Monitoring filters

This guide describes how to configure filters when you use the Monitoring API. You use filters to specify monitored resources, metric types, group definitions, and time series.

You can also use a filter to configure an alerting policy that monitors processes running on your systems. For information about those filters, see Process-health filters.

Before you begin

If you aren't familiar with metrics, time series, and monitored resources, see Metrics, Time Series, and Resources.

If you aren't familiar with labels, see Labels for an introduction.

Using filters

You can use filters in the Monitoring API to do the following:

  • Select the specific time series data that is returned from a list API request. The filter can select time series based on the data's project, group, monitored resource properties, and metric properties. For more information and examples, see Retrieving time series data.

  • Assign resources to a Group based on the resources' properties and the project to which they belong. For more information and examples, see Defining group membership.

  • Select resources within a group based on the resources' properties and the project to which they belong. For more information and examples, see Listing group members.

  • List particular metric types. For more information and examples, see Listing metric descriptors.

  • List particular monitored resource types. For more information and examples, see Listing monitored resource descriptors.

Filter selectors

A filter consists of at least one selector, which is a filter keyword. The following simple examples illustrate the different selectors:

  • project: Matches when the metrics of the specified project are visible to the scoping project of a metrics scope mentioned in the name parameter.

    Use the project selector when a Google Cloud project can view the metrics of multiple Google Cloud projects or AWS accounts and you only want metrics for single project. For example, if the metrics scope for Project-A includes Project-B, then a match occurs when name has a value of Project-A and you use the following filter:

        project = "Project-B"
    
  • group: Matches resources belonging to one Group.

    • Matches the group with the identifier group-id:

      group.id = "group-id"
      
  • resource: Matches monitored resources of a particular type or having particular label values.

    • Matches all monitored resources that are Compute Engine virtual machine (VM) instances:

      resource.type = "gce_instance"
      
    • Matches all resources whose zone begins with europe-:

      resource.labels.zone = starts_with("europe-")
      
  • metric: Matches a particular metric type or time series with with a particular label that matches a specific value.

    • Matches a specific metric:

      metric.type = "compute.googleapis.com/instance/cpu/usage_time"
      
    • Matches time series with a label named instance_name whose value starts with gke-hipster or gke-nginx:

      metric.labels.instance_name = monitoring.regex.full_match("^gke-(hipster|nginx).*")
      

The following table shows which selectors are permitted in filters based on the Monitoring API call:

Filter purpose project selector group selector resource selector metric selector
Define groups yes yes1
List group members yes yes
List time series yes yes yes yes2
List metric descriptors yes yes
List monitored resource descriptors yes
1 The resource selector has additional options when used for defining group membership.
2 When listing time series, you must specify exactly one metric type.

The following sections show examples of typical uses of monitoring filters. See Filter syntax for a complete discussion of the available filter objects and operators.

Retrieving time series data

Method: projects.timeSeries.list
Filter objects: project, group.id, resource.type, resource.labels.[KEY], metric.type, metric.labels.[KEY]

A time series is a list of time-stamped data points of a metric type from a specific monitored resource. For details, see The metric model. The metric type is specified by a metric descriptor, and the monitored resource is specified by a monitored-resource descriptor.

The filter specified in the list command must include a metric selector and that selector must specify exactly one metric type:

  • Return all time series for a particular metric type:

    metric.type = "compute.googleapis.com/instance/cpu/usage_time"
    
  • Return all time series for a specific group. The group selector only works with aligned time series data; see Group selector for more information:

    metric.type = "compute.googleapis.com/instance/cpu/usage_time" AND
        group.id = "2468013579"
    
  • Return all time series from a specific Compute Engine instance:

    metric.type = "compute.googleapis.com/instance/cpu/usage_time" AND
        metric.labels.instance_name = "my-instance-name"
    
  • Return all time series from Compute Engine instances whose names start with frontend-:

    metric.type = "compute.googleapis.com/instance/cpu/usage_time" AND
        metric.labels.instance_name = starts_with("frontend-")
    
  • Return all time series from Compute Engine instances whose names start with gke-hipster or gke-nginx:

    metric.type = "compute.googleapis.com/instance/cpu/usage_time" AND
        metric.labels.instance_name = monitoring.regex.full_match("^gke-(hipster|nginx).*")
    

Defining group membership

Method: projects.groups
Filter objects: project, resource.type, resource.labels.key, metadata.system_labels.[KEY], metadata.user_labels.[KEY]

A group can contain any number of resources, as specified by a filter. Group membership is dynamic; more or fewer resources might match the filter every time the filter is evaluated. The name parameter in the Group object specifies the group and the scoping project of a metrics scope. If the project selector is used in the filter, then it must specify a project whose metrics are visible to the scoping project.

  • All Compute Engine virtual machine (VM) instances in Europe:

    resource.type = "gce_instance" AND resource.labels.zone = starts_with("europe-")
    

Listing group members

Method: projects.groups.members.list
Filter objects: project, resource.type, resource.labels.[KEY]

Use a filter to limit which group members you retrieve. The name parameter specifies a scoping project of a metrics scope and a group defined in that project. If the project selector is used in the filter, then it must specify a project whose metrics are visible to the scoping project.

  • All the Pub/Sub topic resources that belong to project my-project:

    project = "my-project" AND resource.type = "pubsub_topic"
    

Listing metric descriptors

Method: projects.metricDescriptors.list
Filter objects: project, metric.type

Use a filter to limit which metric descriptors you retrieve:

  • To only return the Compute Engine metric descriptors, use the following filter:

    metric.type = starts_with("compute.googleapis.com")
    

See Metric List for a complete list of the metric types defined by Monitoring. For an overview of how metrics are named, see Naming conventions and requirements.

Listing monitored resource descriptors

Method: projects.monitoredResourceDescriptors.list
Filter objects: resource.type

Use a filter to limit which monitored resource descriptors you retrieve:

  • To only retrieve the Pub/Sub monitored resource descriptors, use the following filter:

    resource.type = starts_with("pubsub")
    

See Monitored Resource List for a complete list of the monitored resource types defined by Monitoring.

Reference: filter syntax

For an overview of filters with examples, see Using filters.

A monitoring filter is a string consisting of up to four types of selectors:

<monitoring_filter> ::=  <project_selector> AND <group_selector> AND <resource_selector> AND <metric_selector>

The filter matches an item if all of the included selectors match the item. As described in the following sections, some selectors can have multiple comparisons joined by AND or OR.

Depending on the filter's purpose, certain selectors might be required, optional, or prohibited. For example, the filter that defines the resources in a group cannot contain a metric selector, because groups doesn't contain metric types or time series. On the other hand, the filter used to list time series must contain a metric selector. The order of the selectors in the filter doesn't matter, but comparisons for different selectors must not be intermingled.

Comparisons

Filters and their selectors are built from comparisons. Each comparison has the following form:

[OBJECT] [OPERATOR] [VALUE]

These elements are described below:

  • [OBJECT]: selects a value to be tested, one of the following:

    project
    group.id
    metric.type
    metric.labels.[KEY]
    resource.type
    resource.labels.[KEY]
    metadata.system_labels.[KEY]
    metadata.user_labels.[KEYSTRING]
    

    [KEY] is a name, such as zone or instance_id. [KEYSTRING] can be name, but if it contains special characters then it must be quoted with quotation marks (").

  • [OPERATOR]: a comparison operator, one of the following:

    =            # equality (case-sensitive)
    > < >= <=    # numeric ordering
    !=           # not equal
    :            # "has" substring match and test for key (case-sensitive)
    
  • [VALUE]: a literal value or a built-in function call, one of the following

    <string>     # "a Unicode string". Don't use apostrophes (`'`) to quote strings.
    <bool>       # true or false
    <number>     # 0, -2, 123456, 3.14156
    <function>   # operators on the right side of '=' or '!=':
                 #   starts_with(<string>)
                 #   ends_with(<string>)
                 #   has_substring(<string> [, ignore_case=false])
                 #   one_of(<string>,...,<string>) for up to 100 strings
                 #   monitoring.regex.full_match(<RE2-string>)
    

    Except in the case of timeSeries.list, the has_substring filter takes an optional second argument, which specifies whether the match ignores case or not. The default value is false, so the default match is case-sensitive:

    • Case-sensitive: display_name=has_substring("Demo")
    • Case-sensitive: display_name=has_substring("Demo", false)
    • Case-insensitive: display_name=has_substring("Demo", true)

    Only the has_substring(<string>) form is supported for timeSeries.list.

    The monitoring.regex.full_match filter takes a regular-expression string in RE2 syntax.

You can use the following operators to group or modify comparisons. OR has higher precedence than AND. The operators must be written in upper case:

(...)        # grouping comparisons
AND          # conjunction (optional but recommended)
OR           # disjunction

The AND operator can be omitted between operators, but it is clearer and less error-prone to include it.

The comparison x = one_of("a", "b", "c") is equivalent to

(x = "a" OR x = "b" OR x = "c")

Only in group definitions, you can use the unary negation operator, NOT, before a comparison, but not with an exists operator (:) or before a parenthesized expression:

NOT          # negates the following comparison

Filter selectors

Use selectors to limit the filter selections to certain items. In the following sections, braces are used to show repetition. For example, the notation <x> {OR <y>} means you can write any of the following:

<x>
<x> OR <y>
<x> OR <y> OR <y>
<x> OR <y> OR <y> OR <y>
...

Project selector

A project selector limits the filter selection to items belonging to a single project or to any one of a set of projects. Each project can be specified by its ID or its number:

<project_selector> ::= project '=' (<number> | <string>) {OR project '=' (<number> | <string>)}

If your project selector has more than a single comparison, enclose the entire selector in parentheses for better readability. For example:

(project=12345 OR project="my-project-id") AND resource.type="gce_instance"

Group selector

A group selector limits the filter selection to items belonging to a single group:

<group_selector> ::= group.id '=' <string>

For example, the following filter could be used to retrieve a time series from each of the VM instances in a group:

group.id = 12345 AND
    resource.type = "gce_instance" AND
    metric.type = "compute.googleapis.com/instance/disk/read_bytes_count"

The group selector is only permitted in filters passed to projects.timeSeries.list. In addition, group selection requires aligned data; that is, the call to projects.timeSeries.list must include values for perSeriesAligner and alignmentPeriod. This is because group membership is itself a kind of time series that must be joined with the metric data, and supplying alignment parameters gives you control over how that join happens. For more information on alignment parameters, see Aggregating data.

Resource selector

A resource selector limits the filter selection to resources—or items associated with resources—that have a specific resource type or label values:

<resource_selector> ::= <resource_type_expression>
                      | <resource_label_expression>
                      | <resource_type_expression> AND <resource_label_expression>

<resource_type_expression> ::= resource.type '=' <string>
                             | resource.type ':' <string>
                             | resource.type '=' starts_with '(' <string>')'
                             | resource.type '=' ends_with '(' <string> ')'

<resource_label_expression> ::= <r_label_comparison> {AND <r_label_comparison>}
                              | <r_label_comparison> {OR <r_label_comparison>}

<r_label_comparison> ::= resource.labels.[KEY] '=' (<string> | <bool>)
                       | resource.labels.[KEY] ':' <string>
                       | resource.labels.[KEY] '=' (starts_with | ends_with) '(' <string> ')'
                       | resource.labels.[KEY] ('=' | '>' | '<' | '>=' | '<=') <number>

If you use more than one <r_label_comparison> in your selector, enclose them all in parentheses for better readability. For example, the following filter could be used to define a group that includes all VM instances in the USA and Europe:

resource.type = "gce_instance" AND
  (resource.labels.zone = starts_with("us-") OR resource.labels.zone = starts_with("europe-"))

Resource selector for group definitions

Resource selectors used to define group membership use extensions to the <resource_selector> syntax:

  1. You include filters based on the value of metadata system labels, metadata.system_labels.[KEY], and metadata user labels, metadata.user_labels.[KEYSTRING]. The keys for metadata.user_labels should be quoted because they can contain special characters such as hyphens.

    When a selector contains a metadata filter and a resource filter, you must combine them with AND; you can't use OR. For example, a chart with the following selector displays the CPU utilization for all VM instances with a machine type of e2-medium or e2-micro:

    metric.type="compute.googleapis.com/instance/cpu/utilization"
    resource.type="gce_instance" AND
    (metadata.system_labels."machine_type"="e2-medium" OR
     metadata.system_labels."machine_type"="e2-micro")
    
  2. You can use the not-equals operator (!=) to compare resource types, resource labels, and metadata. The operator can be used when comparing strings, numbers, Booleans, or the substring functions. For example, resource.type!=starts_with("gce") is true if the resource type does not start with "gce".

  3. You can use a single NOT operator before a resource comparison. For example, NOT resource.labels.zone="europe" is true if the resource's zone doesn't include "europe". You cannot use NOT before an exists operator (:) or a parenthesized expression.

  4. You can use the exists operator (:) to test for the existence of keys. For example, the comparison resource.labels:zone is true if the label key zone is present in the resource.

For example, one of the platform resource metadata keys for VM instances is spot_instance. The following filter selector chooses instances that are spot instances:

    resource.type = "gce_instance" AND metadata.system_labels.spot_instance = true

Metric selector

A metric selector specifies certain metrics or metric descriptors by limiting the metric type and metric labels. When listing time series data, the metric selector must specify a single metric type:

<metric_selector> ::= <metric_name_expression> [AND <metric_label_expression>]

<metric_name_expression> ::= metric.type '=' <string>
                           | metric.type ':' <string>
                           | metric.type '=' starts_with '(' <string> ')'
                           | metric.type '=' ends_with '(' <string> ')'

<metric_label_expression> ::= <metric_label_comparison> {[AND] <metric_label_comparison>}
                            | <metric_label_comparison> {OR <metric_label_comparison>}

<metric_label_comparison> ::= metric.labels.[KEY] '=' <string> | <bool>
                            | metric.labels.[KEY] ':' <string>
                            | metric.labels.[KEY] '=' starts_with '(' <string> ')'
                            | metric.labels.[KEY] '=' ends_with '(' <string> ')'
                            | metric.labels.[KEY] ('=' | '>' | '<' | '>=' | '<=') <number>

For example, the following filter could be used to retrieve a time series for a specific database instance:

metric.type = "cloudsql.googleapis.com/database/state" AND
  (metric.labels.resource_type = "instance" AND
   metric.labels.resource_id = "abc-123456")

Examples

In the filtering examples, we use the following metric descriptor, monitored resource descriptor, and virtual machine instance, simplified for illustration:

# Metric descriptor:
{ "name": "projects/my-project-id/metricDescriptors/compute.googleapis.com%2Finstance%2Fdisk%2Fread_bytes_count"
  "type": "compute.googleapis.com/instance/disk/read_bytes_count",
  "labels": [ { "key": "device_name",
                "description": "The name of the disk device." } ] }

# Monitored resource descriptor:
{  "name": "monitoredResourceDescriptors/gce_instance"
   "type": "gce_instance",
   "labels": [
     { "key": "instance_id",
       "description": "The instance ID provide by Google Compute Engine." },
     { "key": "zone",
       "description": "The Google Cloud Platform zone hosting the instance."
     } ] }

# Resource descriptor for a virtual machine instance.
{ "type": "gce_instance",
  "instance_id": "1472038649266883453",
  "zone": "us-east-1b",
  "disks": [ "log_partition" ],
  "machine_type": "n1-standard-2",
  "tags": { "environment": "bleeding-edge",
            "role": "frobulator" },
  "project_id": "my-project-id" }

Metric retrieval examples

To request the disk read bandwidth usage for all instances and all devices, define a filter as shown below. This returns, for each instance, a separate time series reporting the read bandwidth for each device:

metric.type = "compute.googleapis.com/instance/disk/read_bytes_count"

To refine the request to query for the read bandwidth for only the disk device known as "log_partition" on each instance, define the filter as shown below. This filter returns, for each instance, at most one time series, depending on whether a device of that name exists on that instance:

metric.type = "compute.googleapis.com/instance/disk/read_bytes_count" AND
metric.labels.device_name = "log_partition"

To restrict the request to just a single instance, specify that instance:

resource.type = "gce_instance" AND
resource.labels.instance_id = "1472038649266883453" AND
metric.type = "compute.googleapis.com/instance/disk/read_bytes_count" AND
metric.labels.device_name = "log_partition"

Filtering with groups

The following examples illustrate the use of the group selector in filters to restrict monitored resources to those in a specific group. See Group definitions for information on the selectors used to define group membership.

{ "name": "projects/my-test-project/groups/024681012",
  "display_name": "My Redis Cluster",
  "filter": "metadata.user_labels.role=redis" }

In a call to the projects.timeSeries.list method, the following filter requests the disk read bandwidth usage for all Compute Engine instances in a particular group. The group must be defined in the scoping project of a metrics scope specified in the method's name parameter:

resource.type = "gce_instance" AND
group.id = "024681012" AND
metric.type = "compute.googleapis.com/instance/disk/read_bytes_count"