A time window, not a wall of logs
A journal is a record of events, not a diagnosis. The first useful filter is usually time, followed by the component you meant to inspect.
Begin with one question
“Why did this service stop?” is a more useful starting point than “show every error.” Decide which service and which attempt matter. An old failure, an unrelated warning, and a recent successful restart can all appear in one unfiltered terminal window. Reading them together makes it surprisingly easy to invent a sequence that never happened.
journalctl --unit=example.service --since "20 minutes ago" --no-pager
example.service is a placeholder: replace it with the unit you are investigating. The journal may be restricted to members of an administrative group. A lack of access is not evidence that no events were recorded; likewise, a service can write to a separate log instead of the system journal.
Make the interval explicit
A relative time is convenient while looking at a fresh incident. For a repeatable comparison, use an explicit start and end. Note the timezone as well. A timestamp copied from another machine may not represent the same local clock reading, and two machines' clocks may disagree.
journalctl --unit=example.service --since "2026-11-20 09:00:00 UTC" --until "2026-11-20 09:10:00 UTC" --utc --no-pager
The dates above are an example, not a report of an actual event. When reproducing a problem, record the test's own start and finish. That makes it easier to distinguish evidence from a previous attempt, especially when requests are retried automatically.
Keep this boot separate from an earlier one
Boot boundaries are another source of confusion. Use the current boot when that is the scope of your question, or list recorded boots before selecting an earlier one:
journalctl --list-boots
journalctl -b --unit=example.service --no-pager
A journal can be volatile or subject to retention limits. A gap in the available history does not establish that a machine was quiet. Check what was actually retained before drawing a conclusion from an absence.
Read state alongside history
A log line saying that something failed is a statement about a moment. It does not necessarily describe the current process. Compare the relevant journal interval with current service state, configuration, and a small functional check. A running process is not by itself proof that its work succeeds, either.
Finally, share only the lines needed to support the question. Logs can contain account names, request addresses, headers, and secrets. Preserve useful timestamps and error messages, but remove sensitive values before posting a snippet elsewhere.