feat(metrics)!: expose identify metrics for connected peers only

Previously we would increase a counter / gauge / histogram on each received identify information. These metrics are missleading, as e.g. they depend on the identify interval and don't represent the set of currently connected peers.

With this commit, identify information is tracked for the currently connected peers only. Instead of an increase on each received identify information, metrics represent the status quo (Gauge).

Example:

```
\# HELP libp2p_libp2p_identify_remote_protocols Number of connected nodes supporting a specific protocol, with "unrecognized" for each peer supporting one or more unrecognized protocols...
\# TYPE libp2p_libp2p_identify_remote_protocols gauge
libp2p_libp2p_identify_remote_protocols_total{protocol="/ipfs/id/push/1.0.0"} 1
libp2p_libp2p_identify_remote_protocols_total{protocol="/ipfs/id/1.0.0"} 1
libp2p_libp2p_identify_remote_protocols_total{protocol="/ipfs/ping/1.0.0"} 1
libp2p_libp2p_identify_remote_protocols_total{protocol="unrecognized"} 1
\# HELP libp2p_libp2p_identify_remote_listen_addresses Number of connected nodes advertising a specific listen address...
\# TYPE libp2p_libp2p_identify_remote_listen_addresses gauge
libp2p_libp2p_identify_remote_listen_addresses_total{listen_address="/ip4/tcp"} 1
libp2p_libp2p_identify_remote_listen_addresses_total{listen_address="/ip4/udp/quic"} 1
\# HELP libp2p_libp2p_identify_local_observed_addresses Number of connected nodes observing the local node at a specific address...
\# TYPE libp2p_libp2p_identify_local_observed_addresses gauge
libp2p_libp2p_identify_local_observed_addresses_total{observed_address="/ip4/tcp"} 1
```

Pull-Request: #3325.

Author: mxinden

Cargo.lock

@@ -12,4 +12,4 @@ hyper = { version = "0.14", features = ["server", "tcp", "http1"] }
 libp2p = { path = "../../libp2p", features = ["async-std", "metrics", "ping", "noise", "identify", "tcp", "yamux", "macros"] }
 log = "0.4.0"
 tokio = { version = "1", features = ["rt-multi-thread"] }
-prometheus-client = "0.20.0"
+prometheus-client = "0.21.0"

examples/metrics/Cargo.toml

@@ -33,7 +33,7 @@ const METRICS_CONTENT_TYPE: &str = "application/openmetrics-text;charset=utf-8;v
 
 pub(crate) async fn metrics_server(registry: Registry) -> Result<(), std::io::Error> {
     // Serve on localhost.
-    let addr = ([127, 0, 0, 1], 0).into();
+    let addr = ([127, 0, 0, 1], 8080).into();
 
     // Use the tokio runtime to run the hyper server.
     let rt = tokio::runtime::Runtime::new()?;

examples/metrics/src/http_service.rs

@@ -1,9 +1,29 @@
 ## 0.13.0 - unreleased
 
+- Previously `libp2p-metrics::identify` would increase a counter / gauge / histogram on each
+  received identify information. These metrics are misleading, as e.g. they depend on the identify
+  interval and don't represent the set of currently connected peers. With this change, identify
+  information is tracked for the currently connected peers only. Instead of an increase on each
+  received identify information, metrics represent the status quo (Gauge).
+
+  Metrics removed:
+  - `libp2p_identify_protocols`
+  - `libp2p_identify_received_info_listen_addrs`
+  - `libp2p_identify_received_info_protocols`
+  - `libp2p_identify_listen_addresses`
+
+  Metrics added:
+  - `libp2p_identify_remote_protocols`
+  - `libp2p_identify_remote_listen_addresses`
+  - `libp2p_identify_local_observed_addresses`
+
+  See [PR 3325].
+
 - Raise MSRV to 1.65.
   See [PR 3715].
 
 [PR 3715]: https://github.com/libp2p/rust-libp2p/pull/3715
+[PR 3325]: https://github.com/libp2p/rust-libp2p/pull/3325
 
 ## 0.12.0
 

misc/metrics/CHANGELOG.md

@@ -27,7 +27,8 @@ libp2p-ping = { workspace = true, optional = true }
 libp2p-relay =  { workspace = true, optional = true }
 libp2p-swarm = { workspace = true }
 libp2p-identity = { workspace = true }
-prometheus-client = "0.20.0"
+prometheus-client = { version = "0.21.0" }
+once_cell = "1.16.0"
 
 [target.'cfg(not(target_os = "unknown"))'.dependencies]
 libp2p-gossipsub =  { workspace = true, optional = true }