Start at the beginning on East of the Sun, West of the Moon

Mastodon server

Wed, 31 Dec 2025 00:00:00 +0000

Status

Active since December 2022.

Moved from AWS to Hetzner in November 2025.

What

My personal Mastodon server. The accounts on there are all me, at this time.

Where

https://strangeweb.page/

I wanted to keep this relatively separate from other projects, so I registered a .page domain.

How

When I migrated to Hetzner, I simplified the setup a bit:

A single VM

More than enough CPU, RAM, storage, and of course an IPv4 and an IPv6 address.

Access via ssh locked down

Only my home IP address can get in. Easily updated when that changes (which is rare).

The other ports available

Port 80 (http, in practice really only for LetsEncrypt, but also the classic http to https redirect), and of course 443 (https) for the main attraction.

Still using the S3 bucket

I plan to migrate that to different object storage in the near future.

Terraform code

It creates the VM, updates the DNS (A and AAAA records which now live at deSEC.io), as well as reverse DNS managed by Hetzner, then throws cloud-init at it.

The cloud-init setup

This takes care of basic conveniences (things like emacs, jq, rsync, and others) but also installs dependencies for Mastodon and prepare it most of the way (skipping the interactive configuration of Mastodon itself).

Code and related things

Mastodon instructions for setting up your own server

The Infrastructure as Code at https://codeberg.org/.

Previously

The setup in AWS (from 2022-2025) was as follows:

Basic Virtual Private Cloud (VPC)

Pretty standard setup. A subnet for public facing things (which is the only things in here right now), and one for resources that don't connect out to the internet. Right now there aren't any of the latter, but it is good to be prepared

A bastion server

A locked down instance that is intended to be the only way to get into the VPC with ssh, with an IP address that is separate from the actual service.

The actual Mastodon server

An S3 bucket

This is where Mastodon stores media associated with posts (in the timeline of my accounts, or my own), with a CloudFront shield in front of it

Terraform code

This managed all of the above.

For Mastodon upgrades

I created a script to temporarily bump up the instance size, because it was unhappy for some of those steps but perfectly content running in a smaller instance during normal operations.

RSS feed fix

Sun, 28 Dec 2025 00:00:00 +0000

Looks may be deceiving

When I set up a static website my focus around the RSS feed was that I wanted to have the full content, so once I had that, I moved on.

Today I noticed that the links back to the posts in the RSS feed (or the link for the site itself) didn't actually work, they were missing the https://is-here.com/ prefix. 🤔

Thankfully this was a simple fix

All that was missing was a baseURL parameter in my config.yaml file. It was there originally, but it got lost when I migrated from the S3 bucket to statichost.eu and revamped my config.yaml file.

Running the local hugo preview with --baseURL http://localhost:1313/ to override that, and all is well again in this part of the world.

Teletekst Bot

Sun, 28 Dec 2025 00:00:00 +0000

Status

Active since July 2024

What

Monitor the news ("Nieuws" in Dutch) such as it is shared by the NOS Teletekst, by checking the references from the index pages at 101, 102, and 103, and then scan for any other pages from 104 to 199.

When a new page is found, or an existing page is updated (using a relatively basic heuristic), post the new or updated page to the Mastodon account. It also attempts to detect when a page is moved around.

Because occasionally pages are included for sport (in general) or soccer (voetbal), it will use an appropriate #TeletekstXYZ hashtag for that, so people can mute that. The same goes for the content updates, which can be a bit much especially for emerging stories, and get tagged with #TeletekstUpdate.

For content updates the posts also includes comparison of the previous and current version with color-coding for words or lines that have been removed, were added, or stayed the same. It's similar to git diff but has been adapted to make more sense in this text context.

If the content was unchanged and only the page number or title changed, the existing Mastodon post is updated in place.

Where

https://mstdn.social/@teletekst (originally at https://botsin.space/@teletekst, but that server shut down at the end of 2024).

How

There are two Cron jobs:

Every 5 minutes a program walks all the pages from 101 to 199, using the next page references. If all pages are exactly identical, discard the results. It needs to gracefully handle broken next page links and other quirks.
Also every 5 minutes, but a minute later, a second program checks if changes were found. If so, look for new pages, deletes pages, moved pages, and modified pages. Two pages are treated differently, namely the "Kort nieuws binnenland" and "Kort nieuws buitenland", where one or two short national or international news items are dumped.

Disclaimer

This is not an (official or endorsed) NOS.nl service!

Code

https://codeberg.org/jochie/Teletekst

Backups, What a Concept

Sat, 27 Dec 2025 00:00:00 +0000

How it started

For about three years I ran a single-user Mastodon instance at strangeweb.page in AWS, where the non-IaC bits were mostly manually put together. It was also a big example of "Do as I say, don't do as I do" because I wasn't maintaining any backups.

How it's going

Recently I moved that instance from AWS EC2 to a Hetzner VM, which meant I had to actually go through the steps of taking the configuration and data from one host to another. As part of that exercise I cobbled together a small shell script that takes those vital bits and copies them to a safe place.

In my case that's object storage at a different provider, avoiding a single point of failure.

The script

  #!/bin/sh

  . "$0.env"

  echo "$(date): Creating a DB snapshot"

  # Create a backup of the mastodon tables:
  sudo -u mastodon \
     pg_dump \
     -Fc mastodon_production \
     -f /home/mastodon/backup.dump

  echo "$(date): Creating a Redis snapshot"

  # Create a snapshot of the redis data:
  redis-cli SAVE

  for src in \
      /home/mastodon/live/.env.production \
      /var/lib/redis/dump.rdb \
      /home/mastodon/backup.dump; do
      echo "$(date): Backing up $src"
      aws s3 \
          --endpoint "https://$ENDPOINT" \
          cp \
          --quiet \
          "$src" \
          "s3://$BUCKET/snapshot/mastodon/"
  done

  echo "$(date): Done"

The .env file contains two settings: ENDPOINT and BUCKET. The ENDPOINT is necessary because the slightly older awscli version on the host was ignoring the endpoint_url directive in ~/.aws/config.

It's a fairly simple file, pointing to an object storage bucket at Scaleway:

[default]
region = nl-ams
output = json
services = scw-nl-ams

The other missing piece is ~/.aws/credentials with credentials for that bucket/account.

Disclaimer

Going back to that "Do as I say, don't do as I do" bit, I have not yet performed the ultimate test of restoring a test-server from the backed up data. So if I'm missing something, maybe let me know so I can fix it before it's too late? 😅

I also know that this only maintains a single snapshot unless the object storage has version support and it's enabled.

Moving to Europe (part 2)

Sun, 13 Jul 2025 00:00:00 +0000

The saga continues

After I moved my (three) domains, most of the DNS, and my static websites to Europe I got side-tracked. An interaction with a former colleague nudged me to pick that back up.

Services

When I last wrote about it I thought I would land on UpCloud or Scaleway for a virtual server (or two), but now I'm actually working with a virtual server at Hetzner. They do things a little bit differently, so I'm still getting used to that. Both have projects, which feels like the mental model equivalent of AWS VPCs? Hetzner has performance/usage graphs integrated into their web UI, whereas Scaleway has a hosted Grafana service (for which you need to create a dedicated user). Hetzner's virtual servers seem more generous with their root volumes, but then it may not be as easy to change that out later, so choose wisely?

That said, for something that mostly needs to be compute, storage, and a web presence, Hetzner's offerings seem quite nice.

Currently I'm putting together a "services" instance that will host things like webmail (Roundcube), bookmarks (Readeck), passwords (Vaultwarden, or something along those lines), etc. This would be the same Services project hinted at in the Read Later project page. The goal here will be to make it relatively painless to blow it away and spin it up elsewhere, so a layer of Terraform, some cloud-init fun, and automatically generated and managed LetsEncrypt certificates are going to be part of that.

Object Storage

For this, Scaleway still seems to be the most reasonable price, in particular for buckets with not a lot of content. Hetzner starts at 4.99 Euro for even an empty bucket (which then gets you 1TB of storage and traffic), by comparison. That's a non-starter for a bucket where I expect to have just a few dozen GB of data. For completeness, UpCloud's buckets also start at 5 Euros which then includes 250GB of storage, and states there are no data transfer fees, period?

Open questions

For moving my mail server I still need to do some research to find out which of these 3 (or some alternative I've overlooked or discarded unfairly) is the better pick in terms of IP reputation, or spam abuse complaint handling. Anyone know?

Previously

Moving to Europe

Read Later

Sun, 13 Jul 2025 00:00:00 +0000

Status

Active since July 2025.

What

A self-hosted bookmark/read-later service

Where

https://readeck.is-here.com/

How

This service is setup as one of several virtual hosts on a virtual machine that is dedicated to a couple of these self-hosted things (the project page for is still WIP/AWOL for now).

To try and keep the maintenance requirements relatively low I used the docker approach, only sharing the data with the host system so that it is preserved over restarts and can be backed up regularly:

  docker run \
         --detach \
         --rm -ti \
         -p 6080:8000 \
         -v /var/www/readeck:/readeck \
         codeberg.org/readeck/readeck:latest

A fragment of the Apache configuration file:

  ProxyPreserveHost On
  ProxyPass / http://0.0.0.0:6080/
  ProxyPassReverse / http://0.0.0.0:6080/
  RequestHeader set X-Forwarded-Proto https

Disclaimer: I don't have a lot of experience with the proxy or proxy_http modules, so it's possible that could be done better. There are more complete instructions on the Readeck Deployment documentation page.

Integrating it:

For Firefox I use this Readeck add-on.
There is no iOS (or Android, for that matter) app but the (self-hosted) website can be added as a pseudo-app on iOS by using Safari's "Add to Home Screen".
That pseudo-app doesn't get you a "Share to …" entry, but thankfully there are acceptable workarounds (listed in the FAQ) in the form of shortcuts. It took a little bit of fidgeting to figure out how to configure the shortcut, but after that it was relatively smooth sailing.
Although it's somewhat redundant, I also created a script to check my Mastodon bookmarks and import new ones into Readeck. So if I'm in a hurry, I'll quickly bookmark a ~~post~~ toot, otherwise I'll use the "Send Page + Tags To Readeck" shortcut.

Code and related things

Originally I had planned to migrate from Instapaper (the read-later service I used for ~15 years) to a self-hosted Wallabag instance, but before the dust had even settled on that (meaning figuring out what browser add-on to use or how to integrate it on my phone) this post by Shom pointed me towards Readeck. That's why the scripts below have a migration path from Instapaper -> Wallabag -> Readeck, and not Instapaper -> Readeck.

My collection of scripts in this read-later repository.

wallabag_import.py: Iterate over the Instapaper entries and add them one by one to my (also self-hosted) Wallabag service. For the ~17k bookmarks that I had that took multiple hours and possibly got my IP in trouble with a few websites.
readeck_wallabag.py: Initiate a migration from Wallabag to Readeck (handled entirely by Readeck after that).
reprocess_readeck.py: Loop over all bookmarks in a Readeck collection, re-adding them and deleting the original ones as it goes.
bookmark_readeck.py: Check my Mastodon bookmarks and add any new ones to Readeck

WARN Act Bot

Sun, 13 Jul 2025 00:00:00 +0000

Status

Active since around April 2023.

What

As I understand it, there are WARN Acts in most states (maybe all) that require employers of companies over a certain size to give a heads-up to the state when they plan to let go for one reason or another. This could be for permanent closure of a place of business, redundancies for whatever reason, and some other reasons I can't think of right now. Many states make the notices available in some form on the web. Sadly they don't seem to have considered standardizing that, so automation based on the content found online is a challenge.

In my case, I decided to automate checking the notices for the California EDD, and post the incremental changes (summarized where possible) on the Fediverse.

Where

https://sfba.social/@warn_act_ca (originally at https://botsin.space/@warn_act_ca, but that server shut down at the end of 2024).

How

The setup (in AWS) is as follows:

An S3 bucket to maintain state from one run to another
A few Systems Manager Parameter Store entries for the knowledge of where the bot is and the token to post to its timeline
A Lambda that is triggered from EventBridge on a regular basis that checks the California website
An SQS queue through which to manage the posts to the bot account, with a Lambda consumer that does that work
Terraform code to create these resources and some tangentially related ones

Cost

$0.00!

Actually, the dedicated AWS account in which this runs had a bill for (literally) a few pennies one time, apparently related to AWS Cost Explorer of all things, but otherwise the monthly bills have been $0.00 since I started.

The reason is that all of the service usage (S3, Lambda, SNS, SQS, CloudWatch, and data transfer) are all flying well under the limits of the permanent AWS free tiers. The biggest risk is probably S3 where the it's running up a bill of about $0.005 per month, so each month that is rounded down to $0.00 but when you request an annual total it lists that as ~$0.05. 😁

Code

https://codeberg.org/jochie/WARN

New search engine, who dis?

Tue, 29 Apr 2025 00:00:00 +0000

Anything but Google

Like many others I've become disenchanted with Google search. So many, in fact, that its dominance is starting to decline! Instead, I've been using DuckDuckGo as my default search for the last several years, on desktop and on my phone (with its dedicated application).

A new player enters the field

Several posts (the Daring Fireball blog, Ian Betteridge and Bart Busschots in the Fediverse) recently nudged me to try out Kagi. The idea behind their offering is that aside from a 100 searches/month free tier you pay a monthly fee and in return their interface is clean, sparse, and lacks ads.

Of course if you're paying, you need to have an account. Their FAQ states that they only collect the bare minimum of data to run the service and for now I'm giving them the benefit of the doubt on that. Hypothetically it also allows you to set some personal preferences, but I haven't looked into that yet.

The first few searches

My usual first searches are vanity searches. That may seem silly, but I would hope that I know reasonably well what to expect there.

Kagi did not disappoint:

My LinkedIn profile
Several Facebook profiles with similar names (my own isn't public, and covered in dust)
Some references to the time that I contributed to the DGD ecosystem (parts of the KernelLib, maintaining the network package)
A YouTube account with all of 2 playlists (Dutch artists, and 80s/90s music, unsurprising)
A reference to the WARN ACT bot, one of the two that I run on the Fediverse
Some currently inactive static websites with my name on it still
A reference to one of few mailing lists that I still vaguely read or contribute to
My name in a list of Patreon contributors of the DTNS podcast

Unexpected blasts from the past:

An ancient SourceForge.net profile that I had long since forgotten about (created in 2004)
A reference to the time that I created a WordPress plugin (not sure I still have that file, and I doubt it still works)
A name credit in iPhone in Action (2008) as found in part on this website for a modified bookmarklet to show the source of a web page on the iPhone? I honestly don't remember that, but seeing as the two authors were the people I worked with at Skotos that makes perfect sense
And on that note, it found an archived version of that Skotos website as well, and some related pages
An obscure reference to a Perl module from 2002 where I contributed some testing and a snippet of code

Weirdest find:

My name in an Engadget post from October 2007 by Erica Sadun around an iPhone rumor. This predates my Twitter account, so I have no recollection how I might have passed on the link to that The Register article to her?!

All in all, some interesting finds and I hope it will prove to be equally useful with other types of searches. Fingers crossed!

Moving to Europe

Sun, 13 Apr 2025 00:00:00 +0000

Shifting things around

For reasons, I am slowly but surely moving most of my online resources to European providers. This has meant a journey of discovery, finding suitable ones for:

Domain registration
DNS hosting
Some place for a few static (placeholder) websites
Virtual servers of some kind
Object storage or something similar
Git repository hosting

Resources that helped

Most of my information has come from european-alternatives.eu. Some test accounts were needed here and there because it isn't necessarily obvious how well they may cooperate with automation tools like Terraform and how much click-ops will be required, and how well does their web UI work in that case.

Obstacles I encountered

There were some hiccups with getting the European companies to play nice with my US credit cards.

One disappointment was because the company (Infomaniak 🇨🇭) seems to be incapable of sending SMS to US phone numbers. Pity, because they looked promising otherwise.

Another Nope was finding out that OVH 🇫🇷 pushes you to OVH US if you're in the US or Canada, a wholly owned subsidiary of OVH Group, defeating the purpose of going with European companies.

Choices/solutions

Service	My Pick	Based in
Domain registration	https://www.openprovider.com/	🇳🇱
DNS	https://desec.io/	🇩🇪
Static websites	https://www.statichost.eu/	🇸🇪
Git repository hosting	https://codeberg.org/	🇩🇪

Of these, statichost.eu was a nice find, because it will let you point to a git repository (public or private), and run with it, with support for either fully static content, or one of several static site generators, and even pick a specific version to avoid surprises.

Still to do

That leaves virtual servers, for my self-hosted mail-server and the one-person-Mastodon instance. Most likely I'll pick one of UpCloud (🇫🇮) or Scaleway (🇫🇷), after sitting down again to do a side-by-side comparison.

For object storage Scaleway seems to have a reasonably priced solution, but that also needs another look.

Hello again, world?

Mon, 13 Nov 2023 00:00:00 +0000

From Twitter to Mastodon

For many years Twitter was my main social media account. My Facebook account has not been visited in many months and Instagram has mostly been kept alive to follow some friends and family's postings. Though I originally did post from time to time, the last few years it had turned mostly into a read-only account.

Then last year around the time of the change in ownership, I gave up on it and began an effort to migrate my network to Mastodon or, more accurately the Fediverse. Initially at mastodon.social, the flagship server for the Mastodon software, with an account that I had created in May 2022 but began using in earnest in October. Later I migrated to an account on a personal server.

Starting to feel an itch

As I was posting on my Mastodon account, I regularly found myself writing more than was going to fit in the space of the default 500 characters. Sure I could change that default (although it is a little bit more work than just changing a configuration file entry), but that felt wrong. To me, the fact that I was bumping against that ceiling suggested I was looking for a different medium for those posts.

Static Site Generator to the rescue

While I did feel the itch to write some longer form things, the idea of running (potentially vulnerable) infrastructure for this that would need to be maintained did not appeal to me.

Comes along Hugo (docs, themes, etc at the official website) which comes with a generous set of themes that can be tweaked as needed, allows me to write posts and other pages in Org Mode syntax, and has built-in support for deploying its output into an AWS S3 bucket.

More details on the project page.

Website

Mon, 13 Nov 2023 00:00:00 +0000

Status

Active in its current form since November 2023 and its current location since March 2025.

What

My personal website.

Where

Right here.

Content

Everything on here has been generated once by a so-called SSG and then dropped in place.

The content is generated with Hugo. I picked the Ananke theme, and made some minor modifications:

Modified the default Hugo RSS template (layouts/_default/rss.xml):
- Change the feed to include full posts instead of just the summary
- Add <category> tag(s)
Copied and modified the default single (post/project) page (layouts/_default/single.html):
- Include an optional "Updated …" bit after the original post date
- Move the context menu (with the table of contents and related posts) before the content, which places it to the left of the text, or at the top of page instead of at the bottom in a small screen situation
Copied and adapted the main page (layouts/index.html) just to make the text there not be centered.
Copied and adapted the footer (layouts/partials/site-footer.html) just so I could make the copyright years be a range instead of just the current year. I tried to put the beginning of the range in the copyright configuration setting but then that ended up looking funny in the RSS source

Beyond that, I've left the theme as-is and I'm taking advantage of its (and Hugo's) features.

Hosting

Until March 2025 the static site was hosted in an AWS S3 bucket, made available through a Cloudfront distribution to make it available via https.

At that point, I moved it to statichost.eu, which specializes in hosting SSGs, directly from Git repositories, and hosting it with http and https.

Keeping up with the Joneses?

As so often happens, I looked away for one minute (alright, closer to 4-5 months) and the version of Hugo that I get when I do a fresh install (from Brew) is a bit different than what I started with, and that is going to break some things.

To give myself a stable work environment and breathing space I decided to run the hugo command inside a quick'n'dirty container:

The Dockerfile

  FROM cibuilds/hugo:0.119.0

  EXPOSE 1313

A super simple Makefile

  hugo-docker:
          docker build --tag myhugo .

  server-beta:
          docker-helper server --bind 0.0.0.0 --environment beta

  server-live:
          ./docker-helper server --bind 0.0.0.0 --environment live

  deploy-beta:
          ./docker-helper --environment beta && ./docker-helper deploy --environment beta

  deploy-live:
          ./docker-helper --environment live && ./docker-helper deploy --environment live

And a shell script to glue it all together:
```
  #!/bin/sh

  if [ $# -eq 0 ]; then
      echo "Usage: $0 <hugo-arguments>" 1>&2
      exit 1
  fi

  DIR=$(git rev-parse --show-toplevel)
  docker run -v $HOME/.aws:/home/circleci/.aws -v $DIR:$DIR -p 1313:1313 -w $(pwd) -it myhugo hugo $*
```
That script fires up the container with only the current Git repo exposed within the container, and starting off at the current working directory as well, then running the hugo version that I grabbed. It also shares ~/.aws for the deployment steps.

Now I can run a command like this and I can sanity check that the site is going to look ok with that particular version, and then later figure out what needs to be fixed as I nudge the version along at my own pace:

make server-beta

And then deploy when I'm happy:

make deploy-live

Noting that now that I've moved to statichost.eu the deploy-* targets are obsolete, but it's still useful to test locally with server-*.

About All This

Mon, 01 Jan 0001 00:00:00 +0000

The Person

Erwin is a Dutch citizen, living in the United States of America. He considers himself a coder, a programmer, a software engineer, whatever title gets the point across.

Most of his Github (and a few of the Codeberg) repositories are private, but he shares a few scripts written over the years, for fun or for practical purposes.

His time on ~~Twitter~~ "X" has come to an end and instead you can find him hang out on the Fediverse which, for all its own flaws, has been a much more attractive and friendly alternative.

This Domain

The is-here.com domain was registered a long time ago, for reasons mostly forgotten. Domains and email addresses with it can be funny wordplay, until you find yourself needing to spell out your email address to people that had no idea that domain names can have dashes in them, and that they are actually significant.

The Disclaimer

Any opinions on this website are Erwin's, not those of his employers, past or present.

Contacting Erwin

Mon, 01 Jan 0001 00:00:00 +0000

If you want to contact Erwin, you are invited to reach out via the Fediverse (Mastodon, e.a.). You can find a link to his active account at the top or bottom of all pages. Other options may work as well, but as they tend to involve email in some form or another, you will find that the response time is a bit lacking.