Right out of school, Christopher Nguyen and Lee Liu co-founded their first startup, a job board that was acquired after 18 months. Their next company was a Groupon-style site that powered eBay’s daily deals, using a marketing tool that created targeted, personalized emails. But three years in, they decided to throw in the towel on that business.
The co-founders were part of Y Combinator, and the advice they got was that “to be successful as a startup, you need to be growing at something like 10% per week,” says Liu. “And we were barely growing 5% per month.”
After batting around ideas like building a Slackbot, they decided to start their third company in logging, which seemed like a stronger market. “It was one of those things where we have maybe 4 to 5 months to prove out this tech and after that, we’ll be out of money and have to fold and shut down,” says Liu.
A server logging product, LogDNA is “like the black box on airplanes,” says Liu, the company’s CTO. “When an airplane crashes, all these investigators are trying to figure out what’s on the black box so they can recreate what the plane was doing in the air. We basically do that, but for apps. If your iPhone app crashes or your web app crashes or you have errors, we can trace out or look into what’s happening just prior to that. We capture audit events and logs from either your infrastructure or your apps or various services you use online.”
Liu had used different logging tools at his second startup. “They all had 80% of what I needed; they were never quite ideal,” says Liu. “I was missing this one thing, which was JSON parsing.” So he ended up building his own simple tool to solve this problem, without the overhead of a full set of features. “It worked for me, and I got the results I needed really, really fast,” he adds. “We built that out of our need because the tools online were either way too expensive or just didn’t do the things we needed them to do.”
In bringing the LogDNA product to market, Liu polled other startups to find out what their specific needs were. “We asked a lot of our early customers, who were YC founder friends, what kind of logs they use,” says Liu. “They all had different problems, so we definitely took all their concerns and created LogDNA. We wrote features that customers cared about.”
A Nail-Biter Startup Story
LogDNA was launched in February 2016, when the company had just one month of runway left. “When we launched, we didn’t even have credit card payments working, because we still had to build it,” Liu says. “But I figured, ‘Hey, we have a two-week-free trial, so I’ve got two weeks to build this. We should be all right.’”
Luckily, he was right. LogDNA launched on Product Hunt and got 800 signups in one day. The next day, it was featured on HackerNews, and that led to 2,000 signups. They were on to something. People had used tools like this before, but they weren’t as polished as LogDNA. “One person told us, ‘You guys are like the iPhone of logging. Everything else feels like a BlackBerry,’” says Liu.
Still, the company’s funds were dwindling. A couple of early investors saved the day with well-timed checks. “I’m literally watching our bank account going down close to zero,” says Liu, “and then the first check hits, and I’m like, ‘We’ll be okay, at least for another month.’ And then a second, and a third, and a fourth. We’re like, ‘Okay, we’ll be fine.’”
The company ended up raising $1.3 million in the spring of 2016, reached over 2,000 customers in 2017, and raised $25 million in 2018.
Growth and Scaling
With its customer-driven development, LogDNA quickly gained traction. “And logs are such that over time, they always grow,” says Liu. “Every one of our customers is some kind of startup or some sort of growth company. If we can get someone early on, we’ll grow with them throughout their journey. That part’s actually really awesome.”
The company has also started targeting enterprises, which tend to have more custom requirements. But there’s still just one product, and every customer is charged per gigabyte of log volume. “We have a really nice-looking log viewer, and we have graphing and metrics that show trends over time,” says Liu. “And then we have alerts that this app is crashing three per hour, but now it’s crashing 300 per hour, so something’s wrong. I feel that we’ve finally gotten to a point where we’re on par with our competitors, modernizing the logging landscape.”
The special sauce is the user interface. “We call them quality-of-life improvements, tiny little things that make it more delightful to use,” says Liu. “And I think a lot of our customers feel that these guys actually give it some real thought.”
Being able to manage everything through the command line makes Packet bare metal feel like cloud software. It's really cool that they can spin up a physical machine in eight minutes. It's like a breath of fresh air.
Plus, the search speed and live tail are incredibly fast, which helps accelerate debugging for customers. “Our biggest use case is debugging apps, so a lot of people keep LogDNA open throughout the day,” says Liu.
Partnering with Packet
As part of the Y Combinator incubator, LogDNA took advantage of the credits given by various cloud providers. After moving from one to another to the next, and using up all the credits, the company realized it needed to find a cost-effective long-term solution that would provide dedicated service. “With traditional public cloud providers, you need a lot of servers to get a baseline performance because the servers are not dedicated,” says Liu.
When Liu found Packet at the end of 2017, he quickly saw that it checked a number of boxes: dedicated servers, the ability to customize, and all at really good prices. Plus, “Packet’s API allows us to manage servers not just their web interface, but through automated scripts and things like that,” he says. “The control plane is really awesome. Being able to manage everything through the command line makes Packet bare metal feel like cloud software. It’s really cool that they can spin up a physical machine in eight minutes. It’s like a breath of fresh air.”
And “Packet’s metal plus local disk is really, really fast, and therefore we need less of them to power our stack,” says Liu. “Everything leads to the fact that this is super cost-effective for us.”
LogDNA began migrating to Packet in the second half of 2018.
The LogDNA team has been so pleased with the partnership that “we actually modeled some of our own support after what Packet did,” says Liu. “For example, we have a shared Slack channel with them. I can invite anybody to the channel, and we can chat and relay issues and problems, and they have their entire support team on there. It was just a really great experience, and the turnaround time is really, really fast. We started doing that with some of our own customers.”
Currently, LogDNA is in Packet’s New Jersey location, because it has the SOC and PCI compliance that the company’s customers require. Europe will likely be added next. And they’re watching for opportunities where they might be able to expand to new locations with Packet.
Another focus in the near term is to develop intelligent sampling. “We recognize that some of our larger customers, the ones that are doing tons of terabytes of logs per day, pay us a lot,” says Liu. “And to be fair, no one can really read or scan through ten terabytes of logs anyway. But you still want to get good signals out of that log data without having to necessarily store all of it.”
The team has already developed some prototypes, which on the low end can save customers two-thirds off their bill. “And on the high end, it’s almost nine-tenths,” says Liu. “These are substantial numbers of how much savings we can achieve. And if you’re willing to dial up what we call the fuzziness threshold, it can get really, really high.”
All of these efforts are driven by LogDNA’s mission, which is: How do we help companies achieve zero downtime? “Right now, I would say we are very reactive to issues like something went wrong, the operator gets an alert, someone looks at the alert, debugs the problem, and then fixes it,” says Liu. “To get to zero downtime, you need to be very proactive. You need to find problems before they occur. And that’s ultimately where we want to get to."