Skip to main content

Using AI active stack analysis

Digging into the active stack when a transaction slows down is time-consuming even for experienced developers. It involves reading multiple stack dumps, finding common patterns, and judging severity. WhaTap's AI active stack analysis (Beta) automates this so that users without stack analysis experience can immediately grasp the cause and response direction.

Prerequisites​

  • Project member or higher permission
  • A transaction to analyze has been collected (visible on the hitmap or transaction list)

How to use it​

Path: Transaction trace > Active stack tab > Active stack analysis button

  1. Select the transaction to analyze from the hitmap or transaction list.
  2. In the transaction detail, select the Active stack tab.
  3. Click the Active stack analysis button at the top.
  4. Review the AI analysis in four sections.
  5. Click the No.N stack number in each section to view the original stack.

Analysis result structure​

AI analyzes the collected stack dumps and presents the result in four sections.

Table | AI analysis result structure
SectionContents
Overall summaryNumber of stacks analyzed and a summary of the overall execution flow
Common patternsRecurring execution patterns (pattern name, count, description)
Detected issuesPerformance problems + severity (High/Medium/Low) + the evidence stacks
RecommendationsPrioritized, concrete improvement directions

Severity criteria​

Table | Severity criteria
SeverityCriteria
HighDeadlock, full blocking, or the same blocking recurring 4+ times
MediumI/O waits, 2–3 lock contentions, slow external calls
LowSingle-occurrence waits, light optimization opportunities

High issues like deadlocks or full blocking take time to spot manually; AI flags them automatically with severity.

When to use it​

During incident response (most useful)​

Shorten trace/stack interpretation time in step ③ Root cause analysis of the Incident response scenario.

  1. Pick a slow transaction on the hitmap and enter the trace.
  2. On the Active stack tab, click the Active stack analysis button.
  3. AI presents results (seconds to tens of seconds).
  4. Start from High issues, then drill down to the original stack.

Release verification​

After a deploy, run AI analysis on traces during a response-time spike. You can quickly tell whether the regression is a code-level bottleneck, I/O, or lock contention. Apply this in step ② Acute regression detection of the Release verification scenario.

Periodic performance tuning​

Scan for latent bottlenecks (Medium/Low) that don't show up in normal times, quarter by quarter, and feed them into the improvement backlog. Fold the results into the quarterly retro of the Performance reporting scenario.

Understand the limits of AI analysis​

AI output is a first hypothesis, not the final conclusion. Use it this way:

  • For High issues AI surfaces → cross-check against the original stack before acting
  • Business context (recent deploys, traffic events) is not visible to AI — combine with timeline info
  • Repeated analysis may not be fully consistent (LLM nature) — validate with observable metrics (response time, error rate)

Adjacent AI analysis features​

WhaTap provides several AI analysis features beyond "stack." Combining them as needed amplifies the effect.

Table | Adjacent AI analysis features
FeatureWhen
AI active stack analysis (this guide)Identify bottlenecks in a transaction trace
AI thread dump analysisInterpret thread blocking and lock contention in Instance Performance Management
AI SQL tuning guideIdentify inefficient queries and get execution-plan based improvements — complements DB realtime visibility
AI browser error stack analysisCode-level cause for frontend errors — complements RUM
WhaTap AI Chatbot / MCPNatural-language queries across WhaTap data — MCP integration

Next steps​

Last updated on