Virtual garbage
A cloud-cost scare, and the ops lesson underneath it. Juan opens Google Cloud and shows the bill climbing, from around four thousand a month to six, nine, thirteen thousand a day, on a trajectory to blow past a million for the month. Digging in, he finds two causes. The backend is genuinely heavier now, from all the new features. But the bigger one is 'virtual garbage': every deploy was saving data to the cloud and never deleting the old, so it piled up, like recording a video, publishing it to YouTube, and never erasing it from your own memory. They were storing roughly four to five hundred gigabytes when only about fourteen were actually in use, four hundred gigabytes sitting there doing nothing. Delete, delete, delete, and the cost dropped back toward nine thousand, still a bit high because of the heavier backend. The teaching is quiet and universal: a deploy that leaves its artifacts behind isn't finished, it's leaking, and you pay rent on garbage until you sweep it.
// trace: where this idea came from
- ↳ video diary @ 1:02:11 (el costo diario trepó a 13.000, iba a pasar de un millón en el mes)
- ↳ video diary @ 1:02:53 (basura virtual, cada deploy guardaba y nunca borraba, como no borrar el video de tu memoria)
- ↳ video diary @ 1:03:20 (400-500 GB almacenados cuando solo se usaban 14, 400 ocupando espacio sin hacer nada)
- ↳ Entry 245-1: Three branches, one Frankenstein (la cadencia de deploys semanales que iba dejando la basura)
A scare, and the lesson under it. Juan opens Google Cloud on camera, because he can and it doesn’t scare him, and shows the bill climbing: from around four thousand a month it had jumped to six, then nine, then thirteen thousand a day, and it looked like it would keep going ▸ 1:01:39. On that trajectory the month would end north of a million, which is exactly the kind of thing that makes you stop and read the invoice ▸ 1:02:11.
borra el deploy viejo o pagas renta por la basura →
Two causes. The backend is genuinely heavier now, from all the features he’s been bolting on. But the bigger one he calls virtual garbage: every weekly deploy was saving its data to the cloud and never deleting the old, so it accumulated. It’s like recording a video, publishing it to YouTube, and then never erasing it from your own memory ▸ 1:02:53. The numbers were stark: they were storing roughly four to five hundred gigabytes when only about fourteen were actually in use, four hundred gigabytes sitting there doing nothing but costing money ▸ 1:03:20. Once they saw it, the fix was blunt: delete, delete, delete, and the daily cost settled back toward nine thousand, still a touch high because of the heavier backend ▸ 1:03:47. The teaching is quiet and universal. A deploy that leaves its artifacts behind isn’t finished, it’s leaking, and the cloud will happily charge you rent on garbage until someone remembers to sweep…