One Person, One SQLite File, One Small Machine: A Weekly Site for Solo Writers

How big a server does someone who only wants to post once a week really need? We ran an EmDash blog site locally and measured the size, the memory and the backup.

We wanted to settle one very concrete question. If it is just you, writing one post a week, with nobody to hire and no wish to worry about a server every month, how heavy is this EmDash site, really?

“Light” is a word that is easy to say in passing. We decided not to say it, and to measure instead.

What we did: create a Node + SQLite blog site with the official scaffolding (EmDash 1.4.0), load the template’s 8 sample posts and 1 page, build it, start it, and look at what was left on the machine. Everything below was measured on our own Mac, not on an actual small cloud host, so treat the numbers as a reference, not a promise.

First, the number that feels a bit uncomfortable

node_modules is 748 MB.

These are the dependencies used for development and building, and a machine that only runs the site does not need them. What the build produces is the dist directory, at 33 MB. When you put something on a small machine, dist is what you put there, plus the dependencies needed at runtime. We did not do a “production dependencies only” install this time, so we cannot give a number for what runtime actually needs.

One more thing, since it is where we waited the longest: the first dependency install, in an environment with a restricted network, took more than seven minutes. That time depends mostly on the network and does not describe your situation.

How much memory does it use once running?

A build takes about 25 seconds. Then:

HOST=127.0.0.1 PORT=4401 node ./dist/server/entry.mjs

Right after starting, the process takes about 115 MB. After we requested the home page, the post list and the RSS feed with curl more than twenty times, it rose to around 137 MB. On our machine, page responses fell between a few milliseconds and a few tens of milliseconds (the first request for the post list took 44 milliseconds, and the rest were mostly single-digit).

To be clear: this is the situation where a handful of people never visit at the same time. There was no load test and no simulated concurrency.

The startup log has two lines that made us look twice: one from Node saying SQLite is an experimental feature, and one saying cache.set() was called but caching is not enabled. We did not look into the first. The second looks like a reminder that you can configure a cache, which we did not.

Where does the data live?

In one file: data.db.

  • An empty template site: about 0.9 MB
  • After loading 8 posts and 1 page: about 1.6 MB
  • Compressed with gzip: about 63 KB

That is the content of the whole site, without images. When we loaded the seed we added --skip-media, so no images were downloaded, and how much space images take and where they go was not measured this time.

A backup is one command

We originally wanted to use emdash site export to export the whole site, and found that it needs the address and token of a running instance, which means logging in first. We did not go through the login step this time, so we used a plainer method:

sqlite3 data.db ".backup /somewhere/backup-$(date +%H%M).db"

The backup file appeared at about 1.6 MB. We opened it and counted: 8 rows in ec_posts and 1 row in ec_pages, the same as the original database.

Note that what we verified is only that “the data in the backup is correct”. We did not actually put it on another machine, bring the site up from it, and visit it again.

What does it say about its own state?

npx emdash doctor

In the output, the database, the migrations (90 applied, none pending), the collections and the datetime storage all pass, with one warning:

WARN  users: no users — complete setup wizard at /_emdash/admin

It means we loaded only content and created no users. That is correct, since we never went through that step.

Who is this scenario actually for?

After reading this far, our own conclusion is a little more conservative than our opening.

It fits when you are the only writer, post once or twice a week, and have all your content in one file. Backing up means copying one file, the machine runs a single process, and it takes somewhere above a hundred MB of memory. You do not need to know much about operations, and you do not need a database service.

What we are not sure about:

  • How to handle storage and backups once the images pile up
  • Whether the build can finish on a genuinely small cloud host (say 1 core, 1 GB). We did not measure the peak memory of the build. If the machine is too small, consider building elsewhere and uploading dist
  • The experience of the full path where you write and publish posts in the admin. We did not log in to the admin. This time the content was loaded from the command line

So if you plan to use it seriously, we suggest following the steps above on your own machine and measuring once yourself before deciding whether to buy that “small machine”.