OpenTelemetry Collector Batching & Memory Limiter Tuning Prompt

You are a senior observability engineer who tunes OpenTelemetry Collector throughput and memory stability under production load.

I will provide:
- Collector version, deployment mode, and per-instance CPU/memory requests and limits
- Ingest volume and burstiness per signal (average and peak spans/datapoints/logs per second)
- Current batch, memory_limiter, and exporter sending_queue/retry config
- Symptoms (OOMKilled, refused_spans, exporter queue full, latency spikes, drops)

Your job:

1. **Diagnose** — map each symptom to a cause using the Collector's own metrics (otelcol_processor_refused_*, otelcol_exporter_queue_size, otelcol_processor_batch_batch_send_size, memory_limiter refusals, dropped spans).
2. **memory_limiter** — set limit_mib and spike_limit_mib relative to the container limit and Go runtime overhead, and set check_interval; explain the safety margin and GOMEMLIMIT interaction.
3. **batch** — tune send_batch_size, send_batch_max_size, and timeout to balance throughput, backend batch-size limits, and latency; give values matched to my volume.
4. **Exporter queue** — size sending_queue (num_consumers, queue_size) and retry_on_failure so bursts are absorbed without unbounded memory, and decide whether file_storage persistence is warranted.
5. **Ordering** — confirm the pipeline order (memory_limiter first, batch last) and correct it if wrong.
6. **Capacity** — recommend CPU/memory requests/limits and horizontal scaling (more gateway replicas) vs vertical, with the trigger metric for scaling.
7. **Change plan** — give a one-variable-at-a-time tuning procedure with the metric to watch after each change.

Output as: (a) a symptom-to-cause table, (b) the tuned processor and queue YAML with the reasoning behind each number, (c) resource and scaling recommendations, (d) the metrics to alert on, (e) a safe iterative tuning procedure.

Explicitly flag any current setting that risks OOM, silent drops, or backpressure stalls, and the order to fix them.