launchpad-dev team mailing list archive
-
launchpad-dev team
-
Mailing list archive
-
Message #04674
single notification-based-changes considered harmful
Just a heads up, we've got what *looks* like a death-by-a-thousand cuts.
Take +filebug with an apport UUID.
The attachments from apport are attached to the bug as part of the bug
creation. Each attachment raises a notification; that is listened for
and generates a mail notification insertion.
So, a package with 50 subscribers and an apport attachment with 20
generates 1000 INSERTs into the system.
Now, there is a storm feature request to group the inserts, but the
flush before a select would almost certainly mean that we still end up
doing 20 batch inserts (and its still 1000 rows).
A better way to structure this is to raise a *single* event at the end
of adding all attachments & creating the bug, which will generate 50
notification entries: still a lot, but nowhere near as bad.
It may be, due to the contextual friction that we really don't want
in-process events for this sort of thing at all : but that would be a
very broad conclusion to draw, and I hesitate to do so.
I am however convinced that we shouldn't be raising many events of any
sort where nontrivial work (and db access is nontrivial) occurs as a
result of raising it.
-Rob
Follow ups