From: Oguzhan Ozmen <oozmen@bloomberg.net>
Date: Wed, 6 May 2026 19:17:55 +0000 (+0000)
Subject: rgw/multisite: fix endpoint unreachable detection in RGWRESTConn sync paths
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=28f2a5e22afaae6794a1f6fe766b61ad0afe8024;p=ceph.git

rgw/multisite: fix endpoint unreachable detection in RGWRESTConn sync paths

The checks that decide whether to call set_endpoint_unconnectable()
were comparing against -EIO, but the actual error codes returned on
connection failure changed after commit 37352a9074 ("rgw: change
rgw_http_error_to_errno default to -ERR_INTERNAL_ERROR").

- complete_request() -> wait() returns req_data->ret which is set to
  rgw_http_error_to_errno(0) = -ERR_INTERNAL_ERROR when http_status is 0
  (TCP connect failed). Fix the six call sites to check -ERR_INTERNAL_ERROR.

- forward_request() returns tl::unexpected(-ERR_SERVICE_UNAVAILABLE)
  when http_status == 0 (no HTTP response received at all). Fix the two
  forward/forward_iam conditionals to check -ERR_SERVICE_UNAVAILABLE.

Without this fix, connection failures are never detected in the sync
paths, so set_endpoint_unconnectable() is never called and the
IP failover / retry logic is effectively dead.

The .h coroutine paths were already fixed by dbb409e21b9 ("rgw: fix
endpoint detection in RGWRESTConn") but that commit missed all .cc
sync paths.

Signed-off-by: Oguzhan Ozmen <oozmen@bloomberg.net>
---

diff --git a/src/rgw/rgw_rest_conn.cc b/src/rgw/rgw_rest_conn.cc
index 46f6685c6fd..ce3425f6e8c 100644
--- a/src/rgw/rgw_rest_conn.cc
+++ b/src/rgw/rgw_rest_conn.cc
@@ -329,7 +329,7 @@ auto RGWRESTConn::forward(const DoutPrefixProvider *dpp, const rgw_owner& uid,
     auto result = req.forward_request(dpp, key, info, max_response, inbl, outbl, y);
     if (result) {
       return result;
-    } else if (result.error() != -EIO) {
+    } else if (result.error() != -ERR_SERVICE_UNAVAILABLE) {
       return result;
     }
     set_endpoint_unconnectable(endpoint);
@@ -360,7 +360,7 @@ auto RGWRESTConn::forward_iam(const DoutPrefixProvider *dpp, const req_info& inf
     auto result = req.forward_request(dpp, key, info, max_response, inbl, outbl, y, service);
     if (result) {
       return result;
-    } else if (result.error() != -EIO) {
+    } else if (result.error() != -ERR_SERVICE_UNAVAILABLE) {
       return result;
     }
     set_endpoint_unconnectable(endpoint);
@@ -415,7 +415,7 @@ int RGWRESTConn::complete_request(const DoutPrefixProvider* dpp,
                                   real_time *mtime, optional_yield y)
 {
   int ret = req->complete_request(dpp, y, &etag, mtime);
-  if (ret == -EIO) {
+  if (ret == -ERR_INTERNAL_ERROR) {
     ldout(cct, 5) << __func__ << ": complete_request() returned ret=" << ret << dendl;
     set_endpoint_unconnectable(req->get_endpoint());
   }
@@ -578,7 +578,7 @@ int RGWRESTConn::complete_request(const DoutPrefixProvider* dpp,
                                   optional_yield y)
 {
   int ret = req->complete_request(dpp, y, etag, mtime, psize, pattrs, pheaders);
-  if (ret == -EIO) {
+  if (ret == -ERR_INTERNAL_ERROR) {
     ldout(cct, 5) << __func__ << ": complete_request() returned ret=" << ret << dendl;
     set_endpoint_unconnectable(req->get_endpoint());
   }
@@ -629,7 +629,7 @@ int RGWRESTConn::get_resource(const DoutPrefixProvider *dpp,
     }
 
     ret = req.complete_request(dpp, y);
-    if (ret == -EIO) {
+    if (ret == -ERR_INTERNAL_ERROR) {
       set_endpoint_unconnectable(endpoint);
       if (tries < NUM_ENPOINT_IOERROR_RETRIES - 1) {
         ldpp_dout(dpp, 20) << __func__  << "(): failed to get resource. retries=" << tries << dendl;
@@ -683,7 +683,7 @@ int RGWRESTConn::send_resource(const DoutPrefixProvider *dpp, const std::string&
     }
 
     ret = req.complete_request(dpp, y);
-    if (ret == -EIO) {
+    if (ret == -ERR_INTERNAL_ERROR) {
       set_endpoint_unconnectable(endpoint);
       if (tries < NUM_ENPOINT_IOERROR_RETRIES - 1) {
         ldpp_dout(dpp, 20) << __func__  << "(): failed to send resource. retries=" << tries << dendl;
@@ -742,7 +742,7 @@ int RGWRESTReadResource::read(const DoutPrefixProvider *dpp, optional_yield y)
   }
 
   ret = req.complete_request(dpp, y);
-  if (ret == -EIO) {
+  if (ret == -ERR_INTERNAL_ERROR) {
     conn->set_endpoint_unconnectable(req.get_endpoint());
     ldpp_dout(dpp, 20) << __func__ << ": complete_request() returned ret=" << ret << dendl;
   }
@@ -809,7 +809,7 @@ int RGWRESTSendResource::send(const DoutPrefixProvider *dpp, bufferlist& outbl,
   }
 
   ret = req.complete_request(dpp, y);
-  if (ret == -EIO) {
+  if (ret == -ERR_INTERNAL_ERROR) {
     conn->set_endpoint_unconnectable(req.get_endpoint());
     ldpp_dout(dpp, 20) << __func__ << ": complete_request() returned ret=" << ret << dendl;
   }