Topic: cost
-
Which trace sampling strategy keeps rare failures visible in a low-traffic service?
Open question: sampling guidance is written for services with thousands of traces per second, where one percent is still a representative sample; for a service with a few requests per second, which combination of head sampling, tail sampling, per-route rates and retention has kept the one failing trace a week available at a cost the team accepted?
-
Log sampling for high-volume events: keep every error, sample the repetitive lines
Sampling drops a fraction of similar log events on purpose; the useful forms are one-in-N, burst-then-rate per period, per-level rules that leave warnings and errors untouched, and pipeline sampling keyed on a request ID so a whole request is kept or dropped together, with the applied rate written into the surviving events.
-
Cost allocation tags: a tagging standard that survives the billing export
Cloud bills can be grouped only by tags that exist on the resource, are activated for billing, and are spelled consistently; define a short mandatory key set, enforce it at creation with policy, apply it through infrastructure code, and reconcile untagged spend every month.
-
SMS pitfalls: GSM-7 versus UCS-2 encoding and message segments
An SMS carries 140 bytes: 160 GSM-7 characters or 70 UCS-2 characters. One character outside GSM-7 (a curly quote, an emoji) switches the whole message to UCS-2; longer messages are split into segments of 153 or 67 characters that providers typically bill separately. Count segments before sending and normalise punctuation.
-
Time and cost budgets for model calls in agents
Each agent run gets a token, step and time budget enforced in code. The usage fields of every response are read, stable content goes into a cacheable prefix, and offline work goes to the batch endpoint. A run without a budget ends at a timeout, not as planned.
Machine-readable: JSON