The monitoring and alerting feature can be used to monitor metrics in charts, providing real-time monitoring at the hour, day, and month granularity. Users can configure monitoring rules and monitoring frequency to set up monitoring.
The alerting methods currently supported by the system include email, Feishu, WeCom, Webex, DingTalk, and Webhook. To use this module, an enterprise domain administrator must complete the integration configuration for the office platform. For detailed steps, see Office Platform Integration.
To view and manage the monitoring configurations that have already been set up in the current project, see Monitoring Management @Monitoring Management
Common application scenarios

• Detecting changes in dimension items: For example, if a company has developed offline stores in multiple locations and needs to monitor changes in city coverage based on the list of cities and stores, dimension value change monitoring can be selected, with the monitoring dimension set to "city". When a new city appears in the list or a city disappears, a corresponding alert notification will be received.
• Monitoring abnormal metric fluctuations: For example, to monitor fluctuations in the company's sales revenue, when the fluctuation amount rises or falls beyond the expected value, an alert notification is sent so that relevant analysts can promptly investigate the cause.
• Intelligent fluctuation detection (supported only for line charts): For example, to monitor daily profit changes and use a time series anomaly detection algorithm to check for abnormal fluctuations in the data. If abnormal fluctuations are detected, an alert will be sent.
  

2.1 Offline data monitoring #

Step 1 Before configuring monitoring, the following prerequisites must be met:

• The base table must be updated regularly.
• The date filter must be set to 'recent' or 'recent with data'.
• Save the chart.
  

Step 2 Click 'Create Monitoring' to open the settings dialog.

Step 3 Enter the monitoring name, select the monitoring time range and trigger factors, and configure the alerting rules, monitoring time, and notification methods.

• Caution when selecting the time range. For example, in the screenshot below, if the time range is set to the selected payment date range, the monitored data will include all rows currently displayed in the chart. If the time range is set to the most recent time point, only the row with the payment date of 2022-05-30 will be monitored.

Step 4 Click "Send Test" to view the test information in the configured monitoring receiver application and adjust the parameters.

Step 5 Click Confirm to successfully create a new monitoring task. For more historical monitoring, click [Monitoring List] to view them in "Monitoring Management" under the project center.

2.2 Real-time data monitoring #

The following prerequisites must be met:

• The base table must be real-time data from Kafka.
• The time partition field must be aggregated.
• The date filter must be set to recent/recent with data.
• Save the chart.

Other operational steps are the same as above.

3.1 Time range #

3.1.1 Date dimension #

The selectable dates for the time range are related to the date dimensions in the dimension bar of the monitoring chart. For example, if the chart dimensions in the screenshot select p_date and payment date, then in the monitoring time range options, you can choose the range for these two date dimensions.

• Date dimension range

Using the payment date in the screenshot below as an example:
(1) Selecting the most recent time point means only monitoring the data for the latest payment date, that is, the data in the row for 2022-01-15.
(2) Selecting the most recent N time points and setting N to 5 means monitoring the data for the five rows from 2022-01-11 to 2022-01-15.
(3) Selecting the chosen time range means monitoring all data retrieved under the current payment date filter conditions.

3.1.2 Most recent time point #

The range for the most recent one or n time points depends on the date type of the time field, supporting 1/5/15 minutes, year-month-day-hour, year-month-day, year-week, and year-month.
For example, if the date type of the time field is set to "year-month-day-hour-minute" with a granularity of "15 minutes", and the condition is set to "most recent 1 time point", then the next time monitoring is triggered, the current data will be compared with the data from 15 minutes earlier.

If the date type is another type, such as seconds, biweek, bimonth, and so on, the following rules apply:

• If the date type is datetime: seconds, 10 minutes, 20 minutes, and so on, then the range for the most recent 1 time point is 1 second
• If the date type is date: biweek, bimonth, quarter, and so on, then the range for the most recent 1 time point is 1 day

3.2 Trigger factors #

3.2.1 Indicator value changes #

Indicator value change monitoring supports configuring multiple alarm rules. You can choose to trigger an alarm when any rule or all rules are met. Indicator conditions are required, while dimension conditions are optional.

• All rules / Any rule: Multiple alarm rules can be added to the monitoring configuration at the same time.
  When all rules are selected, an alarm will be triggered only when all rule conditions are met simultaneously. For example, as shown in the figure below, an alarm is triggered when the profit for the most recent day fluctuates by more than 5% compared to the previous day and the cost fluctuates by more than 10% compared to the previous day.

When any rule is selected, an alarm will be triggered when any one of the rules is met. For example, as shown in the figure below, an alarm is triggered when either the profit for the most recent day fluctuates by more than 5% compared to the previous day or the cost fluctuates by more than 10% compared to the previous day.

• Dimension filtering
  Dimension filtering can be configured for each rule, allowing you to monitor specified dimension items in the chart.
  For example, as shown in the screenshot below:
  When the OR condition is selected: data that meets any of the city or province filters will be monitored.
  When the AND condition is selected: data that meets both the city filter and the province filter will be monitored.
• Criteria for indicator monitoring
  Each rule supports configuration for metric filtering, allowing you to monitor a single metric or multiple metrics, or set multiple conditions for the same metric.
  For example, see the screenshot below:
  When any condition is selected: An alert is triggered if the profit for the most recent day meets any one of the following conditions—fluctuates by more than 5% compared to the previous day or the original value is greater than 4,000,000.
  When all conditions are selected: An alert is triggered only if the profit for the most recent day simultaneously meets both conditions—fluctuates by more than 5% compared to the previous day and the original value is greater than 4,000,000.
• Metric monitoring condition configuration
  All data / Any data:
  The default is any data, meaning an alert is triggered immediately when any data meets the condition. When all data is set, an alert is triggered only when all data under the specified dimension conditions meet the criteria.
  Metric selection:
  The metrics placed in the metric bar on the chart page are optional and support comparison and period-over-period derived fields.
  Comparison range:
  Supports comparison with 1 day ago, 7 days ago, 30 days ago, absolute value, original value, previous time point; comparison with the average of the previous 7 days or 30 days (these two options are available only when the first dimension in the chart's dimension bar is a partition field).
  Comparison method:
  Ratio type: fluctuation exceeds, increase exceeds, decrease exceeds
  Absolute value / original value: >, >=, <, <=, =
  Comparison criteria:
  Value: You can enter a fixed value.
  %: You can enter a percentage value.
  Metric: Supports comparison between the currently monitored metric and other metrics (including comparison and period-over-period derived fields). For example, as shown in the screenshot, an alert is triggered when the original value of profit < the original value of cost.

3.2.2 Dimension value changes #

Dimension value change monitoring can monitor a specific dimension value in the current chart. An alert is triggered when the enumeration value of the dimension for the current day changes compared to 1 day ago, 7 days ago, or 30 days ago.

• Monitored dimensions
  The dimensions that can be monitored are the dimension items of the current chart (monitoring of hidden dimensions is supported).
• Comparison conditions
  Supports comparison with 1 day ago, 7 days ago, and 30 days ago.

3.2.3 Intelligent fluctuation detection #

Intelligent fluctuation detection monitors indicator fluctuations over a period of time using the S-H-ESD algorithm for time series anomaly detection. An alert is triggered if an abnormal fluctuation is detected.

• Monitoring indicators

The selectable monitoring indicators are those in the current chart's indicator bar. Monitoring is supported for hidden fields and derived fields from analysis features, such as year-over-year and month-over-month fields.

3.3 Monitoring schedule #

3.3.1 Scheduling #

• Daily: Monitoring is triggered at the specified time point each day.
  • The default is a time picker for quick selection of 24-hour time points. After selecting a custom time, minute-level precision is supported. The minimum monitoring interval is 10 minutes.
  • UTC time zone selection is supported.
• Hourly: Monitoring is triggered every N hours within the selected time range.
  • Monitoring can be triggered every 1, 2, 3, 4, 6, 8, or 12 hours.
  • UTC time zone selection is supported.
• Minute: Monitoring is triggered every N minutes within the selected time range.
  • Monitoring can be triggered every 10, 15, 20, or 30 minutes.
  • UTC time zone selection is supported.
• Weekly: Monitoring can be triggered at a fixed time point on a specified day of the week.
  • The weekly range supports selection from Monday to Sunday.
  • It supports triggering monitoring at N hours and N minutes on the selected day. The default is a time picker for quick selection of 24-hour time points. After selecting a custom time, minute-level precision is supported. The minimum monitoring interval is 10 minutes.
  • UTC time zone selection is supported.
• Monthly: Monitoring can be triggered at a specified time point on the Nth day of each month.
  • The monthly range supports triggering monitoring on the 1st to 31st day of each month. If the 31st is selected and a month does not have 31 days, monitoring will not be triggered.
  • It supports triggering monitoring at N hours and N minutes on the selected day. The default is a time picker for quick selection of 24-hour time points. After selecting a custom time, minute-level precision is supported. The minimum monitoring interval is 10 minutes.
  • UTC time zone selection is supported.

Custom time: Monitoring can be triggered on working days, trading days, or every day.

3.3.2 Triggered upon data synchronization completion #

When the chart's dataset is synchronized on a daily basis, monitoring can be configured to trigger upon completion of data synchronization.

• Triggered when data synchronization is complete; supports UTC time zone selection
  • This option is only visible for charts with a daily update frequency (that is, datasets with a daily synchronization cycle or automatic synchronization). After configuration, triggering monitoring requires:
    • Completion of synchronization for the T-N time partition data
    • Reaching the earliest trigger time
• Example: The subscription trigger date is 2020-01-02, the trigger date is set to T-1, and the earliest trigger time is 10:00
  • If the dataset completes synchronization at 9:00 on 2020-01-01, the monitoring notification will wait until 10:00 to be sent
  • If the dataset completes synchronization at 11:00 on 2020-01-01, the monitoring notification will be delayed until 11:00

3.4 Notification methods #

The product currently supports the following alert notification methods: email, Feishu, WeCom, Webex, DingTalk, and Webhook. The specific alert notification method depends on the office platform integration configuration completed by your company's system administrator. If no notification methods are available here, please refer to Office Platform Integration for configuration. For IM group notifications, multiple group IDs can be added, separated by commas, and each group must have a bot. To send monitoring results to Feishu via Webhook, you need to parse and construct the request using a standardized output script.
For Webhook output, when the data meets a user-configured condition, an HTTP request will be sent to the specified recipient (Webhook address). The following provides the specification for the output script for your reference.

Alert message #

{
  "secret": "xxx",//Key used for authentication by the business party
  "event": {//Message content
    "channel": "aeolus_monitor",//Used to distinguish monitoring messages from other types of messages
    "timestamp": 1610941345327,//Timestamp of this message
    "title": "monitor_alarm",//monitor_alarm: alarm message for monitoring
    "payload": {
      "status": 1,//0: alarm triggered, 1: alarm not triggered

      "alarm_rule": [//Specific alarm content and alarm conditions; this field is empty when the alarm is not triggered
        {
          "metric_monitor_condition": [//Constraint conditions for metrics
            "指标筛选条件满足任一条件:",
            "「sum(is_active_tea) 规则：>0.0」",
            "「sum(is_active_aeolus) 规则：>0.0」"
          ],
          "dimension_monitor_condition": [//Constraint conditions for dimensions
            "维度筛选条件满足所有条件",
            "tea_app_name in [Lantern]"
          ],
          "data_link": "https://...."//Link to the file containing the complete alarm content
        },
        ......
      ],
      "report_link": "'https://data.bytedance.net/aeolus/#/dataQuery?appId=555371&rid=39452'",//Chart link
      "date_range": {//Time range for chart queries
        "p_date_start": "2021-10-07",//Business start date
        "p_date_end": "2021-10-07"//Business end date
      },
      "suggestion": "通知管理员",//Handling suggestions
      "monitor_name": "监控_1008_5"//Monitoring name
    }
  }
}

Edit monitoring notification #

{
  "secret": "xxx", //Key used for authentication by the business party
  "event": {//Message content
    "channel": "aeolus_monitor", //Used to distinguish monitoring messages from other types of messages
    "timestamp": 1610941345327, //Timestamp of this message
    "title": "monitor_creation", //Create monitoring: monitor_creation, update monitoring: monitor_update, delete monitoring: monitor_delete,
    "payload": {
      "monitor_status": 1,//Monitoring status. 0: enabled, 1: paused'
      "name": "监控_1008_3", //Monitoring name
      "hour": "10:00",//Time when monitoring is activated
      "trigger_type": 0,//Trigger type. 0: scheduled, 1: base table update
      "notified": "{\"lark\":\"duxingwang.001\"}",//Feishu users to be notified
      "compute_type": 0,//Trigger factors, 0: "dimension + metric changes", 1: "dimension value changes", 2: "fluctuation anomaly detection"
      "extra": {//Additional information, currently stores alarm suggestions and whether the base table is kafka. json format'
        "external_header": "",//Can be ignored
        "need_confirm": false,//Whether alarm confirmation is required
        "show_condition": true,//Whether to display monitoring conditions in the alarm
        "suggestion": "通知管理员",//Handling suggestions
        "timezone": "UTC+08:00",//Time zone
        "urgent_alarm": {
          "emails": "",
          "times": ""
        }
      },
      "shield_duration": 0,//Mute time span, in hours
      "lark_group_id": "{}",//Feishu group IDs to notify, comma-separated string
      "notice_type": -1,//Push method
      "frequency": 0,//Scheduled frequency. 0: daily, 1: hourly
      "date_filter_conf": {},
      "rule_conf": {//Rule configuration, json format
        "rule_list": [//Rule list
          {
            "metric": {//Metric alarm conditions
              "rule_list": [//List of metrics to be satisfied
                {
                  "dim_id": "sum_1586869275872",
                  "op_unit": 0,
                  "op_type": ">",
                  "compare_type": 3,
                  "op_value": 0
                }
              ],
              "op": 1//1: logical AND, 0: logical OR
            },
            "dimension": {//Dimension alarm conditions
              "where_list": [
                {
                  "option": {
                    "is_report_filter": false,
                    "filter_pattern": "Accurate",
                    "custom_list": []
                  },
                  "val": [
                    "EasyShoot,galaxy"
                  ],
                  "pre_relation": "and",
                  "val_option": {
                    "true_all_selected": false,
                    "preview_val": [],
                    "excluded_val": []
                  },
                  "id": "1586869275867",
                  "unique_id": 211008185453007,
                  "op": "in"
                }
              ],
              "op": 1//1: logical AND, 0: logical OR
            }
          }
        ],
        "op": 1//1: logical AND, 0: logical OR
      },
      "data_ready": {},
      "owner_email_prefix": "duxingwang.001",//Creator
      "webhooks": [//Configured webhook id
        14
      ],
      "shield_start_time": "",//Mute alarm start time
      "domain_id": 1,
      "report_id": 39452//Chart id
    }
  }
}

data_link format #

["tea_app_name", "is_active_tea"]//Column names of the chart
["1", "2"]//The first chart record that triggered the alert
["3", "5"] //The second chart record that triggered the alert
......

3.5 Handling suggestions #

You can add text and URLs.

3.6 Alert results #

• Display monitoring conditions in the alert
  • After selecting this option, the alert card will include the monitoring conditions configured for monitoring

3.7 Send test #

Immediately send a notification message for this configuration condition to preview the monitoring effect (test messages will not be sent to groups)

3.8 Monitoring list management #

Access the monitoring list:
You can access the monitoring list through the dropdown menu of the visual monitoring page.

Alternatively, click Project Center - Scheduled Tasks - Monitoring Management to enter.

• Roles that can view and edit monitoring

Users who can view and edit monitoring: monitoring creator, project administrator, owner of the chart associated with the monitoring, administrator of the dashboard where the chart associated with the monitoring is located.
To add collaborators for monitoring of charts within a dashboard, grant dashboard management permissions to others; this operation is not supported for personal dashboards.

• Monitoring and subscriptions support batch 'transfer ownership'
  
  Note: If the monitoring recipient is empty, the monitoring will not take effect

4.1 DAU growth rate fluctuation monitoring #

Data: DAU status of each app on both Android and iOS platforms
Monitoring scenario: monitor products with a scale of over 10 million where the year-over-year growth for the month exceeds 10 percent
Configuration parameters:

If any app meets the criteria, an alert will be triggered:

4.2 Monitoring changes in app store ranking #

Data: China Apple App Store Ranking - Overall Ranking
Monitoring scenario: monitor daily changes in the TOP 10 apps
Configuration parameters:

If any app meets the criteria, an alert will be triggered:

4.3 City profit monitoring, alert if less than the national average profit from one day ago #

• Chart data: date, city, profit
• Monitoring requirement: compare each city's profit for the current day with the national average profit from one day ago; trigger an alert if it is lower than the national average
• Implementation method:
  • Create a national average profit field ({filter=true:fixed [date]:avg([profit])})
  • Place the national average profit field in the metrics bar
  • Use the period-over-period analysis function to calculate the original value of the national average profit from one day ago
  • Configure monitoring with the condition: original profit value < period-over-period original value of the national average profit from one day ago
• Send a test to check whether the monitoring configuration meets the requirements
  

4.4 Preventing alerts caused by data spikes #

• Applicable scenario
• Monitor network latency in real time; trigger an alert if it exceeds 50 ms. Due to instability caused by special circumstances, the value suddenly spiked to 120 ms at a certain moment, triggering an alarm. However, when troubleshooting, the data had already returned to the normal range. To avoid false alarms caused by transient spikes, you can set the alarm to trigger only if the data for the past 3 minutes all exceed 50 ms.
• Implementation
  • In the dimension conditions, select the time range as the past 3 minutes.
  • For the metric condition, select "All data" and set the condition to network latency > 50.
    

4.5 Common issues #

4.5.1 Configuring fluctuation compared to N days ago: Alarm condition met but no alarm triggered #

• Taking the profit field in the chart below as an example, if the configuration for profit fluctuating by more than 5% compared to 1 day ago is not effective,
  • you can first configure the period-over-period growth rate of the profit field in the chart.
  • Set up monitoring, select the monitoring metric as period-over-period growth rate, and set the condition to absolute value > 5%.
• This can also achieve monitoring for fluctuations exceeding 5% compared to the previous day.
  
  This method can be used to monitor fluctuations compared to any of the previous N days.

4.5.2 Configuring monitoring compared to X days ago: Alarm condition met but no alarm triggered #

When performing monitoring operations to compare with X days ago, note the following two points:

• If a non-p_date field is involved, the date field in the non-p_date field must be placed as the first dimension. For example, if payment date is used as a non-p_date field, ensure that payment date is the first dimension.
• After selecting a non-p_date date field, the system will automatically shift both the p_date and non-p_date date fields as a whole. For example, if the date is based on [payment date], the past 1 day is [03 - 06]. In practice, this means comparing the data for p_date March 5 and payment date March 5 with the data for p_date March 6 and payment date March 6, rather than comparing the data for p_date March 6, payment date March 5, and payment date March 6.

To more conveniently and accurately complete the monitoring setup for comparison with X days ago, it is recommended to first configure the comparison rules for previous X days through Analysis - Period-over-period settings. After completing the comparison rule configuration, use the original value or absolute value of the period-over-period field in the monitoring rule settings to ensure that monitoring operates accurately according to the configured comparison rules.