If you don't have any knowledge of an heuristic that fits to the content of the graph then you have to choose Dijkstra.
A* was developed for road maps, where for road distances the constraint applies that the direct distance is always shorter than going via another node. This constraint does not apply for general weighted graphs.
If you don't know such an additional constraint / heuristic of the content of your graph then you have to use Dijkstra
Further keep in mind that road map graphs are so huge that it is worth using A*.
If your graph is not huge, then probably it is even not worth thinking whether to find any heuristic. Such a wrong heuristic could even make things worse.
So you could use Dijkstra, and only if you have a performance problem you could start thinking to find a heuristic.