Files
poseidon/internal/runner/kubernetes_manager.go
2024-09-18 10:43:38 +02:00

126 lines
4.1 KiB
Go

package runner
import (
"context"
"fmt"
"github.com/openHPI/poseidon/internal/kubernetes"
"github.com/openHPI/poseidon/internal/nomad"
"github.com/openHPI/poseidon/pkg/dto"
"github.com/openHPI/poseidon/pkg/storage"
"github.com/openHPI/poseidon/pkg/util"
appv1 "k8s.io/api/apps/v1"
"strconv"
"time"
)
type KubernetesRunnerManager struct {
*AbstractManager
apiClient kubernetes.ExecutorAPI
reloadingEnvironment storage.Storage[*alertData]
}
func NewKubernetesRunnerManager(apiClient *kubernetes.ExecutorAPI, ctx context.Context) *KubernetesRunnerManager {
return &KubernetesRunnerManager{
AbstractManager: NewAbstractManager(ctx),
apiClient: *apiClient,
reloadingEnvironment: storage.NewLocalStorage[*alertData](),
}
}
// Load recovers all runners for all existing environments.
func (k *KubernetesRunnerManager) Load() {
log.Info("Loading runners")
newUsedRunners := storage.NewLocalStorage[Runner]()
for _, environment := range k.ListEnvironments() {
usedRunners, err := k.loadEnvironment(environment)
if err != nil {
log.WithError(err).WithField(dto.KeyEnvironmentID, environment.ID().ToString()).
Warn("Failed loading environment. Skipping...")
continue
}
for _, r := range usedRunners.List() {
newUsedRunners.Add(r.ID(), r)
}
}
// TODO MISSING IMPLEMENTATION
//k.updateUsedRunners(newUsedRunners, true)
}
func (k *KubernetesRunnerManager) loadEnvironment(environment ExecutionEnvironment) (used storage.Storage[Runner], err error) {
used = storage.NewLocalStorage[Runner]()
runnerJobs, err := k.apiClient.LoadRunnerJobs(environment.ID())
if err != nil {
return nil, fmt.Errorf("failed fetching the runner jobs: %w", err)
}
for _, job := range runnerJobs {
r, isUsed, err := k.loadSingleJob(job, environment)
if err != nil {
log.WithError(err).WithField(dto.KeyEnvironmentID, environment.ID().ToString()).
WithField("used", isUsed).Warn("Failed loading job. Skipping...")
continue
} else if isUsed {
used.Add(r.ID(), r)
}
}
err = environment.ApplyPrewarmingPoolSize()
if err != nil {
return used, fmt.Errorf("couldn't scale environment: %w", err)
}
return used, nil
}
func (k *KubernetesRunnerManager) loadSingleJob(deployment *appv1.Deployment, environment ExecutionEnvironment) (r Runner, isUsed bool, err error) {
configTaskGroup := deployment.Spec.Template
if err != nil {
return nil, false, fmt.Errorf("%w, %s", nomad.ErrorMissingTaskGroup, deployment.Name)
}
isUsed = configTaskGroup.Annotations[nomad.ConfigMetaUsedKey] == nomad.ConfigMetaUsedValue
portMappings, err := k.apiClient.LoadRunnerPortMappings(deployment.Name)
if err != nil {
return nil, false, fmt.Errorf("error loading runner portMappings: %w", err)
}
newJob := NewKubernetesDeployment(deployment.Name, portMappings, k.apiClient, k.onRunnerDestroyed)
log.WithField("isUsed", isUsed).WithField(dto.KeyRunnerID, newJob.ID()).Debug("Recovered Runner")
if isUsed {
timeout, err := strconv.Atoi(configTaskGroup.ObjectMeta.Annotations[nomad.ConfigMetaTimeoutKey])
if err != nil {
log.WithField(dto.KeyRunnerID, newJob.ID()).WithError(err).Warn("failed loading timeout from meta values")
timeout = int(nomad.RunnerTimeoutFallback.Seconds())
go k.markRunnerAsUsed(newJob, timeout)
}
newJob.SetupTimeout(time.Duration(timeout) * time.Second)
} else {
environment.AddRunner(newJob)
}
return newJob, isUsed, nil
}
func (k *KubernetesRunnerManager) markRunnerAsUsed(runner Runner, timeoutDuration int) {
err := util.RetryExponential(func() (err error) {
if err = k.apiClient.MarkRunnerAsUsed(runner.ID(), timeoutDuration); err != nil {
err = fmt.Errorf("cannot mark runner as used: %w", err)
}
return
})
if err != nil {
log.WithError(err).WithField(dto.KeyRunnerID, runner.ID()).Error("cannot mark runner as used")
err := k.Return(runner)
if err != nil {
log.WithError(err).WithField(dto.KeyRunnerID, runner.ID()).Error("can't mark runner as used and can't return runner")
}
}
}
func (k *KubernetesRunnerManager) onRunnerDestroyed(r Runner) error {
k.usedRunners.Delete(r.ID())
environment, ok := k.GetEnvironment(r.Environment())
if ok {
environment.DeleteRunner(r.ID())
}
return nil
}