How can I identify the indexing bottleneck in Elasticsearch using the nodes stats API before optimization?

Before optimizing an Elasticsearch cluster, I need to confirm where the indexing bottleneck resides. I plan to query the nodes stats API for metrics such as indexing_pressure, thread_pool queue sizes, and merge throttling times, while also sampling hot threads to see CPU contention. However, I am unsure which combination of these metrics most reliably indicates the limiting factor under sustained write load.

Which specific metric from the nodes stats API should I monitor to detect indexing pressure, and how can I correlate it with the hot threads output to validate the bottleneck?