When the connection is getting closed, we call several functions that perform some cleanup (__wt_session_array_walk, __wt_txn_activity_drain, etc). Each function returns an error code that is ignored - if one function does not return 0, we keep going but we do save the error code through WT_TRET.

Because of this, an application may observe a non-zero error code when closing the connection without knowing exactly why and this makes the diagnosis complicated. This ticket should improve this, here are a few possible solutions:

• We could log a message after each call that is not successful.
  • We could enhance the WT_TRET macro, create a new one or we could only add those logs inside the connection.close() function.
• We could expand the WT_SESSION struct with a new struct to save information about each non-zero error code encountered inside a function, and then print that new struct at some point, for example at the end of the function.
• Use what has been added in SPM-3810 - there might be more work needed to make it work with connection APIs.