[3.5] agency lock left behind (#10022)

* short timeout issue and discarded agency lock removal

* short timeout issue and discarded agency lock removal

* no hot backup in 3.5.0

arangod/Cluster/ClusterInfo.cpp

@@ -4311,8 +4311,9 @@ arangodb::Result ClusterInfo::agencyHotBackupLock(std::string const& backupId,
 
   if (!rv->slice().isObject() || !rv->slice().hasKey("results") ||
       !rv->slice().get("results").isArray() || rv->slice().get("results").length() != 2) {
-    return arangodb::Result(TRI_ERROR_HOT_BACKUP_INTERNAL,
+    return arangodb::Result(
-                            "invalid agency result while acuiring backup lock");
+      TRI_ERROR_HOT_BACKUP_INTERNAL,
+      "invalid agency result while acquiring backup lock");
   }
   auto ar = rv->slice().get("results");
 
@@ -4366,6 +4367,8 @@ arangodb::Result ClusterInfo::agencyHotBackupLock(std::string const& backupId,
     std::this_thread::sleep_for(std::chrono::duration<double>(wait));
   }
 
+  agencyHotBackupUnlock(backupId, timeout, supervisionOff);
+
   return arangodb::Result(
       TRI_ERROR_HOT_BACKUP_INTERNAL,
       "timeout waiting for maintenance mode to be activated in agency");
@@ -4436,11 +4439,10 @@ arangodb::Result ClusterInfo::agencyHotBackupUnlock(std::string const& backupId,
     if (result.successful()) {
       if (!result.slice().isArray() || result.slice().length() != 1 ||
           !result.slice()[0].hasKey(modepv) || !result.slice()[0].get(modepv).isString()) {
-        return arangodb::Result(TRI_ERROR_HOT_BACKUP_INTERNAL, std::
+        return arangodb::Result(
-                                                                       string("invalid JSON from agency, when desctivating supervision mode:") +
+          TRI_ERROR_HOT_BACKUP_INTERNAL,
-                                                                   result
+          std::string("invalid JSON from agency, when deactivating supervision mode:") +
-                                                                       .slice()
+          result.slice().toJson());
-                                                                       .toJson());
       }
 
       if (result.slice()[0].get(modepv).isEqualString("Normal")) {

arangod/Cluster/ClusterMethods.cpp

@@ -4047,6 +4047,13 @@ arangodb::Result hotBackupCoordinator(VPackSlice const payload, VPackBuilder& re
     double timeout = (payload.isObject() && payload.hasKey("timeout"))
                          ? payload.get("timeout").getNumber<double>()
                          : 120.;
+    // unreasonably short even under allowInconsistent
+    if (timeout < 2.5) {
+      auto const tmp = timeout;
+      timeout = 2.5;
+      LOG_TOPIC("67ae2", WARN, Logger::BACKUP)
+        << "Backup timeout " << tmp << " is too short - raising to " << timeout;
+    }
 
     using namespace std::chrono;
     auto end = steady_clock::now() + milliseconds(static_cast<uint64_t>(1000 * timeout));
@@ -4057,6 +4064,7 @@ arangodb::Result hotBackupCoordinator(VPackSlice const payload, VPackBuilder& re
     // We specifically want to make sure that no other backup is going on.
     bool supervisionOff = false;
     auto result = ci->agencyHotBackupLock(backupId, timeout, supervisionOff);
+
     if (!result.ok()) {
       // Failed to go to backup mode
       result.reset(TRI_ERROR_HOT_BACKUP_INTERNAL,
@@ -4069,6 +4077,7 @@ arangodb::Result hotBackupCoordinator(VPackSlice const payload, VPackBuilder& re
       LOG_TOPIC("352d6", INFO, Logger::BACKUP)
           << "hot backup didn't get to locking phase within " << timeout << "s.";
       auto hlRes = ci->agencyHotBackupUnlock(backupId, timeout, supervisionOff);
+
       return arangodb::Result(TRI_ERROR_CLUSTER_TIMEOUT,
                               "hot backup timeout before locking phase");
     }