The HTTP metrics histogram has a lot of things that in a production environment are likely not useful, and will cause memory bloat.
http_server_request_duration_milliseconds_bucket{http_method="POST",http_route="/api/v2/series",http_scheme="http",http_status_code="202",http_target="/api/v2/series",net_host_name="datadog-receiver",net_host_port="8080",net_protocol_version="1.1",net_sock_peer_addr="10.43.128.181",net_sock_peer_port="51112",otel_scope_name="github.com/Cyprinus12138/otelgin",otel_scope_version="0.49.0",user_agent_original="datadog-agent/7.52.0",le="0"} 0
The net_sock_peer_addr
and net_sock_peer_port
and user_agent_original
is likely to make pain happen when there are millions of unique combinations of these items.
Perhaps a way to filter the metrics as an option, either by marking what to include or exclude might be useful?
It's probably OK for traces, as those are more or less stand-alone, but the size metrics would generally be used per-endpoint I would think, not per something-the-user-controls-and-can-explode-my-server-memory.
If you'd like me to take a stab at this, I'm likely to implement a callback function that will take the metric keys and return a new list that has whatever I want in there. This would allow me to filter as well as add should I want to.