fix client side binds in TPU (#7455)

* fix client side binds in TPU
* add changelog entry

Author: alexpyattaev

CHANGELOG.md

@@ -40,6 +40,7 @@ Release channels have their own copy of this changelog:
   * `--skip-poh-verify`
 * Deprecated snapshot archive formats have been removed and are no longer loadable.
 * Using `--snapshot-interval-slots 0` to disable generating snapshots has been removed. Use `--no-snapshots` instead.
+* Validator will now bind all ports within provided `--dynamic-port-range`, including the client ports. A range of at least 25 ports is recommended to avoid failures to bind during startup.
 
 #### Changes
 * `--transaction-structure view` is now the default.

gossip/src/node.rs

@@ -10,9 +10,8 @@ use {
         find_available_ports_in_range,
         sockets::{
             bind_gossip_port_in_range, bind_in_range_with_config, bind_more_with_config,
-            bind_to_with_config, bind_two_in_range_with_offset_and_config,
-            localhost_port_range_for_tests, multi_bind_in_range_with_config,
-            SocketConfiguration as SocketConfig,
+            bind_two_in_range_with_offset_and_config, localhost_port_range_for_tests,
+            multi_bind_in_range_with_config, SocketConfiguration as SocketConfig,
         },
         PortRange,
     },
@@ -202,13 +201,17 @@ impl Node {
         let (alpenglow_port, alpenglow) =
             bind_in_range_with_config(bind_ip_addr, port_range, socket_config)
                 .expect("Alpenglow port bind should succeed");
-        // These are client sockets, so the port is set to be 0 because it must be ephimeral.
-        let tpu_vote_forwarding_client =
-            bind_to_with_config(bind_ip_addr, 0, socket_config).unwrap();
-        let tpu_transaction_forwarding_client =
-            bind_to_with_config(bind_ip_addr, 0, socket_config).unwrap();
-        let quic_vote_client = bind_to_with_config(bind_ip_addr, 0, socket_config).unwrap();
-        let rpc_sts_client = bind_to_with_config(bind_ip_addr, 0, socket_config).unwrap();
+        // These are "client" sockets, so they could use ephemeral ports, but we
+        // force them into the provided port_range to simplify the operations.
+        let (_, tpu_vote_forwarding_client) =
+            bind_in_range_with_config(bind_ip_addr, port_range, socket_config).unwrap();
+        let (_, tpu_transaction_forwarding_client) =
+            bind_in_range_with_config(bind_ip_addr, port_range, socket_config).unwrap();
+        let (_, quic_vote_client) =
+            bind_in_range_with_config(bind_ip_addr, port_range, socket_config).unwrap();
+
+        let (_, rpc_sts_client) =
+            bind_in_range_with_config(bind_ip_addr, port_range, socket_config).unwrap();
 
         let mut info = ContactInfo::new(
             *pubkey,

net-utils/src/sockets.rs

@@ -12,7 +12,7 @@ use {
     },
 };
 // base port for deconflicted allocations
-const BASE_PORT: u16 = 5000;
+const BASE_PORT: u16 = 2000;
 // how much to allocate per individual process.
 // we expect to have at most 64 concurrent tests in CI at any moment on a given host.
 const SLICE_PER_PROCESS: u16 = (u16::MAX - BASE_PORT) / 64;
@@ -26,7 +26,7 @@ const SLICE_PER_PROCESS: u16 = (u16::MAX - BASE_PORT) / 64;
 #[allow(clippy::arithmetic_side_effects)]
 pub fn unique_port_range_for_tests(size: u16) -> Range<u16> {
     static SLICE: AtomicU16 = AtomicU16::new(0);
-    let offset = SLICE.fetch_add(size, Ordering::Relaxed);
+    let offset = SLICE.fetch_add(size, Ordering::SeqCst);
     let start = offset
         + match std::env::var("NEXTEST_TEST_GLOBAL_SLOT") {
             Ok(slot) => {
@@ -44,7 +44,7 @@ pub fn unique_port_range_for_tests(size: u16) -> Range<u16> {
     start..start + size
 }
 
-/// Retrieve a free 20-port slice for unit tests
+/// Retrieve a free 25-port slice for unit tests
 ///
 /// When running under nextest, this will try to provide
 /// a unique slice of port numbers (assuming no other nextest processes
@@ -54,7 +54,7 @@ pub fn unique_port_range_for_tests(size: u16) -> Range<u16> {
 /// When running without nextest, this will only bump an atomic and eventually
 /// panic when it runs out of port numbers to assign.
 pub fn localhost_port_range_for_tests() -> (u16, u16) {
-    let pr = unique_port_range_for_tests(20);
+    let pr = unique_port_range_for_tests(25);
     (pr.start, pr.end)
 }