The title of this blog is a quote from James Clear’s Atomic Habits book. The tl;dr of this book is a set of strategies to get 1% “better” each day.
If that’s the case (and I certainly agree), here are some of the recent votes I cast and why.
Completed a higher ed management curriculum in strategy and innovation. I cast this vote so that I could learn new models for expressing existing or new ideas in a way that more effectively conveys the substance to stakeholders. I also met a ton of wildly successful business leaders and faculty that I otherwise would not have had access to.
Worked with our engineering and management teams to re-plan a major project so that we can incorporate additional engineering resources and adjust to more aggressive timelines. I cast this vote because our existing plan would not meet business objectives and that’s not my style.
Proposed structural changes for Red Hat engineering as a whole to set ourselves up for a more sustainable future (firm believer that structure dictates behavior). I cast this vote because our strategic goals were drifting out of alignment with incentives and goals for the IC levels.
Met with Emerging Tech group to identify ways for them to support our managed services growth and technical strategy. I cast this vote because while our CTO office incubates fantastic new and innovative technologies, we have to leverage those skills in different ways in the future.
Worked with an internal team on strategy documents to draft a position paper on “Open Services”. I cast this vote because I believe firmly in Red Hat’s business model and want to see its core ethos carried forward into the managed services world.
Kept up my mentoring schedule. I cast this vote because I want to remember how my mentors affected me and pay it forward. Oliver Parker, here’s to you sir.
Worked with a small group of folks to create a coloring book on Site Reliability Engineering called Reliability Nightmares. Red Hat has over the years delivered a number of coloring books for topics we care about. I cast this vote because “culture eats strategy for breakfast” and I felt this would be a fun tool in fostering a movement.
I’m still digesting the rest of this book, but wanted to get some basic thoughts about this particular quote written down.
I wrote this post in July 2017, and I just found it (4 years later) in my Drafts folder. What a great journey down memory lane it was to read this today. Ask 2017 me if he knew what he’d be into 4 years on, and…yeah. Life moves pretty fast.
We have been doing a lot of hiring lately — I am lucky to be at such a company. It feels like that’s all we’ve done over the time that I’ve been at Red Hat. In every interview I am routinely asked what is it like to work at Red Hat. Mostly I’d pass on a few relevant anecdotes, and move on.
As I’ve just come up on my 8 year anniversary at Red Hat, I thought I would write some of this stuff down to explain more broadly what it’s like to work at Red Hat, more specifically in the few groups I’ve been in, more specifically my personal experience in those groups…
How did I get here?
In 2007, I met Erich Morisse at a RHCA training class at Red Hat’s Manhattan office. Erich was already a Red Hatter, and somehow I ended up with his business card. Fast forward a year or so, and…
I got married in 2008 in Long Island, NY. Within 3 months of that, I applied at Red Hat and flew to Raleigh for the interview. Within 4 months of that, my wife and I moved to Raleigh, and I started at Red Hat on July 20, 2009, as a Technical Account Manager in Global Support Services. I was so excited that they’d have me that the 15% pay cut didn’t bother me.
Life as a TAM
I think there are still TAMs in Red Hat, but I won’t pretend to know what their life is like these days. My experience was filled with learning, lots of pressure and lots of laughs. We had a great group of TAMs…I am still in contact with a few, even 6+ years later. As a TAM we are ultimately tasked with keeping Red Hat’s largest accounts happy with Red Hat (whatever the customer’s definition of happy is). That can mean a variety of things. I personally found that the best way to build and maintain a good relationship was to be onsite with the customer’s technical team as much as possible. While that meant a lot of 6:00am flights, I think it ended up being worth it if only to build up the political capital necessary to survive some of the tickets I’ll describe below.
At the time, TAMs carried about 4-6 accounts, and those accounts largely came from the same vertical, whether it was government, military, national labs, animation studios, and my personal favorite FSI (financial services industry). I gravitated towards the FSI TAMs for a few reasons:
They were the most technical
They had the most pressure
I felt I’d learn from them
I ended up moving to that sub-group and taking on some of the higher profile banks, stock exchanges and hedge funds as my accounts. Supporting those accounts was very challenging for me. I was definitely in over my head, but actually that is where I thrive. For whatever reason, I naturally gravitate towards pressurized, stressful situations. I think there was an experience at a previous job at a datacenter operator (where we were constantly under pressure) that made me learn how to focus under duress and and eventually crave pressure.
I’ll relay two stories from my time as a TAM that I will never forget.
2010: Onsite for a major securities exchange platform launch (moved from Solaris to RHEL). This led to one of the nastiest multi-vendor trouble tickets I was ever on. That ticket also introduced me to Doug Ledford (now one of the Infiniband stack maintainers) and Steven Rostedt (realtime kernel maintainer, sadly now over at VMware). In retrospect I cam see how much I grew during the lifetime of that ticket. I was getting access to some of the best folks in the world (who were also stumped). Helping debug along with them was truly an honor. I think we went through over 40 test kernels to ultimately fix it.
2011: A customer purchases a fleet of server gear that has buggy NICs in every aspect. Firmware is terrible. Drivers are not stable or performant. While the hardware issues were not on my plate, certainly the drivers in the kernel that Red Hat was shipping were very much my responsibility. In this situation, I made several trips out to the customer to ensure them that everything was being done to remedy the situation. I knew this was a serious issue when each time out there I was presenting to higher and higher ranking management. We worked with that vendor daily for quite a while. They fixed bugs in both firmware and driver (upstream), Red Hat kernel folks backported those patches and we tested everything onsite. I don’t know if we got to 40 kernels, but it was at least 20. Plus a dozen or so firmware flashes across roomfuls of machines. This scenario taught me:
I needed to up level my public speaking experience if I was going to be in rooms with highest levels of management. To do this I joined local Toastmasters club along with another TAM. That other TAM founded Red Hat’s own chapter of Toastmasters, and I was the first to speak at it.
I should get more hands on experience with high end hardware itself so that I could relate more to the customer’s Ops folks. I ended up working with some gear loaned to me by Red Hat Performance team. They always seemed to have the cool toys.
More about tc, qdiscs, network buffers, congestion algorithms and systemtap than I’d care to admit.
At time time, I felt like I barely survived. But feedback I received was that I did manage to make the best of bad situations, and the customers are still customers so…mission accomplished. I also became the team lead of the FSI TAMs, and began concentrating on cloning myself by writing documentation, building an onboarding curriculum and interviewing probably 3 people a week for a year.
Becoming a performance engineer
After working with those exchanges, I knew a thing or two about what their requirements were. I got a kick out of system tuning, and wanted to take that to the next level. My opportunity came in a very strange way. Honestly, this is how it happened…I subscribed to as many internal technical mailing lists as I could. Some were wide open and I began monitoring them closely to learn (I still do this).
One day a slide deck was sent out detailing FY12 plans for the performance team. I noted buried towards the end of the deck that they planned on hiring. So, I reached out to the director over there and we had about an hour long conversation as I paced nervously in my laundry room (it’s the only place I could hide from my screaming infants). At the time, that team was based in Westford, MA. I flew up there and did a round of interviews. Within a few days, I was hired and planning my transition out of the support organization.
I believe what got me the job was that I had learned so much low level tracing, and debugging hackery while supporting the FSI sector that I ended up doing very similar work to what was being done on the performance team. And that experience must have shone through.
Being a performance engineer
I remember my first project as a performance engineer: help the KVM team to see if they could use ebtables to build anti-spoofing rules into our hypervisor product called Red Hat Enteprise Virtualization. I remember thinking to myself…oh shit…what is RHEV? What is ebtables? I was under pressure again. Good. Something familiar, at least. To help out the RHEV team I had to quickly learn all of the guts of both topics as well as build load/scale tests to prove out whether it would work or not. I’ll skip to the punchline though…ebtables is abandonware, even 6 years ago. No one cares to fix anything and it’s been on the guillotine for a long time. Based on the issues encountered, I might have been the first (only?) person to really performance and scale test it.
This initial experience was not unlike most experiences on the performance team:
You generally have no clue what the next project will require, so you get very good at soaking up new material.
Don’t be surprised…you are likely the first person to performance or scale test a feature. Get used to it. Developers develop on their laptops.
Most of that is still true to this day — although as time went on, I learned to be more proactive and to engage not only with developers about what they’re working on, but also religiously reading LWN, attending conferences like LinuxCon and like I mentioned, subscribing to as many mailing lists as possible.
The biggest project (not for long) I had on this team was the initial bringup of RHEL7. I look back with great fondness on the years 2012-2014 as I was able to see the construction of the world’s leading Linux distribution from a very unique vantage point: working with the very people who “make RHEL feel like RHEL”. That is … debating over kernel configs…backwards compatibility discussions…working with partners to align hardware roadmaps…GA/launch benchmark releases…can we do something like kSplice…will we reduce CONFIG_HZ.
This last bit brings me to the part of RHEL7 that I had the most to do with…timers. As the vast majority of financial transactions happening on stock exchanges occur on RHEL, we had to pay very close attention to the lowest levels of performance. Timers are an area only the smartest, bravest kernel developers fear to tread. Our goal was to build NOHZ_FULL and test the hell out of it. Nowadays we take this feature for granted in both the financial industry as well as telco where without nohz_full (I am told), all the worlds packets will be a few microseconds late. And that is not good.
You can see some of my nohz_full work here (or read the RHEL docs on the subject, as I wrote those too).
While Red Hat was not my first job, I do consider Red Hat my first (job) love. It is the first job I had that I’d call career-worthy, in that I could see myself working here for a while (there was plenty of work and the company was growing).
I wrote this post in May 2019, shortly after changing roles into the Red Hat Service Delivery SRE team, which is the group building Red Hat’s -as-a-Service tech, at the time largely OpenShift itself (but since has grown wildly). The external version you’re reading has been very lightly edited for internal stuff not relevant to the overall message.
I’ve found myself going back to it often to remind myself of my core own core values, sharing and re-sharing it internally when I feel we get off course, and also putting it in front of new hires so they understand where I am coming from. As expected, some of my thinking has evolved in the intervening years, but I feel the content has aged fairly well.
Service Delivery: Reliable, Secure, Performant, and…Boring?
They have a saying in the American sport of baseball: “a well-umpired game is one in which you don’t even remember seeing the umpires”. In baseball, it’s the umpire’s job to make all aspects of a game reliable (predictable, that is – with little variance), secure (dependable and without prejudice), performant (keep the pace of play) and otherwise as “boring” as possible.
What could this mean from the viewpoint of Red Hat Service Delivery? How does this map to running managed services? As employees running a managed service, we’re the umpires. It’s our job to offer dependable service that meets our personal standards as well as contractual SLAs. I don’t know about you, but my standards for a service, e.g. one I’d be happy to pay for, is higher than the SLA from the company. In other words, we should aim to exceed our customer’s expectations.
Building mature services takes time. It also takes teamwork and communication. We should endeavor to have the same levels of integrity and trust in each other that we hope to earn from our customers. It’s what we do under pressure that helps build trust. And while we should strive for perfection, it is much more important that if we fall down seven times, we get up eight.
How do we do this?
Above all, it means integrity, dependability and trustworthiness. That our customers can trust we’ll do the right thing, even when no one’s looking. And when we slip up, we work diligently to fix it so it’ll never happen again.
How do we achieve this?
A silly (yet applicable) example from my personal life: I’ve been a customer of Google Music for many years. Because they’ve let the app languish, I decided to look for alternatives. I know that Spotify is very popular. However, as a smaller company (and in light of the depth of experience that Google has running managed services) I thought Spotify’s service might not live up to my expectations. I was wrong. Because Spotify is so focused on their one product (streaming music), and chiefly concerned with user experience and design, it felt like they cared about me as a user. Spotify exceeded my expectations not only with their focus on me, but their service is rock solid too (which I did not anticipate!).
Let’s be like Spotify, and exceed our customer’s expectations.
We’ve got demands coming from customers, internal product teams, the shifting sands of upstream and our competitive landscape itself. How do we exceed customer expectations, and reliably serve those many masters?
I think there’s a tractable approach – these groups do have something in common.
First, they want a reliable, secure and performant substrate upon which to build their product or business. They want our managed services to be boring – but not TOO boring. We need to make sure to fix bugs before adding new features. This does not mean we stop adding new features – it means we add them responsibly. Together we build the CI/CD workflows, coding practices and monitoring that de-risks the rollout of new features.
Let’s get away from the waterfall mindset, and ship code hourly. Let’s build the confidence and scaffolding that lets us test things in production.
Code Quality and Testing
How do we build that confidence? Some will come with experience, yes. But the majority needs to come from thoughtfully designed implementations, well-written, defensive coding practices, pre-mortems, and significant, thorough, end to end test coverage. These are the attributes that Red Hat preaches to our customers who engage in the Red Hat Open Innovation Labs.
Let’s hold each other to these standards – clearly communicated constructive criticism is a gift.
With the right testing in place, we can establish backpressure mechanisms back into our upstreams and infrastructure providers. A well-honed and managed CI pipeline for each component, along with a comprehensive end-to-end test suite should be at the forefront of our minds when writing code or fixing production issues. Pre-mortems leading to new tests. Post-mortems leading to new tests. Rinse and repeat.
Observability, Contributing Code, and Competition
An observable system is one that can be easily understood. What has it done, what is it doing, what will it do. Observable systems have the scaffolding built in to make it easy for anyone to answer these questions (not just the author of the code). We’ll be more productive, from developers through support, if our systems can be easily understood. Developing systems that interact responsibly with each other takes Systems Thinking. Watch this 60-second video for a quick overview.
How does this tie back to contributing code? If you understand just a bit more of the entire system than the code you’re responsible for, the code you’re responsible for will improve. You will begin to know what counters need to be put in place, what tracepoints will be helpful, how to create useful (actionable) logs that make your system understandable. We’ve all supported or debugged a poorly instrumented program.
Let’s build observable systems and dashboards that help enable higher service levels, reduce MTTR and exceed our customer’s expectations.
One of my favorite quotes shouldn’t surprise you (I did just spend a decade working on performance):
Let’s be scientists.
Another upside to observable systems – we can’t let competitors know our customers better than we do. Observable systems lead to “digital exhaust” that can drive customer happiness in a variety of ways: proactively addressing bugs, prioritizing feature enhancements, and driving down our cost-to-serve through automating away toil are just the start. Our platforms, software, and customers are trying to telegraph us what to do – all we need to do is build systems that know how to listen.
Let’s build systems that tell us which features we should dark-launch first!
Finally, we are encouraged by the engineering management of adjacent and dependent teams to provide feedback as well as contributing code-fixes directly. If you’ve identified a bug and can fix it, just go fix it. Feel empowered (and highly encouraged) to do so.
How do we learn?
First – if you don’t have time to improve how you work, you have a shutdown problem. You will not be able to sustain or grow in any meaningful way. I think we’ve all eschewed strategic improvements for tactical ones and sometimes regretted it. While there is a time and place for “just fix it”, a well-honed and consistent workflow pays off in multiples.
We need to anticipate growth. This means turning on new services becomes second nature. We need to have sufficient confidence in our processes that change becomes good, something we are all comfortable with, and that creates an environment and mindset to help us move fast(er).
The GitOps workflow embodied in [our internal] SRE contract is how we do that in a supportable way, at scale.
Last “New Goal”: empower individual teams to decide when to stop doing something. If you can back it up by providing supporting metrics, (e.g. deprecate feature-X because it is never used), that should be an easy discussion with PM, engineering and BU teams.
Be on the lookout for these opportunities. Not only to reduce overhead and deliver simpler, more supportable systems, but when we stop doing something, we can start doing something else. Something a little cooler.
Our first Pre-Mortem: what we can’t succeed without
Let’s game out our first pre-mortem. What are some things that, if we don’t do, we will fail?
Predictable release timing, non-waterfall
Extremely well-tested code
Upstream integration testing and release gates
Participation in product development
Observability – EVERY SERVICE gets a dashboard (or two) and alerts
Low overhead, mesh communications between teams
Approach software development with SLO’s top of mind
Commit to eliminate a bit of toil each week with good engineering
Surprises always happen, but we should have clear product requirements to make sure we’re going in the right direction
Invest in our infrastructure – building features is awesome, but we must make sure we also invest in our infrastructure, build a solid one and improve it each sprint
Clear understanding of how each of the many components in the system are supported. When it breaks, who is page, responsible for triage, fixing it, etc.
To close out – have you ever had a leak in your roof? Building shiny features is like installing solar panels to stop the leak. Solar panels themselves are a great idea…but they won’t fix your roof.
I tried hard to make it work. But I could never get the results I wanted. The crisp video, clean audio, the BOKEH. Herein lies my adventures (full send, some might say) over the past ~2 months in getting a proper home office A/V setup. It’s not perfect yet, but I’m fairly happy with the improvement. I’ll close this write-up with some of my next steps / gaps.
Starting with some basic tenets:
Whatever I end up with has to augment all of my ingrained behaviors; I’m not going to relearn how to use a computer (change operating system) for the sake of this.
I’m stuck with the room I’ve got. The walls have odd angles which makes lighting difficult.
I don’t know anything about this, so there will be lots of trial, error, and Googling.
I started by scouring YouTube for videos about the ultimate camera setup for WFH. It turns out the center of knowledge in this area is Twitch streamers. There are also a few companies posting setups who are more business focused, feels like they’re using these videos to promote their consulting businesses (which are around video editing, or promoting small business). Twitch streamers also know a lot about Open Broadcaster Software, but I’ll get to that in a bit.
After seeing this blog come across my Twitter, I impulse-buy a GoPro Hero 8 Black. Turns out GoPro is completely ignoring Linux and their webcam firmware is really, really beta even on Mac and Windows. Returned it.
After watching a lot more of these videos, I started seeing a trend towards a particular Mirrorless DSLR camera, the Canon EOS M50. I’m trying to stay with models that folks recommend for WFH/streaming, and that I can find SOMETHING saying they support Linux / have anecdotal evidence of it working. So I bought a Canon M50. I had to buy it from a place I hadn’t shopped at before (Best Buy), because a lot of people are trying to up their WFH A/V game, making components scarce.
So I’ve got the camera. I also need a micro HDMI –> USB 3 cable, so I got one of those. Elsewhere in the YouTube blackhole, I came across the term “dummy battery”. This is a hollow battery that lets you plug the camera directly into wall power to avoid having to use real batteries and their runtime limitations. Canon dummy batteries were sold out everywhere, including Canon themselves, although I did place an order with them directly (their web purchasing experience is stuck in early 2000s). It was on backorder, so I eventually canceled that order and bought a knockoff dummy battery for 20% of the price of the real one. Was I worried that something 20% cheaper would instantly fry my camera? Yep. But I am in full send mode now.
So I have the camera, the HDMI cable, the dummy battery. I probably need a tripod, right? I think cameras need tripods. OK, I got a tripod. Not sure what I’ll use it for, but I can always return it. Turns out the tripod was a key investment in making sure the camera is stable, level and oriented properly for the right angle.
Next I probably need a memory card, right? Cameras need memory cards? OK, I’ll get a memory card. But which one? I’m planning to put down 4k video @ 60 fps so “sorting by least expensive” is probably not the right move. Turns out there is a whole disaster of categorization of memory card performance. I ended up reading, and re-reading this page a few times. What a nightmare. The least consumer-friendly thing since 802.11 spec naming. Anyway I ended up buying a 128GB class 10 card and it seems to be fine.
I then have to connect the camera to my computer. The videos suggest HDMI, but Canon has recently announced USB support for live-streaming. This blog in particular was referenced in a few videos. Lets try that. OK, now I am into the land of having to configure and load out of tree kernel modules. How do you do this again? It’s been a while. OK, got it. Whew. How badly do I want the BOKEH? This is getting dicey and doesn’t feel right.
Well, it actually works. But it is FAR from ideal. The fact that I’ve got to run gphoto2 on boot, and install all this goop to wire things together … there has to be a better way?
I began using this setup for the first time in real video calls (we use Google Meet where I work). The quality was infinitely improved (making sure my Meet config used 720p and after literally days of experimentation with camera settings). People immediately noticed the difference too. However, sometimes the camera would shut off? I noticed a trend of it shutting off after about a half hour? But I have the dummy battery. What’s going on?
I will spare you the suspense. If I’d found this page, I’d never have bought the Canon M50. It shuts off after a half hour on purpose. There’s no way around it. Also the Canon M50 is not compatible with Magic Lantern firmware. What is Magic Lantern? Something I don’t want to deal with. Gone are my days of CyanogenMod on my Palm Pre (by the way, the Palm Pre was better than any phone I’ve owned before or since, don’t @ me).
So, my Canon shuts off. That’s just about the worst flaw it could have. Back to Best Buy. But not before the 14 day return policy expires, so I can’t do it online, I now have to physically visit a Best Buy to haggle, or eBay it. Despite COVID situation, I decided to mask it and haggle. Luckily they took the camera back, yay. If you’re interested in why it turns off, I found this post which is probably right…yeesh!
Next, what camera should I go for? Elgato’s page made it super easy to find cameras of a similar price-point that didn’t have any glaring flaws.
After much more fact-checking, I decided to get a Sony A6100. This time Amazon had it in stock, and significantly cheaper than other sites (price gouging?). The Sony arrives and it is immediately more useful (because it doesn’t shut off). Incidentally, I had to buy a different dummy battery for the Sony. An off-brand, that has thus far not fried my camera. Tripod, memory card and HDMI cable were compatible.
Next, how to connect this to my computer. The Sony also supports USB, but I’m not happy with the quality. It also uses a lot of CPU. What solutions are there? After many a sleepless night on YouTube learning what a “capture card” is, I went to find the highest end model, and found an Elgato 4K60 PRO PCI capture card. Wow, this was a total waste of time. Not only does it have zero Linux support but it feels like Elgato is actively ignoring Linux across the board, which is possibly corroborated by my experience with their CamLink 4K as well(?). I end up searching my email for that phrase in case some other intrepid Red Hatter is further along than me.
It turns out that an engineer whom I hired ~4 years ago, and who I allegedly sit next to in the office (it’s been 6+ months, and I’m full sending my home office, maybe I’ll never see him again), wrote a post about capture cards in May 2020! I have learned to trust this person’s opinion on these sorts of niche devices, because he has described his enviable Plex setup (it has a dedicated air conditioner), his GPU bitcoin mining, Threadripper prowess, and other things to me over lunch which indicate he has a nack for understanding things at the level of detail that I need for my BOKEH. His post pointed to a Magewell USB Capture HDMI 4K Plus Device. In particular he had tested it on Fedora and noted how hands-off it was for setup. His complaint was noise about a built-in fan. After balking at the price, I decide that it can be returned, so I get one. It turns out he was right, and it works great! One thing though is that I haven’t heard the fan at all, which I guess is good. Thanks to Sebastian for this tip.
However its really expensive. And the YouTubers are telling me about Elgato CamLink 4k which is 1/3rd the price. I get a CamLink 4K and decide that if it works, I’ll return the Magewell to save the money. Hook up the CamLink, the kernel sees it as a uvc device, but I see nothing but a black screen in OBS and Google Meet. After an hours worth of effort and Google turning up several complaints on this topic (admittedly not a huge effort on my part), I decide to trade money for my weekends/evenings back, and stick with the Magewell. Sebastian was right again. Hat-tip.
On to Audio. Last year I hired an engineer who turned out to be in two metal bands. On video calls with him, he had what looked like a very serious A/V setup in his home office. If you’re into Dual-Violin US Folk Metal, maybe check them out. Anyway, this person gave me some guidance on audio and I ended up going with a Blue Yeti USB Mic. This is one aspect of this journey that worked on the first try. However I could not help but think maybe there’s a way to improve? At conferences, when they were in-person, presenters get a lavalier mic. I bought one and it wasn’t any better. Also it was annoying to have a cable dangling from my collar all day. Returned.
For years I’ve been using Bose active noise cancelling headphones (at the office which is an open-office disaster / cacophony). At home I also bought Bose ones, but the in-ear model. The only thing I don’t like is that they’re wired, so I’m tied to the desk. One thing I do like is that they’re wired, so I never have to worry about batteries like I do with the ones I have at the office. I also have a pair of noise cancelling Galaxy Buds (which I love). I decide to try those. Ahh, my workstation doesn’t have Bluetooth. Back to Amazon to get a cheap bluetooth dongle. And now the Galaxy Buds work. But the experience sucks, for a few reasons:
I have to worry about batteries
They disconnect if I walk down the hall
Pairing is less than ideal in Linux where I have ~5 audio devices
I notice a lot of CPU usage tied back to the Bluetooth device…not good.
I have the Video and Audio basically squared away at this point. What’s next? Lighting.
The majority if howto videos indicate that if you don’t have proper lightning, it won’t matter your camera or software. I begin to research lighting, and found that Elgato Key Lights are a popular thing. You apparantly need two of them, they’re $200 each, and they’re out of stock. So, nope. I have a spare floor lamp and decide to use that. This is much better than the ceiling fan light I had which was casting a very scary shadow on my face 🙂 So the lamp is to my east-south-east, pointed towards the ceiling and I’m OK with the results. This is an area I may eventually want to improve, but maybe I’m nearing diminishing returns time-wise?
Conferences have gone virtual, and I have several presentations lined up, which are all typically recorded. So now I need to figure out how to record myself. According to YouTubers, I need to figure out what OBS stands for. The Open Broadcaster Software is (apparently) an open source streaming/recording application available for Windows, Mac and Linux. I am now watching EposVox masterclass on OBS. It’s complicated but not terrible to do basic stuff. You can see my first stab at integrating my efforts into a recorded presentation here.
After watching that video back, I have a few areas to improve:
I keep having to pause the recording to load the next section of content into my brain. I have to keep looking down to pause. There are apparently things called Elgato Stream Decks, which are a pad of hotkeys which are used by game streamers to automate certain operating system or OBS operations. OBS also supports hotkeys. Here is what my OBS canvas looks like including a few video sources and scenes:
I am not looking into the camera often enough. Yes, I have to look at my monitor to do demos and whatnot (expected), but I am also looking at my monitor to check my notes. I want to avoid that somehow. It turns out that Phone-based Teleprompters are a thing, and in-expensive. I bought one. It’s freaking cool. Mounts directly to the camera lens and displays the “script” overlaying the lens itself. So you are staring directly at the lens, and have a “Star Wars intro” style scrolling text of your content. Cannot recommend this product enough for professional delivery of recorded content. It even comes with a Bluetooth remote to control scrolling speed and start/stop. That dongle comes in handy again!
I want to involve whiteboards in my presentations. In the office, I have whiteboards available to me everywhere. I need one at home. But due to the size and shape of my office, I really don’t have a wall within camera-range to mount one. So I went with one on wheels. I haven’t used it in any presentations yet, but I’ve been using it for keeping notes and so far loving it.
I have to learn how to do basic video editing. After some Googling for the state of the art on Linux, I found Kdenlive which isn’t terrible to learn, after watching a few beginners videos.
I realize the audio is out of sync with the video. OBS let me insert a delay to help match then up. 300ms seems to be perfect.
In the original version of this video, the audio was super low. So I had to learn how to convert an mkv (OBS default) to an mp4, so I can work with the audio independently of the video, and boost the audio gain (by +20dB if you’re curious). Thanks to some quick tips from Langdon White I am able to achieve this. At this point my various experiments and YouTube deep dives are starting to pay off. I am smiling, finally 🙂
For some reason, when I turn off the camera, the zoom level resets to 16mm. But I want it at 18mm. So every time I turn the camera power on, I have to dial the zoom back in manually. Not a huge deal since it’s just once a day.
CPU usage in Chrome…brings the computer to a crawl. My workstation has 16 cores and 64G RAM. Sigh…so now all my Google Meets occur in Firefox. Not too bad, just annoying when it comes to screensharing since I really do not want to use Firefox as my primary browser.
Lens: according to photography snobs, if I don’t get a better lens, they’ll throw me out of the subreddit. This will probably have to wait until after my winning lotto ticket shows up.
After talking with some coworkers who are also upleveling their WFH A/V setups and thus learning what OBS stands for, I come to find out that OBS has some noise filtering options built in. I could have used that to filter out some background noise (e.g. from my kids or my workstation fans).
Conclusion / Final Hardware list
So, in the end, my hardware and software setup as of this posting is:
I have to say, this has been a really fun project. It’s an area I had zero knowledge of going in – just a personal goal to improve my WFH A/V. It’s also an area of somewhat daunting complexity, hardware opinions (nerd fights), and an endless compatibility matrix. That’s part of why I went the route of buying stuff and returning it .
I hope this post helps someone who is looking to improve their home office video quality avoid newb mistakes and just get it done. Also, I do realize that there are likely cheaper options across the board. But at least you have the laundry list of stuff that worked for me, within my given constraints, and can possibly phase your purchases like I did over a couple months.
I came really close to making some large purchases to improve my video call situation, and I may still do so, but I did spend some time and found a few quick wins maybe will help others:
I use Fedora as my workstation. Won’t change it.
I have a Logitech C930e. Logitech doesn’t publish anything on how to tweak the camera on Linux. Figure out how to tweak it.
I like (love) working in the dark. So I never have any lights on. That has to change.
I have a window behind me in my home office. Shutting the blinds is not enough. Repositioning my desk won’t work in the space I’ve got. Get a blackout curtain.
My webcam is sitting on top of my monitor, dead center. This makes it really awkward to look directly at. It’s about 8″ above my eye-line. I don’t think I’m going to change this. My eyeline has to remain at the center of the monitor or I get neck pain.
Here are the tweaks I made that do seem to have improved things:
dnf install v4l2ucp v4l-utils
v4l2-ctl --set-ctrl zoom_absolute=125 # this helps with the "know your frame" part https://www.youtube.com/watch?v=g2wH36xzs_M. This camera has a really wide FoV, so this shrinks it down a bit.
v4l2-ctl --set-ctrl tilt_absolute=-36000 # this helps tilt the center of the camera frame down towards where I'm sitting (8" below camera).
v4l2-ctl --set-ctrl sharpness=150 # This seemed to help the most. I tried a bunch of values and 150 is best for my office.
Lighting: Instead of having my desk lamp illuminate my keyboard, turn it 180degrees to bounce off the white ceiling. Big improvement.
Lighting: You can’t work in the dark anymore.
Auto-focus: I have a TV just to my right. When whatever’s on changes brightness, it makes the camera autofocus freak out. I typically keep the TV muted. Now I’ll pause it while on calls.
Microphone: I have an external USB mic (Blue Yeti). Turns out I had it in the wrong position relative to how I’m sitting. Thanks to a co-workers “webcam basics” slides for that tip (confirmed in Blue Yeti docs).
Despite that, after recording a video meeting of just myself I still didn’t like how the audio turned out. So I bought an inexpensive lavalier microphone from Amazon, figured it’s worth a try.
One thing I cannot figure out how to do is bokeh. I think that remains a gap between what I’ve got now and higher end gear.
Red Hat does managed services. We do DevOps in the style of the Google SRE Book. We’re a decent sized and growing org that supports customers 24/7. Our team is at the forefront of operationalizing OpenShift 4’s revolutionary set of capabilities, and are a way-to-market for many of Red Hat’s newest portfolio offerings.
What does a day-in-the-life of a Red Hat SRE look like… You remember these, don’t you?
If you drop into one of our Slack channels, you might find some interesting discussions on
Building and maintaining CI/CD pipelines.
Writing Operators in golang to handle managed services use-cases.
Handling upgrades of managed OpenShift clusters.
Chatting with cloud provider technical teams on weird quirks of their API.
Debugging bizarre, nearly intractable production issues using the scientific method.
Ensuring our managed OpenShift platforms are the most secure offerings possible.
Teams building OperatorHub.io tooling, certification programs and Operator SDK training (delivered dozens of Operator trainings for our customers)
Basically, if you’ve had any interaction with OpenShift 4, you’ve likely consumed those services.
I’d been in my previous role for 7 years, and celebrated my 10th anniversary at Red Hat by being acquired by Big Blue. My previous team (Red Hat Performance and Scale) afforded me endless technical challenges, opportunities to travel, present, help shape product and build engineering teams from the ground up. Perhaps most importantly, I had the opportunity to mentor as many Red Hatters as I possibly could.
Red Hat Service Delivery allows me to broaden my technical and architecture skill set to areas outside of performance, scale and optimization, while letting me apply the many hard-fought lessons from prior chapters in my career.
Hopefully $subject makes a bit more sense now. Onward!
I’m CC’d on mails when things get slow, but never when things work as expected or are fast…oh well. Like an umpire in baseball, if we are doing our jobs, we are invisible.
Subject: Docker operations slowing down
I reach for my trusty haiku for this type of thing:
Ah but in this scenario, it is something more…sinister (my word). What could be more sinister than DNS, you say? It’s the magical QoS system by which a cloud provider creatively rents you resources. The system that allows for any hand-wavy repackaging of compute or network or disk into a brand new tier of service…
Platinum. No, super Platinum. What’s higher than platinum? Who cares, we are printing money and customers love us because we have broken through their antiquated finance process. We will gladly overpay via OpEx just to avoid that circus.
But I digress…
In this scenario, I was working with one of our container teams folks who had a report of CI jobs failing and someone had debugged a bit and pinned the blame on docker. I watch the reproducer run. It is running
docker run --rm fedora date
in a tight loop. I watch as docker daemon gets through its 5000th loop iteration, and…still good to go. On average, ~3 seconds to start a container and delete it. Not too bad, certainly not something that a CI job shouldn’t be able to handle. I continue to stare at tmux and then it happens…WHAM! 82 seconds to start the last container. Ahh, good. Getting a reproducer is almost always the hardest part of the process. Once we have a tight debug loop, smart people can figure things out relatively quickly.
I am looking at top in another window, and I see systemd-udev at the top of the list…what the…
As much as I would love to blame DNS for this, I have a hunch this is storage related now, because the reproducer shouldn’t be doing anything on the network. Now I am running ps in a loop and grepping for ” D “. Why? Because that is the process state when a thread is waiting on I/O. I know this because of several terribly painful debugging efforts with multipath in 2010. Looking back, it may have been those situations that have made me run screaming from filesystem and disk performance issues ever since 🙂
From man ps:
PROCESS STATE CODES
Here are the different values that the s, stat and state output specifiers (header "STAT" or "S") will display to describe the state of a process:
D uninterruptible sleep (usually IO)
R running or runnable (on run queue)
S interruptible sleep (waiting for an event to complete)
T stopped by job control signal
t stopped by debugger during the tracing
W paging (not valid since the 2.6.xx kernel)
X dead (should never be seen)
Z defunct ("zombie") process, terminated but not reaped by its parent
Normally, processes oscillate between R and S, often imperceptibly (well, at least not something you see very often in top). You can easily trace this with the systemtap script sleepingBeauties.stp if you really need to. This script will print a backtrace of any thread that enters D state for a configurable amount of time.
That is interesting to me. udevd is in the kernel’s path for allocate/de-allocate storage devices. I am now convinced it is storage. kworker is a workqueue kernel thread that fires when the kernel’s writeback watermarks (dirty pages) are hit. For my extreme low latency work, I documented how to shove these in a corner in my Low Latency Tuning Guide for Red Hat Enterprise Linux 7.
I know that if this does not complete in < 5 seconds, something is terribly hosed. Aaaaaand it hangs. This process now shows up in my ps loop looking for D state processes. So I have it narrowed down. Something is wrong with the storage on this VM, and it only shows up after 5000 containers are started (well, I am told it varies by a few thousand here and there).
This may seem like a tangent but I promise it is going somewhere:
Nearly two years ago, when we were first standing up openshift.com version 3 on AWS, we ran into a few eerily similar issues. I remember that our etcd cluster would suddenly start freaking out (that is a technical term). Leader elections, nodes going totally offline…And I remember working with our AWS contacts to figure it out. At the time it was a little less well-known, and today just by googling it appears fairly well understood. The issue with this reproducer turns out to be something called a BurstBalance. BurstBalance is AWS business logic interfering with all that is good and holy. If you purchase storage, you should be able to read and write from it, no?
As with all public cloud, you can do whatever you want…for a price. BurstBalance is the creation of folks who want you to get hooked on great performance (gp2 can run at 3000+ IOPS), but then when you start doing something more than dev/test and run into these weird issues, you’re already hooked and you have no choice but to pay more for a service that is actually usable. This model is seen throughout public cloud. For example, take the preemptible instances on GCE or the t2 instance family on AWS.
I have setup my little collectd->graphite->grafana dashboard that I use for this sort of thing. You can see things are humming along quite nicely for a while, and then…yeah.
Once the reproducer exhausts the gp2 volume’s BurstBalance, things go very, very badly. Why? Simple. Applications were not written to assume that storage would ever slow down like this. Issues in docker cascade back up the stack until finally a user complains that it took 5 minutes to start their pod.
The reason is that we have not paid our bounty to the cloud gods.
Here is BurstBalance and the magical AWS QoS/business logic in action.
You can see it looks a lot like my grafana graphs…quota is exhausted, and the IOPS drop to a trickle.
What would happen then if we did kneel at the alter of Bezos and pay him his tithe? I will show you.
The reproducer is chugging along, until it slams into that magical AWS business logic. Some QoS system somewhere jumps for joy at the thought of earning even more money. This time, we will pay him his fee…for science.
You can see that our reproducer recovers (lower is better) once we flip the volume type to provisioned IOPS (io1)…this was done on the fly. We set the io1 volume to 1000 IOPS (mostly random choice…) which is why it is slightly higher after the recovery than it was before the issue occurred. gp2 can crank along really, really fast. That is, until…
The take aways from this debugging session are:
Regardless of cloud provider, you pay a premium for both performance and determinism.
If you think you are saving money up front, just wait until the production issues start rolling in which, conveniently, can easily be solved by simply clicking a little button and upgrading to the next tier. Actually, it is brilliant and I would do the same if I had the unicorn QoS system at my disposal, and was tasked with converting that QoS system into revenue.
I now must proactively monitor BurstBalance and flip volumes to io1 instead of let them hit the wall in production. Monitoring for this (per AWS documentation, use CloudWatch) is an additional fee appears to be included in their CloudWatch free tier.
Perhaps we flip all volumes to io1 proactively and then flip them back when the critical period is over.
One thing I ran out of time to verify is what happens to my BurstBalance if I flip to io1, then back to gp2? Is my BurstBalance reset? Probably not, but I haven’t done the leg work yet to verify.
We will do less I/O when using overlay2 (might just delay the inevitable).
All super critical things (like etcd) get io1 out of the gate. No funny business.