The Gardener's Relentless Hoe: On the Weed You Mistake for a Crop

There is a piece of received wisdom in our world that is so seductive, so perfectly aligned with our desire for order, that it has become nearly sacrosanct: Log everything. It is the first commandment handed down to new stewards of systems. It is the safety net we are told to weave beneath every tightrope walk of code we deploy. We are to capture every event, every metric, every tremor and whisper from our machines. The promise is that this vast, immutable record will be our oracle, our truth-teller when things go wrong.

And so, we build our observability stacks with the fervor of archivists, collecting petabytes of data with the unshakable belief that the answer must be in there, somewhere. We invest in powerful tools to sift through this digital sediment, confident that we are building a perfect map of our reality. This is not an incorrect impulse. It is, however, an incomplete one.

The critique lies not in the act of logging itself, but in the unexamined faith we place in the resulting pile. We mistake the sheer volume of data for insight. We confuse the presence of information with the presence of understanding. A field of data, much like a field of plants, requires constant, discerning cultivation. Without it, the useful crop is quickly choked out by the weed of noise.

The most pernicious weed is the log entry you never needed. It is the debug message from a function that was never unstable, the verbose output from a process that has run flawlessly for years, the high-cardinality label that fractures your metrics into a million useless fragments. This data grows with relentless, mindless fecundity, costing storage, bandwidth, and, most crucially, cognitive load. It is the thicket you must now hack through when the true problem—a rare, subtle race condition—is hiding just out of sight.

The gardener does not let every seed grow. They know that an unmanaged plot becomes a wilderness, and a wilderness is not a source of sustenance but a place of danger. The art, then, is not in the relentless collection, but in the ruthless curation. It is in asking, with a hoemaker’s cold eye, “Does this log line help me diagnose a known failure mode? Does this metric directly reflect a user-facing state? Will I ever realistically query for this?” If the answer is no, it is a weed. And a weed must be pulled.

True reliability is not born from the anxiety that compels us to record everything for a day that might never come. It is forged by the disciplined understanding of what truly matters to the system’s health. It is about knowing the vital signs so intimately that you can ignore the static. The goal is not a perfect record of everything that happened, but a clear window into why what happened, mattered. The rest is just chaff, growing in the shadow of the wheat, waiting to be cut down.

Notes & further reading

A few pages I came back to while writing this: