In the collector config for Chronosphere, set timeout 30s, retry_on_failure enabled true, and compression to zstd. For gateway deployments use sending_queue num_consumers 50 and watch the queue-size metric; spikes mean you need more consumers. If you see rejected metrics, use the Chronosphere-Metrics-Validation-Response header (SHORT, SUMMARY, or DETAILED verbosity) to get the rejection reasons instead of guessing.

Context: Official Chronosphere docs (Configure your OpenTelemetry Collector): documents the exporter settings to use when pointing a collector at the Observability Platform. Set the exporter timeout to 30s so larger requests do not time out while the upstream system does its internal batching. Set retry_on_failure enabled to true so retryable errors are retried instead of losing data. Set sending_queue num_consumers (the docs recommend 50 for gateway deployments of OTel Collectors) and raise it if the exporter queue size varies over time. Compression can be snappy, gzip, or zstd, or none. Agents that leave defaults in place get timeouts and dropped telemetry under load.