MonitoredResourceUtil#getResource non-uniformly throws exceptions on missing/malformed values

[artem-karpenko]

I use MonitoredResourceUtil#getResource to load resource in our log4j2 appender implementation (basically migrated logback implementation). Container in which application is running is missing some of the expected labels for "k8s_container" resource type, namely the zone is missing which leads to getCloudRunLocation to throw NPE.

Shouldn't getResource behave lenient against this kind of problems and try to initialize and work with as much data as possible? Especially considering that missing most of the labels will indeed not cause any issues - it's only some of them that lead to exceptions.

[minherz]

hi Artem, thank you for the question. The resource type that you mention "k8s_container" expects to be identified because the process is running in the managed environment i.e. GKE. It means that all metadata is expected to be present. The lack of the metadata means either the resource type was identified incorrectly or there is incompatibility between the library expectations and the environment. In both cases we would like to know about it. Passing empty values hides the problem.
You are right about code throwing NPE is incorrect behavior and has to be fixed. Having said that, can you elaborate more about the deployment which enforces a container without the HOSTNAME envionrment variable. It sounds an interesting corner case and can be useful for us in future.

[minherz]

Hi Artem, we have recently refactored the implementation that populates MonitoredResource with resource type and metadata (#708). Please, use the latest (i think 3.3.0) version of the library.