Section Infrastructure
What to collect, and what collecting it costs
Agents and polling, metrics against logs against traces, cardinality, sampling, and where the bill comes from.
For a separate operational view of time, ownership and team activity, see Monitask's practical guide.
This section is about the machinery: what gets collected, by what means, at what resolution, and what the resulting bill is made of. It is the most technical part of the site and the one where the vendor material is thickest, because almost every question here has a product attached to it.
The articles cover polling against agents, the three shapes of telemetry and which question each answers, cardinality and why it decides the invoice, sampling, whether to run it yourself, the disagreement between synthetic checks and real user data, and instrumenting software you did not write and cannot change.
The thread from the first section runs through all of it. Every one of these decisions fixes a proxy in place: the collection interval decides which events are allowed to exist, the sampling rule decides which cases survive to be examined, and the retention policy decides how long any of it can be asked about. None of those are settings. They are choices about what will be knowable later.
No product is named in this section. The selections do that, after the class of problem has been established.
Published
- 01
Polling and agents collect different truths
Nothing failed and nothing was misconfigured: the collection interval decided which events were allowed to exist at all.
Read - 02
Metrics, logs and traces answer different questions
Three shapes of data, not three products to buy, and the useful audit is which shape the last five incidents actually needed.
Read - 03
Cardinality is the number that decides the bill
Series count is the product of label values rather than the sum, so one line of code can multiply an invoice by five figures.
Read
Everything in this section
- 6 min readPolling and agents collect different truths
- 6 min readMetrics, logs and traces answer different questions
- 6 min readCardinality is the number that decides the bill
- 6 min readSampling: what you throw away on purpose
- 6 min readSelf-hosted or managed: the work moves, it does not vanish
- 6 min readSynthetic checks and real user data disagree, and both are right
- 6 min readInstrumenting an application you did not write
Further context
Start from the class of problem, not the list of tools
Every selection here names the situation first and the criteria second. Product names come last, and each one carries the line describing what it costs you.