Documentation
¶
Overview ¶
Package instance provides a mini Prometheus scraper and remote_writer.
Index ¶
- Variables
- func Hostname() (string, error)
- func MarshalConfig(c *Config, scrubSecrets bool) ([]byte, error)
- func MarshalConfigToWriter(c *Config, w io.Writer, scrubSecrets bool) error
- type BasicManager
- type BasicManagerConfig
- type Config
- type ConfigValidator
- type DiscoveredGroups
- type ErrInvalidUpdate
- type Factory
- type GroupChannel
- type GroupManager
- type HostFilter
- type Instance
- type ManagedInstance
- type Manager
- type MetricValueCollector
- type MockManager
- type NoOpInstance
Constants ¶
This section is empty.
Variables ¶
var (
// DefaultRelabelConfigs defines a list of relabel_configs that will
// be automatically appended to the end of all Prometheus
// configurations.
DefaultRelabelConfigs = []*relabel.Config{
{
SourceLabels: model.LabelNames{"__meta_kubernetes_pod_node_name"},
TargetLabel: "__host__",
Action: relabel.Replace,
Separator: ";",
Regex: relabel.MustNewRegexp("(.*)"),
Replacement: "$1",
},
}
DefaultConfig = Config{
HostFilter: false,
WALTruncateFrequency: 1 * time.Minute,
RemoteFlushDeadline: 1 * time.Minute,
WriteStaleOnShutdown: false,
}
)
Default configuration values
var (
// DefaultBasicManagerConfig is the default config for the BasicManager.
DefaultBasicManagerConfig = BasicManagerConfig{
InstanceRestartBackoff: 5 * time.Second,
}
)
Functions ¶
func Hostname ¶ added in v0.4.0
func Hostname() (string, error)
Hostname retrieves the hostname identifying the machine the process is running on. It will return the value of $HOSTNAME, if defined, and fall back to Go's os.Hostname.
func MarshalConfig ¶
func MarshalConfig(c *Config, scrubSecrets bool) ([]byte, error)
MarshalConfig marshals an instance config based on a provided content type.
func MarshalConfigToWriter ¶
func MarshalConfigToWriter(c *Config, w io.Writer, scrubSecrets bool) error
MarshalConfigToWriter marshals a config to an io.Writer.
Types ¶
type BasicManager ¶ added in v0.5.0
type BasicManager struct {
// contains filtered or unexported fields
}
BasicManager creates a new BasicManager, implementing the Manager interface. BasicManager will directly launch instances and perform no extra processing.
Other implementations of Manager usually wrap a BasicManager.
func NewBasicManager ¶ added in v0.5.0
func NewBasicManager(cfg BasicManagerConfig, logger log.Logger, launch Factory, validate ConfigValidator) *BasicManager
NewBasicManager creates a new BasicManager. The launch function will be invoked any time a new Config is applied.
The lifecycle of any ManagedInstance returned by the launch function will be handled by the BasicManager. Instances will be automatically restarted if stopped, updated if the config changes, or removed when the Config is deleted.
The validate function will be called before creating an instance. If the config is not valid, launch will not be called. validate is optional.
func (*BasicManager) ApplyConfig ¶ added in v0.5.0
func (m *BasicManager) ApplyConfig(c Config) error
ApplyConfig takes a Config and either starts a new managed instance or updates an existing managed instance. The value for Name in c is used to uniquely identify the Config and determine whether the Config has an existing associated managed instance.
func (*BasicManager) DeleteConfig ¶ added in v0.5.0
func (m *BasicManager) DeleteConfig(name string) error
DeleteConfig removes a managed instance by its config name. Returns an error if there is no such managed instance with the given name.
func (*BasicManager) ListConfigs ¶ added in v0.5.0
func (m *BasicManager) ListConfigs() map[string]Config
ListConfigs lists the current active configs managed by BasicManager.
func (*BasicManager) ListInstances ¶ added in v0.5.0
func (m *BasicManager) ListInstances() map[string]ManagedInstance
ListInstances returns the current active instances managed by BasicManager.
type BasicManagerConfig ¶ added in v0.5.0
type BasicManagerConfig struct {
InstanceRestartBackoff time.Duration
}
BasicManagerConfig controls the operations of a BasicManager.
type Config ¶
type Config struct {
Name string `yaml:"name" json:"name"`
HostFilter bool `yaml:"host_filter" json:"host_filter"`
ScrapeConfigs []*config.ScrapeConfig `yaml:"scrape_configs,omitempty" json:"scrape_configs,omitempty"`
RemoteWrite []*config.RemoteWriteConfig `yaml:"remote_write,omitempty" json:"remote_write,omitempty"`
// How frequently the WAL should be truncated.
WALTruncateFrequency time.Duration `yaml:"wal_truncate_frequency,omitempty" json:"wal_truncate_frequency,omitempty"`
RemoteFlushDeadline time.Duration `yaml:"remote_flush_deadline,omitempty" json:"remote_flush_deadline,omitempty"`
WriteStaleOnShutdown bool `yaml:"write_stale_on_shutdown,omitempty" json:"write_stale_on_shutdown,omitempty"`
}
Config is a specific agent that runs within the overall Prometheus agent. It has its own set of scrape_configs and remote_write rules.
func UnmarshalConfig ¶
func UnmarshalConfig(r io.Reader) (*Config, error)
UnmarshalConfig unmarshals an instance config from a reader based on a provided content type.
func (*Config) ApplyDefaults ¶
func (c *Config) ApplyDefaults(global *config.GlobalConfig) error
ApplyDefaults applies default configurations to the configuration to all values that have not been changed to their non-zero value. ApplyDefaults also validates the config.
func (Config) MarshalYAML ¶
func (c Config) MarshalYAML() (interface{}, error)
func (*Config) UnmarshalYAML ¶
func (c *Config) UnmarshalYAML(unmarshal func(interface{}) error) error
type ConfigValidator ¶ added in v0.5.0
type ConfigValidator func(c *Config) error
A ConfigValidator should validate a Config and return an error if there is a problem. ConfigValidator may mutate the Config.
type DiscoveredGroups ¶
type DiscoveredGroups = map[string][]*targetgroup.Group
DiscoveredGroups is a set of groups found via service discovery.
func FilterGroups ¶
func FilterGroups(in DiscoveredGroups, host string) DiscoveredGroups
FilterGroups takes a set of DiscoveredGroups as input and filters out any Target that is not running on the host machine provided by host.
This is done by looking at two labels:
__meta_kubernetes_pod_node_name is used first to represent the host machine of networked containers. Primarily useful for Kubernetes. This label is automatically added when using Kubernetes service discovery.
__address__ is used next to represent the address of the service to scrape. In a containerized envirment, __address__ will be the address of the container.
If the discovered address is localhost or 127.0.0.1, the group is never filtered out.
type ErrInvalidUpdate ¶ added in v0.5.0
type ErrInvalidUpdate struct {
Inner error
}
ErrInvalidUpdate is returned whenever Update is called against an instance but an invalid field is changed between configs. If ErrInvalidUpdate is returned, the instance must be fully stopped and replaced with a new one with the new config.
func (ErrInvalidUpdate) As ¶ added in v0.5.0
func (e ErrInvalidUpdate) As(err interface{}) bool
As will set the err object to ErrInvalidUpdate provided err is a pointer to ErrInvalidUpdate.
type Factory ¶ added in v0.5.0
type Factory func(c Config) (ManagedInstance, error)
Factory should return an unstarted instance given some config.
type GroupChannel ¶
type GroupChannel = <-chan DiscoveredGroups
GroupChannel is a channel that provides discovered target groups.
type GroupManager ¶ added in v0.5.0
type GroupManager struct {
// contains filtered or unexported fields
}
A GroupManager wraps around another Manager and groups all incoming Configs into a smaller set of configs, causing less managed instances to be spawned.
Configs are grouped by all settings for a Config *except* scrape configs. Any difference found in any flag will cause a Config to be placed in another group. One exception to this rule is that remote_writes are compared unordered, but the sets of remote_writes should otherwise be identical.
GroupManagers drastically improve the performance of the Agent when a significant number of instances are spawned, as the overhead of each instance having its own service discovery, WAL, and remote_write can be significant.
The config names of instances within the group will be represented by that group's hash of settings.
func NewGroupManager ¶ added in v0.5.0
func NewGroupManager(inner Manager) *GroupManager
NewGroupManager creates a new GroupManager for combining instances of the same "group."
func (*GroupManager) ApplyConfig ¶ added in v0.5.0
func (m *GroupManager) ApplyConfig(c Config) error
ApplyConfig will determine the group of the Config before applying it to the group. If no group exists, one will be created. If a group already exists, the group will have its settings merged with the Config and will be updated.
func (*GroupManager) DeleteConfig ¶ added in v0.5.0
func (m *GroupManager) DeleteConfig(name string) error
DeleteConfig will remove a Config from its associated group. If there are no more Configs within that group after this Config is deleted, the managed instance will be stopped. Otherwise, the managed instance will be updated with the new grouped Config that doesn't include the removed one.
func (*GroupManager) ListConfigs ¶ added in v0.5.0
func (m *GroupManager) ListConfigs() map[string]Config
ListConfigs returns the UNGROUPED instance configs with their original settings. To see the grouped instances, call ListInstances instead.
func (*GroupManager) ListInstances ¶ added in v0.5.0
func (m *GroupManager) ListInstances() map[string]ManagedInstance
ListInstances returns all currently grouped managed instances. The key will be the group's hash of shared settings.
type HostFilter ¶
type HostFilter struct {
// contains filtered or unexported fields
}
HostFilter acts as a MITM between the discovery manager and the scrape manager, filtering out discovered targets that are not running on the same node as the agent itself.
func NewHostFilter ¶
func NewHostFilter(host string) *HostFilter
NewHostFilter creates a new HostFilter
type Instance ¶
type Instance struct {
// contains filtered or unexported fields
}
Instance is an individual metrics collector and remote_writer.
func New ¶
func New(reg prometheus.Registerer, globalCfg config.GlobalConfig, cfg Config, walDir string, logger log.Logger) (*Instance, error)
New creates a new Instance with a directory for storing the WAL. The instance will not start until Run is called on the instance.
func (*Instance) Run ¶
func (i *Instance) Run(ctx context.Context) error
Run starts the instance, initializing Prometheus components, and will continue to run until an error happens during execution or the provided context is cancelled.
Run may be re-called after exiting, as components will be reinitialized each time Run is called.
func (*Instance) TargetsActive ¶ added in v0.5.0
func (i *Instance) TargetsActive() map[string][]*scrape.Target
TargetsActive returns the set of active targets from the scrape manager. Returns nil if the scrape manager is not ready yet.
type ManagedInstance ¶ added in v0.5.0
type ManagedInstance interface {
Run(ctx context.Context) error
Update(c Config) error
TargetsActive() map[string][]*scrape.Target
}
ManagedInstance is implemented by Instance. It is defined as an interface for the sake of testing from Manager implementations.
type Manager ¶ added in v0.5.0
type Manager interface {
// ListInstances returns all currently managed instances running
// within the Manager. The key will be the instance name from their config.
ListInstances() map[string]ManagedInstance
// ListConfigs returns the config objects associated with a managed
// instance. The key will be the Name field from Config.
ListConfigs() map[string]Config
// ApplyConfig creates a new Config or updates an existing Config if
// one with Config.Name already exists.
ApplyConfig(Config) error
// DeleteConfig deletes a given managed instance based on its Config.Name.
DeleteConfig(name string) error
// Stop stops the Manager and all managed instances.
Stop()
}
Manager represents a set of methods for manipulating running instances at runtime.
type MetricValueCollector ¶
type MetricValueCollector struct {
// contains filtered or unexported fields
}
MetricValueCollector wraps around a Gatherer and provides utilities for pulling metric values from a given metric name and label matchers.
This is used by the agent instances to find the most recent timestamp successfully remote_written to for pruposes of safely truncating the WAL.
MetricValueCollector is only intended for use with Gauges and Counters.
func NewMetricValueCollector ¶
func NewMetricValueCollector(g prometheus.Gatherer, match string) *MetricValueCollector
NewMetricValueCollector creates a new MetricValueCollector.
type MockManager ¶ added in v0.5.0
type MockManager struct {
ListInstancesFunc func() map[string]ManagedInstance
ListConfigsFunc func() map[string]Config
ApplyConfigFunc func(Config) error
DeleteConfigFunc func(name string) error
StopFunc func()
}
MockManager exposes methods of the Manager interface as struct fields. Useful for tests.
func (MockManager) ApplyConfig ¶ added in v0.5.0
func (m MockManager) ApplyConfig(c Config) error
ApplyConfig implements Manager.
func (MockManager) DeleteConfig ¶ added in v0.5.0
func (m MockManager) DeleteConfig(name string) error
DeleteConfig implements Manager.
func (MockManager) ListConfigs ¶ added in v0.5.0
func (m MockManager) ListConfigs() map[string]Config
ListConfigs implements Manager.
func (MockManager) ListInstances ¶ added in v0.5.0
func (m MockManager) ListInstances() map[string]ManagedInstance
ListInstances implements Manager.
type NoOpInstance ¶ added in v0.5.0
type NoOpInstance struct{}
NoOpInstance implements the Instance interface in pkg/prom but does not do anything. Useful for tests.
func (NoOpInstance) TargetsActive ¶ added in v0.5.0
func (NoOpInstance) TargetsActive() map[string][]*scrape.Target
Source Files