Add fsync calls to reduce probability of unexpected state after a crash.

3 files changed, 36 insertions, 11 deletions

diff --git a/searchlib/src/vespa/searchlib/common/indexmetainfo.cpp b/searchlib/src/vespa/searchlib/common/indexmetainfo.cpp
index 10015bb658a..837c38eb340 100644
--- a/searchlib/src/vespa/searchlib/common/indexmetainfo.cpp
+++ b/searchlib/src/vespa/searchlib/common/indexmetainfo.cpp
@@ -1,6 +1,7 @@
 // Copyright 2017 Yahoo Holdings. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
 
 #include "indexmetainfo.h"
+#include <vespa/vespalib/io/fileutil.h>
 #include <vespa/vespalib/util/stringfmt.h>
 #include <vespa/vespalib/util/guard.h>
 #include <cassert>
@@ -310,6 +311,7 @@ IndexMetaInfo::save(const vespalib::string &baseName)
 {
     vespalib::string fileName = makeFileName(baseName);
     vespalib::string newName = fileName + ".new";
+    vespalib::unlink(newName);
     vespalib::FilePointer f(fopen(newName.c_str(), "w"));
     if (!f.valid()) {
         LOG(warning, "could not open file for writing: %s", newName.c_str());
@@ -350,6 +352,7 @@ IndexMetaInfo::save(const vespalib::string &baseName)
             newName.c_str(), fileName.c_str());
         return false;
     }
+    vespalib::File::sync(vespalib::dirname(fileName));
     return true;
 }
 
diff --git a/searchlib/src/vespa/searchlib/transactionlog/domain.cpp b/searchlib/src/vespa/searchlib/transactionlog/domain.cpp
index 88c2dd9ecc3..1caff132779 100644
--- a/searchlib/src/vespa/searchlib/transactionlog/domain.cpp
+++ b/searchlib/src/vespa/searchlib/transactionlog/domain.cpp
@@ -3,6 +3,7 @@
 #include "domain.h"
 #include <vespa/vespalib/util/stringfmt.h>
 #include <vespa/vespalib/util/closuretask.h>
+#include <vespa/vespalib/io/fileutil.h>
 #include <vespa/fastos/file.h>
 #include <algorithm>
 #include <thread>
@@ -60,6 +61,7 @@ Domain::Domain(const string &domainName, const string & baseDir, Executor & comm
     _sessionExecutor.sync();
     if (_parts.empty() || _parts.crbegin()->second->isClosed()) {
         _parts[lastPart].reset(new DomainPart(_name, dir(), lastPart, _defaultCrcType, _fileHeaderContext, false));
+        vespalib::File::sync(dir());
     }
 }
 
@@ -294,6 +296,7 @@ void Domain::commit(const Packet & packet)
             _parts[entry.serial()] = dp;
         }
         dp = _parts.rbegin()->second;
+        vespalib::File::sync(dir());
     }
     dp->commit(entry.serial(), packet);
     cleanSessions();
@@ -310,6 +313,7 @@ bool Domain::erase(SerialNum to)
             _parts.erase(it);
         }
         retval = retval && dp->erase(to);
+        vespalib::File::sync(dir());
     }
     if (_parts.begin()->second->range().to() >= to) {
         _parts.begin()->second->erase(to);
diff --git a/searchlib/src/vespa/searchlib/transactionlog/translogserver.cpp b/searchlib/src/vespa/searchlib/transactionlog/translogserver.cpp
index 4c3c5609a93..bfca137ba06 100644
--- a/searchlib/src/vespa/searchlib/transactionlog/translogserver.cpp
+++ b/searchlib/src/vespa/searchlib/transactionlog/translogserver.cpp
@@ -336,6 +336,29 @@ void TransLogServer::exportRPC(FRT_Supervisor & supervisor)
     rb.ReturnDesc("syncedto", "Entry synced to");
 }
 
+namespace {
+
+void
+writeDomainDir(std::lock_guard<std::mutex> &guard,
+               vespalib::string dir,
+               vespalib::string domainList,
+               std::map<vespalib::string, std::shared_ptr<Domain>> &domains)
+{
+    (void) guard;
+    vespalib::string domainListTmp(domainList + ".tmp");
+    vespalib::unlink(domainListTmp);
+    std::ofstream domainDir(domainListTmp.c_str(), std::ios::trunc);
+    for (const auto &domainEntry : domains) {
+        domainDir << domainEntry.first << std::endl;
+    }
+    domainDir.close();
+    vespalib::File::sync(domainListTmp);
+    vespalib::rename(domainListTmp, domainList, false, false);
+    vespalib::File::sync(dir);
+}
+
+}
+
 void TransLogServer::createDomain(FRT_RPCRequest *req)
 {
     uint32_t retval(0);
@@ -351,12 +374,9 @@ void TransLogServer::createDomain(FRT_RPCRequest *req)
         try {
             domain = std::make_shared<Domain>(domainName, dir(), _commitExecutor, _sessionExecutor,
                                               _domainPartSize, _defaultCrcType, _fileHeaderContext);
-            {
-                Guard domainGuard(_lock);
-                _domains[domain->name()] = domain;
-            }
-            std::ofstream domainDir(domainList().c_str(), std::ios::app);
-            domainDir << domain->name() << std::endl;
+            Guard domainGuard(_lock);
+            _domains[domain->name()] = domain;
+            writeDomainDir(domainGuard, dir(), domainList(), _domains);
         } catch (const std::exception & e) {
             LOG(warning, "Failed creating %s domain. Exception = %s", domainName, e.what());
             retval = uint32_t(-1);
@@ -385,12 +405,10 @@ void TransLogServer::deleteDomain(FRT_RPCRequest *req)
                 Guard domainGuard(_lock);
                 _domains.erase(domainName);
             }
-            vespalib::rmdir(Domain::getDir(dir(), domainName).c_str(), true);
-            std::ofstream domainDir(domainList().c_str(), std::ios::trunc);
+            vespalib::rmdir(Domain::getDir(dir(), domainName), true);
+            vespalib::File::sync(dir());
             Guard domainGuard(_lock);
-            for (DomainList::const_iterator it(_domains.begin()), mt(_domains.end()); it != mt; it++) {
-                domainDir << it->first << std::endl;
-            }
+            writeDomainDir(domainGuard, dir(), domainList(), _domains);
         } catch (const std::exception & e) {
             msg = make_string("Failed deleting %s domain. Exception = %s", domainName, e.what());
             retval = -1;