Cache the anonymous page. Do not cache the cart and call it faster.
Worker limits, open files, gzip or brotli, and a cache for HTML that has no cookie. A 502 is not a tuning job. That is the support page.
Only after the access log says NGINX is the wait
Anonymous WordPress pages, bypass on cookie, wp-admin and the cart. A purge path so a published post is not stuck for an hour.
Cache headers, gzip or brotli, open_file_cache. This helps when the time is in file transfer, not in PHP.
worker_processes above the CPU count, or a cache in front of a logged-in shop. Both look clever and break the site.
A cache block in the generated vhost will vanish
Hestia and DirectAdmin rewrite the site file. The cache config goes in the template or an include they keep. Otherwise the next rebuild is the outage.
€55 an hour, after a look at where the time goes
Minimum 1 hour. If the wait is PHP or MariaDB, this page is the wrong job and we stop rather than add a cache in front of a slow query.
Anonymous pages slow, logged-in pages fine?
That is the case a NGINX cache can help. Tell us the panel. No root password in the first message.