How Data‑Driven Ops Boosted Our Front‑End Team’s Code Review Efficiency

Background

The front‑end of our data‑platform experienced explosive growth: code submissions doubled to triple year‑over‑year as the business expanded, the team grew from 2‑3 to about 10 members, and the architecture evolved from a simple React+TypeScript stack to a monorepo micro‑frontend approach with Pro‑Code integration.

Why Data‑Driven Operations?

Rapid business growth created many pain points—excessive code‑review cycles, unclear root causes of delays, and fragmented feedback. A data‑driven approach provides a higher‑level view to surface patterns, predict issues, and guide improvements.

Clear Goals

Use real development‑efficiency data to visualize team‑wide and detailed metrics, enabling precise problem identification and targeted efficiency gains.

Full‑Chain Data‑Driven Workflow

1. Design Metrics

Key indicators include code‑quality scores, requirement cycle time, agile iteration delivery, development activity efficiency, and release latency. The article focuses on the "development activity efficiency" metric, i.e., code‑review frequency and handling time.

2. Instrumentation (Data Collection)

We instrumented the workflow to capture fields such as review submission time, review completion time, submitter, reviewer, merge status, round number, and related branch/requirement info.

3. Tool Overview

The DT‑Hornet tool automates data collection via custom NPM commands ( req, cr, pub) and integrates with the corporate ARMS monitoring platform using a Trace‑SDK instance.

/** Efficiency tool core class (monitoring related) */
class HornetCore {
  private static traceInstance: TraceSdkType = null;
  public static initTrace = (): void => {
    HornetCore.traceInstance = new TraceLiteSdk({ pid: '<研发效能工具PID>' });
  };
  public static sendTraceLog = async (type: string, params?: TraceLogParamType): Promise<void> => {
    try {
      const uid = DYNAMIC_NAMES.ADMIN_WORK_ID;
      const c1 = DYNAMIC_NAMES.ADMIN_NAME;
      const c2 = DYNAMIC_NAMES.PROJECT_NAME;
      const c3 = DYNAMIC_NAMES.BRANCH_NAME;
      let c4: any, c5: any, c6: any, c7: any, c8: any;
      const c9 = HornetCore.isLatestVersion.toString();
      const c10 = HornetCore.version;
      switch (type) {
        case TRACE_LOG_TYPE.CR: {
          const { targetBranch, crAdmins, crAdminCnt, currentReqName, crSequence } = params || {};
          [c4, c5, c6, c7, c8] = [targetBranch, crAdmins, crAdminCnt, currentReqName, crSequence];
          break;
        }
        case TRACE_LOG_TYPE.CR_HANDLING: {
          const { targetBranch, currentReqName, crCount, crStartTime, crEndTime } = params || {};
          [c4, c5, c6, c7, c8] = [targetBranch, currentReqName, crCount, crStartTime, crEndTime];
          break;
        }
        default:
          [c4, c5, c6, c7, c8] = [null, null, null, null, null];
      }
      const logParams = { type, uid, c1, c2, c3, c4, c5, c6, c7, c8, c9, c10 };
      return HornetCore.traceInstance.log(logParams);
    } catch (e) {
      BasicMessage.error(`上报数据错误: ${SPLIT_VALUE.BREAK}${e}`);
      return Promise.reject();
    }
  };
}

4. Monitoring Platform Integration

ARMS provides a Trace‑SDK that receives the logged data, stores it in HBase, and makes it queryable via the FBI platform.

5. Data Processing

Real‑time data is synced to offline hourly tables, then periodically copied to a dedicated DataWorks table to avoid the 30‑day retention limit of ARMS. SQL scripts generate detailed, handling, and summary tables for code‑review analysis.

6. Report Design

We built dashboards on FBI showing overall code‑review health (average handling time, frequency), per‑developer rankings, project‑level breakdowns, and drill‑down tables for custom queries.

7. Insight Cases

Overall trend analysis revealed a rise in code‑review volume matching business growth.

Abnormal alerts highlighted projects with excessive review time or frequency, prompting targeted investigations.

Project‑level efficiency comparisons guided resource allocation and process adjustments.

Thinking: How to Leverage Data‑Driven Ops

Data‑driven ops help validate design decisions, avoid over‑reliance on raw data, and continuously improve R&D efficiency. By automating information sync, integrating engineering pipelines, and adding real‑time alerts, the workflow becomes largely unattended.