Simon Riggs <simon@2ndQuadrant.com> writes:
> Only one setting will be best for the whole cluster, so neither the
> user nor the DBA gains if a user sets this to a different value than
> the one that has been determined to be optimal.
> Since we wait while holding the lock it is actually harmful to
> everyone if anybody sets a stupid value and might even be considered a
> denial of service attack.
> So there is a very good reason to make this SIGHUP, not just a whim.
Hmm. If a malicious user could hurt performance for other sessions with
a bad setting of commit_delay, then USERSET is clearly a bad idea.
But it still seems like it could be SUSET rather than SIGHUP.
regards, tom lane