Where operating systems keep their logs, and how to query each one

What it is

"Check the logs" means a different store and a different command on each system.

System	Where logs live	Query command
Linux with systemd	the binary journal, plus any files under /var/log for services that bypass it	journalctl -u NAME --since "1 hour ago"
Linux with rsyslog	text files under /var/log (syslog on Debian/Ubuntu, messages on RHEL; per-facility files set in /etc/rsyslog.conf and /etc/rsyslog.d)	grep, less; on systemd hosts rsyslog usually receives its messages from the journal, so journalctl sees them too
macOS	the unified log, a binary store not visible as plain files	log show --predicate 'process == "NAME"' --last 1h or log stream to follow live
Windows	the Event Log service's channels (Application, System, Security, plus per-application channels)	Get-WinEvent -FilterHashtable @{LogName='Application'; StartTime=(Get-Date).AddHours(-1)} (-LogName and -FilterHashtable cannot be combined)
AIX	the binary error log (hardware and software errors) and, separately, syslog if configured	errpt for the error log; for syslog, tail -f the file named in /etc/syslog.conf (the default file has no active rules)
FreeBSD	text files under /var/log, rotated by newsyslog, routed by /etc/syslog.conf	grep, less, or tail -f /var/log/messages

Why it matters

The binary stores (systemd's journal, macOS's unified log, the Windows Event Log, AIX's error log) cannot be read with grep; each needs its own query tool, and each tool has its own filter syntax — a unit name on Linux, a predicate expression on macOS, a filter hash table or XPath on Windows. Text-file logging (plain rsyslog, FreeBSD, most AIX application logging) is portable to any text tool but loses the structured fields the binary stores provide, such as the process's boot ID or the exact facility/severity pairing.

How to apply

• Identify the store before writing a query: a service unit's journalctl -u and a plain file's tail are not interchangeable, and running the wrong one against the wrong system silently returns nothing rather than an error.
• Bound every query by time (--since, --last, -FilterHashtable @{StartTime=...}) before bounding by content; unbounded log reads are the most common way an agent script hangs or floods its own output.
• On macOS, log show and log stream must run from an admin account or with sudo, and message arguments marked private appear as <private> either way; plan for that in unattended scripts.
• On Windows, use Get-WinEvent, not the older Get-EventLog: only the former reads the per-application channels, and Get-EventLog does not exist in PowerShell 7.

Pitfalls

• Assuming AIX's errpt is a general application log; it is a hardware/software error log, not a place most third-party daemons write routine messages.
• Reading only /var/log/syslog on a systemd host: Debian 12 and later no longer install rsyslog by default, so the file may not exist and the journal is the only store.
• Querying the Windows Security log from a non-elevated session; reading it needs administrator rights, and the script gets an access-denied error, not an empty result.