CTWHow CTW provisions full-stack game observability in seconds with Grafana Cloud
When a new video game launches on G123, the engineers who keep it running don’t build a monitoring setup for it. They don’t wire up log pipelines, stand up dashboards, or configure alerts by hand. They click a button, and the entire observability stack — data sources, folders, dashboards, and alerts — appears in seconds. Not long ago, that same work took days.
CTW Corporation operates G123, one of the world’s largest platforms for HTML5, browser-based games built on popular Japanese anime IP. Players have access to more than 40 live titles instantly, with no downloads or installs, and the platform has served roughly a billion users since its launch, with paying players across 191 countries. Behind it sits a deliberately lean cloud team — about ten people across Tokyo and Shanghai, with a newly opened New York office — responsible for everything from the underlying multi-cloud infrastructure up to the platform-as-a-service layer that partner studios use to deploy their games.
That one-click launch is the surface of a much deeper shift. Standardizing on Grafana Cloud as a single, centralized observability platform and migrating onto it from a patchwork of Datadog – a self-hosted Elasticsearch-Logstash-Kibana (ELK) stack – and three cloud consoles did far more than consolidate tools. It reset how CTW works, and let the team push observability beyond its own walls. What was once a manual practice owned by a handful of SREs is now a shared capability that 84 engineers use every day — and one CTW extends directly to the dozens of game studios building on its platform, many of which had never run real observability before.
“Our goal is to provide secure and scalable cloud services,” says Yin Yannan, the company’s Head of Infrastructure. “And with Grafana Cloud in place, that’s exactly what we are able to do now.”
A fragmented before-state across five places to look
Originally, CTW’s observability was spread thin. Datadog handled metrics and APM, mostly for the platform’s own services. A self-hosted ELK stack carried the logs because at the time, sending CTW’s log volume to a commercial SaaS tool was prohibitively expensive. On top of that, metrics and status lived in the AWS, Alibaba Cloud, and GCP consoles. With infrastructure spanning three clouds, an investigation could mean checking five different places and hoping the answer was in one of them.
Two problems compounded the sprawl. The first was cost and operational load. Running ELK in-house meant the team owned storage, scaling, upgrades, performance tuning, and reliability a heavy burden for a group that prides itself on a lean, infrastructure-as-code approach. “As our log volume expanded, we had to maintain the whole thing,” Yin says. “The ELK stack just cost a lot of energy.”
The second was coverage. Many games ran on a VM-style deployment and kept their logs on their own servers, never shipping them to CTW. “Games have their logs, and we don’t have a picture of what’s going on there,” Yin recalls. When something went wrong, the team monitored what metrics it could, took screenshots, and shared findings with studios manually over chat and operations reports. Controlling a cost spike often meant dropping data — which left the team blind exactly when it needed to see. “Sometimes we’d think it was the game’s fault, and the game thought it was our platform’s fault,” Yin says. “It was a chaotic thing, and not very efficient.”
Consolidating on one open platform
CTW had already adopted Grafana Cloud in November 2022 for a consolidated place to visualize their systems, and the decision to further centralize on it came down to a few clear reasons.
Cost was the first; it was far more reasonably priced at CTW’s volume than the alternative. Second, many developers in CTW’s ecosystem already knew Grafana as they were using the dashboard capabilities on a regular basis, so the baseline expertise was there. Finally, it expanded their monitoring capabilities. The team needed to track in-browser game clients and gain deeper application-level insights, prompting them to adopt Grafana Cloud Frontend Observability and distributed tracing.
Now CTW has moved off both ELK and Datadog.
Day to day, the most impactful change was adopting Grafana Cloud Logs, which directly addressed CTW’s biggest pain point. The self-hosted ELK stack had been a major operational burden for the lean team — requiring a handful of SREs to manage storage, scaling, upgrades, and reliability — while leaving significant coverage gaps. By replacing ELK with Grafana Cloud Logs, CTW centralized all log sources (including Kubernetes, VMs, and public clouds) into a single managed platform, accelerating incident troubleshooting from day one.
Alongside their new logging platform, CTW transferred their metrics off Datadog onto Grafana Cloud Metrics and were finally able to provide add distributed tracing to their team via Grafana Cloud Traces. Crucially, by deploying Grafana Alloy agents onto the VM servers, CTW pulled in telemetry from the games that used to keep their logs stranded locally.
Consolidation didn’t just unify the data, it simplified the team’s work.,. “The big difference is the way we work,” Yin says. “We used to have logs and metrics scattered everywhere. Now we have a unified platform, and our work style just changed.” Five places to look became one shared view, turning what used to be a scramble across consoles into a single, predictable workflow.
However, the connection between CTW and Grafana Labs has deepened for reasons beyond just the technical wins of Grafana Cloud.
For Yin, the relationship matters just as much. “With the Grafana Labs team, it’s more like having an old friend,” he says. “We just talk, we just solve things. We like to work with people who love their own product.” And the open philosophy resonates with how CTW sees itself: “We’re a platform facing different game studios. It’s like opening a chocolate box — you never know what the next one will look like. We just embrace the new things. We’re very open to changes.” Or, as Yin puts it more simply: “We love open and, with Grafana’s open agentic observability platform partnered with their open and supportive staff, that’s exactly what we’ve gotten from this partnership.”
Observability as code, provisioned in one click
What reforged the daily output was treating observability as code. CTW’s SRE team codified its monitoring best practices and visualization standards into reusable templates, then wired them into the company’s in-house PaaS, Publisher, that studios utilize to deploy and operate games. . Through the Grafana Cloud API, provisioning observability became just another built-in step in that workflow. “Everything is set up in one click,” Yin says. “The databases, the servers, the CDNs — the entirety of Grafana’s set.” When a game launches or is decommissioned, its full observability stack is created or torn down automatically, replacing the manual, error-prone setup the team used to do by hand. That is what took CTW from days to seconds per game.
Multi-tenancy is enforced where necessary, on the server side, not the frontend. Each game gets its own read tokens, and Grafana Cloud scopes every query to that game’s labels through access policies, meaning a studio can only ever see its own data. Per-game data sources, folder-level permissions, and access tied to CTW’s internal HR and identity systems round out the isolation, so when an external contractor is released from a project, access is revoked automatically.
The payoff shows up most in incidents. Where the team once gathered screenshots from scattered systems, it now works from one shared view. “From the platform side, finding the root cause is really quick now, within minutes,” Yin says. The benefit reaches the studios, too: partners who used to download log files to their laptops to analyze can now watch real-time dashboards directly. “They just use Grafana now,” Yin says. “Their work style completely changed.”
Turning cost spikes into a managed practice
On a platform where every new game multiplies telemetry, it is essential that cost discipline is included. Because CTW can’t dictate what logs or metrics each studio produces, usage can spike two, three, or four times overnight. Adaptive Metrics and Adaptive Logs allow the team to analyze what’s actually used, drop or sample what isn’t, and keep error-level logs intact.
In May 2026, CTW combined Adaptive Metrics with an aggressive source-side collection cleanup and cut its active metric series by about 40% without losing visibility. That headroom allowed the team to reinvest rather than simply slash costs.: It is now piloting Fleet Management to centrally orchestrate its Alloy agents as the fleet grows and has also improved the conversations with studios.
“Some studios don’t even know what they’re shipping; they just want to have it,” Yin says. “Now we can tell them what they actually need.”
Providing observability as a service for all customer studios
On a platform that hosts dozens of studios, a major opportunity is making observability an element they can simply switch on. CTW now offers its capabilities — including Grafana Cloud Frontend Observability — as click-to-apply options inside Publisher, with the configuration already done. A studio can now just opt in and the telemetry starts flowing.
This is essential because most game developers don’t live in observability. “The frontend observability part isn’t emphasized enough by the games,” Yin says. “Some developers from the studios don’t really understand what’s going on there.”
So CTW’s plan is to provide a direct answer rather than the raw data — processing the frontend signals and surfacing the conclusion, or what specifically needs to be addressed , instead of another dashboard to interpret. It is observability democratized to teams that never considered it: studios that once analyzed log files on their own laptops now get ready-to-use insight without needing an observability specialist of their own.
For CTW, that is also a business advantage.
Observability has become part of the platform’s selling point. . “If you deploy games onto our platform, you don’t need an SRE team, you don’t need DevOps,” Yin says. “We can provide everything for you.” Alongside localization, promotion, and hosting, CTW can now offer technical support and a full observability stack as added value — deepening its partnerships with studios and giving them one less thing to build.
What’s next: closing the loop with AI
CTW’s focus for the year ahead is AI, and specifically connecting observability to the rest of the software lifecycle. Developers across the company already work in AI coding tools — like Claude Code and Cursor — and the Grafana Assistant and the Grafana MCP server have changed how they operate. “We integrate the MCP server into our agents, and it can do everything, literally everything,” Yin says. “We use it to analyze logs and metrics, then look into the code and modify the part that might be wrong.”
The goal is to move from AI helping a human investigate, to AI participating in the full cycle: investigation, diagnosis, and remediation with engineers reviewing and approving the results.
In Yin’s view, this is possible not because of a bolted-on assistant, but because of the underlying openness. He highly praises Grafana Assistant for being more precise than the AI in other tools, which often hit a ceiling: they analyze logs and metrics, but stop there. “It has no access to our code,” Yin says. “So the MCP server, really — that’s the full circle of it.” The Grafana MCP server allows CTW to connect any AI agent, using their model of choice, to both its observability data and source code. As a result, an investigation can seamlessly flow from symptom to root cause to a proposed code change without switching between dashboards and coding tools.
The transparency is the focus point. CTW’s teams can lean on Grafana Assistant where its answers are most precise, or wire the MCP server into the model they prefer, and CTW gets to decide which, rather than being locked to one vendor’s AI. For a platform whose business is absorbing whatever stack the next studio brings, an observability layer that bends to its choice of tools and now its choice of AI is the natural fit.
