Worker exhaustion during blocking I/O calls in Swan Lake runtime

Concurrency Model Constraints

Ballerina's Swan Lake architecture employs a worker-based concurrency model where requests are managed via lightweight strands. While this design aims to reduce OS-level thread switching overhead, the interaction between the event loop and blocking I/O operations can impact service availability.

Resource Contention

When a service executes synchronous blocking calls within a high-concurrency environment, the available worker pool may become saturated. This behavior can lead to increased latency and request queuing without necessarily triggering a crash or a specific runtime exception.

Given the isolation of state within workers, there is uncertainty regarding the optimal configuration of worker thread limits in ballerina.toml to prevent total service unavailability during I/O spikes.

• What is the specific threshold at which the worker pool triggers request queuing?
• How does the runtime prioritize new incoming requests when all available strands are blocked by synchronous I/O?