Resource Detectors produce blank Exception() after 5 seconds

[jeremydvoss]

Describe your environment Resource Detector only get 5 seconds by default to run. However, instead of ending the process after 5 seconds, a blank "Exception()" is creating resulting in the follow warning: Exception in detector <opentelemetry.resource.detector.<RESOURCE DETECTOR CLASS> object at 0x000002B955D43D60>, ignoring Note that the Exception string is blank resulting in 2 whitespaces: Exception__in... This blank exception is causing confusion among Azure customers. And perhaps more importantly, the process still hangs.

  Steps to reproduce

1. Create a resource detector that sleeps for more than 5 seconds (try 10-20):

from opentelemetry.sdk.resources import Resource, ResourceDetector

from time import sleep

class SleepingResourceDetector(ResourceDetector):
    # pylint: disable=no-self-use
    def detect(self) -> "Resource":
        sleep(20)
        attributes = {}
        return Resource(attributes)

2. Set resource detector entry point in pyproject.toml

[project.entry-points.opentelemetry_resource_detector]
sleeping = "opentelemetry.resource.detector.sleeping:SleepingResourceDetector"

3. Install package to install entry point
4. Point sdk to resource detector: export OTEL_EXPERIMENTAL_RESOURCE_DETECTORS=sleeping
5. Create a Resource: Resource.create() This will eventually call get_aggregated_resources

What is the expected behavior? The "concurrent future" setup should gracefully exit the process and not print out a confusing blank warning.

What is the actual behavior? Process hangs and confusing warning message with blank error.

Additional context Add any other context about the problem here.

[ocelotl]

Reopening since #3645 did not fully fix this issue.

[aabmass]

This seems similar to https://github.com/open-telemetry/opentelemetry-python/issues/3309 when processors or exporters block, we have no way to cancel them.

[jeremydvoss]

Some options for adding a timeout inside resource detectors:

• Add subtype like ResourceDetectorWithTimeout
• Unified Resource Detector Timeout env var
• Add method to base class getTimeout() that can be but doesn't need to be used by detectors

[jeremydvoss]

A couple more issues:

• While processes run in parallel, timeout applies sequentially. Detectors waited for last can take the longest
• We wait for all to finish before allowing app to continue. Even if we let the process finish, we should discard its result and let the app continue in parallel. @aabmass mentioned this may be possible by changing how the with block works. Will look through the docs.