K8sAutoscaleNode triggers autoscale events for a resource on a Kubernetes cluster. The node also outputs points for the triggered events.
Example:
// Target 100 requests per second per host var target = 100.0 var min = 1 var max = 100 var period = 5m var every = period stream |from() .measurement('requests') .groupBy('host', 'deployment') .truncate(1s) |derivative('value') .as('requests_per_second') .unit(1s) .nonNegative() |groupBy('deployment') |sum('requests_per_second') .as('total_requests') |window() .period(period) .every(every) |mean('total_requests') .as('total_requests') |k8sAutoscale() // Get the name of the deployment from the 'deployment' tag. .resourceNameTag('deployment') .min(min) .max(max) // Set the desired number of replicas based on target. .replicas(lambda: int(ceil("total_requests" / target))) |influxDBOut() .database('deployments') .measurement('scale_events') .precision('s')
The above example computes the requests per second by deployment and host. Then the total_requests per second across all hosts is computed per deployment. Using the mean of the total_requests over the last time period a desired number of replicas is computed based on the target number of request per second per host.
If the desired number of replicas has changed, Kapacitor makes the appropriate API call to Kubernetes to update the replicas spec.
Any time the k8sAutoscale node changes a replica count, it emits a point. The point is tagged with the namespace, kind and resource name, using the NamespaceTag, KindTag, and ResourceTag properties respectively. In addition the group by tags will be preserved on the emitted point. The point contains two fields: old
, and new
representing change in the replicas.
Available Statistics:
Property methods modify state on the calling node. They do not add another node to the pipeline, and always return a reference to the calling node. Property methods are marked using the .
operator.
Cluster is the name of the Kubernetes cluster to use.
node.cluster(value string)
CurrentField is the name of a field into which the current replica count will be set as an int. If empty no field will be set. Useful for computing deltas on the current state.
Example:
|k8sAutoscale() .currentField('replicas') // Increase the replicas by 1 if the qps is over the threshold .replicas(lambda: if("qps" > threshold, "replicas" + 1, "replicas"))
node.currentField(value string)
Only one decrease event can be triggered per resource every DecreaseCooldown interval.
node.decreaseCooldown(value time.Duration)
Only one increase event can be triggered per resource every IncreaseCooldown interval.
node.increaseCooldown(value time.Duration)
Kind is the type of resources to autoscale. Currently only "deployments", "replicasets" and "replicationcontrollers" are supported. Default: "deployments"
node.kind(value string)
KindTag is the name of a tag to use when tagging emitted points with the kind. If empty the point will not be tagged with the resource. Default: kind
node.kindTag(value string)
The maximum scale factor to set. If 0 then there is no upper limit. Default: 0, a.k.a no limit.
node.max(value int64)
The minimum scale factor to set. Default: 1
node.min(value int64)
Namespace is the namespace of the resource, if empty the default namespace will be used.
node.namespace(value string)
NamespaceTag is the name of a tag to use when tagging emitted points with the namespace. If empty the point will not be tagged with the resource. Default: namespace
node.namespaceTag(value string)
Replicas is a lambda expression that should evaluate to the desired number of replicas for the resource.
node.replicas(value ast.LambdaNode)
ResourceName is the name of the resource to autoscale.
node.resourceName(value string)
ResourceNameTag is the name of a tag that names the resource to autoscale.
node.resourceNameTag(value string)
ResourceTag is the name of a tag to use when tagging emitted points the resource. If empty the point will not be tagged with the resource. Default: resource
node.resourceTag(value string)
Chaining methods create a new node in the pipeline as a child of the calling node. They do not modify the calling node. Chaining methods are marked using the |
operator.
Create an alert node, which can trigger alerts.
node|alert()
Returns: AlertNode
Select the bottom num
points for field
and sort by any extra tags or fields.
node|bottom(num int64, field string, fieldsAndTags ...string)
Returns: InfluxQLNode
Combine this node with itself. The data are combined on timestamp.
node|combine(expressions ...ast.LambdaNode)
Returns: CombineNode
Count the number of points.
node|count(field string)
Returns: InfluxQLNode
Compute a cumulative sum of each point that is received. A point is emitted for every point collected.
node|cumulativeSum(field string)
Returns: InfluxQLNode
Helper function for creating an alert on low throughput, a.k.a. deadman's switch.
Example:
var data = stream |from()... // Trigger critical alert if the throughput drops below 100 points per 10s and checked every 10s. data |deadman(100.0, 10s) //Do normal processing of data data...
The above is equivalent to this Example:
var data = stream |from()... // Trigger critical alert if the throughput drops below 100 points per 10s and checked every 10s. data |stats(10s) .align() |derivative('emitted') .unit(10s) .nonNegative() |alert() .id('node \'stream0\' in task \'{{ .TaskName }}\'') .message('{{ .ID }} is {{ if eq .Level "OK" }}alive{{ else }}dead{{ end }}: {{ index .Fields "emitted" | printf "%0.3f" }} points/10s.') .crit(lambda: "emitted" <= 100.0) //Do normal processing of data data...
The id
and message
alert properties can be configured globally via the 'deadman' configuration section.
Since the AlertNode is the last piece it can be further modified as usual. Example:
var data = stream |from()... // Trigger critical alert if the throughput drops below 100 points per 10s and checked every 10s. data |deadman(100.0, 10s) .slack() .channel('#dead_tasks') //Do normal processing of data data...
You can specify additional lambda expressions to further constrain when the deadman's switch is triggered. Example:
var data = stream |from()... // Trigger critical alert if the throughput drops below 100 points per 10s and checked every 10s. // Only trigger the alert if the time of day is between 8am-5pm. data |deadman(100.0, 10s, lambda: hour("time") >= 8 AND hour("time") <= 17) //Do normal processing of data data...
node|deadman(threshold float64, interval time.Duration, expr ...ast.LambdaNode)
Returns: AlertNode
Create a node that can set defaults for missing tags or fields.
node|default()
Returns: DefaultNode
Create a node that can delete tags or fields.
node|delete()
Returns: DeleteNode
Create a new node that computes the derivative of adjacent points.
node|derivative(field string)
Returns: DerivativeNode
Compute the difference between points independent of elapsed time.
node|difference(field string)
Returns: InfluxQLNode
Produce batch of only the distinct points.
node|distinct(field string)
Returns: InfluxQLNode
Compute the elapsed time between points
node|elapsed(field string, unit time.Duration)
Returns: InfluxQLNode
Create an eval node that will evaluate the given transformation function to each data point. A list of expressions may be provided and will be evaluated in the order they are given. The results are available to later expressions.
node|eval(expressions ...ast.LambdaNode)
Returns: EvalNode
Select the first point.
node|first(field string)
Returns: InfluxQLNode
Flatten points with similar times into a single point.
node|flatten()
Returns: FlattenNode
Group the data by a set of tags.
Can pass literal * to group by all dimensions. Example:
|groupBy(*)
node|groupBy(tag ...interface{})
Returns: GroupByNode
Compute the holt-winters (https://docs.influxdata.com/influxdb/latest/query_language/functions/#holt-winters) forecast of a data set.
node|holtWinters(field string, h int64, m int64, interval time.Duration)
Returns: InfluxQLNode
Compute the holt-winters (https://docs.influxdata.com/influxdb/latest/query_language/functions/#holt-winters) forecast of a data set. This method also outputs all the points used to fit the data in addition to the forecasted data.
node|holtWintersWithFit(field string, h int64, m int64, interval time.Duration)
Returns: InfluxQLNode
Create an HTTP output node that caches the most recent data it has received. The cached data are available at the given endpoint. The endpoint is the relative path from the API endpoint of the running task. For example, if the task endpoint is at /kapacitor/v1/tasks/<task_id>
and endpoint is top10
, then the data can be requested from /kapacitor/v1/tasks/<task_id>/top10
.
node|httpOut(endpoint string)
Returns: HTTPOutNode
Creates an HTTP Post node that POSTS received data to the provided HTTP endpoint. HttpPost expects 0 or 1 arguments. If 0 arguments are provided, you must specify an endpoint property method.
node|httpPost(url ...string)
Returns: HTTPPostNode
Create an influxdb output node that will store the incoming data into InfluxDB.
node|influxDBOut()
Returns: InfluxDBOutNode
Join this node with other nodes. The data are joined on timestamp.
node|join(others ...Node)
Returns: JoinNode
Create a node that can trigger autoscale events for a kubernetes cluster.
node|k8sAutoscale()
Returns: K8sAutoscaleNode
Create an kapacitor loopback node that will send data back into Kapacitor as a stream.
node|kapacitorLoopback()
Returns: KapacitorLoopbackNode
Select the last point.
node|last(field string)
Returns: InfluxQLNode
Create a node that logs all data it receives.
node|log()
Returns: LogNode
Compute the mean of the data.
node|mean(field string)
Returns: InfluxQLNode
Compute the median of the data. Note, this method is not a selector, if you want the median point use .percentile(field, 50.0)
.
node|median(field string)
Returns: InfluxQLNode
Compute the mode of the data.
node|mode(field string)
Returns: InfluxQLNode
Compute a moving average of the last window points. No points are emitted until the window is full.
node|movingAverage(field string, window int64)
Returns: InfluxQLNode
Select a point at the given percentile. This is a selector function, no interpolation between points is performed.
node|percentile(field string, percentile float64)
Returns: InfluxQLNode
Create a new node that samples the incoming points or batches.
One point will be emitted every count or duration specified.
node|sample(rate interface{})
Returns: SampleNode
Create a new node that shifts the incoming points or batches in time.
node|shift(shift time.Duration)
Returns: ShiftNode
Compute the difference between min
and max
points.
node|spread(field string)
Returns: InfluxQLNode
Create a node that tracks number of consecutive points in a given state.
node|stateCount(expression ast.LambdaNode)
Returns: StateCountNode
Create a node that tracks duration in a given state.
node|stateDuration(expression ast.LambdaNode)
Returns: StateDurationNode
Create a new stream of data that contains the internal statistics of the node. The interval represents how often to emit the statistics based on real time. This means the interval time is independent of the times of the data points the source node is receiving.
node|stats(interval time.Duration)
Returns: StatsNode
Compute the standard deviation.
node|stddev(field string)
Returns: InfluxQLNode
Compute the sum of all values.
node|sum(field string)
Returns: InfluxQLNode
Select the top num
points for field
and sort by any extra tags or fields.
node|top(num int64, field string, fieldsAndTags ...string)
Returns: InfluxQLNode
Perform the union of this node and all other given nodes.
node|union(node ...Node)
Returns: UnionNode
Create a new node that filters the data stream by a given expression.
node|where(expression ast.LambdaNode)
Returns: WhereNode
Create a new node that windows the stream by time.
NOTE: Window can only be applied to stream edges.
node|window()
Returns: WindowNode
© 2015 InfluxData, Inc.
Licensed under the MIT license.
https://docs.influxdata.com/kapacitor/v1.3/nodes/k8s_autoscale_node/