From dddf7424b4f079277df4ba06c7e520678df4b06d Mon Sep 17 00:00:00 2001
From: Chrysostomos Nanakos <chris@include.gr>
Date: Fri, 26 Sep 2025 20:16:30 +0300
Subject: [PATCH] fix: resolve stuck peer refresh state preventing block
 discovery

This prevents peers from becoming permanently invisible to block discovery when
they fail to respond to WantHave requests.

Part of https://github.com/codex-storage/nim-codex/issues/974
---
 codex/blockexchange/engine/engine.nim     | 16 +++++++---------
 codex/blockexchange/peers/peercontext.nim | 12 ++++++++++--
 2 files changed, 17 insertions(+), 11 deletions(-)

diff --git a/codex/blockexchange/engine/engine.nim b/codex/blockexchange/engine/engine.nim
index 3166b808..20effd59 100644
--- a/codex/blockexchange/engine/engine.nim
+++ b/codex/blockexchange/engine/engine.nim
@@ -206,17 +206,15 @@ proc refreshBlockKnowledge(self: BlockExcEngine) {.async: (raises: [CancelledErr
     # want list in the coarsest way possible instead of over many
     # small updates.
     #
-    if peer.refreshInProgress:
-      trace "Peer refresh in progress", peer = peer.id
-      continue
 
     # In dynamic swarms, staleness will dominate latency.
-    if peer.lastRefresh < self.pendingBlocks.lastInclusion or peer.isKnowledgeStale:
-      peer.refreshRequested()
-      # TODO: optimize this by keeping track of what was sent and sending deltas.
-      #   This should allow us to run much more frequent refreshes, and be way more
-      #   efficient about it.
-      await self.refreshBlockKnowledge(peer)
+    if peer.isKnowledgeStale or peer.lastRefresh < self.pendingBlocks.lastInclusion:
+      if not peer.refreshInProgress:
+        peer.refreshRequested()
+        # TODO: optimize this by keeping track of what was sent and sending deltas.
+        #   This should allow us to run much more frequent refreshes, and be way more
+        #   efficient about it.
+        await self.refreshBlockKnowledge(peer)
     else:
       trace "Not refreshing: peer is up to date", peer = peer.id
 
diff --git a/codex/blockexchange/peers/peercontext.nim b/codex/blockexchange/peers/peercontext.nim
index f6eeea46..326468f1 100644
--- a/codex/blockexchange/peers/peercontext.nim
+++ b/codex/blockexchange/peers/peercontext.nim
@@ -27,7 +27,7 @@ export payments, nitro
 
 const
   MinRefreshInterval = 1.seconds
-  MaxRefreshBackoff = 36 # 3 minutes
+  MaxRefreshBackoff = 36 # 36 seconds
 
 type BlockExcPeerCtx* = ref object of RootObj
   id*: PeerId
@@ -45,7 +45,15 @@ type BlockExcPeerCtx* = ref object of RootObj
   activityTimeout*: Duration
 
 proc isKnowledgeStale*(self: BlockExcPeerCtx): bool =
-  self.lastRefresh + self.refreshBackoff * MinRefreshInterval < Moment.now()
+  let
+    staleness = self.lastRefresh + self.refreshBackoff * MinRefreshInterval < Moment.now()
+
+  if staleness and self.refreshInProgress:
+    trace "Cleaning up refresh state", peer = self.id
+    self.refreshInProgress = false
+    self.refreshBackoff = 1
+
+  staleness
 
 proc isBlockSent*(self: BlockExcPeerCtx, address: BlockAddress): bool =
   address in self.blocksSent