Realtime Query Dashboard Results

On 1/7/2011 1:52 PM, THOMPSON, JARED (ATTBAPCO) wrote:

Just looking for understanding/ideas.

I assume when people use dashboards they are not being queried every
second for updating but maybe every minute?

Are there any tools that work good on top of postgres?

(I see in the stock market (though I am looking at/for production data)
they seem to use tools that frequently update their dashboards.

What is a realistic timeframe to expect query updates for a dashboard?

I think it would depend on how busy the box is, and how expensive
getting updates are.

-Andy

On 7 Jan 2011, at 22:02, "THOMPSON, JARED (ATTBAPCO)" <JT060b@ATT.COM> wrote:

I assume when people use dashboards they are not being queried every
second for updating but maybe every minute?

Are there any tools that work good on top of postgres?

(I see in the stock market (though I am looking at/for production data)
they seem to use tools that frequently update their dashboards.

What is a realistic timeframe to expect query updates for a dashboard?

Having written a dashboard on top of PostgreSQL (screenshot at http://blog.agilebase.co.uk/2010/03/31/dashboard-preview/ ), I can at least state my decisions:
Charts are updated once a day overnight, or cached whenever someone looks at them in the system underlying the dashboard, so they are at most one day old. A chart is also updated when a user clicks on it to drill down to the data.

Of course what you decide depends on what the business use case is and what demands there are on the system. In my cases so far the slowest charts take 1 or 2 seconds to generate by SQL so if necessary, each could be loaded in in real time over AJAX, though that hasn't been needed yet.

Regards
Oliver Kohll

On Fri, Jan 7, 2011 at 5:21 PM, Oliver Kohll - Mailing Lists <
oliver.lists@gtwm.co.uk> wrote:

Charts are updated once a day overnight, or cached whenever someone looks
at them in the system underlying the dashboard, so they are at most one day
old. A chart is also updated when a user clicks on it to drill down to the
data.

Of course what you decide depends on what the business use case is and what
demands there are on the system. In my cases so far the slowest charts take
1 or 2 seconds to generate by SQL so if necessary, each could be loaded in
in real time over AJAX, though that hasn't been needed yet.

We found for our use case that customers don't like "1 day old" reports.
We've moved much of our dashboard statistics to trigger-based materialized
view, so whenever they reload, they get current numbers and the cost to
compute those numbers has already been amortized over the metric-gazillion
updates and inserts that comprise them. :) Computing them on-the-fly is
just too time consuming and people don't like to wait, either.