TTL Cache

CircleCI's caching lets you cache directories based on checksum files say package.json or go.mod. However in some scenarios it makes sense to have a cache with a TTL. This is helpful when the cache needs to be expired in some time period in order to generate a newer cache so that builds continue to use the cache effeciently. A good example of this would be things like webpack builds, golangci-lint, eslint etc.

tip

Use a cache with a TTL everywhere that an operation is based on many files in the repo and the inbuilt command manages its own cache. E.g) a TTL cache is very useful in the case of golangci-lint.

Usage

Simply use Swissknife ttl cache instead of restore_cache and save_cache.

Include the orb in your config as follows.

version: 2.1

orbs:

swissknife: roopakv/[email protected] // Grab the latest version from https://orb.swissknife.dev

Restore Cache

The following step is used to restore cache. The cache key prefix is a string similar to the key you would pass to restore_cache or save_cache.

- swissknife/ttl-cache:

type: restore

cache-key-prefix: go-mod-v1-{{ checksum "go.mod" }}-{{ checksum "go.sum" }}

Save Cache

The following step is used to save cache. The cache key prefix is a string similar to the key you would pass to restore_cache or save_cache. The save-cache-directory-1, and similarly 2 and 3 are the directories to save in the cache. Currently ttl-cache allows saving upto 3 directories, All directories saved are saved recursively. These are very similar to the paths field in the native save_cache command.

- swissknife/ttl-cache:

type: save

cache-key-prefix: go-mod-v1-{{ checksum "go.mod" }}-{{ checksum "go.sum" }}

save-cache-directory-1: /path/to/save

save-cache-directory-2: /another/path/to/save

Branch based cache behavior

In many cases including the branch in the cache key can be somewhat helpful and the ttl-cache. When including the branch in the cache there are two ways

• simple branch inclusion
• trust only base branch

The simple branch inclusions means that there is a cache per branch. In many cases this is simple and just works.

The trust only base branch is very helpful in certain cases. Basically caches generated on the base branch (master, main etc.) are restored. This way a branches for a pull request get the benefit of caches, but also a cache generated on that branch is never used by the base branch keeping the base branch clean. This approach is the default approach and is best for things like lint cache, build cache etc.