feat!: split relay configuration between production and staging (#2425)

- [x] regular tests
- [x] netsim (nothing to do, it uses its own local relays)
- [ ] ~~CLI integration tests~~

## Breaking Changes

- `iroh_net::defaults` is now split into `prod` and `staging`
- added `iroh_net::RelayMode::Staging`
- `iroh_net::discovery::dns:: N0_DNS_NODE_ORIGIN` is now
`N0_DNS_NODE_ORIGIN_PROD` and there is now `N0_DNS_NODE_ORIGIN_STAGING`

Closes #2420

iroh-cli/src/config.rs

@@ -10,7 +10,7 @@ use std::{
 
 use anyhow::{anyhow, bail, Context, Result};
 use iroh::net::{
-    defaults::{default_eu_relay_node, default_na_relay_node},
+    defaults,
     relay::{RelayMap, RelayNode},
 };
 use iroh::node::GcPolicy;
@@ -62,6 +62,10 @@ pub(crate) struct NodeConfig {
 
 impl Default for NodeConfig {
     fn default() -> Self {
+        #[cfg(not(test))]
+        use defaults::prod::{default_eu_relay_node, default_na_relay_node};
+        #[cfg(test)]
+        use defaults::staging::{default_eu_relay_node, default_na_relay_node};
         Self {
             // TODO(ramfox): this should probably just be a relay map
             relay_nodes: [default_na_relay_node(), default_eu_relay_node()].into(),

iroh-dns-server/examples/publish.rs

@@ -4,7 +4,7 @@ use anyhow::{bail, Result};
 use clap::{Parser, ValueEnum};
 use iroh_net::{
     discovery::{
-        dns::N0_DNS_NODE_ORIGIN,
+        dns::N0_DNS_NODE_ORIGIN_PROD,
         pkarr_publish::{PkarrRelayClient, N0_DNS_PKARR_RELAY},
     },
     dns::node_info::{to_z32, NodeInfo, IROH_TXT_NAME},
@@ -85,7 +85,10 @@ async fn main() -> Result<()> {
     match args.env {
         Env::Default => {
             println!("   cargo run --example resolve -- node {}", node_id);
-            println!("   dig {} TXT", fmt_domain(&node_id, N0_DNS_NODE_ORIGIN))
+            println!(
+                "   dig {} TXT",
+                fmt_domain(&node_id, N0_DNS_NODE_ORIGIN_PROD)
+            )
         }
         Env::Dev => {
             println!(

iroh-dns-server/examples/resolve.rs

@@ -6,7 +6,7 @@ use hickory_resolver::{
     AsyncResolver,
 };
 use iroh_net::{
-    discovery::dns::N0_DNS_NODE_ORIGIN,
+    discovery::dns::N0_DNS_NODE_ORIGIN_PROD,
     dns::{node_info::TxtAttrs, DnsResolver},
     NodeId,
 };
@@ -45,7 +45,7 @@ async fn main() -> anyhow::Result<()> {
     let (resolver, origin) = match args.env {
         Env::Default => (
             iroh_net::dns::default_resolver().clone(),
-            N0_DNS_NODE_ORIGIN,
+            N0_DNS_NODE_ORIGIN_PROD,
         ),
         Env::Dev => (
             resolver_with_nameserver(LOCALHOST_DNS.parse()?),

iroh-gossip/examples/chat.rs

@@ -298,7 +298,8 @@ impl FromStr for Ticket {
 fn fmt_relay_mode(relay_mode: &RelayMode) -> String {
     match relay_mode {
         RelayMode::Disabled => "None".to_string(),
-        RelayMode::Default => "Default Relay servers".to_string(),
+        RelayMode::Default => "Default Relay (production) servers".to_string(),
+        RelayMode::Staging => "Default Relay (staging) servers".to_string(),
         RelayMode::Custom(map) => map
             .urls()
             .map(|url| url.to_string())

iroh-net/src/defaults.rs

@@ -4,11 +4,6 @@ use url::Url;
 
 use crate::relay::{RelayMap, RelayNode};
 
-/// Hostname of the default NA relay.
-pub const NA_RELAY_HOSTNAME: &str = "use1-1.relay.iroh.network.";
-/// Hostname of the default EU relay.
-pub const EU_RELAY_HOSTNAME: &str = "euw1-1.relay.iroh.network.";
-
 /// The default STUN port used by the Relay server.
 ///
 /// The STUN port as defined by [RFC
@@ -24,34 +19,88 @@ pub const DEFAULT_HTTPS_PORT: u16 = 443;
 /// The default metrics port used by the Relay server.
 pub const DEFAULT_METRICS_PORT: u16 = 9090;
 
-/// Get the default [`RelayMap`].
-pub fn default_relay_map() -> RelayMap {
-    RelayMap::from_nodes([default_na_relay_node(), default_eu_relay_node()])
-        .expect("default nodes invalid")
-}
+/// Production configuration.
+pub mod prod {
+    use super::*;
+
+    /// Hostname of the default NA relay.
+    pub const NA_RELAY_HOSTNAME: &str = "use1-1.relay.iroh.network.";
+    /// Hostname of the default EU relay.
+    pub const EU_RELAY_HOSTNAME: &str = "euw1-1.relay.iroh.network.";
+
+    /// Get the default [`RelayMap`].
+    pub fn default_relay_map() -> RelayMap {
+        RelayMap::from_nodes([default_na_relay_node(), default_eu_relay_node()])
+            .expect("default nodes invalid")
+    }
+
+    /// Get the default [`RelayNode`] for NA.
+    pub fn default_na_relay_node() -> RelayNode {
+        // The default NA relay server run by number0.
+        let url: Url = format!("https://{NA_RELAY_HOSTNAME}")
+            .parse()
+            .expect("default url");
+        RelayNode {
+            url: url.into(),
+            stun_only: false,
+            stun_port: DEFAULT_STUN_PORT,
+        }
+    }
 
-/// Get the default [`RelayNode`] for NA.
-pub fn default_na_relay_node() -> RelayNode {
-    // The default NA relay server run by number0.
-    let url: Url = format!("https://{NA_RELAY_HOSTNAME}")
-        .parse()
-        .expect("default url");
-    RelayNode {
-        url: url.into(),
-        stun_only: false,
-        stun_port: DEFAULT_STUN_PORT,
+    /// Get the default [`RelayNode`] for EU.
+    pub fn default_eu_relay_node() -> RelayNode {
+        // The default EU relay server run by number0.
+        let url: Url = format!("https://{EU_RELAY_HOSTNAME}")
+            .parse()
+            .expect("default_url");
+        RelayNode {
+            url: url.into(),
+            stun_only: false,
+            stun_port: DEFAULT_STUN_PORT,
+        }
     }
 }
 
-/// Get the default [`RelayNode`] for EU.
-pub fn default_eu_relay_node() -> RelayNode {
-    // The default EU relay server run by number0.
-    let url: Url = format!("https://{EU_RELAY_HOSTNAME}")
-        .parse()
-        .expect("default_url");
-    RelayNode {
-        url: url.into(),
-        stun_only: false,
-        stun_port: DEFAULT_STUN_PORT,
+/// Staging configuration.
+///
+/// Used by tests and might have incompatible changes deployed
+pub mod staging {
+    use super::*;
+
+    /// Hostname of the default NA relay.
+    pub const NA_RELAY_HOSTNAME: &str = "staging-use1-1.relay.iroh.network.";
+    /// Hostname of the default EU relay.
+    pub const EU_RELAY_HOSTNAME: &str = "staging-euw1-1.relay.iroh.network.";
+
+    /// Get the default [`RelayMap`].
+    pub fn default_relay_map() -> RelayMap {
+        RelayMap::from_nodes([default_na_relay_node(), default_eu_relay_node()])
+            .expect("default nodes invalid")
+    }
+
+    /// Get the default [`RelayNode`] for NA.
+    pub fn default_na_relay_node() -> RelayNode {
+        // The default NA relay server run by number0.
+        let url: Url = format!("https://{NA_RELAY_HOSTNAME}")
+            .parse()
+            .expect("default url");
+        RelayNode {
+            url: url.into(),
+            stun_only: false,
+            stun_port: DEFAULT_STUN_PORT,
+        }
+    }
+
+    /// Get the default [`RelayNode`] for EU.
+    pub fn default_eu_relay_node() -> RelayNode {
+        // The default EU relay server run by number0.
+        let url: Url = format!("https://{EU_RELAY_HOSTNAME}")
+            .parse()
+            .expect("default_url");
+        RelayNode {
+            url: url.into(),
+            stun_only: false,
+            stun_port: DEFAULT_STUN_PORT,
+        }
     }
 }

iroh-net/src/discovery/dns.rs

@@ -9,8 +9,10 @@ use crate::{
     Endpoint, NodeId,
 };
 
-/// The n0 testing DNS node origin
-pub const N0_DNS_NODE_ORIGIN: &str = "dns.iroh.link";
+/// The n0 testing DNS node origin, for production.
+pub const N0_DNS_NODE_ORIGIN_PROD: &str = "dns.iroh.link";
+/// The n0 testing DNS node origin, for testing.
+pub const N0_DNS_NODE_ORIGIN_STAGING: &str = "staging-dns.iroh.link";
 const DNS_STAGGERING_MS: &[u64] = &[200, 300];
 
 /// DNS node discovery
@@ -44,9 +46,17 @@ impl DnsDiscovery {
         Self { origin_domain }
     }
 
-    /// Create a new DNS discovery which uses the [`N0_DNS_NODE_ORIGIN`] origin domain.
+    /// Create a new DNS discovery which uses the [`N0_DNS_NODE_ORIGIN_PROD`] origin domain and in testing
+    /// uses [`N0_DNS_NODE_ORIGIN_STAGING`].
     pub fn n0_dns() -> Self {
-        Self::new(N0_DNS_NODE_ORIGIN.to_string())
+        #[cfg(not(any(test, feature = "test-utils")))]
+        {
+            Self::new(N0_DNS_NODE_ORIGIN_PROD.to_string())
+        }
+        #[cfg(any(test, feature = "test-utils"))]
+        {
+            Self::new(N0_DNS_NODE_ORIGIN_STAGING.to_string())
+        }
     }
 }
 

iroh-net/src/dns.rs

@@ -381,7 +381,7 @@ async fn stagger_call<T, F: Fn() -> Fut, Fut: Future<Output = Result<T>>>(
 pub(crate) mod tests {
     use std::sync::atomic::AtomicUsize;
 
-    use crate::defaults::NA_RELAY_HOSTNAME;
+    use crate::defaults::staging::NA_RELAY_HOSTNAME;
 
     use super::*;
     const TIMEOUT: Duration = Duration::from_secs(5);

iroh-net/src/dns/node_info.rs

@@ -12,7 +12,8 @@
 //! - `<z32-node-id>` is the [z-base-32] encoding of the [`NodeId`].
 //!
 //! - `<origin-domain>` is the domain name of the publishing DNS server,
-//!   [`N0_DNS_NODE_ORIGIN`] is the server operated by number0.
+//!   [`N0_DNS_NODE_ORIGIN_PROD`] is the server operated by number0 for production.
+//!   [`N0_DNS_NODE_ORIGIN_STAGING`] is the server operated by number0 for testing.
 //!
 //! - `TXT` is the DNS record type.
 //!
@@ -28,7 +29,8 @@
 //! [z-base-32]: https://philzimmermann.com/docs/human-oriented-base-32-encoding.txt
 //! [RFC1464]: https://www.rfc-editor.org/rfc/rfc1464
 //! [`RelayUrl`]: iroh_base::node_addr::RelayUrl
-//! [`N0_DNS_NODE_ORIGIN`]: crate::discovery::dns::N0_DNS_NODE_ORIGIN
+//! [`N0_DNS_NODE_ORIGIN_PROD`]: crate::discovery::dns::N0_DNS_NODE_ORIGIN_PROD
+//! [`N0_DNS_NODE_ORIGIN_STAGING`]: crate::discovery::dns::N0_DNS_NODE_ORIGIN_STAGING
 
 use std::{
     collections::{BTreeMap, BTreeSet},

iroh-net/src/endpoint.rs

@@ -28,7 +28,7 @@ use tracing::{debug, info_span, trace, warn};
 use url::Url;
 
 use crate::{
-    defaults::default_relay_map,
+    defaults,
     discovery::{Discovery, DiscoveryTask},
     dns::{default_resolver, DnsResolver},
     key::{PublicKey, SecretKey},
@@ -85,9 +85,15 @@ pub struct Builder {
 
 impl Default for Builder {
     fn default() -> Self {
+        // Use staging in testing
+        #[cfg(not(any(test, feature = "test-utils")))]
+        let relay_mode = RelayMode::Default;
+        #[cfg(any(test, feature = "test-utils"))]
+        let relay_mode = RelayMode::Staging;
+
         Self {
             secret_key: Default::default(),
-            relay_mode: RelayMode::Default,
+            relay_mode,
             alpn_protocols: Default::default(),
             transport_config: Default::default(),
             concurrent_connections: Default::default(),
@@ -118,7 +124,8 @@ impl Builder {
     pub async fn bind(self, bind_port: u16) -> Result<Endpoint> {
         let relay_map = match self.relay_mode {
             RelayMode::Disabled => RelayMap::empty(),
-            RelayMode::Default => default_relay_map(),
+            RelayMode::Default => defaults::prod::default_relay_map(),
+            RelayMode::Staging => defaults::staging::default_relay_map(),
             RelayMode::Custom(relay_map) => {
                 ensure!(!relay_map.is_empty(), "Empty custom relay server map",);
                 relay_map
@@ -369,7 +376,7 @@ impl Endpoint {
 
     // # Methods relating to construction.
 
-    /// Returns the builder for an [`Endpoint`].
+    /// Returns the builder for an [`Endpoint`], with a production configuration.
     pub fn builder() -> Builder {
         Builder::default()
     }

iroh-net/src/magicsock.rs

@@ -2750,7 +2750,7 @@ mod tests {
     use iroh_test::CallOnDrop;
     use rand::RngCore;
 
-    use crate::{defaults::EU_RELAY_HOSTNAME, relay::RelayMode, tls, Endpoint};
+    use crate::{defaults::staging::EU_RELAY_HOSTNAME, relay::RelayMode, tls, Endpoint};
 
     use super::*;
 

iroh-net/src/netcheck.rs

@@ -785,7 +785,7 @@ mod tests {
     use tokio::time;
     use tracing::info;
 
-    use crate::defaults::{DEFAULT_STUN_PORT, EU_RELAY_HOSTNAME};
+    use crate::defaults::{staging::EU_RELAY_HOSTNAME, DEFAULT_STUN_PORT};
     use crate::ping::Pinger;
     use crate::relay::RelayNode;
 

iroh-net/src/netcheck/reportgen.rs

@@ -1128,7 +1128,7 @@ mod tests {
 
     use super::*;
 
-    use crate::defaults::{default_eu_relay_node, default_na_relay_node};
+    use crate::defaults::staging::{default_eu_relay_node, default_na_relay_node};
 
     #[test]
     fn test_update_report_stun_working() {

iroh-net/src/netcheck/reportgen/probes.rs

@@ -466,7 +466,7 @@ fn sort_relays<'a>(
 mod tests {
     use pretty_assertions::assert_eq;
 
-    use crate::defaults::default_relay_map;
+    use crate::defaults::staging::default_relay_map;
     use crate::netcheck::RelayLatencies;
 
     use super::*;

iroh-net/src/relay/map.rs

@@ -14,8 +14,10 @@ use super::RelayUrl;
 pub enum RelayMode {
     /// Disable relay servers completely.
     Disabled,
-    /// Use the default relay map, with relay servers from n0.
+    /// Use the default relay map, with production relay servers from n0.
     Default,
+    /// Use the staging relay servers from n0.
+    Staging,
     /// Use a custom relay map.
     Custom(RelayMap),
 }

iroh/src/node/builder.rs

@@ -183,13 +183,19 @@ fn mk_external_rpc() -> BoxedServerEndpoint {
 
 impl Default for Builder<iroh_blobs::store::mem::Store> {
     fn default() -> Self {
+        // Use staging in testing
+        #[cfg(not(any(test, feature = "test-utils")))]
+        let relay_mode = RelayMode::Default;
+        #[cfg(any(test, feature = "test-utils"))]
+        let relay_mode = RelayMode::Staging;
+
         Self {
             storage: StorageConfig::Mem,
             bind_port: None,
             secret_key: SecretKey::generate(),
             blobs_store: Default::default(),
             keylog: false,
-            relay_mode: RelayMode::Default,
+            relay_mode,
             dns_resolver: None,
             rpc_endpoint: mk_external_rpc(),
             rpc_port: None,
@@ -210,13 +216,19 @@ impl<D: Map> Builder<D> {
         docs_storage: DocsStorage,
         storage: StorageConfig,
     ) -> Self {
+        // Use staging in testing
+        #[cfg(not(any(test, feature = "test-utils")))]
+        let relay_mode = RelayMode::Default;
+        #[cfg(any(test, feature = "test-utils"))]
+        let relay_mode = RelayMode::Staging;
+
         Self {
             storage,
             bind_port: None,
             secret_key: SecretKey::generate(),
             blobs_store,
             keylog: false,
-            relay_mode: RelayMode::Default,
+            relay_mode,
             dns_resolver: None,
             rpc_endpoint: mk_external_rpc(),
             rpc_port: None,