Launch HN: Opstrace (YC S19) – open-source Datadog

Launch HN: Opstrace (YC S19) – open-source Datadog | Dark Hacker News

316 points by spahl 5 years ago | 114 comments

Hi HN!

Seb here, with my co-founder Mat. We are building an open-source observability platform aimed at the end user. We assemble what we consider the best open source APIs and interfaces such as Prometheus and Grafana, but make them as easy to use and featureful as Datadog, with for example TLS and authentication by default. It's scalable (horizontally and vertically) and upgradable without a team of experts. Check it out here: http://opstrace.com/ & https://github.com/opstrace/opstrace

About us: I co-founded dotCloud which became Docker, and was also an early employee at Cloudflare where I built their monitoring system back when there was no Prometheus (I had to use OpenTSDB :-). I have since been told it's all been replaced with modern stuff—thankfully! Mat and I met at Mesosphere where, after building DC/OS, we led the teams that would eventually transition the company to Kubernetes.

In 2019, I was at RedHat and Mat was still at Mesosphere. A few months after IBM announced purchasing RedHat, Mat and I started brainstorming problems that we could solve in the infrastructure space. We started interviewing a lot of companies, always asking them the same questions: "How do you build and test your code? How do you deploy? What technologies do you use? How do you monitor your system? Logs? Outages?" A clear set of common problems emerged.

Companies that used external vendors—such as CloudWatch, Datadog, SignalFX—grew to a certain size where cost became unpredictable and wildly excessive. As a result (one of many downsides we would come to uncover) they monitored less (i.e. just error logs, no real metrics/logs in staging/dev and turning metrics off in prod to reduce cost).

Companies going the opposite route—choosing to build in-house with open source software—had different problems. Building their stack took time away from their product development, and resulted in poorly maintained, complicated messes. Those companies are usually tempted to go to SaaS but at their scale, the cost is often prohibitive.

It seemed crazy to us that we are still stuck in this world where we have to choose between these two paths. As infrastructure engineers, we take pride in building good software for other engineers. So we started Opstrace to fix it.

Opstrace started with a few core principles: (1) The customer should always own their data; Opstrace runs entirely in your cloud account and your data never leaves your network. (2) We don’t want to be a storage vendor—that is, we won’t bill customers by data volume because this creates the wrong incentives for us. (AWS and GCP are already pretty good at storage.) (3) Transparency and predictability of costs—you pay your cloud provider for the storage/network/compute for running Opstrace and can take advantage of any credits/discounts you negotiate with them. We are incentivized to help you understand exactly where you are spending money because you pay us for the value you get from our product with per-user pricing. (For more about costs, see our recent blog post here: https://opstrace.com/blog/pulling-cost-curtain-back). (4) It should be REAL Open Source with the Apache License, Version 2.0.

To get started, you install Opstrace into your AWS or GCP account with one command: `opstrace create`. This installs Opstrace in your account, creates a domain name and sets up authentication for you for free. Once logged in you can create tenants that each contain APIs for Prometheus, Fluentd/Loki and more. Each tenant has a Grafana instance you can use. A tenant can be used to logically separate domains, for example, things like prod, test, staging or teams. Whatever you prefer.

At the heart of Opstrace runs a Cortex (https://github.com/cortexproject/cortex) cluster to provide the above-mentioned scalable Prometheus API, and a Loki (https://github.com/grafana/loki) cluster for the logs. We front those with authenticated endpoints (all public in our repo). All the data ends up stored only in S3 thanks to the amazing work of the developers on those projects.

An "open source Datadog" requires more than just metrics and logs. We are actively working on a new UI for managing, querying and visualizing your data and many more features, like automatic ingestion of logs/metrics from cloud services (CloudWatch/Stackdriver), Datadog compatible API endpoints to ease migrations and side by side comparisons and synthetics (e.g. Pingdom). You can follow along on our public roadmap: https://opstrace.com/docs/references/roadmap.

We will always be open source, and we make money by charging a per-user subscription for our commercial version which will contain fine-grained authz, bring-your-own OIDC and custom domains.

Check out our repo (https://github.com/opstrace/opstrace) and give it a spin (https://opstrace.com/docs/quickstart).

We’d love to hear what your perspective is. What are your experiences related to the problems discussed here? Are you all happy with the tools you’re using today?

brodouevencode 5 years ago |

We use [insert very large application performance monitoring tool here] for workloads running in [insert very, very large cloud provider here] and after examining our deployments, concluded that we were spending nearly $13k/mo for data transfer out expenditures because the monitoring agents have crazy aggressive defaults. Seems like running our own (which may be worthwhile) would alleviate anything like that.

nrmitchi 5 years ago | |

Tip, if you happen to be using datadog, make sure datadog agent logs are disabled from being ingested into datadog.

If you can disable them at the agent level and avoid the data out that would be even better.

At a previous employer the defaults were quite literally half of our log volume, that we were paying for. I was doing a sanity check before renewing our datadog contract and was very not-pleased to discover that.

fat-apple 5 years ago | | |

We’re about to release a Datadog compatible API so you can point your Datadog agent at Opstrace instead (stay tuned for the blog post). Our goal is to be able to tell you exactly how much data the agent is sending and how much that is costing you (and for example what services/containers are responsible for the bulk of the cost). Here’s a list of the PRs: https://github.com/opstrace/opstrace/pulls?q=is%3Apr+is%3Acl...

mdaniel 5 years ago | | |

I even opened a support ticket for their stupid python agent logging its connection refused tracebacks on every metrics poll and was told "too bad"

They really don't give one whit about log discipline or allowing the user to influence the agent's log levels

brodouevencode 5 years ago | | |

Lol no, the other really large one. Five minute Cloudwatch polling defaults are just overkill even in production.

cyberpunk 5 years ago | |

Can hurt yourself that way too -- happened to us, but with not a lot of data, and all down to Thanos aggregating/reducing/whatever-ing meeeeeeeeeelions of metrics inside a s3 bucket to the tune of about 7k a month :/

spahl 5 years ago | |

Yes that is frustrating indeed. On top of paying your external vendor, you are punished by the egress cost you have to pay to your infrastructure cloud provider. This is one of the problems we wanted to solve. Feel free to contact me seb@opstrace.com.

alexchamberlain 5 years ago | |

It feels like the large monitoring applications should run aggregators in large cloud providers to reduce traffic for everyone.

jgehrcke 5 years ago | | |

Haha, sure. I suppose that for example AWS has little incentive for allowing for example Datadog to to offer a special per-AZ endpoint. But hey. Here we come into play :).

tailspin2019 5 years ago |

Nicely designed site, great logo, but after clicking around a bit (and looking at GitHub) I’m confused by what this product actually is.

DataDog has a UI. Does Opstrace? Or is it just a CLI/API based tool?

If you actually have a UI element to your product you’re doing a huge disservice to yourself by not actually showing this anywhere...

EDIT: I don’t mean to sound negative, I’m wondering if positioning this against Datadog is going to create immediate, potentially incorrect, expectations in people’s minds as to what this product might provide.

From first impressions I’d say this is much closer to Prometheus (which does have a UI but it’s so basic it may as well not - but then the UI is not the point of Prometheus).

sciurus 5 years ago |

It looks like you're largely selling a fancy installer for software primarily developed by another company, Grafana Labs. They offer both open source, hosted SaaS, and paid-for "enterprise" versions of their software.

Why should someone choose Opstrace over purchasing from them directly?

englambert 5 years ago | |

Our installer is indeed an important part of what we’re offering and we’re continuously evolving our operator to manage the ongoing maintenance. But in terms of being a feature complete, Open Source Datadog, you’re right that we have a long way to go to achieve our vision. As mentioned in other replies, we are working on other interesting components as well, such as a new collaborative UI (https://news.ycombinator.com/item?id=25996154), API integrations (https://news.ycombinator.com/item?id=25994268), and more.

That being said, in case you couldn’t tell, we love software from Grafana Labs. It’s popular for a reason. However, we want it to be as easy to install and maintain as clicking a button, i.e., as simple as Datadog. So one problem we are trying to solve today is that while, yes, you can stitch together all of their OSS projects yourself (and many, many people do), it’s a non-trivial exercise to set up and then maintain. We’ve done it ourselves, seen friends go through it; we’d like to stop everyone from becoming a subject matter expert and reinventing the wheel. (Especially since when our friends do it themselves they always skimp on important things like, say, security.) Bottom line—we’re inspired by Grafana Labs. We strive to also be good OSS stewards and contribute to the overall ecosystem like they have.

Another way to solve the “stitching-it-together” problem, as you mentioned, is of course pay Grafana Labs for their SaaS (which I’ve done in the past) or one of their on-prem Enterprise versions. However, these are not open source. The former is hosted in their cloud account and single-tenant; the latter have no free versions. We think Opstrace provides a lot of value, but we understand that it’s not for everyone.

dudeinjapan 5 years ago |

Hi there, at TableCheck (www.tablecheck.com) we recently adopted Lightstep.

In a nutshell, running all these various components (Grafana, etc) is a royal pain in the neck. Even if `opstrace create` spawns them easily, the problem is running/maintaining them. We want someone to run these for us as a SaaS/PaaS and we're happy to pay them.

Re: your principles:

(1) The customer should always own their data --> we agree. However, we are happy for you to be a custodian of that data.

(2) We don’t want to be a storage vendor --> neither do we. We want storage to be someone else's problem. We're happy for you to use a cloud platform like AWS/GCP and charge us a 50% markup.

(3/4) Transparency, predictability of costs, open source --> all excellent.

jarym 5 years ago |

Very exciting! Question: your homepage says it’ll always be Apache 2 but what will you do if someone like AWS rebrands your work (looking over at Elastic here)?

fat-apple 5 years ago | |

Mat here (Seb's Cofounder). Great question. We are not only building a piece of infrastructure but a complete product with its own UI and features, rather than a standalone API. Our customer is the end-user more than the person wanting to build on top of it. GitLab and others have shown that when you do that the probability of being forked or just resold goes down drastically.

ensignavenger 5 years ago | | |

Gitlab is open-core, so that gives them a lot of closed source features to sell. Do you plan to be open-core like Gitlab?

hangonhn 5 years ago |

Damn. That's one hell of a set of credentials for the founders.

I was the engineer who was heavily involved with monitoring at my last job and a lot of what this is doing aligns with what I would have done myself. At my new job, I work on different stuff but I can see we're going to run into monitoring issues soon too. I'm so, so, so glad this is an option because I do not want to rebuild that stuff all over again. Getting monitoring scalable and robust is HARD!

englambert 5 years ago | |

Hey, thank you. :-) That’s kind of how we feel -- it seems like everyone is building tooling around Prometheus, and frankly, we hope that collective effort can hopefully be redirected to more impactful value creation for our industry. On a personal note, most of us on the team have been there in one way or another, struggling to actually monitor our own work. We’ve had surprise Datadog bills and felt the pain of scaling Prometheus. (In fact, I’m planning a blog post about this struggle, so stay tuned.) It feels like this problem should already be solved, but it’s not. So we’re trying to fix it.

crazy5sheep 5 years ago | | |

Prometheus is great, the main problem is the bloat of metrics it's collecting. one really needs to carefully define the rules to scrape, compute, reduce and filter the ones that are not needed and the ones that need to precompute.

boundlessdreamz 5 years ago |

1. It would be great if you can integrate with https://vector.dev/. Also saves you the effort of integrating with many sources

2. When opstrace is setup in AWS/GCP, what is the typical fixed cost?

fat-apple 5 years ago | |

Great questions!

(1) As it stands today, you can already use https://vector.dev/docs/reference/sinks/prometheus_remote_wr... to write metrics directly to our Prometheus API. You can also use https://vector.dev/docs/reference/sinks/loki/ to send your logs to our Loki API. Vector is very cool in our opinion and we’d love to see if there is more we can do with it. What are your thoughts?

(2) As for cost, our super early experiments (https://opstrace.com/blog/pulling-cost-curtain-back) indicate that ingesting 1M active series with 18-month retention is less than $30 per day. It is a very important topic and we've already spent quite a bit of time on exploring this. Our goal is to be super transparent (something you don’t get with SaaS vendors like Datadog) by adding a system cost tab in the UI. Clearly, the cost depends on the specific configuration and use case, i.e. on parameters such as load profile, redundancy, and retention. A credible general answer would come in the shape of some kind of formula, involving some of these parameters -- and empirically derived from real-world observations (testing, testing, testing!). For now, it's fair to say that we're in the observation phase -- from here, we'll certainly do many optimizations specifically towards reducing cost, and we'll also focus on providing good recommendations (because as we all know cost is just one dimension in a trade-off space). We're definitely excited about the idea of providing users useful, direct insight into the cost (say, daily cost) of their specific, current Opstrace setup (observation is key!). We've talked a lot about "total cost of ownership" (TCO) in the team.

stevemcghee 5 years ago |

FWIW, I was able to play with a preview and found it straightforward to set up and it kinda just did what I expected. I'm happy to see them taking next steps here. Good luck opstrace!

tamasnet 5 years ago |

This looks very promising, thank you and congrats!

Also, please don't forget about people (like me) who don't run on $MAJOR_CLOUD_PROVIDER. I'd be curious to try this e.g. on self-operated Docker w/ Minio.

nickbp 5 years ago | |

Hi, this is Nick Parker from the Opstrace team. I personally have my own on-prem arm64/amd64 K3s cluster, including a basic 4-node Minio deployment, so I’m very interested in getting local deployment up and running myself. We’re a small team and we’ve been focusing on getting a couple well-defined use-cases in order before adding support for running Opstrace in custom and on-prem environments. It turns into a bit of a combinatorial explosion in terms of supporting all the possibilities. But we definitely want to support deploying to custom infrastructure eventually.

e12e 5 years ago | | |

This looks like an interesting product. We're figuring out our monitoring stack - but have also found Loki/Grafana - and we're looking at Victoria Metrics rather than Cortex. Our hope is that the combination will turn out to be able to scale down as well as up, and be possible to fit in with Docker swarm/compose on-prem and at digital ocean. Also looks like vector might be a good option for collecting data.

Will keep an eye out, to see if optrace might be a fit for us.

https://victoriametrics.github.io/

tamasnet 5 years ago | | |

Good to know, and thanks for confirming. I totally get that supporting all combinations can be a real challenge. I guess containerization helps, but that's becoming it's own smorgasbord of almost-compatible bits.

filereaper 5 years ago | | |

Happy to hear you're now with the old Mesosphere gang at Opstrace Nick!

sneak 5 years ago |

> We will always be open source, and we make money by charging a per-user subscription for our commercial version which will contain fine-grained authz, bring-your-own OIDC and custom domains.

Seems to me that these are at odds. If you're open source, why does anyone have to pay for these things?

If you're open core, I think it's mighty misleading to say things like "We will always be open source" because then not only is it untrue on its face, but also if someone contributes useful features to the open source project that compete with or supplant your paid proprietary bits, you are incentivized to refuse to merge their work - extremely not in the spirit of open source.

My perspective, which you asked for, is that open core is dishonest, and that you should be honest with yourselves about being a proprietary software vendor if that's indeed your plan, and stop with the open source posturing.

If I've misunderstood you, then I apologize.

zaczekadam 5 years ago |

Hey, I think this might be the coolest product intro I've read.

My two points - right now docs are clearly targeting users familiar with the competition but for someone like me who does not know similar products, a 'how it works' section with examples would be awesome.

Fingers crossed!

jgehrcke 5 years ago | |

Jan-Philip here, from the Opstrace team. Thanks for these kind words! For sure, you’re right, we can do a much better job at describing how things work. Providing great documentation is one of our top priorities :-)!

arianvanp 5 years ago |

Your mascot is almost exactly identical to https://scylladb.com/ 's mascot. Is there any connection; or a happy accident?

englambert 5 years ago | |

Chris here, from the Opstrace team. As it turns out, it’s just a happy coincidence. When we discovered theirs we fell in love with it as well. They have many different versions of their monster (https://www.scylladb.com/media-kit/)... similarly you’ll see several new versions of our mascot, Tracy the Octopus, over time!

mrwnmonm 5 years ago |

Man, I was hoping someone would do this. Thanks very much. Please please please, care about the design. I don't know why open source projects always have bad design.

Wish you all the best. and Congratulations!

fat-apple 5 years ago | |

Thank you! Yes, design is very near and dear to our hearts! If you’re interested in giving me some early feedback on our UX, email me mat@opstrace.com.

mrwnmonm 5 years ago | | |

Would love to. But will wait for the demo.

snissn 5 years ago |

hi! Some quick perspective - my thoughts looking into this are "ok cool what metrics do i get for free? cpu load? disk usage? the hard to find memory usage?" and i just get lost in your home page without any examples of what the dashboard looks like

nickbp 5 years ago | |

Just to answer the question about what metrics are included, you can write and read any kind of custom metrics and log data from your applications, and have to build useful dashboards yourself. When first deployed, the user tenants (you can create any number of tenants to partition your data) are empty (you start with a clean slate) and ready for you to send any metrics/logs to it. You also have to add your own dashboards to interpret the data you've sent.

Opstrace does ship with a "system" tenant designed for monitoring the Opstrace system itself. This tenant has built-in dashboards that we've designed to show you the health of the Opstrace system.

Incidentally, having sharable "dashboards" across people/teams/organizations is something we are also working on, so people don't have to re-invent dashboards all the time.

We also have some guidelines for you to ingest metrics from Kubernetes clusters (https://opstrace.com/docs/guides/user/instrumenting-a-k8s-cl...) and are building native cloud metrics collection. Feel free to follow along in GitHub: https://github.com/opstrace/opstrace/issues/310.

spahl 5 years ago | |

We totally agree our website is way too wordy and we are working on explaining our vision through various ways. Screenshots of course, but also things like short videos. We actually just did one of our quickstart https://youtu.be/XkVxYaHsDyY. It’s not perfect but we will get there:-)

Thanks for the feedback, we appreciate it!

spahl 5 years ago | | |

Sorry I don't know what happened here is a new youtube link: https://www.youtube.com/watch?v=ooqBn1Q-y2Q

It's from this page: https://opstrace.com/docs/quickstart

tmzt 5 years ago |

You mentioned Loki in your post. I evaluated it for our company and was reasonably impressed with the simplicity of setup and efficient storage. Where it failed us was the difficulty searching by customer identifiers or other "high cardinality" labels, or full-text. There's a longstanding issue [1] on the Github for this. Are you doing anything to improve log search versus an Elasticsearch cluster, for instance?

More broadly, how are you contributing to the upstream projects?

[1] https://github.com/grafana/loki/issues?page=7&q=is%3Aissue+i...

jgehrcke 5 years ago | |

JP from Opstrace here. Great questions!

> Are you doing anything to improve log search versus an Elasticsearch cluster, for instance?

No. You're right, Loki is not designed for building up an index for full-text search. The premise here is that you won't typically need that and that in exchange for not having to build up that index, you get other advantages (such as being able to rely on an object store for both payload and index data!). If, on the other hand, in special situations, you need to "grep-search" your logs, this is absolutely doable with Loki! Loki does not neglect this use case. The opposite is true; everyone is already excited about the performance characteristics that Loki already has today when it comes to ad-hoc processing full text. For example, see https://twitter.com/Kuqd/status/1336722211604996098 and definitely have a look at https://grafana.com/blog/2020/12/08/how-to-create-fast-queri.... I'm sure Cyril is happy to answer your questions, too!

> how are you contributing to the upstream projects?

We're reporting issues and try to contribute as much as we can! Of course, this effort has only started. So far, we've contributed to Loki's Fluentd plugin (https://github.com/grafana/loki/pulls?q=is%3Apr+author%3Ajge...), and our testing efforts have helped reveal edge cases; see for example https://github.com/grafana/loki/issues/2124 and https://github.com/grafana/loki/issues/3085.

We're excited to substantially contribute to both Loki and Cortex in the future!

rubiquity 5 years ago |

Nice. I've talked myself out of starting a monitoring product at least a few dozen times. As you point out, customers either get to choose between being gouged or run their own spaghetti.

On top of bad UX, I do think the storage layer is where customers are really getting hit by these companies. The big players are using very unoptimized ingestion and querying layers and pretending like tiered storage never happened. Developers share some of the blame too by not being at all pragmatic about how long and how much to keep. It's a tough nut to crack.

What's the plan for commercial? They run it themselves and pay per user? If so, that's refreshing.

nickbp 5 years ago | |

That's the plan! The incumbent SaaS providers are effectively charging a premium over the underlying storage. Their business is really reselling storage. Removing that premium via a self-hosted system then greatly reduces the need to structure your applications to fit the cost of monitoring it. This also means that any negotiated discounts and features they may use (e.g., S3 Standard-IA) are also applicable to data in Opstrace.

We will also have a blog post about bad UX in a couple weeks… stay tuned. What are some of your biggest gripes about UX?

GeneralTspoon 5 years ago |

This looks super cool!

We just moved away from Datadog because their log storage pricing is too high for us. We moved to BigQuery instead. But the interface kind of sucks.

Would love to get this up and running. A couple of questions:

1. Is it possible to setup outside of AWS/GCP? I would like to set this up on a dedicated server.

2. If not - then do you have a pricing comparison page where you give some example figures? e.g. to ingest 1 billion log lines from Apache per month it will cost you roughly $X in AWS hosting fees and $Y per seat to use Opstrace

fat-apple 5 years ago | |

Currently you can only deploy to AWS and GCP, but we do intend to extend support to on-prem/dedicated servers in due course (see https://news.ycombinator.com/item?id=25992237). Until now we’ve been focussing completely on building a scalable, reliable product by standing on the shoulders of these cloud providers where we can take advantage of services like S3, RDS, and elastic compute.

We've done a deep dive into the cost model for metrics and posted more about it here: https://opstrace.com/blog/pulling-cost-curtain-back. We are still working on a full cost analysis for logs - I'd be happy to send it to you once we have it (feel free to email me mat@opstrace.com to chat about your use case). Our goal is to be super transparent (see https://news.ycombinator.com/item?id=25992081) with cost and to have a page on our website that helps someone determine what to expect (probably some sort of calculator with live data). Our UI will also show you exactly what your system is currently costing you with some breakdown for teams or services so you know who/what is driving your monitoring cost. We're doing user testing on our to-be-released UI now and would love to have people like yourself give us early feedback (since you mentioned the BigQuery interface).

mleonhard 5 years ago |

> opstrace create -c CONFIG_FILE_PATH PROVIDER CLUSTER_NAME

> opstrace destroy PROVIDER CLUSTER_NAME

> opstrace list PROVIDER

I want to keep cluster config in source control, track deployment changes in code reviews, and automate deployments. Do you have any plans to add an 'apply' command to support this?

$ opstrace apply -c CONFIG_FILE_PATH [--dry-run] PROVIDER CLUSTER_NAME

jgehrcke 5 years ago | |

Hey! JP from Opstrace here. Thanks for reading through things and for sharing your thoughts. The quick reply is that we still have to introduce a proper cluster config diff and mutation design.

An `apply` command might look innocent on the surface. But. Upgrades (including config changes) are hard. Super hard. If it's helping a bit: the entire current Opstrace team has dealt with super challenging platform upgrade scenarios in the most demanding customer environments in the past years. We try to not underestimate this challenge :).

We're super fast moving right now and didn't want to bother with in-place config changes (as you can imagine, we wouldn't really be able to provide solid guarantees around that). We'll work on that and make it nice when the time is right, and when we feel like we can can actually provide guarantees.

mleonhard 5 years ago | | |

Yes, `apply` is hard. It's just as hard as deploying, maintaining, and turning down a service. When adding an `apply` command to a devops tool, the tool authors must think through the entire lifecycle of their service in the user's workflow and make it work well.

The tool creators are the ones with the knowledge to figure these things out. If they don't provide `apply`, then users must research and experiment and learn by making mistakes. This is a colossal waste of effort. Users end up with brittle poorly-documented scripts to do all the things that `apply` would do. These scripts cause ongoing waste of engineering effort, customer frustration from downtime, and lost business growth and revenue.

I spent several weeks making `apply` commands for InfluxDB [0] and Grafana. This proved extremely difficult for Grafana because of deficiencies in its API. Both InfluxDB and Grafana need some work to make them fit into a modern infrastructure-as-code ops environment. Grafana's cofounder and product lead were not interested in my feedback [1] [2].

[0] https://github.com/cozydate/influxdb-apply

[1] https://news.ycombinator.com/item?id=23136582

[2] https://news.ycombinator.com/item?id=23233468

richardw 5 years ago |

Point around incentives: We use Dynatrace. I’m sure it’s an eye-watering price but I do like that everyone who wants a license can get one. I don’t have to consider costs to add an entire dev team and teach them how to use it. It also means an entire dev team knows how to use it for future jobs.

fat-apple 5 years ago | |

We certainly don't want to create an adverse incentive where you would consider limiting the number of devs who had access to the monitoring system. There are trade-offs but we think that per-seat pricing (like GitLab, GitHub) actually does make it much easier to budget and plan for monitoring spend. Generally, a headcount plan is more predictable than the data your application generates. For example, a single engineer can add (and maybe should be adding) far more metrics and logs to their applications to monitor it correctly. They should not also be worried about breaking the budget when doing so. Does this make sense to you - what do you think?

polskibus 5 years ago |

What are your plans on supporting open telemetry? Can I send open telemetry data to opstrace?

fat-apple 5 years ago | |

Yes you can do this today. You cannot send this directly to Opstrace via grpc yet, so you would need to use the Prometheus remote write exporter (https://github.com/open-telemetry/opentelemetry-collector/bl...) with the open telemetry collector. Right now this means only metrics are supported but we will be working on traces as well. Check out the point about tracing on our roadmap https://opstrace.com/docs/references/roadmap

rtkaratekid 5 years ago |

Looking through the docs I'm seeing there will be (at some point) an API. Does this include ways to integrate data coming from non-Opstrace sources? My specific case is an in-house monitor that basically just generates data.

englambert 5 years ago | |

Yes, indeed! Currently you can already using the Prometheus remote_write API as discussed here: https://news.ycombinator.com/item?id=25992392. So if you can collect your in-house metrics with Prometheus, then you could write to Opstrace.

Additionally, we are close to launching a Datadog API as mentioned here: https://news.ycombinator.com/item?id=25994268

So stay tuned to our blog or newsletter for more on this.

Are there other specific APIs you’re interested in?

rtkaratekid 5 years ago | | |

Thanks for the reply! Not necessarily, I work in R&D developing low-level data collection so I'm more just trying to keep my ear to to ground in terms of what's going on in the stack just above what I'm doing :)

ogazitt 5 years ago |

Congrats on launching - looks awesome! It's about time we have an open source datadog :)

Also, it's great to see the early focus on developer experience - "opstrace create".

fat-apple 5 years ago | |

Thank you! We like to simplify but not obfuscate!

opsunit 5 years ago |

Why should I run this instead of renewing my Wavefront contract?

englambert 5 years ago | |

It's hard to answer that concretely without knowing a little bit more about your use cases. Care to share a bit more?

One thing comes to mind: we don't bill by data volume. Wavefront is charging you for the volume of data your applications produce. This can lead to negative outcomes, such as surprise bills from a newly deployed service and a subsequent scramble to find and limit the offenders.

We think this pricing model forms the wrong incentives. Charging by volume means a company is more incentivized to have their customers (you) send you more data, and less incentivized to help them get more value from that data. This is a fundamental change we want to bring to the market--we want our incentives to align with yours, we want to be paid for the value we bring to your company. We charge on a per-user basis. You should monitor your applications and infrastructure the right way, not afraid to send data because it might blow the budget.

opsunit 5 years ago | | |

Wavefront brings a number of things to the table that aren't core competencies we wish to maintain in-house.

I know it can scale to massive volumes without interaction from us.

I know it'll be available when our infrastructure isn't. By being a third party we can be confident that any action on our part (such as rolling an SCP out to an AWS org, despite unit tests) won't impact the observability we rely on to tell us we've screwed that up.

I can plug 100s of AWS accounts and 10s of payers into it and I don't have to think about that in terms of making self-hosted infrastructure available via PrivateLinks or some other such complication.

I pay mid six-figure sums annually for these things to "just work". If you folks believe I can achieve this functionality on a per-seat basis I'd be interested in saving those six figures.

thow_away_4242 5 years ago |

Meanwhile, AWS getting the "AWS Opstrace Service" branding and marketing pages ready.

fat-apple 5 years ago | |

:-) I have talked about the subject in this comment thread: https://news.ycombinator.com/item?id=25991764

alexhutcheson 5 years ago |

One pain point with Prometheus is that is has relatively weak support for quantiles, histograms, and sets[1]:

- Histograms require manually specifying the distribution of your data, which is time-consuming, lossy, and can introduce significant error bands around your quantile estimates.

- Quantiles calculated via the Prometheus "summary" feature are specific to a given host, and not aggregatable, which is almost never what you want (you normally want to see e.g. the 95th percentile value of request latency for all servers of a given type, or all servers within a region). Quantiles can be calculated from histograms instead, but that requires a well-specified histogram and can be expensive at query time.

- As far as I know, Prometheus doesn't have any explicit support for unique sets. You can compute this at query time, but persisting and then querying high-cardinality data in this way is expensive.

Understanding the distribution of your data (rather than just averages) is arguably the most important feature you want from a monitoring dashboard, so the weak support for quantiles is very limiting.

Veneur[2] addresses these use-cases for applications that use DogStatsD[3] by using clever data structures for approximate histograms[4] and approximate sets[5], but I believe its integration with Prometheus is limited and currently only one-way - there is a CLI app to poll Prometheus metrics and push them into Veneur[6], but there's no output sink for Veneur to write to Prometheus (or expose metrics for a Prometheus instance to poll), and you aren't able to use the approximate histogram or approximate set datatypes if you go that route, because they can't be expressed as Prometheus metrics.

It would be extremely useful to have something similar for Prometheus, either by integrating with Veneur or implementing those data structures as an extension to Prometheus.

[1] https://prometheus.io/docs/practices/histograms/

[2] https://github.com/stripe/veneur

[3] https://docs.datadoghq.com/developers/dogstatsd/

[4] https://github.com/stripe/veneur#approximate-histograms

[5] https://github.com/stripe/veneur#approximate-sets

[6] https://github.com/stripe/veneur/tree/master/cmd/veneur-prom...

jgehrcke 5 years ago | |

Thanks for bringing this topic to this thread. I'm a physicist by heart and education myself and observe that in the software/observability industry we like to collect data much more than we're interested in properly processing and interpreting it.

> Understanding the distribution of your data (rather than just averages) is arguably the most important feature you want from a monitoring dashboard, so the weak support for quantiles is very limiting.

So much yes! It's a relief to see that we have people here in this thread (and industry) who understand this :-).

People that have a deep background and experience in experimentation, measurement, and quantification rightfully have to see the nature of the data distribution first before they feel in any way OK about proceeding with aggregates.

Parent commenter knows this, but for people reading along: using aggregates (such as mean, standard deviation, standard error, quantiles, ...) implies dropping information. Going from the full distribution to a simplified representation naturally implies that what we talk about is a lossy transformation of data. Of course, one wants to be smart about _which_ information to drop. It should be intuitive that one can only be smart about this choice when having knowledge about the underlying distribution. Often, data is not normally distributed, not Poisson-distributed, but instead somewhat uniquely distributed based on the use case -- in a way that deserves brief characterization (a quick look is often enough!); which then allows for making informed decisions about which aggregate parameters to look at -- and which pieces of information are fine to drop.

> Histograms require manually specifying the distribution of your data, which is time-consuming, lossy, and can introduce significant error bands around your quantile estimates.

Yes! Great point. Honestly, I was a little bit shocked when I saw how this works in the Prometheus ecosystem. I happen to have an example for this I think: we (Opstrace) have contributed a tiny patch to Cortex where we changed the parameterization of a specific histogram metric, because the upper band was super broad, leading to a blind spot (a lack of resolution) in the range of values that was most interesting to us -- see https://github.com/cortexproject/cortex/issues/2530 and https://github.com/cortexproject/cortex/pull/2540.

I feel like in our industry when people do the readout and perform some basic dashboarding they're often OK with using aggregates that have not been sanity-checked. That might sometimes be a fair approach. If a readout is not useful, after all one will learn about that very fact through incidents :).

> Veneur[2] addresses these use-cases for applications that use DogStatsD[3] by using clever data structures for approximate histograms[4] and approximate sets[5]

This is extremely interesting, thanks for sharing. We'll have a look!

Having performed many different kinds of data analyses in my science career, I'm rather convinced of the idea that there should be an easy way to perform advanced (yet standard) data analyses for irregular distributions in multi-dimensional data, such as clustering (hierarchical, k-means, DBSCAN, you name it), Principal Component Analysis (PCA), and the likes. I have ideas in this regime -- let's see what makes sense and how far we get with Opstrace!

Again, thanks for the links and for sharing your perspective.

tobilg 5 years ago |

Great job, congratulations from an ex-Mesosphere colleague!

nickstinemates 5 years ago |

Congrats! This is really exciting

rockyluke 5 years ago |

Congratulations! You did a really great job.

NSMyself 5 years ago |

Looking good, congrats on launching

fat-apple 5 years ago | |

Thanks for that feedback - we've got a lot of work to do to make that more clear! We're still early in our journey so we’re not there yet, but we’re moving fast. We're working on a new collaborative UI for interacting with your data in a way that solves a lot of problems we've witnessed with current monitoring UIs (let me know if you want more detail). It's in early development and we haven't released it yet, so while Opstrace does have a UI now, it's currently limited to system management (adding/removing users and tenants). For interacting with data, we currently ship a Grafana instance per tenant. The roadmap has some basic information about this (might not be something you stumbled across). Let me know if I can clarify anything else. https://opstrace.com/docs/references/roadmap

tailspin2019 5 years ago | | |

Thanks for clarifying. I think all the great work you’re doing on this is at risk of being diminished by pushing the DataDog comparison so hard when really there is (currently) no comparison in my view.

Not to say you won’t get there, but why not shout about your current strengths right now (if you’re going to talk about the product) and not necessarily where you hope to be.

I only make this point because the “open source DataDog” headline just spins me off in the complete wrong direction as a newcomer and I end up confused and somewhat underwhelmed. With better positioning this wouldn’t have been the case!

That said, the project looks v interesting so keep up the good work and I hope you take my constructive criticism in the way it’s intended!

fat-apple 5 years ago | | |

Thanks for the great feedback, definitely understand where you're coming from and we'll keep that in mind moving forward! Very much appreciate the constructive criticism!

Denzel 5 years ago | |

The headline feels deliberately clickbait-y and disingenuous. I love and support the idea. Aspirationally, the founders may want to compete with Datadog, but OpsTrace overlaps some small percentage of Datadog’s feature set.

I’m surprised the mods haven’t edited this title.

Source: I’m an engineer that’s used, operated and hacked on a medium-sized prom+grafana; and used Datadog at a large, multi-region, global scale.

jgehrcke 5 years ago | | |

I'd love to chime in, too, and say thank you for sharing this perspective. It's totally understandable, and this HN thread wouldn't be complete and genuine w/o this talking point. Based on your experience, we'd certainly appreciate for you to keep an eye on how our product evolves! Super interested in your expertise and opinions. Thanks again for not being afraid of being 'that guy'.

dang 5 years ago | | |

Sorry about that! I didn't see anything wrong with the title but I'm not familiar with Datadog's feature set. If you take the title as aspirational, it may make more sense.

Denzel 5 years ago | | |

No need to be sorry, you do a great job moderating HN dang!

I just felt let down because the promise of the title (which I was excited by) doesn’t match reality in that it’s quite impossible to deliver Datadog’s feature set for metrics, traces, and logs — and tie all three together — on top of prometheus+grafana because the underlying TSDB doesn’t even support the notion of user-customizable indexing. Prometheus indexes all labels, which is why most Prometheus users even have to worry about high cardinality and time series explosion. This is one point highlighting how OpsTrace’s current architecture can not satisfy their positioning; there are many more. I’m sure they’re aware of them.

As an aspiration, the title makes sense. As someone in their target market, I was a bit turned off by the embellishment.

fat-apple 5 years ago | | |

We’re truly sorry that you felt let down! You raise a great point that is worth clarifying though. All logs, metrics and future traces are stored in S3/GCS (and yes metrics are stored in TSDB format). While this does not allow a single query language to ask questions across all these sources in one query (yet), it is absolutely possible to build what Datadog is with a new user interface. To go even further, now that it’s all in one place (S3/GCS) other technologies can be leveraged to create a higher level query language across everything, such as PrestoDB (one would have to build backends for it), or even a new dedicated open source columnar store.

Denzel 5 years ago | | |

> other technologies can be leveraged

That’s my point: your software architecture must evolve to deliver on your promises (open-source Datadog) because it’s impossible to satisfy your promises with the architecture as it exists today.

Nonetheless, I appreciate all the responses. Good luck to OpsTrace!

spahl 5 years ago | | |

It's definitely aspirational. With time and work the features gap (for the important ones) will narrow:-)

What we want to emphasize is that it is possible to build this and have the advantages of a Datadog without the drawbacks.

fat-apple 5 years ago | | |

Thanks for your support! We’re still early in our journey when compared to the featureset Datadog has built over time, but we definitely aspire to bridge the best of Datadog and open source. Based on your experience, what sorts of features would you prioritize to get there?

jgehrcke 5 years ago | |

Jan-Philip from Opstrace here. This is lovely feedback!

> is a royal pain in the neck.

It's fun to see how different people put the same unpleasant experience into words in this thread. Thanks for adding your personal touch. Every time we hear something like that, we're re-assured that we're on the right track.

> Even if `opstrace create` spawns them easily, the problem is running/maintaining them

Yes. You're right. While we can be proud of our setup/installation process already, we know that there's so much more to it. We don't underestimate that. Maybe also see https://news.ycombinator.com/item?id=25998587, where I just commented on the robustness topic.

> However, we are happy for you to be a custodian of that data.

Great.

> We want storage to be someone else's problem.

I share that perspective. We, of course, are happy to let S3/GCS do the actual job.

> We're happy for you to use a cloud platform like AWS/GCP and charge us a 50% markup.

That's great to hear, and I hope you can be enthusiastic about the fact that our markup is _not_ going to be relative to storage volume. It's going to be independent of that.

> Transparency, predictability of costs, open source --> all excellent.

Thanks for sharing. That's incredibly motivating.

Keep an eye on us, and we'd love to hear from you!

xchaotic 5 years ago | | |

This is an interesting perspective. Our current monitoring system charges by the amount of data ingested / stored so there is a perverse incentive to observe less men if we have detailed debug level logs available.

danmur 5 years ago | | |

I'm glad I don't have to pay per-man to observe men :P.

dudeinjapan 5 years ago | | |

Will definitely keep an eye on your service. Again, don't underestimate users willingness to pay for a PaaS product you make, even if you also do a dual PaaS/self-hosted option like Gitlab or MongoDB does, for example. We'd definitely prefer the PaaS, and that's where the big $$$ is made these days.

By the way, we are requiring that any vendor we choose in this area support OpenTelemetry as we've already instrumented our apps with it. Lightstep, Datadog, and others already supporting.

spahl 5 years ago | | |

Makes sense! Thanks for the feedback.

And yeah OpenTelemetry is where lots of the industry is going and so are we. You can already do cool things with us and the collector but we plan on automating a lot and expand on that.

A little bit about that in this commment: https://news.ycombinator.com/item?id=25995424

xyzzy_plugh 5 years ago | |

Lightstep was bananas expensive and had several limitations that lead to us moving away from it. Hopefully it's easier to scrub PII from it these days.

dudeinjapan 5 years ago | | |

Lightstep has recently changed their pricing, it's now the most cost effective option out there. You pay $100 per service with basically unlimited data ingestion, it's a fabulous deal. https://lightstep.com/pricing/

As for scrubbing PII, they are now supporting the OpenTelemetry tracing API which does this as standard. For query endpoints you will see something like "QUERY Users where name=? email=?", i.e. masking with "?" chars as you only care about the keys since those what determine your indexing or lack thereof. (This is handled in the OpenTelemetry application library/plugin level.)

As an aside, PII scrubbing should be done that whether or not you own the cluster, because even if you own the cluster you generally don't want your support staff seeing PII esp. as organization grows larger.

englambert 5 years ago | | |

Heh, yes, the PII scrubbing need is very real. This is definitely a contributing factor to our data ownership commitment--just keep your data in your own account.

On the cost topic, last week we published a blog post analyzing the cost of running Opstrace on AWS (https://opstrace.com/blog/pulling-cost-curtain-back). (In fact, feel free to do a local repro to confirm our results.) As mentioned elsewhere here on HN, we are incentivized to provide total transparency in terms of what you spend on your cloud infrastructure. We haven’t compared ourselves to everyone, but feel confident that letting our customers pay S3 directly is the best deal possible.

BringerOfChaos 5 years ago | |

@dudeinjapan, check out https://grafana.com/products/cloud/ The first line on the page says, "Your observability, managed as a service"

It might not fit your use case...but it might.

dudeinjapan 5 years ago | | |

We used self-hosted Grafana until late 2020. We looked at Grafana cloud but ultimate chose a combination of NewRelic One for metric monitoring and Lightstep for request tracing. Those two were the "best-of-breed" for their respective areas with the best pricing for our use case.

davelester 5 years ago | |

SaaS is a common way for open source companies to create revenue, look no further than WordPress, GitLab, Databricks, DataStax, and many others. Kudos to the opstrace team for taking this path.

There’s nothing inherently dishonest when a company emphasizes their open source strategy. Open source community building is as much about shipping code as it is leading people, and that requires you to be transparent about your intentions. I’ve interpreted opstrace’s release as just that.

I think the concern about neutral project governance is an important one. It’s early days, but from what I’ve seen it seems clear what is being sold vs what is open today. The fact that the project is released under the Apache v2 license means that folks are able to reuse, distribute, and sell the project as they wish — even fork it if they dislike the direction. That said, if governance is a priority for your use I’d definitely look to project in neutral software foundations like the Apache Software Foundation and CNCF.

sneak 5 years ago | | |

Open source and proprietary with source publicly available are not the same thing at all.

fat-apple 5 years ago | |

Thanks for your feedback! As with many in the industry, we are trying our best to figure this out.

Our intention is to be really transparent with how we build and price software, which is why our commercial features will also be public in our repo, but commercially licensed. Transparency is critical in our opinion.

This is the model we’ve seen work for other highly impactful software projects.

We’ve created a ticket to track our addition of commercial code to our repo: https://github.com/opstrace/opstrace/issues/319

sneak 5 years ago | | |

Ahh. The commonly used "open source" term is a synonym in practice for "free software", so if you are a proprietary software vendor that is source-available but calling it "open source" (when really, it's not) is a different type of misleading.

Don't call source-available proprietary software "open source", or say that you'll "always be open source" if you're just going to be source-available for parts.

kazinator 5 years ago | |

I'm gonna put on my St. Ignucius robe here, and say that yes, that behavior is within the limits of "open source".

If they used "free software" language, then we might have a case for posturing.