(trafficserver) branch 9.2.x updated: Fix cache retry assert on ServerAddrSet (#13396)

[eze]

This is an automated email from the ASF dual-hosted git repository.

ezelkow1 pushed a commit to branch 9.2.x
in repository https://gitbox.apache.org/repos/asf/trafficserver.git


The following commit(s) were added to refs/heads/9.2.x by this push:
     new 526ed6bb79 Fix cache retry assert on ServerAddrSet (#13396)
526ed6bb79 is described below

commit 526ed6bb79af5c8177567579d3d5289107e3b5e8
Author: Evan Zelkowitz <e...@apache.org>
AuthorDate: Thu Jul 16 12:19:20 2026 -0600

    Fix cache retry assert on ServerAddrSet (#13396)
    
    A TSHttpTxnServerAddrSet retry can re-enter the cache miss path after
    ATS already holds a cache write lock for the same request. The
    redirect-only prepared-write reuse logic then asserts because this is
    not actually a redirect.
    
    Preserve the existing cache write lock when the retry returns through
    HandleCacheOpenReadMiss.
    
    Backport of #12972 to 9.2.x. The accompanying autest is omitted because
    its plugin infrastructure (#12733) is not present on 9.2.x; enum and
    debug-macro names were adapted to the 9.2.x code.
    
    Fixes: #12971
    
    Co-authored-by: Evan Zelkowitz <e_zelkow...@apple.com>
---
 proxy/http/HttpTransact.cc | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/proxy/http/HttpTransact.cc b/proxy/http/HttpTransact.cc
index 7f9f07e854..88494e24f3 100644
--- a/proxy/http/HttpTransact.cc
+++ b/proxy/http/HttpTransact.cc
@@ -3330,6 +3330,12 @@ HttpTransact::HandleCacheOpenReadMiss(State *s)
     s->cache_info.action = CACHE_DO_NO_ACTION;
   } else if (s->api_server_response_no_store) { // plugin may have decided not 
to cache the response
     s->cache_info.action = CACHE_DO_NO_ACTION;
+  } else if (s->cache_info.write_lock_state == CACHE_WL_SUCCESS && 
s->cache_info.action == CACHE_DO_WRITE) {
+    // Origin retry paths such as TSHttpTxnServerAddrSet() can loop back 
through
+    // HandleCacheOpenReadMiss after we already own the cache write lock. This
+    // is still the same request, so keep the existing write lock instead of
+    // re-preparing a write as if this were a redirect or a new cache miss.
+    TxnDebug("http_trans", "Reusing existing cache write lock while retrying 
origin selection");
   } else {
     HttpTransact::set_cache_prepare_write_action_for_new_request(s);
   }