Metrics Query Expression(MQE) Syntax

MQE is a string that consists of one or more expressions. Each expression could be a combination of one or more operations. The expression allows users to do simple query-stage calculation through V3 APIs.

Expression = <Operation> Expression1 <Operation> Expression2 <Operation> Expression3 ...

The following document lists the operations supported by MQE.

Metrics Expression

Metrics Expression will return a collection of time-series values.

Common Value Metrics

Expression:

<metric_name>

For example: If we want to query the service_sla metric, we can use the following expression:

service_sla

Result Type

The ExpressionResultType of the expression is TIME_SERIES_VALUES.

Labeled Value Metrics

Since v10.0.0, SkyWalking supports multiple labels metrics. We could query the specific labels of the metric by the following expression.

Expression:

<metric_name>{<label1_name>='<label1_value_1>,...', <label2_name>='<label2_value_1>,...',<label2...}

{<label1_name>='<label_value_1>,...'} is the selected label name/value of the metric. If is not specified, all label values of the metric will be selected.

For example: The k8s_cluster_deployment_status metric has labels namespace, deployment and status. If we want to query all deployment metric value with namespace=skywalking-showcase and status=true, we can use the following expression:

k8s_cluster_deployment_status{namespace='skywalking-showcase', status='true'}

We also could query the label with multiple values by separating the values with ,: If we want to query the service_percentile metric with the label name p and values 50,75,90,95,99, we can use the following expression:

service_percentile{p='50,75,90,95,99'}

If we want to rename the label values to P50,P75,P90,P95,P99, see Relabel Operation.

Result Type

The ExpressionResultType of the expression is TIME_SERIES_VALUES and with labels.

Binary Operation

The Binary Operation is an operation that takes two expressions and performs a calculation on their results. The following table lists the binary operations supported by MQE.

Expression:

Expression1 <Binary-Operator> Expression2
Operator Definition
+ addition
- subtraction
* multiplication
/ division
% modulo

For example: If we want to transform the service_sla metric value to percent, we can use the following expression:

service_sla / 100

Result Type

For the result type of the expression, please refer to the following table.

Binary Operation Rules

The following table lists if the different result types of the input expressions could do this operation and the result type after the operation. The expression could be on the left or right side of the operator. Note: If the expressions result on both sides of the operator are with labels, they should have the same labels for calculation. If the labels match, will reserve left expression result labels and the calculated value. Otherwise, will return empty value.

Expression Expression Yes/No ExpressionResultType
SINGLE_VALUE SINGLE_VALUE Yes SINGLE_VALUE
SINGLE_VALUE TIME_SERIES_VALUES Yes TIME_SERIES_VALUES
SINGLE_VALUE SORTED_LIST/RECORD_LIST Yes SORTED_LIST/RECORD_LIST
TIME_SERIES_VALUES TIME_SERIES_VALUES Yes TIME_SERIES_VALUES
TIME_SERIES_VALUES SORTED_LIST/RECORD_LIST no
SORTED_LIST/RECORD_LIST SORTED_LIST/RECORD_LIST no

Compare Operation

Compare Operation takes two expressions and compares their results. The following table lists the compare operations supported by MQE.

Expression:

Expression1 <Compare-Operator> Expression2
Operator Definition
> greater than
>= greater than or equal
< less than
<= less than or equal
== equal
!= not equal

The result of the compare operation is an int value:

  • 1: true
  • 0: false

For example: Compare the service_resp_time metric value if greater than 3000, if the service_resp_time result is:

{
  "data": {
    "execExpression": {
      "type": "TIME_SERIES_VALUES",
      "error": null,
      "results": [
        {
          "metric": {
            "labels": []
          },
          "values": [{"id": "1691658000000", "value": "2500", "traceID": null}, {"id": "1691661600000", "value": 3500, "traceID": null}]
        }
      ]
    }
  }
}

we can use the following expression:

service_resp_time > 3000

and get result:

{
  "data": {
    "execExpression": {
      "type": "TIME_SERIES_VALUES",
      "error": null,
      "results": [
        {
          "metric": {
            "labels": []
          },
          "values": [{"id": "1691658000000", "value": "0", "traceID": null}, {"id": "1691661600000", "value": 1, "traceID": null}]
        }
      ]
    }
  }
}

Compare Operation Rules and Result Type

Same as the Binary Operation Rules.

Aggregation Operation

Aggregation Operation takes an expression and performs aggregate calculations on its results.

Expression:

<Aggregation-Operator>(Expression)
Operator Definition ExpressionResultType
avg average the result SINGLE_VALUE
count count number of the result SINGLE_VALUE
latest select the latest non-null value from the result SINGLE_VALUE
sum sum the result SINGLE_VALUE
max select maximum from the result SINGLE_VALUE
min select minimum from the result SINGLE_VALUE

For example: If we want to query the average value of the service_cpm metric, we can use the following expression:

avg(service_cpm)

Result Type

The different operators could impact the ExpressionResultType, please refer to the above table.

Mathematical Operation

Mathematical Operation takes an expression and performs mathematical calculations on its results.

Expression:

<Mathematical-Operator>(Expression, parameters)
Operator Definition parameters ExpressionResultType
abs returns the absolute value of the result follow the input expression
ceil returns the smallest integer value that is greater or equal to the result follow the input expression
floor returns the largest integer value that is greater or equal to the result follow the input expression
round returns result round to specific decimal places places: a positive integer specific decimal places of the result follow the input expression

For example: If we want to query the average value of the service_cpm metric in seconds, and round the result to 2 decimal places, we can use the following expression:

round(service_cpm / 60 , 2)

Result Type

The different operators could impact the ExpressionResultType, please refer to the above table.

TopN Operation

TopN Operation takes an expression and performs calculation to get the TopN of Services/Instances/Endpoints. The result depends on the entity condition in the query.

  • Global TopN:
    • The entity is empty.
    • The result is the topN Services/Instances/Endpoints in the whole traffics.
    • Notice: If query the Endpoints metric, the global candidate set could be huge, please use it carefully.
  • Service’s Instances/Endpoints TopN:
    • The serviceName in the entity is not empty.
    • The result is the topN Instances/Endpoints of the service.

Expression:

top_n(<metric_name>, <top_number>, <order>)
  • top_number is the number of the top results, should be a positive integer.
  • order is the order of the top results. The value of order can be asc or des.

For example: If we want to query the current service’s top 10 instances with the highest service_instance_cpm metric value, we can use the following expression under specific service:

top_n(service_instance_cpm, 10, des)

Result Type

According to the type of the metric, the ExpressionResultType of the expression will be SORTED_LIST or RECORD_LIST.

Relabel Operation

Relabel Operation takes an expression and replaces the label with new label on its results. Since v10.0.0, SkyWalking supports relabel multiple labels.

Expression:

relabel(Expression, <target_label_name>='<origin_label_value_1>,...', <new_label_name>='<new_label_value_1>,...')

The order of the new label values should be the same as the order of the label values in the input expression result.

For example: If we want to query the service_percentile metric with the label values 50,75,90,95,99, and rename the label name to percentile and the label values to P50,P75,P90,P95,P99, we can use the following expression:

relabel(service_percentile{p='50,75,90,95,99'}, p='50,75,90,95,99', percentile='P50,P75,P90,P95,P99')

Result Type

Follow the input expression.

AggregateLabels Operation

AggregateLabels Operation takes an expression and performs an aggregate calculation on its Labeled Value Metrics results. It aggregates a group of TIME_SERIES_VALUES into a single TIME_SERIES_VALUES.

Expression:

aggregate_labels(Expression, <AggregateType>(<label1_name>,<label2_name>...))
  • AggregateType is the type of the aggregation operation.
  • <label1_name>,<label2_name>... is the label names that need to be aggregated. If not specified, all labels will be aggregated. Optional.
AggregateType Definition ExpressionResultType
avg calculate avg value of a Labeled Value Metrics TIME_SERIES_VALUES
sum calculate sum value of a Labeled Value Metrics TIME_SERIES_VALUES
max select the maximum value from a Labeled Value Metrics TIME_SERIES_VALUES
min select the minimum value from a Labeled Value Metrics TIME_SERIES_VALUES

For example: If we want to query all Redis command total rates, we can use the following expression(total_commands_rate is a metric which recorded every command rate in labeled value): Aggregating all the labels:

aggregate_labels(total_commands_rate, sum)

Also, we can aggregate by the cmd label:

aggregate_labels(total_commands_rate, sum(cmd))

Result Type

The ExpressionResultType of the aggregateLabels operation is TIME_SERIES_VALUES.

Logical Operation

ViewAsSequence Operation

ViewAsSequence operation represents the first not-null metric from the listing metrics in the given prioritized sequence(left to right). It could also be considered as a short-circuit of given metrics for the first value existing metric.

Expression:

view_as_seq([<expression_1>, <expression_2>, ...])

For example: if the first expression value is empty but the second one is not empty, it would return the result from the second expression. The following example would return the content of the service_cpm metric.

view_as_seq(not_existing, service_cpm)

Result Type

The result type is determined by the type of selected not-null metric expression.

IsPresent Operation

IsPresent operation represents that in a list of metrics, if any expression has a value, it would return 1 in the result; otherwise, it would return 0.

Expression:

is_present([<expression_1>, <expression_2>, ...])

For example: When the meter does not exist or the metrics has no value, it would return 0. However, if the metrics list contains meter with values, it would return 1.

is_present(not_existing, existing_without_value, existing_with_value)

Result Type

The result type is SINGLE_VALUE, and the result(1 or 0) in the first value.

Trend Operation

Trend Operation takes an expression and performs a trend calculation on its results.

Expression:

<Trend-Operator>(Metrics Expression, time_range)

time_range is the positive int of the calculated range. The unit will automatically align with to the query Step, for example, if the query Step is MINUTE, the unit of time_range is minute.

Operator Definition ExpressionResultType
increase returns the increase in the time range in the time series TIME_SERIES_VALUES
rate returns the per-second average rate of increase in the time range in the time series TIME_SERIES_VALUES

For example: If we want to query the increase value of the service_cpm metric in 2 minute(assume the query Step is MINUTE), we can use the following expression:

increase(service_cpm, 2)

If the query duration is 3 minutes, from (T1 to T3) and the metric has values in time series:

V(T1-2), V(T1-1), V(T1), V(T2), V(T3)

then the expression result is:

V(T1)-V(T1-2), V(T2)-V(T1-1), V(T3)-V(T1)

Note:

  • If the calculated metric value is empty, the result will be empty. Assume in the T3 point, the increase value = V(T3)-V(T1), If the metric V(T3) or V(T1) is empty, the result value in T3 will be empty.

Result Type

TIME_SERIES_VALUES.

Sort Operation

SortValues Operation

SortValues Operation takes an expression and sorts the values of the input expression result.

Expression:

sort_values(Expression, <limit>, <order>)
  • limit is the number of the sort results, should be a positive integer, if not specified, will return all results. Optional.
  • order is the order of the sort results. The value of order can be asc or des.

For example: If we want to sort the service_resp_time metric values in descending order and get the top 10 values, we can use the following expression:

sort_values(service_resp_time, 10, des)

Result Type

The result type follows the input expression.

SortLabelValues Operation

SortLabelValues Operation takes an expression and sorts the label values of the input expression result. This function uses natural sort order.

Expression:

sort_label_values(Expression, <order>, <label1_name>, <label2_name> ...)
  • order is the order of the sort results. The value of order can be asc or des.
  • <label1_name>, <label2_name> ... is the label names that need to be sorted by their values. At least one label name should be specified. The labels in the head of the list will be sorted first, and if the label not be included in the expression result will be ignored.

For example: If we want to sort the service_percentile metric label values in descending order by the p label, we can use the following expression:

sort_label_values(service_percentile{p='50,75,90,95,99'}, des, p)

For multiple labels, assume the metric has 2 labels:

metric{label1='a', label2='2a'} 
metric{label1='a', label2='2c'}
metric{label1='b', label2='2a'}
metric{label1='b', label2='2c'}

If we want to sort the metric metric label values in descending order by the label1 and label2 labels, we can use the following expression:

sort_label_values(metric, des, label1, label2)

And the result will be:

metric{label1='b', label2='2c'}
metric{label1='b', label2='2a'}
metric{label1='a', label2='2c'}
metric{label1='a', label2='2a'}

Expression Query Example

Labeled Value Metrics

service_percentile{p='50,95'}

The example result is:

{
  "data": {
    "execExpression": {
      "type": "TIME_SERIES_VALUES",
      "error": null,
      "results": [
        {
          "metric": {
            "labels": [{"key": "p", "value": "50"}]
          },
          "values": [{"id": "1691658000000", "value": "1000", "traceID": null}, {"id": "1691661600000", "value": 2000, "traceID": null}]
        },
        {
          "metric": {
            "labels": [{"key": "p", "value": "75"}]
          },
          "values": [{"id": "1691658000000", "value": "2000", "traceID": null}, {"id": "1691661600000", "value": 3000, "traceID": null}]
        }
      ]
    }
  }
}

If we want to transform the percentile value unit from ms to s the expression is:

service_percentile{p='50,75'} / 1000
{
  "data": {
    "execExpression": {
      "type": "TIME_SERIES_VALUES",
      "error": null,
      "results": [
        {
          "metric": {
            "labels": [{"key": "p", "value": "50"}]
          },
          "values": [{"id": "1691658000000", "value": "1", "traceID": null}, {"id": "1691661600000", "value": 2, "traceID": null}]
        },
        {
          "metric": {
            "labels": [{"key": "p", "value": "75"}]
          },
          "values": [{"id": "1691658000000", "value": "2", "traceID": null}, {"id": "1691661600000", "value": 3, "traceID": null}]
        }
      ]
    }
  }
}

Get the average value of each percentile, the expression is:

avg(service_percentile{p='50,75'})
{
  "data": {
    "execExpression": {
      "type": "SINGLE_VALUE",
      "error": null,
      "results": [
        {
          "metric": {
            "labels": [{"key": "p", "value": "50"}]
          },
          "values": [{"id": null, "value": "1500", "traceID": null}]
        },
        {
          "metric": {
            "labels": [{"key": "p", "value": "75"}]
          },
          "values": [{"id": null, "value": "2500", "traceID": null}]
        }
      ]
    }
  }
}

Calculate the difference between the percentile and the average value, the expression is:

service_percentile{p='50,75'} - avg(service_percentile{p='50,75'})
{
  "data": {
    "execExpression": {
      "type": "TIME_SERIES_VALUES",
      "error": null,
      "results": [
        {
          "metric": {
            "labels": [{"key": "p", "value": "50"}]
          },
          "values": [{"id": "1691658000000", "value": "-500", "traceID": null}, {"id": "1691661600000", "value": 500, "traceID": null}]
        },
        {
          "metric": {
            "labels": [{"key": "p", "value": "75"}]
          },
          "values": [{"id": "1691658000000", "value": "-500", "traceID": null}, {"id": "1691661600000", "value": 500, "traceID": null}]
        }
      ]
    }
  }
}

Calculate the difference between the service_resp_time and the service_percentile, if the service_resp_time result is:

{
  "data": {
    "execExpression": {
      "type": "TIME_SERIES_VALUES",
      "error": null,
      "results": [
        {
          "metric": {
            "labels": []
          },
          "values": [{"id": "1691658000000", "value": "2500", "traceID": null}, {"id": "1691661600000", "value": 3500, "traceID": null}]
        }
      ]
    }
  }
}

The expression is:

service_resp_time - service_percentile{p='50,75'}
{
  "data": {
    "execExpression": {
      "type": "TIME_SERIES_VALUES",
      "error": null,
      "results": [
        {
          "metric": {
            "labels": [{"key": "p", "value": "50"}]
          },
          "values": [{"id": "1691658000000", "value": "1500", "traceID": null}, {"id": "1691661600000", "value": "1500", "traceID": null}]
        },
        {
          "metric": {
            "labels": [{"key": "p", "value": "75"}]
          },
          "values": [{"id": "1691658000000", "value": "500", "traceID": null}, {"id": "1691661600000", "value": "500", "traceID": null}]
        }
      ]
    }
  }
}