It is 10 pm, a customer sees a wrong amount, the logs say nothing, and the one variable that explains it is never written anywhere. Reproducing locally takes hours; redeploying with one more log line takes just as long. Dynatrace’s Live Debugger targets exactly this case: read the state of a line of code inside the running application, without stopping it or redeploying it.
This article starts from an operations decision: should developers be allowed to set breakpoints in production? The answer depends less on the tool than on the guardrails. The facts come from the Dynatrace SaaS documentation, read on 12 November 2026, and from the vendor’s announcement. I have not tested the feature on a real environment.
Scope: the documentation describes a Dynatrace SaaS environment with a DPS subscription. I found no equivalent page for Dynatrace Managed, so I do not describe the feature for Managed.
The problem: one more log line means one more deployment
When information is missing, the reflex is to add a log line. That means changing the code, running CI, deploying, waiting for the bug to recur, and starting over if the line was not enough. Every round lengthens MTTR and exposes production to yet another change.
A non-breaking breakpoint reverses the order. You pick a line, the agent collects the application state each time that line runs, and the process keeps serving traffic. You get a snapshot holding the stack trace, local variables, process information and tracing data.
What the documentation guarantees, and what it leaves out
The official page sets precise prerequisites:
- a Dynatrace SaaS environment with a DPS subscription that includes the Code Monitoring capability;
- OneAgent 1.309 or later, with the « Java Live-Debugger » and/or « Node.js Live-Debugger » features turned on;
- Full-Stack, Infrastructure or Discovery monitoring mode (with container code-module injection);
- the Java and Node.js code modules, the only ones listed.
Billing goes through Code Monitoring, measured in container-hours, including for non-containerised workloads. Prices are not on the page, so I quote none. The vendor announcement also mentions collecting data inside third-party libraries and viewing snapshots from VS Code or IntelliJ. I could not confirm those points on the documentation pages I read, so I do not treat them as established.
Live Debugger: turn it on without opening everything at once
Activation has three layers, from broadest to finest.
- In Settings, under Collect and capture, General monitoring settings, OneAgent features, turn on Java Live-Debugger, Node.js Live-Debugger or both, then restart the affected processes.
- Enable Observability for Developers. The scope runs from the whole environment down to a process group, a Kubernetes cluster, a namespace, a host group or a deployment.
- Optional but recommended: enable the debugging module on an ActiveGate (version 1.311 or later). It is not mandatory, but it centralises traffic, especially on Kubernetes.
With the Kubernetes operator, the ActiveGate gets the debugging capability in the DynaKube:
My operating advice: start with one pre-production namespace or a single process group, never with « Enable for entire tenant ». Widening later costs one click; rebuilding trust after an incident costs far more.
Setting a breakpoint that does no harm
In the Live Debugger app, you filter the instances to target, load the source code (local or from a Git provider), then click the gutter next to the line number. Three settings decide the production impact.
Default limits
Each breakpoint has a time limit (7 days by default), a hit limit (100 by default) and a collection level (Medium by default). When a limit is reached, it moves to Disabled. You re-enable it with a right-click, Enable, which resets the counter.
Collection level
Three levels set the object depth, the string width and the number of elements collected per list. On the JVM, Low means depth 2 (128 characters, 10 elements), Medium 5 (512, 15) and High 8 (4,096, 25). On Node.js the values are 2/128/5, 3/512/10 and 5/4,096/20. Start with Low: a light snapshot often answers the question.
Conditions
A condition restricts collection to the cases you care about, such as a single customer or an abnormal amount. The documentation gives these advanced expression examples:
The documented operators are ==, !=, >, >=, <, <=, in, &&, || and parentheses. For the size of a list, use size(), not len or length.
Understanding why no snapshot arrives
A breakpoint has five states: Active, Pending, Disabled, Error and Warning. The common cases from the documentation:
- Active but no snapshot: the line is not executed in the right instance, or the function was already running when you set the breakpoint. Only later calls are covered.
- Pending: OneAgent missing, application outside the filter, source file not loaded, or a line with no executable code. On the JVM, classes compiled without debug information also block it.
- Error: the source file seen in the IDE differs from the deployed one, or collection took too long. Dynatrace checks the file version before it sets the breakpoint.
- Warning: a rate-limiting mechanism samples collection on a very hot path to protect performance.
On Node.js, transpiled code without source maps or a concurrent debugger such as WebStorm’s also causes trouble. The status message, via right-click then Status, gives the exact cause.
Who may set a breakpoint: permissions
The real answer to the opening decision sits in IAM policies. Two permissions exist: dev-obs:breakpoints:set to see agents and place breakpoints, and dev-obs:breakpoints:manage to view and disable those of every user. Conditions let you restrict by namespace, cluster, process group or host. Reading snapshots goes through a separate storage permission.
Separating « set » from « read » helps: a snapshot may contain sensitive variables. The vendor announcement mentions masking and retention controls. Check their actual configuration in the documentation before opening anything in production.
Best practices
- Reserve
breakpoints:setfor a named team and limit it to one namespace or process group. - Keep the default limits or lower them: 100 hits almost always suffice.
- Add a condition whenever traffic is high, rather than relying on rate limiting.
- Record in the incident ticket which breakpoint was set, by whom and why.
- Pair the snapshot with a trace: it answers « which value », the trace answers « which service ». For CPU use, see continuous profiling, a separate third-party tool.
To tie this work to service health, first check pod state in the CrashLoopBackOff and OOMKilled diagnosis, then follow the effect of your fix in a DQL dashboard.
Frequently asked questions
Does Live Debugger stop the application?
No. The documentation describes non-breaking breakpoints: state is collected without stopping the running code. A rate limiter protects heavily used paths.
Which languages are supported?
The prerequisites page lists the Java and Node.js code modules. Nothing else is documented there: check before planning a rollout for another language.
Does it work on Dynatrace Managed?
The documentation I read describes a SaaS environment with DPS. I found no Managed page: treat the feature as unavailable on Managed until the vendor documents it.
Conclusion
Live Debugger replaces the « add a log, deploy, wait » cycle with a targeted, time-boxed read. Its value in production depends on the guardrails: a narrow activation scope, IAM permissions, limits and conditions. To go further, read the Live Debugger documentation.
span.observer is an independent community blog, not affiliated with Dynatrace LLC. Trademarks belong to their owners. Features change quickly: check prerequisites and limits in the official documentation.
Originally published on span.observer. French version: https://span.observer/fr/live-debugger-deboguer-production-sans-redeployer/.



Top comments (0)