Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Retrieving metrics with the Performance Insights API

When Performance Insights is enabled, the API provides visibility into instance performance. Amazon CloudWatch Logs provides the authoritative source for vended monitoring metrics for Amazon services.

Performance Insights offers a domain-specific view of database load measured as average active sessions (AAS). This metric appears to API consumers as a two-dimensional time-series dataset. The time dimension of the data provides DB load data for each time point in the queried time range. Each time point decomposes overall load in relation to the requested dimensions, such as Query, Wait-state, Application, or Host, measured at that time point.

Amazon DocumentDB Performance Insights monitors your Amazon DocumentDB DB instance so that you can analyze and troubleshoot database performance. One way to view Performance Insights data is in the Amazon Web Services Management Console. Performance Insights also provides a public API so that you can query your own data. You can use the API to do the following:

To use the Performance Insights API, enable Performance Insights on one of your Amazon DocumentDB instances. For information about enabling Performance Insights, see Enabling and disabling Performance Insights. For more information about the Performance Insights API, see the Performance Insights API Reference.

The Performance Insights API provides the following operations.

Amazon CLI for Performance Insights

You can view Performance Insights data using the Amazon CLI. You can view help for the Amazon CLI commands for Performance Insights by entering the following on the command line.

aws pi help

If you don't have the Amazon CLI installed, see Installing the Amazon Command Line Interface in the Amazon CLI User Guide for information about installing it.

Retrieving time-series metrics

The GetResourceMetrics operation retrieves one or more time-series metrics from the Performance Insights data. GetResourceMetrics requires a metric and time period, and returns a response with a list of data points.

For example, the Amazon Web Services Management Console uses GetResourceMetrics to populate the Counter Metrics chart and the Database Load chart, as seen in the following image.

All metrics returned by GetResourceMetrics are standard time-series metrics, with the exception of db.load. This metric is displayed in the Database Load chart. The db.load metric is different from the other time-series metrics because you can break it into subcomponents called dimensions. In the previous image, db.load is broken down and grouped by the waits states that make up the db.load.

For information about the counter metrics returned by GetResourceMetrics, see Performance Insights for counter metrics.

The following calculations are supported for the metrics:

For example, assume that a metric is collected for 300 seconds (5 minutes), and that the metric is collected one time each minute. The values for each minute are 1, 2, 3, 4, and 5. In this case, the following calculations are returned:

For information about using the get-resource-metrics Amazon CLI command, see get-resource-metrics.

For the --metric-queries option, specify one or more queries that you want to get results for. Each query consists of a mandatory Metric and optional GroupBy and Filter parameters. The following is an example of a --metric-queries option specification.

{
   "Metric": "string",
   "GroupBy": {
     "Group": "string",
     "Dimensions": ["string", ...],
     "Limit": integer
   },
   "Filter": {"string": "string"
     ...}

Amazon CLI examples for Performance Insights

The following examples show how to use the Amazon CLI for Performance Insights.

Retrieving counter metrics

The following screenshot shows two counter metrics charts in the Amazon Web Services Management Console.

The following example shows how to gather the same data that the Amazon Web Services Management Console uses to generate the two counter metric charts.

For Linux, macOS, or Unix:

aws pi get-resource-metrics \
   --service-type DOCDB \
   --identifier db-ID \
   --start-time 2022-03-13T8:00:00Z \
   --end-time   2022-03-13T9:00:00Z \
   --period-in-seconds 60 \
   --metric-queries '[{"Metric": "os.cpuUtilization.user.avg"  },
                      {"Metric": "os.cpuUtilization.idle.avg"}]'

For Windows:

aws pi get-resource-metrics ^
   --service-type DOCDB ^
   --identifier db-ID ^
   --start-time 2022-03-13T8:00:00Z ^
   --end-time   2022-03-13T9:00:00Z ^
   --period-in-seconds 60 ^
   --metric-queries '[{"Metric": "os.cpuUtilization.user.avg"  },
                      {"Metric": "os.cpuUtilization.idle.avg"}]'

You can also make a command easier to read by specifying a file for the --metrics-query option. The following example uses a file called query.json for the option. The file has the following contents.

[
    {
        "Metric": "os.cpuUtilization.user.avg"
    },
    {
        "Metric": "os.cpuUtilization.idle.avg"
    }
]

Run the following command to use the file.

For Linux, macOS, or Unix:

aws pi get-resource-metrics \
   --service-type DOCDB \
   --identifier db-ID \
   --start-time 2022-03-13T8:00:00Z \
   --end-time   2022-03-13T9:00:00Z \
   --period-in-seconds 60 \
   --metric-queries file://query.json

For Windows:

aws pi get-resource-metrics ^
   --service-type DOCDB ^
   --identifier db-ID ^
   --start-time 2022-03-13T8:00:00Z ^
   --end-time   2022-03-13T9:00:00Z ^
   --period-in-seconds 60 ^
   --metric-queries file://query.json

The preceding example specifies the following values for the options:

It queries for a one-hour time range:

The response looks similar to the following.

{
    "AlignedStartTime": "2022-03-13T08:00:00+00:00",
    "AlignedEndTime": "2022-03-13T09:00:00+00:00",
    "Identifier": "db-NQF3TTMFQ3GTOKIMJODMC3KQQ4",
    "MetricList": [
        {
            "Key": {
                "Metric": "os.cpuUtilization.user.avg"
            },
            "DataPoints": [
                {
                    "Timestamp": "2022-03-13T08:01:00+00:00", //Minute1
                    "Value": 3.6
                },
                {
                    "Timestamp": "2022-03-13T08:02:00+00:00", //Minute2
                    "Value": 2.6
                },
                //.... 60 datapoints for the os.cpuUtilization.user.avg metric
        {
            "Key": {
                "Metric": "os.cpuUtilization.idle.avg"
            },
            "DataPoints": [
                {
                    "Timestamp": "2022-03-13T08:01:00+00:00",
                    "Value": 92.7
                },
                {
                    "Timestamp": "2022-03-13T08:02:00+00:00",
                    "Value": 93.7
                },
                //.... 60 datapoints for the os.cpuUtilization.user.avg metric 
            ]
        }
    ] //end of MetricList
} //end of response

The response has an Identifier, AlignedStartTime, and AlignedEndTime. B the --period-in-seconds value was 60, the start and end times have been aligned to the minute. If the --period-in-seconds was 3600, the start and end times would have been aligned to the hour.

The MetricList in the response has a number of entries, each with a Key and a DataPoints entry. Each DataPoint has a Timestamp and a Value. Each Datapoints list has 60 data points because the queries are for per-minute data over an hour, with Timestamp1/Minute1, Timestamp2/Minute2, and so on, up to Timestamp60/Minute60.

Because the query is for two different counter metrics, there are two elements in the response MetricList.

Retrieving the DB load average for top wait states

The following example is the same query that the Amazon Web Services Management Console uses to generate a stacked area line graph. This example retrieves the db.load.avg for the last hour with load divided according to the top seven wait states. The command is the same as the command in Retrieving counter metrics. However, the query.json file has the following contents.

[
    {
        "Metric": "db.load.avg",
        "GroupBy": { "Group": "db.wait_state", "Limit": 7 }
    }
]

Run the following command.

For Linux, macOS, or Unix:

aws pi get-resource-metrics \
   --service-type DOCDB \
   --identifier db-ID \
   --start-time 2022-03-13T8:00:00Z \
   --end-time   2022-03-13T9:00:00Z \
   --period-in-seconds 60 \
   --metric-queries file://query.json

For Windows:

aws pi get-resource-metrics ^
   --service-type DOCDB ^
   --identifier db-ID ^
   --start-time 2022-03-13T8:00:00Z ^
   --end-time   2022-03-13T9:00:00Z ^
   --period-in-seconds 60 ^
   --metric-queries file://query.json

The example specifies the metric of db.load.avg and a GroupBy of the top seven wait states. For details about valid values for this example, see DimensionGroup in the Performance Insights API Reference.

The response looks similar to the following.

{
    "AlignedStartTime": "2022-04-04T06:00:00+00:00",
    "AlignedEndTime": "2022-04-04T06:15:00+00:00",
    "Identifier": "db-NQF3TTMFQ3GTOKIMJODMC3KQQ4",
    "MetricList": [
        {//A list of key/datapoints
            "Key": {
                //A Metric with no dimensions. This is the total db.load.avg
                "Metric": "db.load.avg"
            },
            "DataPoints": [
                //Each list of datapoints has the same timestamps and same number of items
                {
                    "Timestamp": "2022-04-04T06:01:00+00:00",//Minute1
                    "Value": 0.0
                },
                {
                    "Timestamp": "2022-04-04T06:02:00+00:00",//Minute2
                    "Value": 0.0
                },
                //... 60 datapoints for the total db.load.avg key
                ]
        },
        {
            "Key": {
                //Another key. This is db.load.avg broken down by CPU
                "Metric": "db.load.avg",
                "Dimensions": {
                    "db.wait_state.name": "CPU"
                }
            },
            "DataPoints": [
                {
                    "Timestamp": "2022-04-04T06:01:00+00:00",//Minute1
                    "Value": 0.0
                },
                {
                    "Timestamp": "2022-04-04T06:02:00+00:00",//Minute2
                    "Value": 0.0
                },
                //... 60 datapoints for the CPU key
            ]
        },//... In total we have 3 key/datapoints entries, 1) total, 2-3) Top Wait States
    ] //end of MetricList
} //end of response
			

In this response, there are three entries in the MetricList. There is one entry for the total db.load.avg, and three entries each for the db.load.avg divided according to one of the top three wait states. Since there was a grouping dimension (unlike the first example), there must be one key for each grouping of the metric. There can't be only one key for each metric, as in the basic counter metric use case.

Retrieving the DB load average for top Query

The following example groups db.wait_state by the top 10 query statements. There are two different groups for query statements:

When analyzing database performance, it can be useful to consider query statements that only differ by their parameters as one logic item. So, you can use db.query_tokenized when querying. However, especially when you're interested in explain(), sometimes it's more useful to examine full query statements with parameters. There is a parent-child relationship between tokenized and full queries, with multiple full queries (children) grouped under the same tokenized query (parent).

The command in this example is the similar to the command in Retrieving the DB load average for top wait states. However, the query.json file has the following contents.

[
    {
        "Metric": "db.load.avg",
        "GroupBy": { "Group": "db.query_tokenized", "Limit": 10 }
    }
]

The following example uses db.query_tokenized.

For Linux, macOS, or Unix:

aws pi get-resource-metrics \
   --service-type DOCDB \
   --identifier db-ID \
   --start-time 2022-03-13T8:00:00Z \
   --end-time   2022-03-13T9:00:00Z \
   --period-in-seconds 3600 \
   --metric-queries file://query.json 

For Windows:

aws pi get-resource-metrics ^
   --service-type DOCDB ^
   --identifier db-ID ^
   --start-time 2022-03-13T8:00:00Z ^
   --end-time   2022-03-13T9:00:00Z  ^
   --period-in-seconds 3600 ^
   --metric-queries file://query.json 

This example queries over 1 hour, with a one minute period-in-seconds.

The example specifies the metric of db.load.avg and a GroupBy of the top seven wait states. For details about valid values for this example, see DimensionGroup in the Performance Insights API Reference.

The response looks similar to the following.

{
    "AlignedStartTime": "2022-04-04T06:00:00+00:00",
    "AlignedEndTime": "2022-04-04T06:15:00+00:00",
    "Identifier": "db-NQF3TTMFQ3GTOKIMJODMC3KQQ4",
    "MetricList": [
        {//A list of key/datapoints
            "Key": {
                "Metric": "db.load.avg"
            },
            "DataPoints": [
                //... 60 datapoints for the total db.load.avg key
                ]
        },
               {
            "Key": {//Next key are the top tokenized queries
                "Metric": "db.load.avg",
                "Dimensions": {
                    "db.query_tokenized.db_id": "pi-1064184600",
                    "db.query_tokenized.id": "77DE8364594EXAMPLE",
                    "db.query_tokenized.statement": "{\"find\":\"customers\",\"filter\":{\"FirstName\":\"?\"},\"sort\":{\"key\":{\"$numberInt\":\"?\"}},\"limit\"
:{\"$numberInt\":\"?\"},\"$db\":\"myDB\",\"$readPreference\":{\"mode\":\"primary\"}}"
                }
            },
            "DataPoints": [
            //... 60 datapoints 
            ]
        },
        // In total 11 entries, 10 Keys of top tokenized queries, 1 total key 
    ] //End of MetricList
} //End of response

This response has 11 entries in the MetricList (1 total, 10 top tokenized query), with each entry having 24 per-hour DataPoints.

For tokenized queries, there are three entries in each dimensions list:

When querying, it might be convenient to specify a Group in GroupBy. However, for finer-grained control over the data that's returned, specify the list of dimensions. For example, if all that is needed is the db.query_tokenized.statement, then a Dimensions attribute can be added to the query.json file.

[
    {
        "Metric": "db.load.avg",
        "GroupBy": {
            "Group": "db.query_tokenized",
            "Dimensions":["db.query_tokenized.statement"],
            "Limit": 10
        }
    }
]

Retrieving the DB load average filtered by Query

The corresponding API query in this example is similar to the command in Retrieving the DB load average for top Query. However, the query.json file has the following contents.

[
 {
        "Metric": "db.load.avg",
        "GroupBy": { "Group": "db.wait_state", "Limit": 5  }, 
        "Filter": { "db.query_tokenized.id": "AKIAIOSFODNN7EXAMPLE" }
    }
]

In this response, all values are filtered according to the contribution of tokenized query AKIAIOSFODNN7EXAMPLE specified in the query.json file. The keys also might follow a different order than a query without a filter, because it's the top five wait states that affected the filtered query.