From 4250d9843133d38c7ace1a1b510265a15ebaa0fd Mon Sep 17 00:00:00 2001
From: tv <tv@krebsco.de>
Date: Fri, 30 Oct 2026 15:10:29 +0100
Subject: add dormant-probe-max option

---
 src/state/timing.rs | 21 +++++++++++++++++----
 1 file changed, 17 insertions(+), 4 deletions(-)

(limited to 'src/state')

diff --git a/src/state/timing.rs b/src/state/timing.rs
index 9055ed6..988e430 100644
--- a/src/state/timing.rs
+++ b/src/state/timing.rs
@@ -11,8 +11,10 @@ const SYNC_REPLY_TIMEOUT: Duration = Duration::from_secs(10);
 /// tries to deliver a message (`Transport::delivery_limit`).
 const SYNC_REPLY_TIMEOUT_MAX_FACTOR: u32 = 30;
 /// Upper bound on the wait between probes of an unreachable peer, as a multiple of
-/// `dormant_probe`: an hour at the default keepalive (1m). A returning node HELLOs its own
-/// neighbours, so this only bounds how long one we never hear from stays out.
+/// `dormant_probe`, unless `--dormant-probe-max` says otherwise: an hour at the default
+/// keepalive (1m). A returning node HELLOs its own neighbours, and once one pair across
+/// a partition meets, the rest are probed at once (`handshake::probe_now`), so this
+/// bounds how long one we never hear from stays out.
 const PROBE_MAX_FACTOR: u32 = 6;
 
 #[derive(Clone, Copy, Debug)]
@@ -25,6 +27,8 @@ pub(crate) struct Timing {
     pub(crate) dormant_after: Duration,
     pub(crate) resync_after: Duration,
     pub(crate) dormant_probe: Duration,
+    /// The longest wait between probes (`probe_backoff`).
+    pub(crate) probe_max: Duration,
     /// While catching up: send the source a STATUS after this much silence (`swarm::source`).
     pub(crate) catch_up_probe: Duration,
     /// First wait for an answer to a sync message before sending again; doubles while
@@ -81,6 +85,7 @@ impl Timing {
             // the first sync reply timeout, so a lost message's sender resends it first.
             resync_after: (keepalive * 2).max(sync_reply_timeout * 2),
             dormant_probe: keepalive * 10, // >= 10*: dormant peers cost almost nothing, but still rejoin within minutes
+            probe_max: keepalive * 10 * PROBE_MAX_FACTOR,
             // Seconds, not a keepalive interval: a dead source should cost little time.
             // Only the source gets these, and only while catching up.
             catch_up_probe: (keepalive / 30).clamp(Duration::from_secs(1), Duration::from_secs(5)),
@@ -89,10 +94,18 @@ impl Timing {
         }
     }
 
+    /// `--dormant-probe-max`, if given: the longest wait between probes.
+    pub(crate) fn with_probe_max(mut self, max: Option<Duration>) -> Self {
+        if let Some(max) = max {
+            self.probe_max = max;
+        }
+        self
+    }
+
     /// Waits between probes of an unreachable peer: `dormant_probe`, doubling up to
-    /// `PROBE_MAX_FACTOR` times as much.
+    /// `probe_max` (all of them `probe_max` if that is shorter).
     pub(crate) fn probe_backoff(&self) -> Backoff {
-        Backoff::new(self.dormant_probe, self.dormant_probe * PROBE_MAX_FACTOR)
+        Backoff::new(self.dormant_probe.min(self.probe_max), self.probe_max)
     }
 
     /// Waits for sync answers: `sync_reply_timeout`, doubling up to 30 times as much.
-- 
cgit v1.3.1