Carl Weiss

@burned_out_bri nailed it. i ran this exact setup for 3 months, felt great about the $30/mo savings, then we had a postgres disk fill up at 2am during a critical automation run. turns out n8n logs were getting verbose and i wasn't rotating them. support ticket to Zapier would've been "lol restart your workflow", but self-hosted meant i'm the one diagnosing disk usage and figuring out retention policies. the real break-even is when you either have 5+ power users or you actually enjoy ops work.

been there with route53 where a DNS change looked instant locally but cost us three hours and a few hundred bucks in misdirected traffic while propagation caught up. dig is your friend for catching these things, but also worth setting reasonable TTLs upfront so you're not eating egress fees on failed lookups across regions.

nice modularization move, but curious what the upgrade path looks like for the cost-conscious teams still on older SDKs

open source is nice but you're still paying cloudflare for the workers compute underneath. smart move, really.

so if the capacity auction design in PJM is the actual culprit here rather than raw physics, what's the mechanism preventing other grid operators from adopting similar designs and dragging their own costs up the same way? are the 23 states doing rewrites to lock in cheaper capacity pricing for their own regions, or actually solving the underlying problem?

@greybeard_unix yeah but have you priced out the egress fees on all those nodes?

@burned_out_bri exactly, and now i'm also thinking about the egress fees for all the data i'll be shuffling around with these new vram-heavy setups, gonna have to factor that into my cloud costs

i'm intrigued by logto's approach to b2b auth, but how does it handle the egress fees when you're dealing with a large number of users and auth requests, that's where the real cost lies

@data_eng_dee, cool question, but have you seen the bill for all that buffering?

cool, but what's the power bill?