Processor test suite gaps: no coverage for tokio-dependent exporters or runtime deadlock scenarios

[bryantbiggs]

Summary

While working on #3380 (fix for #2802), I noticed that the existing test suite for BatchSpanProcessor, BatchLogProcessor, and PeriodicReader has significant coverage gaps around tokio runtime interactions. The deadlock bug in #2802 went undetected because no tests exercise exporters that depend on tokio internals.

Gaps identified

1. All runtime-flavor tests use simple exporters

The existing #[tokio::test(flavor = "current_thread")] and #[tokio::test(flavor = "multi_thread", worker_threads = 1)] tests use MockSpanExporter, InMemoryLogExporter, and InMemoryMetricExporter — none of which call tokio::spawn, use tokio I/O, or otherwise depend on the tokio runtime context. These tests pass regardless of whether the runtime context is available on the processor's worker thread, so they cannot catch deadlocks like #2802.

PR #3380 adds initial regression tests with TokioSpawn*Exporter mocks for multi_thread(1), but coverage is still limited.

2. No shutdown() tests with tokio-dependent exporters

The original issue (#2802, reproduction in #3356 (comment)) documented deadlocks on both force_flush() and shutdown(). The regression tests in #3380 only cover force_flush(). The shutdown() path goes through similar code but has different timeout behavior (5s timeout vs. infinite hang) that should be verified.

3. No timeout or cancellation tests

• force_flush() on current_thread runtime with tonic-like exporters hangs forever (recv() has no timeout)
• shutdown() returns Err(Timeout(5s)) but the worker thread stays stuck
• These scenarios aren't tested or documented

4. SimpleLogProcessor has ignored async exporter tests

Two tests in simple_log_processor are #[ignore]d:

• test_simple_processor_async_exporter_with_all_runtime_worker_threads_blocked
• test_simple_processor_async_exporter_with_current_thread_runtime

These suggest known but unaddressed issues with async exporters on constrained runtimes in the simple processor path.

5. No integration tests with real exporters

All processor tests use in-memory or mock exporters. There are no integration tests that exercise the actual OTLP/tonic export path against a mock collector, which is where deadlocks manifest in practice.

Suggested improvements

• Add shutdown() regression tests with TokioSpawn*Exporter mocks for all three processors
• Add tests verifying timeout behavior when export hangs (both force_flush and shutdown)
• Investigate and un-ignore the SimpleLogProcessor async exporter tests
• Consider adding integration tests with tonic exporter against a mock/in-process OTLP collector
• Document the current_thread runtime limitation (exporters using tokio::spawn will deadlock when force_flush/shutdown blocks the only runtime thread)

Context

• Fix PR: fix(sdk): resolve exporter deadlock on constrained tokio runtimes #3380
• Original discussion: fix(sdk): resolve OTLP exporter deadlock on single-threaded tokio runtimes #3356
• Root issue: OTLP MetricExporter deadlock issue #2802