processes sort with parallel worker is broken

[dlax]

See discussion at #291.

From postgres 13, maybe we could rely on leader_pid column of pg_stat_activity view as first sort step on SQL side?

[kmoppel-cognite]

FYI - just remembered that I've seen actually cases where parent pid was higher than background worker / child pid...so this code here can deliver wrong results still https://github.com/dalibo/pg_activity/pull/291/files#diff-c7ba7b4a3a801c37989b9ac81ece8995e3154d2f67325272f24877b5e730cf65R219-R222

[blogh]

FYI - just remembered that I've seen actually cases where parent pid was higher than
background worker / child pid

I agree, I've seen it too. I think changing the sort order when paused could be a valuable
thing to have in the future in which case sorting in the SQL will not help. (I'd also like to add
filters when paused).

When we order by mem / cpu / read / write, the \_ thing is also meaningless because
the worker process is often not listed after their parent.

It was thinking about adding the parent pid and the backend type info in the UI so that we
have a parent pid to sort on in the sorted function. But it adds a lot of text and we would
need a better way to filter what we display. Some of the ideas in #260 might help for that
(group columns and allow to cycle between groups).

It would also be neat to group the leader process and it's childs in a single "ui block", but
it's a lot more complex.

Before changing the output, the switch to "rich" must be finished or cancelled. I'll try to
help there next.

[blogh]

@kmoppel-cognite what do you think about sorting/filtering in paused/interactive mode ?

[kmoppel-cognite]

@kmoppel-cognite what do you think about sorting/filtering in paused/interactive mode ?

@blogh For me personally I've never had the need for such extra features, I've just used "pause" to freeze the UI to be able to copy some SQL or kill some PIDs. And there's also the new --filter syntax, so I'd personally KISS and try to reduce UI logic / code as much as possible.