Re: Missing calls to UnlockBuffers() - unify error handling?

Widening the perspective a bit, our current approach for such error-handling /
shutdown functions seems ... not maintainable. In particular we have a
substantial number of top-level sigsetjmp() handlers that have slightly
different recovery codepaths.  When adding a new process type, how is one
supposed to even know what function one needs to call?

PostgresMain() has this comment:
                /*
                 * NOTE: if you are tempted to add more code in this if-block,
                 * consider the high probability that it should be in
                 * AbortTransaction() instead.  The only stuff done directly here
                 * should be stuff that is guaranteed to apply *only* for outer-level
                 * error recovery, such as adjusting the FE/BE protocol status.
                 */

But a) that clearly hasn't worked, given how long the following block is, and
b) can't really work that well, because we have plenty processes that don't
use transaction, therefore putting cleanup in AbortTransaction() doesn't go
that far.
  

I don't quite know how it should look like, but it seems pretty obvious that
it should be more unified than it is.  My gut feeling is that we should have a
single error recovery function that has a flags argument guiding which
subsystems should be reset and which shouldn't.

Seems we should just reorder the sequence in ProcKill() to first call
LWLockReleaseAll()