Confirm the pattern

Compare the dashboard time range with deployments, backups, scheduled jobs, and visitor traffic. A brief CPU peak is different from constant saturation. A stopped or unreachable VPS can leave gaps rather than a meaningful zero.

Read-only Linux checks

uptime
top -b -n 1

These show load and a process snapshot. Load also reflects tasks waiting for resources; it is not identical to CPU percent. Process names and arguments may contain private information, so redact output before sharing.

Find the bottleneck

Look for expensive queries, too many workers, repeated failed requests, compression, or build jobs. Correlate CPU with RAM, disk I/O, and logs. Do not kill unknown processes or reboot repeatedly as a first step. Test application optimizations in staging and keep a backup before changing production settings.