- Start Date: 2017-03-14
- RFC PR:
- Yarn Issue: yarnpkg/yarn#986
Allow users to bypass the global cache.
For certain situations, it would be preferrable to not use a global cache for
packages because the cache takes space and the step of copying from the cache
to node_modules can take a significant amount of time. Some examples are:
-
Building a Docker image. Images are isolated, so there's no point in having a cache within the image. If anything, it has the negative effect of increasing the image size.
-
Running Yarn in a CI context. This is another example of a potentially isolated environment where we would just want to put the installed packages directly into
node_modules. While sophisticated CI setups might have multiple projects share the same cache, there are certainly setups that don't do that. -
Running Yarn on a long-running server. The cache grows indefinitely over time, eventually reaching max disk size or inode limit. Simply bypassing the cache at run time could be preferrable to having to periodically run
yarn cache clean.
Overall, the desired effects of bypassing the cache are to increase the speed and decrease the disk usage of certain Yarn commands (i.e., install, add, upgrade).
There should be a CLI switch --no-cache that tells Yarn to skip doing any work
with the cache.
Furthermore, there could be an equivalent .yarnrc setting of no-cache that
does the same thing when set to true. This way, the developer doesn't need to
remember to use the CLI switch every time.
We would update the documentation for
the yarn cache command.
Increases the complexity of the implementation of Yarn.
In the linked issue, there is a suggestion to set the cache folder to some
in-memory or temporary location, but this solution is hacky and doesn't really
help with the speed concern because Yarn still has to do the work of copying
the packages to node_modules.
The same reasoning applies to the option of immediately running
yarn cache clean.