Sentry的异常数据处理流程

Posted 2021-09-14 十一点我就睡

Sentry的异常数据处理流程

获取异常

在 Sentry的前端异常监控方案中之前我们说过，Sentry的全局异常获取方式有2种，window.onerror以及unhandledrejection。

异常数据的处理

以unhandledrejection为例 globalhandlers.ts中

    addInstrumentationHandler({
      // eslint-disable-next-line @typescript-eslint/no-explicit-any
      callback: (e: any) => {
        let error = e;

        // dig the object of the rejection out of known event types
        try {
          // PromiseRejectionEvents store the object of the rejection under \'reason\'
          // see https://developer.mozilla.org/en-US/docs/Web/API/PromiseRejectionEvent
          if (\'reason\' in e) {
            error = e.reason;
          }
          // something, somewhere, (likely a browser extension) effectively casts PromiseRejectionEvents
          // to CustomEvents, moving the `promise` and `reason` attributes of the PRE into
          // the CustomEvent\'s `detail` attribute, since they\'re not part of CustomEvent\'s spec
          // see https://developer.mozilla.org/en-US/docs/Web/API/CustomEvent and
          // https://github.com/getsentry/sentry-javascript/issues/2380
          else if (\'detail\' in e && \'reason\' in e.detail) {
            error = e.detail.reason;
          }
        } catch (_oO) {
          // no-empty
        }

        const currentHub = getCurrentHub();
        const hasIntegration = currentHub.getIntegration(GlobalHandlers);
        const isFailedOwnDelivery = error && error.__sentry_own_request__ === true;
        // addEventListener的事件中直接throw Error的时候,shouldIgnoreOnError会为true.即不会上报
        // 原因是在instrumentDOM方法中，用globalDOMEventHandler方法对监听事件包了一层,使得ignoreOnError>0
        if (!hasIntegration || shouldIgnoreOnError() || isFailedOwnDelivery) {
          return true;
        }

        const client = currentHub.getClient();
        const event = isPrimitive(error)
          ? this._eventFromRejectionWithPrimitive(error)
          : eventFromUnknownInput(error, undefined, {
              attachStacktrace: client && client.getOptions().attachStacktrace,
              rejection: true,
            });

        event.level = Severity.Error;

        addExceptionMechanism(event, {
          handled: false,
          type: \'onunhandledrejection\',
        });

        currentHub.captureEvent(event, {
          originalException: error,
        });

        return;
      },
      type: \'unhandledrejection\',
    });

 

整体流程分为以下几个模块讲解

兼容处理

        try {
          // PromiseRejectionEvents store the object of the rejection under \'reason\'
          // see https://developer.mozilla.org/en-US/docs/Web/API/PromiseRejectionEvent
          if (\'reason\' in e) {
            error = e.reason;
          }
          // something, somewhere, (likely a browser extension) effectively casts PromiseRejectionEvents
          // to CustomEvents, moving the `promise` and `reason` attributes of the PRE into
          // the CustomEvent\'s `detail` attribute, since they\'re not part of CustomEvent\'s spec
          // see https://developer.mozilla.org/en-US/docs/Web/API/CustomEvent and
          // https://github.com/getsentry/sentry-javascript/issues/2380
          else if (\'detail\' in e && \'reason\' in e.detail) {
            error = e.detail.reason;
          }
        } catch (_oO) {
          // no-empty
        }

这一块都是在做兼容性处理，适配不同浏览器中的错误类型

过滤条件

          const currentHub = getCurrentHub();
        const hasIntegration = currentHub.getIntegration(GlobalHandlers);
        const isFailedOwnDelivery = error && error.__sentry_own_request__ === true;

        if (!hasIntegration || shouldIgnoreOnError() || isFailedOwnDelivery) {
          return true;
        }

这里的hasIntergration是在判断是否安装了对应的全局监听函数的插件(初始化的时候就会默认加载)
shouldIgnoreOnError : addEventListener的事件中直接throw Error的时候,shouldIgnoreOnError会为true.即不会上报
原因是在instrumentDOM方法中，用globalDOMEventHandler方法对监听事件包了一层,使得ignoreOnError>0
isFailedOwnDelivery即判断是否为Sentry自身的请求错误，若是，则不上报。

核心处理逻辑

        const client = currentHub.getClient();
        const event = isPrimitive(error)
          ? this._eventFromRejectionWithPrimitive(error)
          : eventFromUnknownInput(error, undefined, {
              attachStacktrace: client && client.getOptions().attachStacktrace,
              rejection: true,
            });

export function isPrimitive(wat: any): wat is Primitive {
  return wat === null || (typeof wat !== \'object\' && typeof wat !== \'function\');
}

isPrimitive(error)即在判断error是否为原始数据类型
若为原始数据类型，则处理比较简单

  private _eventFromRejectionWithPrimitive(reason: Primitive): Event {
    return {
      exception: {
        values: [
          {
            type: \'UnhandledRejection\',
            // String() is needed because the Primitive type includes symbols (which can\'t be automatically stringified)
            value: `Non-Error promise rejection captured with value: ${String(reason)}`,
          },
        ],
      },
    };
  }

若为引用数据类型，核心处理函数在tracekit.ts为

    stack = computeStackTraceFromStacktraceProp(ex);
    if (stack) {
      return popFrames(stack, popSize);
    }

具体就在computeStackTraceFromStacktraceProp，popFrames这2个函数中。

function computeStackTraceFromStackProp(ex: any): StackTrace | null {
  if (!ex || !ex.stack) {
    return null;
  }

  const stack = [];
  const lines = ex.stack.split(\'\\n\');
  let isEval;
  let submatch;
  let parts;
  let element;

  for (let i = 0; i < lines.length; ++i) {
    if ((parts = chrome.exec(lines[i]))) {
      const isNative = parts[2] && parts[2].indexOf(\'native\') === 0; // start of line
      isEval = parts[2] && parts[2].indexOf(\'eval\') === 0; // start of line
      if (isEval && (submatch = chromeEval.exec(parts[2]))) {
        // throw out eval line/column and use top-most line/column number
        parts[2] = submatch[1]; // url
        parts[3] = submatch[2]; // line
        parts[4] = submatch[3]; // column
      }
      element = {
        // working with the regexp above is super painful. it is quite a hack, but just stripping the `address at `
        // prefix here seems like the quickest solution for now.
        url: parts[2] && parts[2].indexOf(\'address at \') === 0 ? parts[2].substr(\'address at \'.length) : parts[2],
        func: parts[1] || UNKNOWN_FUNCTION,
        args: isNative ? [parts[2]] : [],
        line: parts[3] ? +parts[3] : null,
        column: parts[4] ? +parts[4] : null,
      };
    } else if ((parts = winjs.exec(lines[i]))) {
      element = {
        url: parts[2],
        func: parts[1] || UNKNOWN_FUNCTION,
        args: [],
        line: +parts[3],
        column: parts[4] ? +parts[4] : null,
      };
    } else if ((parts = gecko.exec(lines[i]))) {
      isEval = parts[3] && parts[3].indexOf(\' > eval\') > -1;
      if (isEval && (submatch = geckoEval.exec(parts[3]))) {
        // throw out eval line/column and use top-most line number
        parts[1] = parts[1] || `eval`;
        parts[3] = submatch[1];
        parts[4] = submatch[2];
        parts[5] = \'\'; // no column when eval
      } else if (i === 0 && !parts[5] && ex.columnNumber !== void 0) {
        // FireFox uses this awesome columnNumber property for its top frame
        // Also note, Firefox\'s column number is 0-based and everything else expects 1-based,
        // so adding 1
        // NOTE: this hack doesn\'t work if top-most frame is eval
        stack[0].column = (ex.columnNumber as number) + 1;
      }
      element = {
        url: parts[3],
        func: parts[1] || UNKNOWN_FUNCTION,
        args: parts[2] ? parts[2].split(\',\') : [],
        line: parts[4] ? +parts[4] : null,
        column: parts[5] ? +parts[5] : null,
      };
    } else {
      continue;
    }

    if (!element.func && element.line) {
      element.func = UNKNOWN_FUNCTION;
    }

    stack.push(element);
  }

  if (!stack.length) {
    return null;
  }

  return {
    message: extractMessage(ex),
    name: ex.name,
    stack,
  };
}

可以看出，核心流程其实就是在对不同的浏览器做兼容处理，以及将数据组合成堆栈的形式

后续

将异常数据进行捕获以及处理之后，就是上报流程了。详情可以参看上一篇文档 Sentry的异常数据上报机制

        currentHub.captureEvent(event, {
          originalException: error,
        });

总结

在大致了解到了整个异常数据处理流程后，发现重点是在对于不同浏览器的兼容性处理上，整个链路比较清晰，中间很多繁琐的处理细节也都没有太多的去关注，毕竟时间和精力相对有限，想的是在对Sentry的整体脉络有了一定了解之后，对自己做监控能有很多启发，当然这也是研究Sentry源码的初衷。