cache.saveCache should raise an error on reserve failure, but doesn't
Describe the bug
https://github.com/actions/toolkit/blob/main/packages/cache/src/cache.ts#L157-L165
The function signature for cache.saveCache mentions:
@returns number returns cacheId if the cache was saved successfully and throws an error if save fails
When there is a failure to reserve the cache key, this exception is caught, and a job warning annotation is set, but the exception is not propagated. Therefore the error is not actually returned as per the defined interface of the function. This means that reservation failures wrapped in a try/catch block will not execute the exception handler in the caller function.
I believe both of these code paths should re-raise the exception. Otherwise it is being swallowed and the only way to know there was a problem, is to check that the cacheId is still -1:
https://github.com/actions/toolkit/blob/main/packages/cache/src/cache.ts#L245-L248
As it stands, we get a log message like this:
Failed to save: Unable to reserve cache with key XXXX, another job may be creating this cache. More details: Cache already exists. Scope: refs/heads/master, Key: XXXX, Version: 50059a988699d11b5543e65bd21ddbb819f05df288843af627d295d329cbda93
But no error is thrown, and the build continues despite the failure.
To Reproduce Steps to reproduce the behavior:
- Use actions/cache in a github actions workflow, using the cache.saveCache function to save a value at a static key.
- Create a race condition using your own creativity, ensuring that multiple actions runs are saving the cache concurrently with the same key.
- Await one of the actions to fail to save the cache.
- See error
Expected behavior
As per the function documentation, the exception should be thrown rather than logged and suppressed.
Screenshots N/A
Desktop (please complete the following information):
- OS: N/A
- Browser N/A
- Version N/A
This is not a client side error. It occurs in the latest released version of actions/cache.
Additional context
N/A
This may be tangentially related to #658