Every now and again I get a hankerin’ to fiddle around with data viz that isn’t inGraphs. This week, I wanted to take a peek at some redliner numbers to see if I could tease something interesting out of them. My goals were pretty simple:
-
Generate a scatter plot that showed all of the data for each run (not just the Redline Number), and
-
Hack it together in 90 minutes or fewer (cuz I’ve got Shit to Do)
I started off taking a look at redliner runs for comm-inbox-bps:
As an artifact of how the chart is created the terminated (orange) line across the top all have NaNs for their QPS values. It’s kind of all over the place, but I believe the lines that can be visually picked out moving up-and-to-the-right are individual redliner runs. I could probably find some way to make this more obvious…but again, I was strictly time-boxing myself.
I also wanted to take a look at voyager-api:
In this case, the visual lines are actually not individual runs…they’re runs for each deployable (read: pillar) within the voyager-api multiproduct. Let’s color them by deployable instead of by step status, shall we?
Ah, now we’re getting somewhere! The higher-QPS pillars (like Messaging) show up pretty clearly as having steeper slopes, with the lower-QPS services (like Learning) having much shallower slopes; so shallow, in fact, that in some cases it’s actually possible for a single node to handle 100% of the traffic in a given fabric. Coolness. I’m not sure exactly how useful it is at present - maybe with more data points and/or a different way of presenting it (for instance, by individual run instead of by pillar). …but it’s interesting, and really quite lovely.