bug-1928847: Update architecture diagram for GCP

[relud]

stage submitter now reads from gcs and pubsub, and writes to the collector's load balancer. webapp and crontabber now show that they read processed crashes from gcs. I split up gcs into raw and processed crashes to limit the number of crossing lines, but kept a single pubsub to also limit crossing lines.

With more crossing lines I couldn't visually track orthogonal lines anymore, so I switched to straight lines and reorganized things as needed, with a secondary goal of data flowing left to right and top to bottom where possible (e.g. data still flows up for telemetry to reduce line crossing).

based on the legend stage submitter writes to crash-reports and processor "writes" to the version string api, because if I switch the arrow to have processor read the api it looks like the load balancer is routing version string api requests to the processor.

[willkg]

I have a couple of thoughts:

1. I find the new diagram really hard to read and reason about.
2. It feels like this isn't an architecture diagram. The arrows switched from being about connections (Thing A connects to Thing B) and turned into data flow.

We should have a data flow diagram for crash data--covered in bug 1769978--but that should be separate from the architecture diagram.

What does everyone else think? Does the new diagram help you understand the Socorro architecture?

[relud]

I find the new diagram really hard to read and reason about.

I don't know how to materially turn this feedback into diagram changes. I agree, but I also thought that about the old diagram. I could hand this off so someone else can give it a shot instead?

I feel like part of the problem is that almost every component accesses both gcs raw/processed and pubsub, and showing those connections creates a lot of overlapping lines. maybe i could separate those and put them in a box that with a note that says "used by every component"?

It feels like this isn't an architecture diagram. The arrows switched from being about connections (Thing A connects to Thing B) and turned into data flow.

I was under the impression the arrows already indicated data flow. Were the arrows are supposed to be like source--uses-->destination? in that case i would remove the legend, and have everything pointing away from kuberenetes components except the gclbs?

[relud]

I updated the diagram to apply my own suggestions from my last comment, plus switch back to orthogonal lines, lmk if that helps?