fix: DList timeout=0 means wait indefinitely

When timeout is 0, treat it as no timeout (wait forever) instead of
using a deadline that has already passed. This allows callers to
distinguish between "wait with timeout" and "wait without timeout".

Signed-off-by: Shawn Wang <shawn.wang@zilliz.com>

Author: sparknack

include/cachinglayer/lrucache/DList.h

@@ -106,6 +106,10 @@ class DList : public std::enable_shared_from_this<DList> {
     IsEmpty() const;
 
     // Reserve loading resource with timeout, called before loading a cell.
+    // When timeout > 0, the request will fail after the specified duration.
+    // When timeout == 0, the request will wait indefinitely (no timeout).
+    // Note: timeout=0 requests use time_point::max() as their deadline, which
+    // means they sort after all finite-deadline requests in the waiting queue.
     folly::SemiFuture<bool>
     ReserveLoadingResourceWithTimeout(const ResourceUsage& size, std::chrono::milliseconds timeout,
                                       OpContext* ctx = nullptr);

src/cachinglayer/lrucache/DList.cpp

@@ -58,7 +58,8 @@ DList::ReserveLoadingResourceWithTimeout(const ResourceUsage& original_size, std
     // If immediate reservation fails, add to waiting queue
     std::unique_lock<std::mutex> lock(list_mtx_);
 
-    auto deadline = std::chrono::steady_clock::now() + timeout;
+    auto deadline =
+        timeout.count() > 0 ? std::chrono::steady_clock::now() + timeout : std::chrono::steady_clock::time_point::max();
     auto [promise, future] = folly::makePromiseContract<bool>();
 
     uint64_t request_id = next_request_id_.fetch_add(1);