BearerToken error handling fail

[hporchet]

When the application of the metadata is in error, the fail method crash : apply() or fail() already called and hide the exception of the apply method.

public class BearerToken extends CallCredentials {
    ...

    @Override
    public void applyRequestMetadata(RequestInfo requestInfo, Executor executor, MetadataApplier applier) {
        executor.execute(() -> {
            try {
                Metadata headers = new Metadata();
                headers.put(META_DATA_KEY, header);
                applier.apply(headers); // crash
            } catch (Throwable e) {
                applier.fail(Status.UNAUTHENTICATED.withCause(e)); // can't be called after crashed, error : apply() or fail() already called
            }
        });
    }
}

[tstirrat15]

@hporchet do you have a reproducible example? I'd like to write a test around this if possible.

It seems as though this could happen if there were nested/sequential withCallCredentials calls or some other piece of code that's making an apply call. Otherwise the apply seems to be just going and doing a hashmap merge, which I wouldn't expect to error in normal operation.

Where are you seeing the problem manifest?

[hporchet]

I have seen this problem when i was on windows and the server return NOT_IMPLEMENTED to all grpc call. The request was correctly send but the not implemented raise a STATUS_EXCEPTION at the apply level.

[tstirrat15]

Hmm... That makes sense. We probably shouldn't be catching all Throwables there either. Lemme see if I can narrow that sufficiently.

[hporchet]

I think a simple log of the error before call fail() on debug mode seems enough, it just need to be easily find when a problem occur.