So based on the total RSS (Resident Set Size) running on Linux, the out-of-the-box usage of yarn start is around 300MB, which definitely seems high:

Just by reducing some of those "npm scripts calling other npm scripts" or wrapper scripts like cross-env, we get down to 123MB, which is an improvement:

Interestingly, though, we are still running two Node processes: one for yarn and one for the main node server. Using npm instead of yarn uses slightly less, but it's still around 92MB:

The only way to get this down is to call node __sapper__/build directory, rather than using an npm/yarn script, which gets us down to 48.6MB:

I did some research, but the only thing I've come across is this blog post, which basically concludes that the npm start pattern is an inherent waste of RAM because it uses about twice as much memory as calling node directly. Unless I'm missing something, I'm seeing the exact same results.

The author does note that the parent npm process seems to decrease in memory over time, and after 48 hours is negligible. Perhaps yarn has the same behavior; without running it for 48 hours, I'm not sure.

Server-side memory usage reported as "twice the size of Pleroma" #971

Description

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions