I’m using a PowerWalker Basic VI 850 SB (international), which can be connected via USB. Hint: vendor does not provide USB-cable (Type B needed).
I’m going to use the Network UPS Tools (NUT) to monitor the status of my UPS and - if needed - keep my server alive.
Main motivation on this is to prevent data loss. Additionally a UPS corrects common power issues, whereas mine is line-interactive. In simple terms: it prevents under-/overvoltage to some degree.
Connected to my server, it signals when a power outage occurs, tries to keep it up until it’s battery is low and shutting it down to prevent damaged RAID-/disks.
NUT provides tools for connecting, monitoring and communicate stats about a UPS. In short: it can be used on a SOHO setup, as well as on a multi-server farm in any 19” server rack (one UPS, many appliances connected to it). Take a look the monitoring diagrams.
NUT utilizes:
upsd the server, which reads data from driver and allows network communicationupsmon monitoring to set some actions based on UPS’ stateA list of compatible power supplies can be found on the website.
Installation on Fedora Server 36 is pretty easy:
dnf install nut nut-client
Setting up /etc/ups/nut.conf, since I use it as a standalone server (without any other connected to it):
MODE=standalone
Quickly scanning for connected and supported devices: nut-scanner -U (-U scans only USB). It outputs a similar data to what we need in our next step. Just append a new UPS with driver, port and a small desc(ription) to ups.conf:
# /etc/ups/ups.conf
# appending driver/port combination with a short description
# nutdev1 was a suggestion by the scanner, it could be everything instead
[nutdev1]
driver = "usbhid-ups"
port = "auto"
desc = "My small office server"
Setting up the upsd daemon to allow the local upsmon to access data (next step):
# /etc/ups/upsd.conf
ACL all 0.0.0.0/0
ACL localhost 127.0.0.1/32
ACCESS grant monitor localhost
ACCESS deny all all
As well as a user (the password will be used only internally):
# /etc/ups/upsd.users
[monuser]
password = nieNg3eeHeixis1phooz
allowfrom = localhost
upsmon master
Optionally: If you want to access your UPS using an admin user (for instance controlling “the beep” or testing battery), you need an additional user:
[admin]
password = mypass
actions = set
instcmds = all
Finally setting up upsmon (with your monuser password):
# /etc/ups/upsmon.conf
MONITOR nutdev1@localhost 1 monuser nieNg3eeHeixis1phooz master
Running upsc nutdev1 as user nut failed, telling me no device were found. A nut-scanner -U told me access denied.
In my case a quick lsusb | grep -i UPS provided:
Bus 005 Device 002: ID 0764:0601 Cyber Power System, Inc. UPS
So I need to allow user nut to access it. Using udev it’s pretty simple:
# /etc/udev/rules.d/90-nut-ups.rules
ACTION=="add", SUBSYSTEM=="usb", ATTR{idVendor}=="0764", ATTR{idProduct}=="0601", MODE="0660", GROUP="nut"
$ udevadm control --reload-rules
$ udevadm trigger
Now test again:
su nut --shell=/bin/bash
$ upsc nutdev1
# should start driver without errors, if fine, stop it (starting as systemd later).
$ upsdrvctl start
$ upsdrvctl stop
Ready to run!
$ systemctl enable nut-driver nut-server nut-monitor
$ systemctl start nut-driver nut-server nut-monitor
Finally, everything runs as expected
To see if everything might work, you can run the following command, which will only display the commands to shutdown the system:
$ upsdrvctl -t shutdown
-t tells the command, to just test and output. Running without might actually shutdown your system. Be aware of data loss, if something isn’t correctly configured (since it might force the system to halt, after a configured timespan).
You could also use this command, to issue a “shutdown for all primary nodes” (standalone is a primary). Warning: this will halt your system - forcefully if needed(!):
$ upsmon -c fsd
The final test is to unplug the cable of the UPS to see if the upsmon reacts correctly. It gracefully shut down my server after 40min (jup, my UPS is oversized). After that, my UPS shutdown, too.
Plugging the power cable in again resulted in a reboot of my server. All set! If your server won’t start again, check BIOS settings to “restart on power loss”.
I used many references for this. Maybe one or two links might help you, too.
I’m really happy with my setup. It just works as expected and helps me getting good sleep at night, without worrying to loose my data ‘cause some neighbour hit the wall.
Now I need to measure the values of my UPS to store them in my metrics server for visualization later. Happily there is a simple tool for this: upsc nutdev1 output quite everything!
6-slim (770 MB) to 7:slim (390 MB)).
I managed to make it smaller by fixing some issues and dropping ffmpeg from the base
image. That said, ffmpeg isn’t needed by default, since it’s only used by the
gifv-optimizer, which itself isn’t enabled by default.
If you want to use the thumbor.optimizer.gifv, you can just add it yourself by
building your own docker image and activate gifv:
FROM mvhirsch/thumbor:7-slim
RUN apt-get update \
&& apt-get install --yes --quiet --no-install-recommends ffmpeg \
&& rm -rf /var/lib/apt/lists/* \
ENV OPTIMIZERS "['thumbor.optimizers.jpegtran', 'thumbor.optimizer.gifv']"
The newest versions are already available to download. Feel free to use them:
mvhirsch/thumbor:7-slimmvhirsch/thumbor:7-alpinemvhirsch/thumbor:7-fundamentalsAs always mvhirsch/thubmor:7-fundamentals comes pre-configured with an AWS S3 storage and pnquant and mozjpeg.
pngquant and mozjpeg are the recommended image optimizers for the web, reducing file size without quality loss.
If you experience any issues, don’t hesitate and raise an issue on github.
]]>You (as a software developer) should read it, too.
What every programmer should know about memory
The document is a bit outdated, since it’s from 2007 - but still worth a reading. Technology did evolve since, but the principals are still the same.
If you find a more recent version, please tell me.
Basic understandings about static RAM and dynamic RAM (DRAM). How it evolved over time and how it works (especially on multi thread processors).
Practical tips on how to optimize memory usage for performance in C.
]]>To keep yourself healthy and happy, it’s important to keep in mind: you are not alone. Phone your parents and family members (brothers, sisters, aunts, uncles, grandparents, …) as well as your friends. We live in a modern highly-technologized world where calling is as simple as writing a WhatsApp. Most of you still know how to dial a phone number ;-)
But also consider using video-calls (via WhatsApp, Jitsi, Hangouts, Facetime, …). Seeing your family and friends will make you and others happy. Talk with them.
To all parents out there: your parents want to see their grandchildren, too. Please do a regularly video-call with them.
Stay healthy by train your body. Eat healthy things (like fruits) but also keep yourself happy by eating what you like (e.g. chocolate) - it’s still okay to do so! Maybe you now have much more time as there are many countries (and many more might coming) are locking down. Now might be the right time to get yourself into shape, let’s start by searching on Youtube.
Learn something new, like a new skill or keep what you love (like reading (ebooks!), doing arts, doing sports, …). Please #StayAtHome. There is plenty of online resources to get your mind and body in a fresh state.
And if you just want to be motivated, more and more artists are playing for you (yes you!) online: Youtube #TogetherAtHome.
Play with your kids, your spouse or other members at your home. Love them, they will love you, too.
]]>Please feel free, to use this image and don’t forget to provide feedback and contribute: github.com/mvhirsch/thumbor-docker
You like this image? Please star it on Github and Dockerhub.
]]>cache:
untracked: true
paths:
- node_modules/
and your next build will be sped up. Super cool, isn’t it?
npm’s way to manage dependencies isn’t always deterministic and sometimes just do really weird. Especially if you mix versions like is it building on Node 8? Node 10? Node 11? In Gitlab-CI you just use different node-containers, they all come with different npm versions pre-packaged. Would you like to manage different caches for it? Gitlab-CI does support that. Welcome to another dependency-maintenance hell.
What about upgrading some modules in a branch? Next time building main will it extract the newer version of that upgraded module? Will npm downgrade it? Are you really sure? Of course you can use package-lock.json. Good luck, tell me if you really happy with it.
Since my days using composer on PHP projects, it’s a naive advice to cache node_modules/ (npm) and vendor/ (composer). Even if the official Gitlab-CI documentation tells you to cache node_modules/ and vendor/. It might be a different machine, different bindings, missing post-install scripts, version-mismatch or other.
Still I need performance, so what to cache? Did you know, npm has it’s own cache? By defaults it’s somewhere at ~/.npm/ (Linux). A simple command will show you: npm cache verify
You can override it with an environment variable (or setting it in ~/.npmrc). So my prefered way is to override npm’s cache and cache that folder in gitlab-ci-runners. Please note that I’m using a relative path, since gitlab-ci-runners can only cache relative paths (not absolute ones).
cache:
paths:
- .npm_cache/
variables:
NPM_CONFIG_CACHE: ".npm_cache"
We can even do it better by telling npm to prefer that cache (it will still take a look at the npm registry, but prefers the downloaded file anyway):
npm install --prefer-offline
Like above, I recommend caching composer’s cache directory instead:
cache:
paths:
- .composer/cache
variables:
COMPOSER_HOME: .composer
Happy Caching!
Update 2025: Gitlab recommend caching ~/.npm now (previously /node_modules); but still recommends to cache vendor/ for PHP projects.
]]>After reading the documentation, I was quickly looking for a Docker image to play around. Sadly, there is no official one. I found two others: APSL and MinimalCompact which did a great job providing a Docker image and fullfilling my needs. Indeed, both images just work fine. But there is more.
As for all in Docker, every image should one run one process (not running circus, which can be configured to run multiple instances of a process). Again, it just works, but it’s bloated.
After looking at the dependencies of both images, I made a decision: both images are good to go, but for my opinion, quiet to large (in size and dependencies). So I decided to build my own.
Finally, my first public Docker image arrived: mvhirsch/thumbor
It comes with the smallest footprint of dependencies I was able to manage (650 MiB instead 1.7 GiB). Still, there is room for improvement (like a planned alpine version of it). But I’m happy to announce a fully working, just scaleable Docker image. Feel free to contribute: https://github.com/mvhirsch/thumbor-docker
Sacrificing the golden calf of coding – Christian Heilmann
Why the confused name …? Because I really thought it’s all about software engineering, development, testing, building cool things (like gems or toys or mobile apps). Surprisingly it’s more about hiring people, which is just annoying - especially if you wanted to learn cool new stuff about software development or make new friendships by hacking code together.
Starting with Steve Wozniak about all and everything within uprising technology (you might be able to watch it on Youtube, too), the congress started well with over 8.000 people attending to 6 stages with different talks. Speaking about Stories, Backend and Frontend development, automation, VR, Security, Testing and lots of other interesting things. So I started into a good session, full of interesting people to listen to! I heard about nginx unit, apache kafka, managing cloud deployments well, beeing better at development, doing nodejs right, - there was a lot of awesome talkers to hear!
A talk that I got inspired was Christian Heilmann’s Sacrificing the golden calf of coding. It’s all about software engineering and to use software (development) right. Because why bother with the legendary Tabs vs Spaces anyway? Stop rebuilding/redefining your work; there even is a machine-learning tool building HTML based on your screenshots.
Another interesting talk was given by Sara Viera Your brain does not have a fix flag, inspired by eslint --fix. What a genius funny little women!
Something very crude was: hey guys, I heard you like to go home early, so we stopped serving coffee since 3pm. Just why? The planned congress appears to be 6pm. Do you want provoke the mob?
Something I want to applaud at is the performance of the wifi. I’ve never seen such a big congress, with a working wifi, but this time it was all about performing well and absolutely solid. But, why the heck did you need to remove the wifi password for all attendees on the second day? Aren’t developers able to type in a simple wifi password (seen on every displayed across the whole location)?
Do you want to work with us?
Apart from the talks on stage, there was a lot to discover on the congress: many companies showed off some technologies and frameworks they work with. And like many, they wanted everyone to work with them. No sorry, I just wanted to look at what you’re building right here (a duck on the monitor is an eye-catcher - thank you for trapping me in). So everywhere I looked or I went they wanted to hire me. By the way, I’m happy at my work and I don’t want to leave.
After all, I saw many people at the conference, talk to most of them and were able to learn some nice tricks. At all, it was a success for me, but I think I’ll never attend (as an attendee) to the WeAreDevelpers World Congress again.
]]>