Dash0

Series A or earlier? Dash0 is 50% off in year one, and 90% if your investor is a Dash0 partner. You get the full platform and every AI feature while being connected with the engineers who wrote the code. Need a feature we don't have yet? You ask the people who build it. Investor not a partner yet? Send us their name. If they join, you move to 90%.

Venture-backed, Series A or earlier? Dash0 is 50% off in year one, and 90% if your investor is a Dash0 partner. ✅ The full platform and every AI feature ✅ Live in an afternoon, no platform engineer needed ✅ Built on OpenTelemetry Full platform and every AI feature. The discount applies to year-one list price.

Venture-backed, Series A or earlier? Dash0 is 50% off in year one, and 90% if your investor is a Dash0 partner. ✅ Access to the full platform and every AI feature ✅ Live in an afternoon, no platform engineer needed ✅ Built on OpenTelemetry Join the program that supports your growth and explore effortless observability. The discount applies to year-one list price.

Venture-backed, Series A or earlier? Dash0 is 50% off in year one, and 90% if your investor is a Dash0 partner. ✅ Access to the full platform and every AI feature ✅ Live in an afternoon, no platform engineer needed ✅ Built on OpenTelemetry Join the program that supports your growth and explore effortless observability. The discount applies to year-one list price.

This week an AI agent attacked our dev environment, and another AI agent caught it. Morpheus, from our friends at Cenobe Cybersecurity, ran a fully agent-driven pentest against us from the outside. Agent0 alerted us the moment it started. From there we just asked it questions: what's being hit, from where, what's failing. The whole investigation happened in a chat. Agent0 was never meant to be a security tool. We built it as an SRE helper. But it already reads all our logs, so pointing it at security took almost no work. That's what excites me. It's never been cheaper to just try something.

Trace #sampling allows you to harness the power of distributed tracing without breaking the bank, but also adds inaccuracies; let's talk about how you fix that. There are two commonplace ways of doing sampling: probabilistic #HeadSampling and #TailSampling. With probabilistic head sampling, it is like you throw a die for every trace, and keep the trace only if the die's face is low enough. With tail sampling, you collect the traces and keep those that are *interesting*, typically slow ones and erroneous ones. Head and tail sampling reduce the accuracy of your data, over or underestimating error rates and latencies of your services. (Btw, creating animations with LLMs is pretty addictive.) What can you do to reduce inaccuracies? You additionally record rate, errors and durations (RED) metrics for your services on the full data before sampling. With #OpenTelemetry, you can do trace sampling and #RED metrics collection with the same pipeline setup. Sometimes you also combine the two methods: tail sampling for interesting traces and probabilistic head sampling for baselines. However, processing the resulting data in alerting and dashboards is usually a challenge, while RED metrics make it easy. I wrote an article about this a little while ago that is still pretty relevant, and there is also an accompanying talk, which I fearlessly link in the main post: * https://lnkd.in/eKJCyuXy * https://lnkd.in/eRaczB4N What else do you do to improve the accuracy of trace sampling?