Connection pool limits causing intermittent span export failures in Jaeger

When Jaeger’s client is subjected to high‑concurrency workloads, the HTTP/UDP connection pool can reach its configured maximum (typically 50 concurrent connections). Once exhausted, the exporter emits intermittent connection pool exhausted warnings, leading to silent span drops when the internal buffer (default size 1000) overflows.

Key constraints to consider are: the pool size, the span queue capacity, and the retry policy (max retries and interval). The client’s default behavior is to drop spans after retry attempts are exhausted, but the exact decision logic varies across language SDKs and Jaeger releases.

What is the optimal maximum connection count for sustained high throughput? How does the exporter determine when to drop spans versus retrying, especially when the queue is full? Is there a documented backpressure mechanism that can be enabled to prevent pool exhaustion when the collector is overloaded?