Client WU usage surprise 😮

[chris.williamson1996]

Over the last 24 hrs my client and myself have had quite the surprise from bubble. It was bound to happen eventually but WU overages finally hit hard but the logic behind it makes little sense.

In short, we have a list of 10,500 “businesses”, we have a backend workflow setup to do a few things.

1. Use google places API to retrieve the placeID based on the city, state, biz name.

2. Retrieve google business photos API to retrieve the photos of the biz from their profile using the “photo referenceID” returned from prior step.

3. retrieve the phone number from the “google places details” api.

4. Trigger a “Dataforseo” API to scrape ALL of the google reviews for each business based on the placeID retrieved in step 1.

5. Schedule a retrieval of the scrape for +1 hr later and create a new review in the db for each scraped review and update the review average value on the biz. (1500 reviews per biz max).

We expected those to be pretty WU heavy of course but the actual results of WU usage made little to no sense in my opinion. In theory you’d expect the google places calls & creation of reviews to be by far heaviest part.

What ended up being the heaviest part? Using make changes to business (rating) results from step 2(list of reviews for that biz):average.

Here’s some statistics from our panel.
Total WU used: 6.679M

retrieve places ID’s: 423? (it updated 10,500 records and spent 423wu? ok.)

Used to retrieve PlaceID’s and photos: 13k

New_review workflow 6.4M

Clarification of my involvement of the project: I was brought in to handle the placesID and tech SEO side of the project, review scrape setup & retrieve was already in existence and instructed to use.

Inside of this new_review flow there are 3 actions:
1: create new review

2: +1 biz total review count & add step 1 to nested “review list”

3: update the business “rating” field. (:average of all reviews in list) (ideally this should’ve happened at the very end a single time not after every review, but again, review part of this was not my setup)

In the end it seems something is off here for a few reasons.

Should a bubble :average operator really account for 84% of 6.6M total WU usage when all other flows are far bigger, I guess it COULD make it this costly since it’s got to pull the full list of reviews as they’re using nested data? However this doesn’t explain why 10500 placesID which included 3x google API calls and 3x data changes + photo creation recursive flow only account for 423 usage.

Additionally the overage charge last night at 8:30PM-9PM when all flows ended was $485.

It wasn’t until 3:05PM today when we got a workload unit spike email & when checked had a new $1,093 overage however if you look at the chart there was 0 usage since roughly 9:30PM last night and according to bubble the WU and overages are all “real time”.
Bubble real time reporting thread

Quick edit: although it doesn’t make a massive difference, the 10,500 total businesses is actually only 9200 because the other 1300 errored out due to dataforseo limit and never made it to the review flow.

[mikeloc]

Dumb question given that it’s you, Chris, but you’re in touch with support, right? On the bright side, you can obviously direct them to this thread where you have done the heavy lifting (and a really great job, of course) of providing all of the details for them.

[lindsay_knowcode]

This sounds painful Chris If I’ve understood your excellent analysis it’s the average calculation database function? Usually you expect that to be computationally cheap.

Interesting find if that is the case.

[chris.williamson1996]

@mikeloc Of course, thank you.

@lindsay_knowcode Right! I can see how it could potentially get costly since they set up the new_review flow to pull and update :average rating based on a nested list every time a review was saved, considering it happened from google scraping there are a substantial amount of reviews but the math still doesn’t math on this situation or on the other flows.

the only thing I can really make any sense of is if :average just takes an absurd amount of time since we are charged for workflow timing.

based on their WU pricing chart the math just doesn’t math. An Update to Workload, Plus More Transparent Calculations

[laura.oppenheimer]

Hey @chris.williamson1996 as @mikeloc said, please be in touch with support. I definitely would like to take a look at why you didn’t get that spike notification until after the fact (it’s not great if you get a notification about something to fix right now if it’s no longer happening).

Success can definitely take a look at what you have set up and where there might be an issue/error - but also wanted to remind you that workload is billed by the tier active on the last day of the month. So most likely you’d be better off upgrading tiers to something that will cover more of your overages (I don’t want to recommend which tier specifically because I’m not sure of your WU you consider “correct” and which ones are in error")

[cmarchan]

Darn!

Definitely will be monitoring how this progresses.

It would be painful for excessive cost in such uneventful processing to be a reason to off load this …. processing …. outside of Bubble.

[chris.williamson1996]

Right! this flow could’ve been done for pennies in comparison external to bubble then just data imported in.

All of my personal apps and the majority of my client apps are still grandfathered in to old plans so this is one of my few experiences with the new WU system and I gotta say it’s pretty disappointing.

The statistics that can be looked at in the breakdowns seem either bugged or misleading in many cases, the overages alerts were like 18 hrs late, overage charges were recalculated way after flows ran so it’s not like we could’ve jumped ship on flows, and logging to really dive into a per action charge just didn’t load (per bubbles typical logging).

I’ve advocated for people to adapt to the new model because businesses tend to have to adapt to change and any good business will be able to navigate these price changes and still profit but charging people on a usage basis on a bugged system or failing to notify them timely is by far enough reason for people to not adapt.

[hugo.vieira]

I’m experiencing a huge increase in WUs usage compared to some month ago, in the same exact workflows and usage type, before, my client app would use 50% of the basic plan WUs for the whole month, now we use 50% in the first couple of days of the month, it feels like searches are much more expensive.

Fun fact is I’m only getting this increase in WU usage since the " Issues with Main Bubble Cluster" started happening regularly, maybe it’s just my imagination

[chris.williamson1996]

UPDATE: Been over a week and they can’t get me any answer of substance outside of admitting their live alert system isn’t working as planned, the email went out a day late, & their seems to be an issue with WU usage.

Since that info they provided the day after my report bubble support has been like it typically is with their long reply times & generic canned responses from now 4 different total agents about “we are looking into it”, and no offered solutions.

It’s like they don’t realize client apps have launch dates that need stuck to for marketing purposes🙄

This situation definitely makes me confident about legacy plans going away…