I don't think the hesitation to push to prod Friday afternoons has anything to do with the working schedule of your customers. It's more about whether you'll be able to roll back on a weekend.
It meant that issues would sit undiscovered over the weekend until Monday morning, and lead to a week starting in chaos.
Much better to get fast feedback while the deploy could be monitored against prod traffic.
I don’t think most people fully appreciated how much work a handful of us did to make sign off faster and deployment less error prone.
>I don't like working weekend [sic] anymore [sic] that [sic] anyone else.
I’ve worked with several people who either confessed to or were accused of having taken up pastimes that would guarantee they weren’t available during times they shouldn’t be expected to be available.
Management gets rewarded for getting more work out of us and there’s always someone pushing boundaries trying to leach a little more work out of the staff they have. These same people seem to think burnout is a big mystery, because to admit they caused the problem would be to admit they did something wrong.
Example, got a feature with a new required field? Add empty field, synthesize a default value for existing rows, start accepting data, validate that data, require the data, use the data.
If you’re doing deployments once a week, you end up juggling two stories so you’re not sitting on your hands the whole time, and two stories divided your attention, so the odds of fucking it up go up anyway even though you’re being careful on paper.
Meanwhile one of your coworkers is doing the same dance on another feature, so you start tripping over each other as team size increases.
Friday deployments make no sense if you have teams that work Monday-Friday.
The whole point of WLB is that we don’t live like monks.
Do you know how every feature toggle in your system works? Do you know the potential blast radius of those toggles? Can you speculate on the surprise surface area that might be additionally exposed by impedance mismatches and side effects (ie, unanticipated shared state)? Hell does every engineer who wrote a feature toggle know this about their own code? You know who I’m talking about.
If you’re having problems in production that can be anticipated, why are you still having those problems? How long should those be tolerated before someone gets fired?