Отладка production в реальном времени

What is live production debugging?

Live production debugging lets engineering teams inspect running application code—variable values, stack frames, branch decisions, and request context—without redeploying, restarting services, or adding permanent log lines. Tools in this category inject read-only instrumentation (often called probes, snapshots, or non-breaking breakpoints) that fire on real traffic and disappear after capture.

This tag is narrower than full observability suites (Datadog, Dynatrace as a whole) and broader than local IDE debuggers. It covers products whose primary job is runtime evidence from live environments, especially when logs and traces are green but the business outcome is wrong.

Problems these tools solve

• Silent failures — HTTP 200 with wrong data; no exception in logs.
• Missing variable state — stack traces point far from where the bad value was set.
• Slow redeploy loops — add log → CI/CD → wait → still not enough data.
• Hard-to-reproduce prod bugs — race conditions, third-party contract drift, idempotency edge cases.

Representative products include Lightrun, HyperProbe, and live-debugger capabilities descended from Rookout (now part of Dynatrace Live Debugger). Datadog Live Debugger and similar features inside APM platforms share the same user task but are usually listed under their parent platform unless the directory entry is specifically about the live-debug feature.

How live production debugging works

1. An in-process agent or SDK attaches to your service.
2. You or an AI agent places a probe / logpoint / snapshot on a suspect line.
3. The next matching request captures runtime state without pausing threads.
4. Evidence feeds root cause analysis; probes expire via TTL, rate limits, or hit caps.

Many newer tools expose MCP servers so coding agents in Cursor or Claude Code can place probes and read captures from the IDE.

How to compare tools in this category

Safety and production impact

Check read-only guarantees, non-blocking hooks, PII redaction, audit trails, approval gates, and documented overhead (many vendors cite under ~1% at moderate RPS on reference workloads).

Deployment model

Some tools default to managed cloud (HyperProbe Free, Lightrun SaaS); others emphasize self-hosted or VPC for regulated environments. Match this to your security review, not marketing badges alone.

Language and platform support

Verify runtimes (Node.js, Java, Python, Ruby, .NET, Go, etc.) and orchestration (Kubernetes operators, serverless constraints).

AI and IDE integration

If you use coding agents, compare MCP coverage, hypothesis-driven investigation flows, and whether RCA is evidence-backed vs log-only guessing.

Frequently asked questions

Is live production debugging the same as APM?

No. APM focuses on metrics, traces, and logs at scale. Live production debugging adds on-demand code-level snapshots when those signals are insufficient.

Do I need to redeploy to debug production?

Not with tools in this tag. Instrumentation is injected in-process; avoid tools that still require a new build for each debug log line.

Is it safe for production traffic?

Vendors design for non-blocking, read-only capture with expiry controls—but your team should still set TTL, rate limits, approval workflows, and redaction rules.

Which tools belong in this tag?

Standalone products like Lightrun and HyperProbe, and directory entries specifically about live debugger features—not entire APM platforms unless scoped to that capability.