Disclosed is a mechanism for handling failover of a data management application for a shared disk file system in a distributed computing environment having a cluster of loosely coupled nodes which provide services. According to the mechanism, certain nodes of the cluster are defined as failover candidate nodes. Configuration information for all the failover candidate nodes is stored preferably in a central storage. Message information including but not limited to failure information of at least one failover candidate node is distributed amongst the failover candidate nodes. By analyzing the distributed message information and the stored configuration information it is determined whether to take over the service of a failure node by a failover candidate node or not. After a take-over of a service by a failover candidate node the configuration information is updated in the central storage.