I have a hobby blog that I'm trying to run for <$5/month. I run it on Github Pages because it's a static site and GH takes care of just about everything. The only thing I don't get is analytics or server logs, I'm planning to build that myself with standard AWS components. I considered other options, but I didn't want to use GA (for many of the reasons mentioned in the blog), other tools like Plausible were outside of my budget, and the open source tools looked like more hassle to self-host than reinventing the wheel myself:
Browser (~3 lines of JS) -> API Gateway -> SQS Queue -> Lambda (ETLs queue into) -> Athena.
I was originally just going to use Postgres, but RDS/Aurora are expensive and running Postgres in EC2 is going to be at least as much work (configuring SSH, process management, backups, monitoring, logging, image building, networking, etc).
My custom plan is ~$3/month all-in provided I stay below ~1M requests per month, and even then it scales very cheaply. Also, this design is highly scalable, although I doubt I'll ever take advantage of it; it's mostly just icing on the cake. The main motivation is that these components are available on the free plan ("forever", not just the first 12 months) and/or the pricing model makes the charges negligible for my super-low-volume use case.
Note that this doesn't give me pretty dashboards; the interface is SQL. Fortunately, there are other analysis tools that I have at my disposal which can plug in to SQL.
Lastly, of course my $5 budget doesn't do justice to the actual value of my time; it's more of a fun challenge. If I really just wanted web analytics, I should shell out the $6/month for plausible or similar.
Your choice is, of course, your choice, but I'd strongly suggest GA for this use case. I understand and agree that Google is in fact evil and we should avoid feeding the beast, but a hobby website using GA is too minor to care about for any of the privacy reasons the author states and if it's fully static pages then ideally the performance will already be light enough that the bloat of GA isn't going to make a difference.
> "...a hobby website using GA is too minor to care about for any of the privacy reasons the author states..."
while it's a whole lot less signal per site, long-tail websites can provide more signal per visit than large ones. knowing you like kazakh folk music more uniquely identifies you for targeting than knowing you read cnn.
and for small sites, you don't often need a lot of data. just knowing number of visits and some basic visitor characteristics (country, time, etc.) on a per-page basis is usually as much as you need to make use of. depending on the business, a large site can and possibly should build custom analytics as a competitive advantage that doesn't have google (and facebook, amazon, etc.) looking over their shoulder.
nobody really needs GA. many don't even need realitme analytics afforded by a js library loaded on every page. you can just do log analysis, à la the original urchin analytics that google bought long ago.
I have a blog hosted on GitHub pages, and I only want to know what pages are viewed the most. Before using GA, I have considered multiple alternative. For self-hosted alternatives, I don't bother to use it because I don't want to maintain a server just for my blog. As for other paid hosted alternative, I avoid them because I want it to be as cheap as possible. So I eventually choose GA, but in a different fashion, because I do care about its bloat. I simply write a short JavaScript using Measurement Protocol[1] to only send pageview to GA, which works very well for me.
I've used GA before for this same blog, and the bloat was perceptible. Not only that, but the GA interface was poor and there was a lot of spam. Adding onto that the various ethical concerns (privacy and Google's questionable politics), I decided to go with something else.
Also, to reemphasize, the primary purpose of the blog and its infrastructure is to serve my own amusement and to indulge my technical interests.
For a light site, Google Analytics may be the majority of its “weight”. It’s not huge, but I’d like to think my readers came to my site for the content and not the analytics so it’s a bit ironic that the ratio doesn’t match that.
I haven't looked at it, but I want to get away from managing (pet) VMs. I want something that is fully git-opsed, and git-opsing VMs is a lot of work. It's much easier to string together a bunch of AWS resources via CloudFormation (or Terraform, although I'm less familiar), and then I can stand it up or tear it down whenever I want.
I've done something similar to this that has worked for me without a lot of effort besides the first time setup. I use Fathom which is open-source and available to use on any hosting service. I use it with DigitalOcean so it's $5/mo and I get to run it on as many websites I have.
Might not be very helpful in your case but I'd recommend thinking about using a simpler setup than SQS Queue + Lambda, if that works for you ️
> Might not be very helpful in your case but I'd recommend thinking about using a simpler setup than SQS Queue + Lambda, if that works for you ️
I did try to find something simpler, and I remember looking at Fathom but I don't recall off the top of my head what my qualm was. Basically anything that involves a VM is more complex than the system I described and anything that involves Docker is considerably more expensive (Fargate/ECS/EKS pricing models).
You could use a VPS with hetzner or scaleway for $5/mo with decent specs, or use an older raspberry pi 2/3. I had metrics services like prometheus, grafana, and matomo analytics running on my old pi before I migrated them to a VPS.
It's a little more complex, yes, but running a VPS with a few services / docker is not really that complex in the grand scheme of things. It opens you to the world of open source and/or just running long-lived processes. You also could host your site on it, removing a dependency on github offering free static hosting forever.
I had to scroll a bit more than this thread would prefer to find the pricing. It says "get started for free" but the lowest pricing tier is $14.00/mo. Looks like "free" is only the 7 day trial.
Browser (~3 lines of JS) -> API Gateway -> SQS Queue -> Lambda (ETLs queue into) -> Athena.
I was originally just going to use Postgres, but RDS/Aurora are expensive and running Postgres in EC2 is going to be at least as much work (configuring SSH, process management, backups, monitoring, logging, image building, networking, etc).
My custom plan is ~$3/month all-in provided I stay below ~1M requests per month, and even then it scales very cheaply. Also, this design is highly scalable, although I doubt I'll ever take advantage of it; it's mostly just icing on the cake. The main motivation is that these components are available on the free plan ("forever", not just the first 12 months) and/or the pricing model makes the charges negligible for my super-low-volume use case.
Note that this doesn't give me pretty dashboards; the interface is SQL. Fortunately, there are other analysis tools that I have at my disposal which can plug in to SQL.
Lastly, of course my $5 budget doesn't do justice to the actual value of my time; it's more of a fun challenge. If I really just wanted web analytics, I should shell out the $6/month for plausible or similar.