The patch for addressing
IMPALA-4765 introduced a new critical data structure 'tableLoadingBarrier_' to prevent using multiple table-loading threads for the same table.
However, the following bug might be difficult to detect with the current logging//metrics. If for some reason an entry in the 'tableLoadingBarrier_' is never removed then that table will never again be loaded and all queries waiting for that table will hang (and will not be cancellable, see IMPALA-915). The only remedy is to restart the catalogd.
We should carefully consider the above possibility and how to debug/identify it from logs, and perhaps enhance the logging.