LEARN. PRACTICE. GROW.Concept and mechanism
Sizing should observe the load that matters. A low monthly average can hide the daily closing peak. Measure volume, latency, queues, and dependencies, including the condition where one node is unavailable. In a simple example, two nodes handling five hundred requests per second receive seven hundred. Normal capacity suffices, but one node cannot support that load within the measured limit. The plan needs additional capacity, supported demand reduction, or another explicit commitment. Counting servers or duplicating health checks does not change the calculation. Acceptance should show service performance under the required condition rather than only nominal resource totals.
Guided application
Artifacts also need identity and trust. An intact package digest does not alone authenticate its publisher; verify the signature using a key with validated provenance. For images, a tag can change. Record the accepted digest to relate testing and execution without confusing immutability with absence of vulnerabilities. Container memory limits can cause OOM even while the host has free memory. Measure the profile and neighboring impact before adjusting. Data in a volume can survive container replacement, but this does not demonstrate recovery after deletion. In a fictional batch exercise, validate peak load, completeness, persistence, and restore with representative data before transferring operational responsibility.
Normal capacity of 1000 does not sustain 700 when failure leaves only 500 available.
Common pitfalls
Average as peak; digest as publisher trust; free host memory as no limit; volume as backup.
Related topics: Automate with evidence · Operate and recover systems · Identity, PAM, and SSH
Relate load, artifact, and state to evidence of the production condition.
Reference: Capacity and overload management · Historical LFCE V3.18 (2018-06-12); exam retired 2022-05-01; technical references inspected 2026-09-30