Operations, Maintenance, and Monitoring
Signals worth paging on
Monitoring collects metrics, logs, and traces; observability is the ability to ask new questions of that data without shipping new code. Distinguish an SLI (a measured indicator such as success rate or latency), an SLO (an internal target for that indicator), and an SLA (a contractual commitment with consequences). Alert on symptoms users feel and on error-budget burn, not on every threshold crossing, or the team learns to ignore the pager.
Routine maintenance
Operational hygiene includes patching, certificate and credential rotation, capacity trending, job and queue health, and log retention. Backups count only when a restore has been tested; an unverified backup is a hypothesis. Maintenance windows should be planned, communicated, and reversible.
Metrics vocabulary
MTTR measures how long recovery takes; MTBF measures the interval between failures; MTTD measures detection. Most monitoring improvements move MTTD and then MTTR, because you cannot repair what you have not noticed.
Traps
Questions praise a monitoring tool that checks only whether a host responds, which says nothing about whether the service works. Others treat a graph as proof of root cause, or assume that a green dashboard means the user journey is healthy. Watch for the option that increases alert volume rather than alert quality.
Worked example
Users report slow checkout while CPU and memory graphs look normal. Dashboard health does not clear the service; the symptom is user-facing latency, so you inspect request traces and dependency timings. That data points to a slow downstream call, and the fix belongs with the dependency owner, not the host.