Hacker News: latencyharbor

How to Turn Customer Outage Reports into SLIs and Alerts for Earlier Detection

latencyharbor — Tue, 06 Oct 2026 19:06:08 +0000

Article URL: https://oneuptime.com/blog/post/2026-09-29-customer-outage-report-to-sli-alert/view

Comments URL: https://news.ycombinator.com/item?id=49982644

Points: 1

# Comments: 0

Back to metal: moving away from cloud

latencyharbor — Mon, 14 Sep 2026 13:04:37 +0000

Article URL: https://backtometal.oneuptime.com

Comments URL: https://news.ycombinator.com/item?id=49696170

Points: 2

# Comments: 0

New comment by latencyharbor in "Composable Tests"

latencyharbor — Tue, 18 Aug 2026 16:41:06 +0000

A dependency graph can help with reporting, but I’d avoid making “test 1 ran successfully” part of test 2’s fixture.

Those are two separate ideas:

1. Execution dependency: if a cheap contract test fails, skip expensive downstream tests because their results won’t be informative. 2. State dependency: test 2 consumes state left behind by test 1.

The first can be useful. The second tends to create order dependence, awkward retries, and failures that are hard to reproduce when a CI runner shards or parallelizes the suite.

A safer model is a DAG of independently reproducible tests: each node declares prerequisites for scheduling/reporting, but creates or restores its own input state. Then a downstream test can be marked “blocked by X” rather than failed, while still being runnable by itself when debugging.

That distinction also keeps the dependency metadata from becoming a hidden setup mechanism.