Proper clearing resources with FallbackCredentialsFactory

[DmitryLukyanov]

I'm using FallbackCredentialsFactory as a helper that provides aws credentials. All works fine, but I have a question about FallbackCredentialsFactory.Reset functionality. It simply clears cachedCredentials and reassigns CredentialsGenerators, but it doesn't properly clear underlying resources. For example, ECSTaskCredentials is disposable, but Reset doesn't clear its resources, instead it simply reassigns it. Same for other credentials generators. Is there expected way to call Reset without possible memory leaks?

[ashishdhingra]

@DmitryLukyanov Looking at the logic in FallbackCredentialsFactory:

• CredentialsGenerator is declared as a delegate.
• CredentialsGenerators is a list of CredentialsGenerator delegates.
• FallbackCredentialsFactory.GetCredentials() iterates through the list of delegates and executed them one by one until credentials are resolved.
  • For each of the executed generator, the exception is thrown and the next generator is tried.
    • BCL AppConfigAWSCredentials throws exception in constructor.
    • AssumeRoleWithWebIdentityCredentials.FromEnvironmentVariables() tries to construct AssumeRoleWithWebIdentityCredentials object, which throws exception in the constructor if certain conditions are not met.
    • GetAWSCredentials(ICredentialProfileSource source) throws exception if the profile could not be found, or, CredentialProfile. GetAWSCredentials() > AWSCredentialsFactory.GetAWSCredentials() throws an exception.
    • EnvironmentVariablesAWSCredentials() throws exception in constructor via call to FetchCredentials().
    • ECSEC2CredentialsWrapper(proxy) throws exception if certain conditions are not met, else returns new instance of ECSTaskCredentials(), OR, DefaultInstanceProfileAWSCredentials.Instance.

In each of the cases, the next generator is tried only if the previous generator returns exception in the construction process (in which case the object is never created), with ECSTaskCredentials being the last. The FallbackCredentialsFactory.Reset() essentially sets the only created instance of cachedCredentials to null, the object would then be claimed by garbage collector eventually.

Please let me know if this provides detailed flow, or, if by any chance certain condition is missed and needs to be evaluated.

Thanks,
Ashish

[DmitryLukyanov]

Hey @ashishdhingra ,

what will happen if let's say we have caught an exception when we call GetCredentials for previously cached ECSTaskCredentials? I assume in this case, the cached value here should be cleared (via Reset?) and sdk should start lookup credentialsGenerator again to find available credentials generator (EC2/maybe newly configured env variables and etc). But if so, the resources in previously created and failed instance of ECSTaskCredentials won't be properly disposed.

UPDATE: it looks like all resources including inner semaphore in ECSTaskCredentials will be eventually disposed as you told above, I just expected calling Dispose directly which I would consider more "best practice"-like approach.
I will mark this question as answered.

[github-actions[bot]]

Hello! Reopening this discussion to make it searchable.