Anyone ever ran into memory consumption issues with Ghost/NodeJS?
I’m running 6 instances (only 1 has any relevant traffic, the rest are very small personal blogs) on a DO droplet (1vcpu, 2gb ram) and it oftens hits the 85% monitoring threshold and emails me about it, while also dropping below it a few minutes later.
The instance was initially a 1vcpu, 1gb ram one, which I upgraded after installing a few more Ghost instances. I thought this might be some leftover issue from the droplet upgrade (they do have some warnings about space not properly upgrading sometimes), but I didn’t get around to migrate it to a newly created droplet of this size to exclude this option.
Services running on the droplet: node, php, nginx, redis, mysql, fail2ban, postfix, etc. Cannot really identify a culprit, looking at htop most processes don’t consume a lot of memory and they all jump up and down in the list showing in the top 10 most resource consuming processes.
As a reference, I run the same DO droplet (1vcpu, 2gb ram) with some small Wordpress sites, same tech stack (except node) and yet the memory consumption is one third of the droplet with the Ghost sites.
@dsecareanu are you able to correlate the memory usage with sites that receive a lot of image uploads?
We’re aware that in certain situations the default memory allocator in Ubuntu can have issues where memory used by the image processor is not freed back to the OS. This can be resolved by using jemalloc as the memory allocator.
For reference we see typical memory usage for reasonably sized sites sitting at ~180-200MB. For very large sites (10s-100s of thousands of posts/tags) this can obviously be higher.
This could be the case (though rather huge library of images and not so much lots of image uploads). The blog with the most traffic also had some issues with theme freezing due to huge image size (almost 1gb of images indexed by the search theme search engine). This has been optimized a while ago, reducing size of images down to 3-400mb, but I haven’t checked lately to see if the size of images grew again out of control. Will check this avenue though, thanks.
So, according to the previous answer above by @Kevin it appears that 32 instances of Ghost running minimally at 200 Megs per each, 6.4 Gigabyes of memory usage is indeed a reasonable expectation…
Those numbers sound reasonable. I run 12 Ghost installs in their own FreeBSD jails, and each runs under 100 MiB. But they’re not hosting that many images beyond hero banners etc.
This is also entirely anecdotal (#GrainOfSalt!), but I’ve found Ghost handles increased traffic much better, and without the same spikes WordPress does. I’d had servers sit at 50% free only to have monitoring alert to “critical” memory use in php-fpm. Ghost is much more predictable.
