Open benesch opened 3 months ago
Copying in some context from https://github.com/MaterializeInc/materialize/pull/27091#issuecomment-2121623020:
This is a light abuse of the parameter, since it's only documented to apply to update-type statements currently
Nah, this seems great. We should just update the documentation! Tracked in #27182.
To fix that I think I need to mock this up as a "real" peek response, which is more work and may be too weird to merge?
Are you sure that's the way to go? You may just want to keep track of the task somewhere in
struct Coordinator
and add some bespoke cancellation handling inhandle_privileged_cancel
. There's some prior art in that function (e.g., subscribes are handled separately from peeks).
@bkirwi notes that the optimizer is currently pondering the same question (should it use stats that are potentially stale but immediately available), and it's quite possibly worth waiting for those questions to resolve before picking this work up again.
Follow up tasks from https://github.com/MaterializeInc/incidents-and-escalations/issues/24:
EXPLAIN FILTER PUSHDOWN
with Ctrl-C (context)statement_timeout
to say that it applies toEXPLAIN
statements too