On 9/25/07, Luke Lonergan <llonergan@greenplum.com> wrote:
> Hi Mark,
>
> I haven't yet looked at what you've done, but I'm an enthusiastic supporter
> of this idea. We're looking to do something that will view running queries
> and allow drill down into those executing at any given time, showing their
> plans and some notion of what operators are being executed.
>
> The idea of a ptop that shows running queries using a curses interface seems
> like a great start.
>
> Our needs for data warehousing workloads are going to be different from the
> OLTP users - our queries hang around long enough to warrant a drill-down.
>
> How far can you take the drill-down piece? Have you thought about how to
> acquire the status in the executor yet? One strategy we've considered is to
> use the same approach as "pstack" on Solaris - it takes a pid and dumps the
> stack of a backend, which clearly shows which executor node is being worked
> on currently. I think pstack uses dtrace underneath the hood...
Hi Luke,
Thanks. Honestly I haven't thought much past what I've thrown
together thus far, so I haven't considered getting status in the
executor. First thought would be to use popen() to call pstack and
dump the output to the screen. Of course other suggestions are
welcome. :)
Regards,
Mark