Programming challenge: Protohackers 3

Protohackers is a server programming challenge, where various network protocols are set as a problem. It has started not so long ago, and the No 3. challenge was just released yesterday, aiming at creating a simple (“Budget”) multi-user chat server. I thought I sacrifice a decent part of my weekend give it a honest try. This is the short story of trying, failing, then getting more knowledge out than I’ve expected.

Definitely wanted to tackle it using Python as that’s my current utility language that I want to know most about. Since the aim of Protohackers, I think, is to go from scratch, I set to use only the standard library. With some poking around documentation I ended up choosing SocketServer as the basis of the work. It seemed suitable, but there was a severe dearth of non-dummy code and deeper explanation. In a couple of hours I did make some progress, though, that already felt exciting:

  • Figured out (to some extent) the purpose of the server / handler parts in practice
  • Made things multi-user with data shared across connections
  • Grokked a bit the lifecycle of the requests, but definitely not fully, especially not how disconnections happen.

Still it was working to some extent, I could make a server that functioned for a certain definition of “functioned”, as the logs attest:

Console log of server messages while trying my Budget Chat Server implementation.
Server logs from trying my Budget Chat Server

Creating a Prometheus metrics exporter for a 4G router

Recently I begun fully remote working from home, with the main network connectivity provided by a 4G mobile router. Very soon I experienced patchy connectivity, not the greatest thing when you are on video calls for half of each day. What does one do then (if not just straight replacing the whole setup with a wired ISP, if possible), other than monitor what’s going on and try to debug the issues?

The router I have is a less-common variety, an Alcatel Linkhub HH441 (can’t even properly link to on the manufacturer’s site, just on online retail stores). At least as it should, it does have a web front-end, that one can poke around in, and gather metrics from – of course in an automatic way.

The Alcatel HH41 LinkHub router that I had at hand to use

Looking at the router’s web interface, and checking the network activity through the browsers’ network monitor (Firefox, Chrome), the frontend’s API calls showed up, so I could collect a few that requested different things like radio receiving metrics, bandwidth usage, uptime, and so on… From here we are off to the races setting up our monitoring infrastructure, along these pretty standard lines:

Programming Taiwan

Taiwan WWII Map Overlays

A while ago I came across the Formosa (Taiwan) City Plans, U.S. Army Map Service, 1944-1945 collection, in the Perry-Castañeda Library Map Collection of the University of Texas in Austin. I’m a sucker for maps, enjoy learning about history a lot, and I have a lot of interest in my current home, Taiwan – so you can call this a magic mix of cool stuff.

There are 26 maps in the collection, made by the US Army by flying over different parts of the island, and mostly I guess stitching together aerial photographs. The maps themselves were not that easy check in an image viewer, since there’s no context, zoom is clumsy, and have no idea where about half the places should be located. Instead, I thought it would be great to have them as an overlay on top of current maps and satellite imagery on Google Maps.

The result is Taiwan City Maps overlays, which does exactly that. Feel free to click the link and explore right now! In the rest of this post, I try to first show how that page was made, and also some history lessons I gained by making it.


Automating the hell out of it

Even before the 4-Hour Work Week made me more serious about this, I really enjoyed automating tasks, that benefit from not needing to remember to do, or would be troublesome to do otherwise. This frees up a lot of time, keeps a bunch of problems away, and it is actually quite fun when the information comes to me instead me going to it.

Now I have automated checking my bank account and credit card balance, updating dynamic IP of server, ebook sales numbers, and network clock synchronizing. There are some general ideas that I summarize, then give an intro to all of those scripts.

Banking script


Most of my scripts are written in bash, because it’s relatively straightforward to hammer out simple stuff, and it is surprisingly simple to do a lot of things once I have thought enough about a problem. The Advanced Bash-Scripting Guide is always on my reading list, but I usually get to check only the parts that are relevant to the given problem. You can get quite far with a few simple constructs.

The most common parts I seem to come across:

  • if-then-else constructs: if [ -f ‘directory ‘]; then echo “Found!”; fi
  • for loops: for f in *.png; do optipng $f; done
  • loading the results of a command into a variable: VAR=$(command)

For most other problems with a little keyword-fu there’s always an answer on StackOverflow or on the web.

Another group of scripts uses Python, when a bit more data-manipulation is needed, like web scraping or JSON parsing. Actually, all of the scripts could be rewritten in Python for consistency, and it would probably be be simpler too, which is something for the future.

As a general tip, most of these scripts need tweaking, and all of them are sort of alpha-beta quality code. To facilitate hacking and reduce heartache of mangled clever code, I keep everything in git repos. I share those repos online, so have to make sure there are no secrets checked in, ever. It helps to strategically use .gitignore, separate files for the secrets, and having an example how that secrets file should look in the inside.

Most of these scripts are run periodically by cron, so it is worth having some basic knowledge about how to schedule it.

Some scripts send me emails under specific circumstances (some after every run, some when new information appears), and for good delivery I have set up postfix to use Gmail as an SMTP relay. This way I’m sure to receive the emails and receive them quickly.


These are the scripts I use most often and the longest. Still, many of them are under development and adjust them whenever I learn how to do things better. I list the links to all their repos, where it can be improved.

Banking account balances

My two main bank accounts are queried once a day for available balance and I’m notified by email. Both accounts needed quite a bit of web scraping (and got them done at two different OpenHack Taipei events). The banks’ websites are pretty awfully organized (iframes within iframes within iframes; not using CSS classes and id), though it doesn’t have to be good for me, it has to be good for the bank.

Cathay United Bank

The cathaycheck (click for repo) script queries the available balance at Cathay United Bank by logging in with curl, and parsing the final page with Beautiful Soup. The script can be a skeleton for any other website where on has to log in and then navigate over a series of pages to get the information. The required HTML variable names can be extracted with the help of the Inspect Element tools in Chrome.

At the moment the credentials is stored in the crontab command, which is not really ideal, should rewrite to use a secrets file, though given that it runs on a server where I’m the only user (and root), for me there’s no practical difference at the moment. I have set it up to receive an email at the end of the day with the current balance.

ANZ Taiwan credit card

The anzcheck (click for repo) script queries my spending with the ANZ Taiwan credit card. Again bash for logging in and Beautiful Soup for parsing the final page. It needs a bit more logic extracting information from a table, because the websites developers added no classes or ids to the items to make it easier to understand – or for them to style, but that’s not my problem.

Just recently updated that it extracts the spending items added to my balance on a given day, so I can will never be caught by surprise again (hopefully). Since many of my charges go to companies that have Chinese names, I quickly run into the problem of having to tell my Heirloom Mailx (that I use to send emails on my ArchLinux box)  that the text I want to mail is plain text, not an attachment. With some hacking the solution was to add a few more commands to “mail” so it knows that the text is UTF-8. From “” in the repo, the parameters needed are:

-S sendcharsets=utf-8 -S ttycharset=utf-8 -S encoding=8bit

I could still extract some more information from the bank’s website, though nothing really urgent.

No-IP address updater

At the Taipei Hackerspace we have a handful of servers running, but the residential internet connection is provided by Chunghwa Telecom only gives us a dynamic IP address. Applying for a static IP seems to be pretty troublesome, so in the meantime I’m using a script on one of the servers to update the IP address associated with our dynamic address.

The no-ip-bash-updater (click for repo) script is forked originally from elsewhere, but I have rewritten it quite a bit so that it

  • needs no extra file to store the current IP address, but compares external IP with a DNS query
  • stores no secrets in the file

It uses a pretty straightforward API call with HTTP authentication, the only real logic in there is to check when that call actually needs to be made.

E-book sales

Recently I have helped a friend to publish an ebook version of How to Start a Business in Taiwan on Leanpub, and of course I want to know when there are any sales are made (disclaimer: I don’t get a cut of the sales, all goes to the author). The leanpubsales (click for repo) script is written in Python, because using JSON there is easier than it would be with bash. The call otherwise is quite simple, just keep an external file around to check if the sales number have increased or not, if yes then send an email. To send an email conditional on the output the the script the “ifne” command from moreutils is very useful (meaning: “if input is not empty”).

The query is run periodically, and lovely to receive the results. I will surely set up a script when I get my own book ideas published on Leanpub.

RTC correction

As a physicist in atomic physics, which is the area of science very much concerned about keeping precise time, keep all my servers’ times synchronized with network time protocol (NTP) using chrony. One difficulty is that the real-time clock (RTC) of those computers is pretty crappy and drifts away. Wouldn’t be a problem if I never restart them, but a pain if I do: after restart it can be tens of seconds away until the time is synchronized again.

Chrony can sync NTP and the RTC, but it doesn’t do that automatically, I have to trigger it manually. Instead I have written up an rtccorrect (click for repo) script that is run every 2 hours or so (could be done just once a day, actually), and eliminates the drift of the RTC.

Server backup

For backing up data between servers rsync has proven invaluable. I have a couple of scripts that do just that, though those are among my oldest ones and at that time I haven’t separated out personal information (way too easy to inline every credential, email, login, and all that), so I need to sanitize that. A couple of  ideas about these backup scripts:

  • sometimes higher transfer speed can be achieved by messing with the ssh algorithms, eg. passing “-e ‘ssh -c arcfour'” to rsync
  • more often there’s even better performance when there’s an rsync daemon running on the remote computer (though with Raspberry Pi, both cases are still frustratingly slow)
  • can exclude some files if no need to transfer them, eg: “–filter=’- *.part'”
  • using rsync not just to transfer but to mirror, the “–delete” (delete at target if doesn’t exist at origin) and “–archive” are pretty useful

For these backups I also use the Dead Man’s Snitch to know when things didn’t work out, e.g having a similar command in the cron list, where is my script’s name, xxxxxxxx is the snitch ID from my account: && curl -s > /dev/null

This way I got to know when my backup server was dying all the time because of bad heatsink, or my host server by flaky hosting company….


I guess there will be just more automation in the future, and maybe many of these scripts can be ported onto a common base so new ones are made much easier. What else do you guys automate?

Programming Taiwan

Barometric recording of Typhoon Soulik

It all started a few weeks ago with Sparkfun having “20%-off” day, when I got myself (among other things) a BMP085 barometric pressure sensor. When it arrived, I have soldered some pins on it, and set it up with an Arduino Nano, to have the readings off it easily.

View of the circuit
BMP085 barometric pressure sensor breakout board from Sparkfun

Originally all I wanted is just some laid back pressure recording, so maybe I can use that to predict the weather a bit. “Pressure falls: bad weather comes, pressure rises: things will clear up”. I was recording for about a week, and nothing really noteworthy came out of that.

Then it was the news, that the year’s first typhoon is on the way to Taiwan, and it was supposed to be a big one. Obvious that I will try to record the barometric pressure pattern of its passing, but wanted to make it more interesting and informative. More visual than just the timeseries plot of pressures.

The Japanese Meteorological Agency (JMA) is a good place to watch for information about typhoons. They list path prediction, typhoon properties like strength, wind speeds, and central pressure, have satellite imagery. Putting these together, two days before the typhoon arrived, I set up a script to download the satellite imagery as it became available.

Satellite picture of Typhoon Soulik and location of Taiwan on 2013-07-12 morning
The morning before the typhoon arrived

The JMA publishes usually 2 satellite images in an hour for our North Western Quadrant (at :00 and :30), one of them covers the whole area, the other covers just the top 80% or so, leaving a dark band on the bottom. Nevertheless, matching up the pressure reading with the satellite pictures would be a good little project for this time.

Friday came, the government gave the afternoon off, though it turned out no landfall happened till everyone supposed to be off anyways, just a bit of on-and-off rain. People stocked up on convenience store food (I now have a good supply of instant noodles:) and water, taped over their glass windows, take in their plants and BBQ equipment from outside – well, those who have planned.

Around 10pm the big rain has arrived, here’s a video of how it looked from my window. Went to sleep later, and got woken up around 3:30am by the rain having changed into pretty darn big wind. Here’s another video of the violent part of the typhoon that time in the morning, that doesn’t even really do it justice. The houses around here are pretty tall, and I wonder if they have protected from the wind, or been artificial canyons channeling it. Some things got broken, though not as much as I expected – which is a very good thing.

In the meantime by the power of the Internet I have checked out the pressure reading, how is it going a few miles away in the Taipei Hackerspace, where I have left the barometric pressure sensor (the geolocation is 25.052993,121.516981)

Here’s the entire recording of the approximately 2 days of typhoon. It was pretty okay weather in the start and end of the plot.

Plot of pressure readings
Pressure reading during the passing of Typhoon Soulik, recorded at the Taipei Hackerspace

The readings have been corrected to sea level (from about 20m height, where the Taipei Hackerspace is), should be good within 1hPa or less.

The the pressure was indeed dropping like a rock, and the dip on the graph coincided with the most violent wind that woke me up. According the JMA, the central area of the typhoon had pressures down to 950hPa, which means that core must have passed pretty close to here, having readings below 958hPa, though probably not directly, as it didn’t stay down there for long.

I made a video syncing up the pressure reading and the satellite picture. The red dot on the video marks the recording location. (Watching it in full screen and HD makes it clearer.)

I would wonder what was the flat part in the readings while the typhoon was leaving. Maybe sign of changing direction, by the look of it.

Either way, this was fun to do, and I am glad that only a few people got hurt here, much fewer then even during the less powerful typhoons. Maybe getting people scared a little (like with this “super typhoon” stuff that went on) helps them keep safe? Just don’t use it too often.

Extra material

I put almost all material used here into a gist: the satellite imagery download script, the plotting, the movie frame generation, the movie generation script, and the complete barometric recording. Because this last part is pretty big (5Mb), Github truncated the rest of the scripts. I guess it’s okay to check check it out. Will add the Arduino sketch to read the sensor and the logging script later.

The satellite imagery weighs about 60Mb, so don’t put it online, but if anyone wants them, let me know.

Keep safe!