DO cost headline: "We're spending $X/day on durable objects"
Status summary#
Spec'd 2026-09-07. Misha gave the headline verbatim; mockup with that morning's real numbers DM'd to him. Implemented; unit tests green. Left: a forced 🧪 TEST RUN from CI to see the real thread, then merge.
Why#
The daily #error-pulse headline (#2590) is a row of DO-hours per account. Nobody thinks in DO-hours.
What changes#
- The headline is exactly this sentence, nothing else:
We're spending $1,023/day on durable objects based on current usage ($161 dev/preview, $862 prd). Current usage = the last complete UTC hour's DO-hours × 24 × $0.005625, per account, summed. The last complete hour, because the probe runs at :41 with ~15–20 min of analytics lag, so the current hour is always partial. - The rest is a table in the thread, opened on purpose: per account the
latest hour, today so far, hours over the ceiling, pinned invocations;
then the incident-docs and workflow-run links. The thread's first
reply, created on the day's first run and rewritten every run after
(
chat.update; found among the bot's replies by its code-block table). Alert replies (over the ceiling in the last two hours, probe could not run) stay as they are, after it. (upsertDetailsReply) -
findOrCreateHeadlinematches the day's headline by the bot's messages since 00:00 UTC startingWe're spending(test runs keep the🧪 TEST RUN —prefix and their own thread). - Probe failure: that account's number cannot be computed, so the
sentence says so in its slot:
($161 dev/preview, prd: probe failed)and the total is the accounts that answered. - Unit tests updated for the sentence and the table. (scripts/ci/do-duration-alert.test.ts)
Out of scope#
- Changing the ceilings or what counts as a breach.
- Fixing whatever is burning prd right now (2026-09-07: 6,385 DO-hours in one hour). See stream-do-wake-loop-runaway.md.