MGM/FST: Adding retry mechanism for failed CTA Frontend DNS resolution

cern-eos · Nov 26, 2024 · 64839cf · 64839cf
1 parent c6953f2
commit 64839cf
Showing 3 changed files with 13 additions and 3 deletions.
diff --git a/common/xrootd-ssi-protobuf-interface b/common/xrootd-ssi-protobuf-interface
diff --git a/fst/XrdFstOfsFile.cc b/fst/XrdFstOfsFile.cc
@@ -3893,7 +3893,12 @@ XrdFstOfsFile::NotifyProtoWfEndPointClosew(uint64_t file_id,
     // If static initialization throws an exception, it will be retried next time
     static XrdSsiPbServiceType service(endPoint, resource, config);
     auto sentAt = std::chrono::steady_clock::now();
-    service.Send(request, response);
+    try {
+      service.Send(request, response, false);
+    } catch (std::runtime_error& err) {
+      eos_static_err("Could not send request to outside service. Retrying with DNS cache refresh.");
+      service.Send(request, response, true);
+    }
     auto receivedAt = std::chrono::steady_clock::now();
     auto timeSpent = std::chrono::duration_cast<std::chrono::milliseconds>
                      (receivedAt - sentAt);

diff --git a/mgm/WFE.cc b/mgm/WFE.cc
@@ -2800,7 +2800,12 @@ WFE::Job::SendProtoWFRequest(Job* jobPtr, const std::string& fullPath,
   // Send the request
   try {
     const auto sentAt = std::chrono::steady_clock::now();
-    service.Send(request, response);
+    try {
+      service.Send(request, response, false);
+    } catch (std::runtime_error& err) {
+      eos_static_err("msg=\"Could not send SSI protocol buffer request to outside service. Retrying with DNS cache refresh.\"");
+      service.Send(request, response, true);
+    }
     const auto receivedAt = std::chrono::steady_clock::now();
     const auto timeSpentMilliseconds =
       std::chrono::duration_cast<std::chrono::milliseconds> (receivedAt - sentAt);