Pivot Tracing

Pivot Tracing lets users define arbitrary metrics over trace data at runtime. It does so by combining two powerful techniques:

1. A Happen-Before operator that allows users to perform queries based on the causal relationship of the events.
2. The ability to instrument code dinamically without having to redeploy.

Sifter: Scalable Sampling for Distributed Tracing

Distributed tracing can be ridiculously expensive if you try to trace a hundred percent of requests. A common technique to reduce costs is to sample only a small portion of the traffic. But naive sampling techniques like uniform sampling will inevitably capture more common-case executions and might miss the more interesting edge cases. Instead, [Sifter’s approach][1] is to bias sampling decisions towards outliers and anomalous traces. This way, anomalous traces have a higher chance of being sampled, and the more uninteresting traces are discarded.

About Deploying on Fridays

Common knowledge says that you don’t deploy on Friday if you want to have a peaceful weekend. Yet, some people will tell you that if you’re not comfortable deploying every day of the week, you’re doing it wrong. They’ll say that deploying shouldn’t be scary and that you probably don’t have enough tests. So, which one is it?