Observability Companies to Watch in 2024

- 16 mins read

Observability is often described as three pillars—logs, metrics, and traces. Many companies have been built around this idea, but others have risen to challenge it.

Let’s see what observability companies are up to in 2024.

I’ve recently been tasked with a project to consolidate observability systems. The primary goals were to decrease context switching between systems and increase debugging velocity across teams.

This project involved researching and evaluating observability companies to find one that provided the best value. About halfway through this research I found Oxide’s RFD 68: Partnership as Shared Values.

This RFD changed the way I was thinking about this research by categorizing a company as either a partner or a vendor, using values to weigh whether you should choose to work with a company. I recommend reading that RFD if you have the time.

After researching and evaluating many observability companies, these are the companies I believe are worth watching in 2024.

  1. Honeycomb
  2. Axiom
  3. Grafana Labs
  4. VictoriaMetrics
  5. ClickHouse
  6. SigNoz
  7. Datadog

Honeycomb

Honeycomb may not need an introduction, as they are a beloved company in the observability space. They are pioneering an Observability 2.0 mindset that abandons the traditional three pillars of observability in favor of arbitrarily-wide structured log events. The high-level idea is that these wide events contain the necessary context to quickly and effectively observe your system. You can think of a wide event as a structured log that contains all the necessary fields you’d want to query, all linked by one or more fields (e.g., trace ID).

Things I Like

The way Honeycomb thinks about observability is refreshing, especially in a market filled with companies looking to charge you more for adding an extra label on your telemetry data. Honeycomb’s Observability 2.0 mindset using structured log events gives you control and freedom over your telemetry data with the ability to query it quickly. It might feel awkward to ditch the three pillars of observability, but when you do Honeycomb will reward you by helping you find those pesky unknown unknowns within your system.

Honeycomb’s BubbleUp feature is one of their key differentiators and it’s a game-changer. BubbleUp takes the best features from anomaly detection, correlation, and visualization, packaging them into an easy-to-use tool that anyone can use to identify outliers in telemetry data. Check out Honeycomb’s sandbox for a demo of BubbleUp.

Honeycomb is a huge advocate for OpenTelemetry, a popular open source collection of APIs and SDKs for telemetry data, with first-class support for the OpenTelemetry Protocol (OTLP) in their APIs and libraries. The leadership at Honeycomb includes industry leaders Charity Majors and Liz Fong-Jones, both of which have extensive observability experience and share a common vision. With both the technical and non-technical sides covered, Honeycomb is well positioned to take their Observability 2.0 mindset mainstream.

Things to Improve

Honeycomb is targeted at teams that create or operate production systems, primarily those that have the ability to either change how software emits telemetry data or decorate emitted telemetry data before shipping it to a destination. Honeycomb’s focus on structured log events can make it difficult for teams to migrate from an Observability 1.0 mindset to Observability 2.0, especially if those teams treat logs, metrics, and traces separately. I would like to see Honeycomb bridge the gap between Observability 1.0 and Observability 2.0.

Honeycomb’s pricing is based on events, measured in events/month. This pricing model is great because it doesn’t penalize teams for sending high-cardinality telemetry data. However, this pricing model can become expensive for teams with separate logs, metrics, and traces looking to migrate to Honeycomb. Instead of being able to send their data as-is and incrementally migrate towards an Observability 2.0 approach, teams must choose to either sample data or dedicate resources to creating structured log events. I believe Honeycomb has a great opportunity here to enhance this experience and better support teams in this scenario.

Partner or Vendor?

I view Honeycomb as a partner. I believe their Observability 2.0 vision is where the industry is headed and they are willing to partner with companies to transform their way of thinking about observability. Honeycomb is one of the biggest advocates for OpenTelemetry and their Pollinators Slack is buzzing with people that have tasted the sweetness of Observability 2.0. I had the opportunity to chat with a few employees from Honeycomb, provide feedback on their product, and test changes that they made based on the feedback. Honeycomb isn’t selling you a product. They are selling a vision of what observability can look like with a tool to turn that vision into reality.

Axiom

Axiom is a newer company on this list. They started with the goal of disrupting industry logging leader Splunk and have delivered an excellent logging offering to do just that. Axiom’s logging offering has expanded into traces, with metrics support coming soon. Their “Stop sampling, observe every event.” tagline tells a story similar to Honeycomb’s Observability 2.0 mindset using structured log events.

Things I Like

Axiom’s text-based Axiom Processing Language (APL), inspired by Microsoft’s Kusto Query Language (KQL), is a powerful, readable, and intuitive query language. The syntax uses pipes to separate operators which should feel familiar to those used to Unix tools.

Here’s an example APL query that extracts the job state from logs using regular expressions, filters by successful jobs, and produces a summary of successful jobs over time.

['app-logs']
| where message contains "transitioned to state"
| extend job_state = extract("transitioned to state (\\w+)", 1, message)
| where job_state == "succeeded"
| summarize num_jobs = count() by bin_auto(_time)

Readable, isn’t it?

Outside of APL, I like that Axiom is thinking about observability similar to Honeycomb with a focus on structured log events. Axiom may have started with the three pillars of observability, but it seems they are pivoting to structured log events, and can be a great competitor if they can execute well.

Axiom’s pricing model is based on ingest, measured in GiB/month. I like this pricing model because it’s predictable and meets companies where they are in their observability journey. Companies can migrate from high volume, narrow structured log events to lower volume, wide structured log events without worrying about increasing their bill.

Things to Improve

Axiom is a newer company, so the obvious thing they can do to improve is expand their customer base and secure a spot in the market. Onboarding more customers will build trust in their brand, test the reliability and scalability of their platform, and expose new use cases that can be used to refine their roadmap.

More concretely, I would like to see Axiom land a first-class metrics experience and expand their philosophy around structured log events. Metrics are listed as coming soon on Axiom’s website, but the platform does support ingesting metrics via their ingest API, with an OpenTelemetry Protocol (OTLP) endpoint also listed as coming soon. If you do ingest metrics today, note that the metrics query experience isn’t yet up to par with competitors, lacking the ability to query histograms and calculate rates for metric values over time.

Partner or Vendor?

I view Axiom as more of a partner than a vendor. Their values align with my own, they support open tooling such as OpenTelemetry, and their Discord is active and growing. I’ve spoken with some of the employees at Axiom and they were a joy to work with. They were the only vendor that didn’t hesitate to enable the enterprise trial and they expressed interest in receiving feedback to better their product. Given their smaller size, the overall experience felt personable.

Grafana Labs

Grafana Labs is the company behind the popular Grafana visualization tool. Grafana Labs has since expanded their offerings into Loki, Mimir, and Tempo for logs, metrics, and traces respectively. Their entire range of offerings is marketed as the “LGTM” stack, which stands for Loki, Grafana, Tempo, and Mimir, and plays on the familiar “looks good to me” message teams use in code reviews.

Things I Like

Grafana Labs is has an open source tier of their offerings that can be self-hosted. In fact, I’d argue that’s what made Grafana Labs popular in the industry. Teams that use Grafana Labs’ offerings in production often ship the open source versions alongside their applications in non-production environments to give teams the same observability experience across production and non-production. That’s a stark difference compared to other SaaS companies where it’s cost prohibitive to send telemetry data from non-production environments and teams must either do without their telemetry data or find another way to visualize and query it.

Grafana itself is a powerful visualization tool. If you haven’t used Grafana yet I encourage you to try it. Grafana can read telemetry data from many different data sources, including the popular metrics tool Prometheus, and visualize that telemetry data using the query language of the data source itself. That means you can visualize Prometheus metrics using its own PromQL syntax and Elasticsearch data using its lucene syntax. This is not to say other offerings like Loki and Tempo aren’t good, but Grafana is the standard when it comes to visualization, so much so that other companies often bundle Grafana with their own offerings.

Things to Improve

Grafana Labs has a SaaS offering called Grafana Cloud that you can use if you don’t want to self host. However, I found the pricing of Grafana Cloud to be expensive when compared with other companies, especially at scale. I would like to see Grafana Cloud become more competitive with their pricing over time.

As noted above, Grafana allows you to visualize telemetry data backed by the query language of the data source itself. This is a benefit when teams are familiar with a given query language and mostly use a few data sources, but can quickly become a burden when context switching between data sources. On top of that, Grafana Labs’ own offerings use slightly different query languages (e.g., PromQL for Mimir, LogQL for Loki, TraceQL for Tempo) which can burden teams with unecessary context switching.

Partner or Vendor?

Overall I struggled to answer this question for Grafana Labs. On one hand they feel like a partner when you look at their open source offerings coupled with their sponsorship of open source projects. On the other hand their Grafana Cloud offering feels like a vended product that attempts to unify logs, metrics, and traces telemetry data through their Grafana visualization interface. The people I spoke to at Grafana were easy to work with but I didn’t end up doing a serious evaluation of their Grafana Cloud offering due to pricing concerns.

VictoriaMetrics

VictoriaMetrics is the company that created the VictoriaMetrics time series database as a more reliable, memory efficient replacement for existing time series databases. They have since expanded into logs with their VictoriaLogs offering, which at the time of this writing is in preview.

Things I Like

VictoriaMetrics is open source and popular with those looking for an efficient self-hosted metrics solution. Speaking of self hosting, getting set up with VictoriaMetrics is quick and intuitive, so be sure to download its all-in-one binary or container image and check it out. I also like that VictoriaMetrics wrote their own time series database after researching the pain points operators were experiencing with other metrics offerings. VictoriaMetrics also has a SaaS offering if you don’t want to self host.

VictoriaMetrics supports a wide range of ingestion protocols including Prometheus, Graphite, OpenTSDB, CSV, and InfluxDB. This wide range of support makes it easier for teams to migrate their workloads to VictoriaMetrics. Generally speaking, VictoriaMetrics is a drop-in replacement for any one of the previously mentioned tools.

Things to Improve

I’d like to see VictoriaMetrics continue to expand into other telemetry data. This is already in development with their VictoriaLogs offering, but I think going the full distance with traces would be beneficial. Right now VictoriaMetrics seems more like a monitoring company than an observability company, but perhaps that will change with time.

Another thing I would like to see changed is for VictoriaMetrics to list pricing information on their website. Currently, you must request pricing information by filling out a form and waiting to be contacted. Trust me, I get it. When you’re a smaller company you want to make sure leads are fruitful and have a good chance of converting into a sale. However, it’s difficult for teams to estimate pricing for growth without posting it on the website.

Partner or Vendor?

I don’t believe I have enough information to answer this question accurately. VictoriaMetrics is one of those companies I’m keeping my eye on and might migrate a few Prometheus workloads over to in due time. If I had to answer this question I would say VictoriaMetrics feels more like a vendor than a partner since their primary product is meant to target existing products rather than carve a new path of its own. I didn’t join the VictoriaMetrics Slack to see how their community building is, nor did I speak to any employees from the company to make me lean towards partner here.

ClickHouse

ClickHouse isn’t exactly an observability company. They are an open source database for online analytical processing (OLAP) workloads. Their claim to fame is that they are fast and efficient. No really, they are fast. They are on this list because their speed and efficiency make their database attractive for storing telemetry data and companies are starting to pop up that use ClickHouse as their storage and querying layer.

Things I Like

ClickHouse has a clear vision and knows exactly what they’re offering. They are focused on ensuring their database remains fast and reliable as you store billions of rows of data inside. I like their dedication to their vision and their astonishing performance. Try loading up a large data set and querying it to see what I mean.

Like most databases, ClickHouse uses SQL as its query language. Teams that are familiar with SQL will feel right at home when querying data and companies looking to use ClickHouse as their storage layer will have stable APIs to build from. I’m not saying that SQL is the best query language, but it’s ubiquity makes it simple to onboard to ClickHouse.

Things to Improve

I don’t have anything specific to say here. It wasn’t exactly fair for me to include ClickHouse on this list since they are focused on OLAP workloads. Instead, I’ll give them a pass and thank them for the work they are doing.

Partner or Vendor?

Definitely vendor. I don’t say that in a bad way either. I say it because ClickHouse knows the target audience for their database and is marketing to that audience well. For the average operator, I imagine purchasing ClickHouse would be mostly transactional in that you’ll buy the database and use it in your workflows, only reaching out to ClickHouse when there are issues or feature requests.

SigNoz

SigNoz is an observability company whose mission is to help you find the signal in the noise. If you couldn’t tell, their name is a portmanteau on the words signal and noise. SigNoz takes all of the open source observability tooling you know and love and glues them together through their visualization tooling.

Things I Like

I have not personally used SigNoz, but I’ve placed them on this list because they are one of those companies that are using ClickHouse to store and query telemetry data and I wanted to keep an eye on their progress. In addition to using ClickHouse, SigNoz leans into OpenTelemetry much like Honeycomb, using it as their ingest layer. Take a look at their architecture to understand how all these components fit together.

Things to Improve

From their demo video, SigNoz reminds me of Grafana Labs. They treat logs, metrics, and traces separately and unite them with their visualization experience. In this sense it may feel like SigNoz has not yet found its own identity, but we’ll have to wait and see. That brings me to the point I was trying to make— SigNoz is not yet well-known or proven. I’d like to see them become more utilized over time and their name one that is frequently brought up in observability conversations.

Partner or Vendor?

I can’t answer this one given that I didn’t use their product. I’ll be keeping an eye on SigNoz to see how their decision to use OpenTelemetry and ClickHouse pans out.

Datadog

Datadog is one of the first companies people think of when talking about observability. They have been around for a while and are the default observability choice for many. Datadog started with metrics and expanded into logs and traces over time, now offering a rich portfolio covering use cases across infrastructure, security, and application development. I put Datadog on this list since they are the current industry leader and I want to watch how they react with increased competition.

Things I Like

No matter your use case Datadog likely has an offering for you. This is attractive to teams that need a single company to meet the diverse needs across teams. While Datadog may not be the best at any one thing, they are good enough across everything and I like that about them.

Datadog is a proven company in the industry. Their wide customer base gives companies confidence that Datadog will be around for years to come and that they can scale to meet their use cases. Datadog is a safe choice for observability and, as the saying goes, no one ever got fired for buying Datadog.

Things to Improve

Everything I said about Datadog sounds great on paper but it all comes at a cost. Datadog’s pricing is… complex to say the least. Go ahead and navigate to their pricing page and click through the different offerings in the sidebar. Let me know if you’re able to accurately estimate what your bill would look like if you migrated to Datadog. Seriously, ping me if you were able to easily do this. Outside of being complex, Datadog’s pricing is expensive! Take a look at what they charge for custom metrics, which would be any metric that is emitted from your in-house application. You know, the applications your company runs to make money. There are many videos and blog posts about Datadog’s expensive pricing so I won’t go into any more detail here. Feel free to research this for yourself.

Outside of pricing, it’s easy to become locked in to Datadog. Datadog has their open source Datadog agent that makes it simple to send them telemetry data. The agent comes with many integrations that you can toggle with a simple boolean in their configuration file. I dislike two things about the agent. First, it’s not clear if there are any billing implications when an integration is enabled, which I won’t dive into any further since I already spent time talking about pricing. Second, the agent treats metrics coming from integrations as standard metrics instead of custom metrics. Standard metrics are far cheaper than custom metrics so if you choose to emit metrics from an application directly instead of using their integration, you’ll pay more for the same exact metrics.

Standard metrics may also be transformed by the agent into a “canonical” format that’s specified by Datadog. For example, say you have a foo application that emits a Prometheus metric named bar_http_requests. You have queries using this metric and you want to create a foo integration in the Datadog agent so others can easily get your application’s metrics into Datadog. Before accepting your foo application as an integration, Datadog may choose to rename your metric to foo.bar.http_requests which is not what your application emits in the first place. Granted, adding the foo. prefix makes sense since it helps prevent collisions between integrations but after that no further transformations should be done. This behavior makes it extremely difficult to migrate away from Datadog since it will require updates to your queries that are using the original metric name.

Partner or Vendor?

Datadog is definitely a vendor. They are focused on increasing their revenue and building their walled garden, not on innovating in the observability space. It’s no secret that people are unhappy with Datadog’s pricing, vendor lock-in, and treatment of the OpenTelemetry community and are looking for a company to supplant them.