suite & task analytics in UI about cylc-ui HOT 11 CLOSED

(I have changed the title slightly, @hjoliver, to allow this to be a cover-all issue relating to suite & task data & analytics reporting, as a single Issue does not seem to exist for this. Hope that's okay!)

from cylc-ui.

I'll change the title further to remove the "utility" bit, since this will be an integral part of the new UI.

from cylc-ui.

See also cylc/cylc-flow#121

from cylc-ui.

(from an email to @matthewrmshin:

At the end of a suite run, or at intervals, we could generate a report showing things like mean queue-wait and run durations for each task, or a graph of the individual values over time, and provide hooks for customization - e.g. to plot some critical measure of model performance over time (number of iterations to convergence, or whatever). This might make it easier to see certain kinds of problems...

)

from cylc-ui.

(I guess a lot of these are related to what users want for Rose Bush (or its successor) as well.)

from cylc-ui.

This was something I looked into several years back as a possible addition to rose bush so there's a few comments I'd like to throw into the mix.

Suite run dbs contain all the info needed for nice summary reports and stats about suite runs

I'd say nearly all the information. It'd be incredibly useful if we had info about the families tasks are members of too. I say this as, for a number of suites, we have large numbers of VER, OPS, ensemble member etc. tasks. It'd be good if we had plots that were grouped by family (e.g. like tree view) that could be expanded out/broken down. The interest would be more that ensemble short forecasts were taking an average of X minutes to run not that ensemble member N was taking X minutes, member N+1 taking Y minutes etc.

I've got an ancient script knocking around somewhere I was using to generate a web page based report containing plots and tables for things like runtime and queue time (useful for highlighting really short jobs taking longer to queue than they run for) and found that for things like the OPS rose-stem suite (very large non-cycling suite) it was better to put things in bins (e.g. 0-5s, 5-10s etc) than go for each task individually - I could then expand out the bins to see what was in it. In the case of our cycling Global operational suite there are circa 1000 tasks so we'd need to think about how best to take information from that and present in a useful manner, something which I think would be aided by things like knowledge of families.

from cylc-ui.

In that case, note cylc/cylc-flow#1524.

from cylc-ui.

[meeting]

• document how to write your own report util?
• requirements are very site specific
• may put something simple in rose bush
• supply a simple generic util

from cylc-ui.

Interestingly, a number of areas round here are playing round with gantt charts for visualising the flow and timings of suites. Nominally speaking, we could integrate something like that into Rose bush (or whatever it becomes when moved to cylc) that could create a page for it.

from cylc-ui.

Notes from discussion to move to cylc-web:

• (Bruno) I think this is possible in the Web GUI, and would be quite useful. We could allow users to print, export as PDF, CSV, etc.
• (Matt) There are 2 parts. We need to collect more relevant data, and then we need to serve the data on the web UI.

from cylc-ui.

Superseded by the more general #262

from cylc-ui.