Detect if it took too long to shutdown, produce coredump in that case

1 files changed, 11 insertions, 2 deletions

diff --git a/filedistribution/src/vespa/filedistribution/model/zkfacade.cpp b/filedistribution/src/vespa/filedistribution/model/zkfacade.cpp
index 8c67dba3150..9bf29ab1854 100644
--- a/filedistribution/src/vespa/filedistribution/model/zkfacade.cpp
+++ b/filedistribution/src/vespa/filedistribution/model/zkfacade.cpp
@@ -9,12 +9,14 @@
 #include <cassert>
 #include <cstdio>
 #include <sstream>
+#include <thread>
 #include <boost/throw_exception.hpp>
 #include <boost/function_output_iterator.hpp>
 
 #include <zookeeper/zookeeper.h>
 #include <vespa/filedistribution/common/logfwd.h>
 #include <vespa/defaults.h>
+#include <vespa/vespalib/util/sync.h>
 
 typedef std::unique_lock<std::mutex> UniqueLock;
 
@@ -253,8 +255,15 @@ ZKFacade::ZKFacade(const std::string& zkservers)
 ZKFacade::~ZKFacade() {
     disableRetries();
     _watchersEnabled = false;
-    zookeeper_close(_zhandle);
-    LOGFWD(debug, "Zookeeper connection closed successfully.");
+    vespalib::Gate done;
+    std::thread closer([&done, zhandle=_zhandle] () { zookeeper_close(zhandle); done.countDown(); });
+    if ( ! done.await(50*1000) ) {
+        LOGFWD(debug, "Zookeeper connection closed successfully.");
+    } else {
+        LOGFWD(error, "Not able to close down zookeeper. Dumping core so you can figure out what is wrong");
+        abort();
+    }
+    closer.join();
 }
 
 const std::string