From ac2f01a6ab5796db67a72a17e6f2e1310d7b6564 Mon Sep 17 00:00:00 2001
From: Kevin Rajan <7121943+kvnloo@users.noreply.github.com>
Date: Fri, 25 Sep 2026 17:25:45 -0500
Subject: [PATCH] docs(offload): correct LMCache MP server and timeout recipe

---
 atom/kv_transfer/offload/README.md | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/atom/kv_transfer/offload/README.md b/atom/kv_transfer/offload/README.md
index f439614697..b496c67ac0 100644
--- a/atom/kv_transfer/offload/README.md
+++ b/atom/kv_transfer/offload/README.md
@@ -145,7 +145,7 @@ access to the ATOM worker allocations:
 lmcache server --host 127.0.0.1 --port 5555 \
   --chunk-size 256 \
   --null-block-id -1 --separate-object-groups \
-  --supported-transfer-mode lmcache_driven --l1-size-gb 64
+  --supported-transfer-mode lmcache_driven --l1-size-gb 64 --eviction-policy LRU
 ```
 
 For example, add the following settings to a DSv4 launch that publishes the
@@ -171,6 +171,14 @@ python -m atom.entrypoints.openai_server \
   }'
 ```
 
+Lookup has two timeout scopes. `lmcache.mp.mq_timeout` is passed to the
+LMCache scheduler/worker adapters for message-queue operations.
+`lmcache.mp.lookup_timeout` starts only after lookup submission returns and
+bounds ATOM's polling loop while the tier keeps returning no answer;
+`lmcache.mp.lookup_poll_interval` controls the sleep between those polls.
+Therefore `lookup_timeout` is not a hard wall-clock bound for a blocking
+adapter submission or status call.
+
 ATOM's configured chunk size must equal the MP server chunk size, and both must
 align to ATOM's PAGE/hash block size. A native prefix is loadable only where
 PAGE KV and a complete STATE checkpoint exist at the same boundary on every TP