Home Monitoring Re-Write Number Four!

Since getting the Tesla Powerwall installed, our trusted Wattson has not been able to display correct figures as it can’t tell if we are importing or exporting until the Powerwall is full.  The Wattson displays a relatively static value of +150W indicating that we’re importing, yet the data from the various other devices in the house contradicts that figure.

So it’s time to say goodbye to Wattson and hand it on to a neighbour and hope they get some use out of it.

Wattson’s demise is a great excuse to upgrade to a tablet and display a lot more information than just whether we’re importing or exporting, so I’ve gone out and bought a Samsung Galaxy Tab A from JL to replace Wattson.

In order to display more information on the tablet, I needed to re-write the home monitoring application and start graphing the data at home rather than relying on PVOutput.  PVOutput is a great website, but it’s limited to a 5 minute picture of what’s going on and I’ve run out of fields to upload data, even though I donate to get extra fields! Wattson has gotten us used to being able to see what’s going on instantly rather than waiting for a snapshot 5 minutes later.

The second re-write I did of the home monitoring application in 2015 has been running well for a few years, but despite what I wrote back then about it being maintainable, it was a pain to add in a new datasource and it was written in my least favourite framework – Mule.

Since then I’ve tried re-writing it in Node.js, but that code was less than elegant and not tested at all… It also relied on a heavy weight MySQL database which I wanted to avoid if possible. HSQLDB may be a bit basic, but it’s served me well for many years and allows me to make changes to the files in a text editor if required.

I did learn something valuable from the Node.js re-write – consolidate the five tables I had before into one large table. I’ve changed the following five tables

to a single table for ease of storing the data and to save space.

The previous database file size was 640MB (note that’s more than 200MB per year as I blogged about the database being 400MB only last year) vs. the new single table layout file size of 240MB. Every field in the database except the composite primary keys are nullable. This allows the data to be stored into the table in any order, after all I can’t guarantee which Arduino will send it’s data first.

The next step was to work out how to convert the database from the original layout to the new layout without having my pc running at 100% for over 2 hours (the first time I loaded the data from the old tables to the new table, this is exactly what happened!). The trick was to not insert based on a select union, but to use the HSQLDB merge functionality. The two hour ETL turned into a three minute ETL. This much improved ETL time allows me to take a copy of the old database (the in use one) at any time, transform it and check the new app is compatible with the schema and can write data into the new layout correctly.

As I’ve mentioned above, the new application is no longer based on Mule and instead is a Spring Boot app.   The home monitoring application receives input using Spring MVC controllers and persists the data to the database against the date and time (rounded to the minute).

At the service layer, there’s also three separate scheduled services, one for uploading PVOutput data once a minute, one for requesting the EE addons status page and scraping the data every hour and one for calling the Tesla Powerwall API every five seconds.

EE addons status page scraping I hear you say… “what’s that for?”  We no longer have fixed line internet and rely on EE 4G internet, which is great until we run out of data two days before the end of the month!  The EE addons status page displays how much data you have used, how much is remaining and how long until the next period.  Since I’ve now got the option to display a lot of different data on the tablet, it seemed sensible to display the EE data allowance too!

For anyone interested in doing something similar, here’s a class I’ve written to read the HTML and trim it to extract the right bits of information. The fields aren’t accessible as I don’t store the information – I simply pass it straight to Splunk via toString.

package uk.co.vsf.home.monitoring.service.ee;

import java.util.regex.Matcher;
import java.util.regex.Pattern;

import org.apache.commons.lang3.StringUtils;
import org.apache.commons.lang3.builder.ReflectionToStringBuilder;
import static org.apache.commons.lang3.StringUtils.*;

public class EeDataStatus {

	private static final String ALLOWANCE_LEFT = "allowance__left";
	private static final String ALLOWANCE_TIMESPAN = "allowance__timespan";
	private static final String BOLD_END = "</b>";
	private static final String BOLD_START = "<b>";
	private static final String SPAN_END = "</span>";
	private static final String SPAN_START = "<span>";
	private static final String DOUBLE_SPACE = "  ";

	private final String allowance;
	private final String remaining;
	private final String timeRemaining;

	public EeDataStatus(final String response) {
		String allowance = response.substring(response.indexOf(ALLOWANCE_LEFT) + ALLOWANCE_LEFT.length());
		allowance = allowance.substring(0, allowance.indexOf(SPAN_END));

		Pattern pattern = Pattern.compile("(\\d+.*\\d*GB)");
		Matcher matcher = pattern.matcher(allowance);

		matcher.find();
		this.remaining = matcher.group();
		matcher.find();
		this.allowance = matcher.group();

		String timespan = response.substring(response.indexOf(ALLOWANCE_TIMESPAN) + ALLOWANCE_TIMESPAN.length());
		timespan = timespan.substring(0, timespan.indexOf(SPAN_END));
		timespan = timespan.substring(timespan.indexOf(SPAN_START) + SPAN_START.length());
		timespan = timespan.replaceAll(BOLD_END, EMPTY).replaceAll(BOLD_START, EMPTY);
		timespan = timespan.replaceAll(CR, EMPTY);
		timespan = timespan.replaceAll(LF, EMPTY);
		timespan = timespan.replaceAll(DOUBLE_SPACE, SPACE);
		timespan = StringUtils.trim(timespan);
		this.timeRemaining = timespan;
	}

	@Override
	public String toString() {
		return new ReflectionToStringBuilder(this).toString();
	}
}

When I tried writing the home monitoring application in Node.js I gave Prometheus a go to see whether that would be a good tool for graphing at home.  It worked well when graphing small sets of data, but when I tried to graph over a years worth of data, it either errored because there was too much data coming back from the query, or took a vast amount of time to refresh the graph.  It’s possible I wasn’t using the tool correctly, but I decided it wasn’t for me in this use case because of the inability to graph large amounts of data and because it’s not as intuitive as the graphing tool I’ve chosen to go with.

So what graphing tool have I chosen?  Splunk 🙂

I chose Splunk for a number of reasons:

  1. I’ll be sending less than 500MB to Splunk a day, so it’s free 😀
  2. It’s incredibly intuitive to search through data in Splunk, so I should be able to give my dad a basic lesson and he can create graphs for himself. I had considered the ELK stack, but the searching language isn’t quite as intuitive…
  3. Splunk doesn’t care about the schema of the data you throw at it.  This makes it easy to work with as I can add/remove fields when required and not have to change a schema.

Writing the data to Splunk uses the ToStringBuilder JSON format and a Log4j socket appender.  The ToStringBuilder format is configured at bootup via the following component.

package uk.co.vsf.home.monitoring;

import org.apache.commons.lang3.builder.ToStringBuilder;
import org.apache.commons.lang3.builder.ToStringStyle;
import org.springframework.stereotype.Component;

@Component
public class ToStringBuilderStyleComponent {

	public ToStringBuilderStyleComponent() {
		ToStringBuilder.setDefaultStyle(ToStringStyle.JSON_STYLE);
	}
}

And I chose the Log4j socket appender because it doesn’t require the use of tokens to talk to Splunk.

<?xml version="1.0" encoding="UTF-8"?>
<Configuration status="warn">
    <Appenders>
        <Socket name="socket" host="SERVER NAME" port="9500">
            <PatternLayout pattern="%m%n"/>
        </Socket>
        <Console name="STDOUT" target="SYSTEM_OUT">
        </Console>
    </Appenders>
    <Loggers>
        <Logger name="uk.co.vsf.home.monitoring" level="info" additivity="false">
            <AppenderRef ref="socket" />
            <AppenderRef ref="STDOUT" />
        </Logger>

	...

</Configuration>

Bringing it all together, we’ve gone from Wattson which displayed only one figure – house load – as shown in the (albeit not great) picture below:

To this 😀

And this complicated device/application diagram

Hopefully this incarnation of the home monitoring application will last a few years, but I suspect I’ll be re-writing it all again at some point 🙂

References
Tesla Powerwall 2 API https://github.com/vloschiavo/powerwall2/
Log4j2 Socket Appender https://logging.apache.org/log4j/2.x/manual/appenders.html#SocketAppender

blitzortung.org Daily Position Cron

As you probably know from my blog, I have a number of Arduino’s around the house for monitoring household and weather metrics and I’m always looking for ways to add more devices and data sets.

My dad happened to find a map one summer on blitzortung.org to view live lightning strikes and we decided to sign up for a lightning detector in December 2015 (I think!).  When you register you use your email address and can see where you are in the list of people who want a device by going back to en.blitzortung.org/cover_your_area.php and filling in your email + the other boxes.

After one year on the waiting list, our position in the queue had moved a bit, but it became tedious checking the list once a week, so I wrote a script that I’ve now made generic enough for anyone.  Simply add it to crontab and take the pain out of checking every week 🙂

GitHub gist: https://gist.github.com/vls29/aac9d3efaf265734dfc1b64c46482160#file-blitzortung-position-sh

#!/bin/bash

email=$1

epc=$(date +%s)
echo $epc

# Country does not seem to be important
res=$(curl --data "info_time=$epc&info_email=$email&info_country=United+Kingdom&info_text=TSqrb" http://en.blitzortung.org/cover_your_area.php)
#echo "-----------------------------"
#echo "HTML Response"
#echo $res
#echo "-----------------------------"

html=$(echo "$res" | grep $email)
#echo "-----------------------------"
#echo "Position Text"
#echo $html
#echo "-----------------------------"

textpositionstart=411+${#email}+1
echo "textpositionstart: $textpositionstart"

position=${html:$textpositionstart:6}
position=$(echo $position | sed 's@^[^0-9]*\([0-9]\+\).*@\1@')
echo "position: $position"

lastpositionfilename=blitzortung-last-position.txt
lastposition=$(cat $lastpositionfilename)
catresult=$?
echo "catresult $catresult"
if [ "$catresult" -eq "1" ]; then
    echo "didn't find last position file"
    lastposition=100000
else
    echo "found last position file"
fi
echo "lastposition: $lastposition"

if [ "$position" -lt "$lastposition" ]; then
    echo $position | mail -s "blitzortung.org position" $email
    echo $position > $lastpositionfilename
else
    echo "position not less than $lastposition: $position"
fi

exit

The only input to the script is the email address you’ve used on the waiting list (assuming you haven’t hardcoded it in the script like I have).  You don’t need country as that doesn’t appear to be used by the site to verify the email address.

Monumental App Update Mess Up!

A few weeks back I received a request to add in the ability to select half days in the Retirement Countdown Clock app (http://blog.v-s-f.co.uk/2016/02/retirement-countdown-clock-app/) and I decided that this was a quick change that wouldn’t take too long, so why not 🙂

Well I made a complete mess up of the update… It started off as seeming like a simple update, but I’d just had a rather large problem on my laptop that killed the SSD, so had not much software installed on the new hard drive. After all the necessary apps were installed, I set about updating the app, adding in the ability to select the half days. It only took about 4 hours in total to make the code changes and test (most of which was updating the runtime target version). I packaged it, tested it on my laptop and old phone, both of which said they would install from fresh and then added a new submission to the store.

Job done 🙂 or so I thought…

Two days after the app was published to the store, I logged in and to my horror I’d received over 11,000 crash reports!!! O.M.G!

All the crash reports were for the new version (2.1.0.0) and all were in exactly the same line of code… I wondered well how come it worked on my laptop and phone then? And the key answer was that it installed the app from fresh and didn’t do an update. I dashed around the house to find anoher phone I hadn’t tested on and updated the app from the store. Lo and behold, it crashed as soon as you tried to open the app from the start screen 🙁

I had all the info I needed in the crash reports to find the particular dodgy line of code – wasn’t handling the previously stored int and converting correctly into a decimal. Less than two hours later, a new submission was sent for approval to the store, but it takes a minimum of a day to get a submission approved… In that time the crash reports topped 20,000.

I learnt a very valuable lesson – don’t rush a change through, even if it seems simple and make sure you test it as if you’ve done an upgrade as well as a fresh install!

Sorry to all those people that downloaded the dodgy update, hopefully you’ve updated to 2.1.1.0 and it’s now working again.