XdsNameResolver blindly propagates XdsClient errors

[ejona86]

If a watch in XdsClient fails, XdsNameResolver will propagate the error directly to the channel without adding context and without setting a proper status.

grpc-java/xds/src/main/java/io/grpc/xds/XdsNameResolver.java

Line 725 in 3b9ff36

listener.onError(error);

A user saw

 [Channel<2441>: (xds:///<snip>)] Failed to resolve name. status=Status{code=NOT_FOUND, description=Requested entity was not found., cause=null}

And accompanying RPC failures:

Caused by: java.lang.RuntimeException: io.grpc.StatusRuntimeException with message: NOT_FOUND: Requested entity was not found.

That RPC failure is very bad, because it isn't clear it is from XdsNR and should be UNAVAILABLE.

As seen in b/221844501.

[YifeiZhuang]

Hi @ejona86 , curious how did you encounter the error? Shall we log the real error and always throw "UNAVAILABLE" to the client, while copying the cause and description details?

[ejona86]

At the least we should wrap and repackage the error. For example: "UNAVAILABLE: Unable to load Route 'Foo'. xDS server returned: NOT_FOUND: Requested entity was not found." This is unrelated to logging.

[ejona86]

This is part of the (now-available) v1.45.0 release.