billflare.

Watch D1’s daily row quotas while testing

Michal Raczka (@michalraczka) warns that D1’s daily row read and write quotas can stop all D1 queries until midnight UTC, notes that indexes trade reads for extra writes, and moves heavy tests to local Miniflare.

Products
D1Workers
Source
Author experience𝕏
Author
Michal Raczka (@michalraczka)

Key points

  • The author writes that once a daily row read or write quota is hit, no D1 operations can run until midnight UTC.
  • He calls indexes a double-edged sword: they can greatly reduce rows read, but every index adds another written row.
  • In his case, hitting the read/write quota during testing was more limiting than the 500 MB storage limit; his solution is local Miniflare for tests that read or write lots of data.

Original sources

  1. Original source · Published

    Read on 𝕏

    Michal Raczka

    @michalraczka · Original 𝕏 post

    D1 is amazing for tinkering with Cloudflare Workers, but you need to watch out for the daily row read/write quotas: - when you hit one, you can't run any D1 operations until midnight UTC - defining indexes is a double-edged sword, it can significantly reduce the number of rows…
Author experienceWorkers +5

Four layers of cost protection for small Cloudflare projects

David Z (@davezfr) writes an X article for vibe coders on budget alerts, stop conditions for paid paths, automatic pausing, and letting a personal agent verify and pause.

Workers

A voting page polling every three seconds: the author revises a $1,000 report.

@keking99 first reports about $1,000 for a voting page polling /api/result every three seconds. A later post reports $816.52 in accumulated usage as of October 4 and cites another screenshot total of $817.14. He later says his support ticket keeps being refused.

Author experienceAccount-wide usage-based products

A budget alert is a horn, not a brake

@GUYUE_LGY points to Cloudflare’s Budget alerts documentation: alerts send email when spend crosses a threshold but do not pause or cap usage. Before letting AI run unattended, the author checks what actually happens at the line.