on a virtual machine. Through investigation, we found that every check of whether a file existed caused the bucket to do a full file list because of a very inefficient S3 storage implementation. We implemented collectstupid and after the initial deploy of large files, subsequent deploys went down to less than 5 minutes. This also cut down memory usage to almost nothing when compared to 13GB.
STUPID_DEPLOY_CACHE: File where md5sums are stored on the remote filesystem
## Feature requests
Feel free to open an issue