custom-kube-scheduler/kubernetes-v1.15.4/pkg/scheduler/algorithm/priorities/custom_resource_allocation.go

/*
Copyright 2020 Achilleas Tzenetopoulos.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/

package priorities

import (
	"encoding/json"
	"fmt"
	"strings"

	_ "github.com/go-sql-driver/mysql"
	client "github.com/influxdata/influxdb1-client/v2"
	"k8s.io/klog"
)

var (
	customResourcePriority = &CustomAllocationPriority{"CustomResourceAllocation", customResourceScorer}
	//customResourcePriority = &CustomAllocationPriority{"CustomRequestedPriority", customResourceScorer}
	// LeastRequestedPriorityMap is a priority function that favors nodes with fewer requested resources.
	// It calculates the percentage of memory and CPU requested by pods scheduled on the node, and
	// prioritizes based on the minimum of the average of the fraction of requested to capacity.
	//
	// Details:
	// (cpu((capacity-sum(requested))*10/capacity) + memory((capacity-sum(requested))*10/capacity))/2
	CustomRequestedPriorityMap = customResourcePriority.PriorityMap
)

func customScoreFn(si scorerInput) float64 {
	return si.metrics["ipc"] / si.metrics["mem_read"] * si.metrics["mem_write"]
}

func onlyIPC(metrics map[string]float64) float64 {
	return metrics["ipc"]
}

func onlyL3(metrics map[string]float64) float64 {
	return 1 / metrics["l3m"]
}

func onlyNrg(metrics map[string]float64) float64 {
	return 1 / metrics["procnrg"]
}

func calculateScore(si scorerInput,
	logicFn func(scorerInput) float64) float64 {

	res := logicFn(si)
	//klog.Infof("Has score (in float) %v\n", res)

	return res
}

func calculateWeightedAverage(response *client.Response,
	numberOfRows, numberOfMetrics int) (map[string]float64, error) {
	// initialize the metrics map with a constant size
	metrics := make(map[string]float64, numberOfMetrics)
	rows := response.Results[0].Series[0]
	for i := 1; i < len(rows.Columns); i++ {
		for j := 0; j < numberOfRows; j++ {
			val, err := rows.Values[j][i].(json.Number).Float64()
			if err != nil {
				klog.Infof("Error while calculating %v", rows.Columns[i])
				return nil, err
			}
			metrics[rows.Columns[i]] += val * float64(numberOfRows-j)
		}
		metrics[rows.Columns[i]] = metrics[rows.Columns[i]] / float64((numberOfRows * (numberOfRows + 1) / 2))
		//klog.Infof("%v : %v", rows.Columns[i], metrics[rows.Columns[i]])
	}
	// TODO better handling for the returning errors
	return metrics, nil
}

func customScoreInfluxDB(metrics []string, uuid string, socket,
	numberOfRows int, cfg Config, c client.Client) (map[string]float64, error) {

	// calculate the number of rows needed
	// i.e. 20sec / 0.5s interval => 40rows
	//numberOfRows := int(float32(time) / cfg.MonitoringSpecs.TimeInterval)
	// merge all the required columns
	columns := strings.Join(metrics, ", ")
	// build the coommand
	var command strings.Builder
	fmt.Fprintf(&command, "SELECT %s from socket_metrics where uuid = '%s' and socket_id='%d' order by time desc limit %d", columns, uuid, socket, numberOfRows)
	q := client.NewQuery(command.String(), cfg.Database.Name, "")
	response, err := c.Query(q)
	if err != nil {
		klog.Infof("Error while executing the query: %v", err.Error())
		return nil, err
	}
	// Calculate the average for the metrics provided
	return calculateWeightedAverage(response, numberOfRows, len(metrics))
}

func customResourceScorer(nodeName string) (float64, error) {
	//return (customRequestedScore(requested.MilliCPU, allocable.MilliCPU) +
	//customRequestedScore(requested.Memory, allocable.Memory)) / 2

	//read database information
	var cfg Config
	err := readFile(&cfg, "/etc/kubernetes/scheduler-monitoringDB.yaml")
	if err != nil {
		return 0, err
	}

	/*-------------------------------------
	//TODO read also nodes to uuid mappings for EVOLVE
	-------------------------------------*/

	// InfluxDB
	c, err := connectToInfluxDB(cfg)
	if err != nil {
		return 0, err
	}
	// close the connection in the end of execution
	defer c.Close()

	//Get the uuid of this node in order to query in the database
	curr_uuid, ok := nodes[nodeName]
	socket, _ := sockets[nodeName]
	cores, _ := cores[nodeName]

	if ok {

		metrics := []string{"c6res"}
		time := 20

		numberOfRows := int(float32(time) / cfg.MonitoringSpecs.TimeInterval)

		// Define Core availability
		r, err := queryInfluxDbCores(metrics, curr_uuid, socket, numberOfRows, cfg, c, cores)
		if err != nil {
			klog.Infof("Error in querying or calculating core availability in the first stage: %v", err.Error())
		}
		average, err := calculateWeightedAverageCores(r, numberOfRows, len(metrics), len(cores))
		if err != nil {
			klog.Infof("Error defining core availability")
		}

		// Select Socket
		results, err := customScoreInfluxDB([]string{"ipc", "mem_read", "mem_write"}, curr_uuid, socket, numberOfRows, cfg, c)
		if err != nil {
			klog.Infof("Error in querying or calculating average for the custom score in the first stage: %v", err.Error())
			return 0, nil
		}
		res := calculateScore(scorerInput{metrics: results}, customScoreFn)

		if sum := average["c6res"] * float64(len(cores)); sum > 1 {
			res = res * sum
		}
		// Select Node

		klog.Infof("Node name %s, has score %v\n", nodeName, res)
		return res, nil
	} else {
		klog.Infof("Error finding the uuid: %v", ok)
		return 0, nil
	}

}

// WARNING
// c6res is not a dependable metric for isnpecting core availability
// Some Systems use higher core states (e.g c7res)
// func findAvailability(response *client.Response, numberOfMetrics, numberOfRows, numberOfCores int, floor float64) (map[string]float64, error) {
// 	// initialize the metrics map with a constant size
// 	metrics := make(map[string]float64, numberOfMetrics)
// 	rows := response.Results[0].Series[0]
// 	for i := 1; i < len(rows.Columns); i++ {
// 		//klog.Infof("Name of column %v : %v\nrange of values: %v\nnumber of rows: %v\nnumber of cores %v\n", i, rows.Columns[i], len(rows.Values), numberOfRows, numberOfCores)
// 		for j := 0; j < numberOfRows; j++ {
// 			//avg, max := 0.0, 0.0
// 			for k := 0; k < numberOfCores; k++ {
// 				val, err := rows.Values[j*numberOfCores+k][i].(json.Number).Float64()
// 				if err != nil {
// 					klog.Infof("Error while calculating %v", rows.Columns[i])
// 					return false, err
// 				}
// 				// if val > floor {
// 				// 	return true, nil
// 				// }
// 				// sum += val

// 				//avg += val / float64(numberOfCores)
// 				avg += val
// 			}
// 			metrics[rows.Columns[i]] += avg * float64(numberOfRows-j)
// 		}
// 		metrics[rows.Columns[i]] = metrics[rows.Columns[i]] / float64((numberOfRows * (numberOfRows + 1) / 2))
// 		if metrics[row.Columns[i]] > 1 {
// 			return true, nil
// 		}
// 		//klog.Infof("%v : %v", rows.Columns[i], metrics[rows.Columns[i]])
// 	}
// 	// TODO better handling for the returning errors
// 	return false, nil
// }