Introduce metricDataDispatcher

Does not compile yet
This commit is contained in:
Jan Eitzinger 2024-08-27 16:44:16 +02:00
parent 56ebb301ca
commit f914a312f5
Signed by: moebiusband
GPG Key ID: 2574BA29B90D6DD5
3 changed files with 138 additions and 114 deletions

View File

@ -181,7 +181,7 @@ func main() {
log.Fatalf("failed to initialize archive: %s", err.Error()) log.Fatalf("failed to initialize archive: %s", err.Error())
} }
if err := metricdata.Init(config.Keys.DisableArchive); err != nil { if err := metricdata.Init(); err != nil {
log.Fatalf("failed to initialize metricdata repository: %s", err.Error()) log.Fatalf("failed to initialize metricdata repository: %s", err.Error())
} }

View File

@ -0,0 +1,131 @@
// Copyright (C) NHR@FAU, University Erlangen-Nuremberg.
// All rights reserved.
// Use of this source code is governed by a MIT-style
// license that can be found in the LICENSE file.
package metricDataDispatcher
import (
"context"
"fmt"
"time"
"github.com/ClusterCockpit/cc-backend/internal/config"
"github.com/ClusterCockpit/cc-backend/internal/metricdata"
"github.com/ClusterCockpit/cc-backend/pkg/archive"
"github.com/ClusterCockpit/cc-backend/pkg/log"
"github.com/ClusterCockpit/cc-backend/pkg/lrucache"
"github.com/ClusterCockpit/cc-backend/pkg/schema"
)
var cache *lrucache.Cache = lrucache.New(128 * 1024 * 1024)
func cacheKey(
job *schema.Job,
metrics []string,
scopes []schema.MetricScope,
) string {
// Duration and StartTime do not need to be in the cache key as StartTime is less unique than
// job.ID and the TTL of the cache entry makes sure it does not stay there forever.
return fmt.Sprintf("%d(%s):[%v],[%v]",
job.ID, job.State, metrics, scopes)
}
// Fetches the metric data for a job.
func LoadData(job *schema.Job,
metrics []string,
scopes []schema.MetricScope,
ctx context.Context,
) (schema.JobData, error) {
data := cache.Get(cacheKey(job, metrics, scopes), func() (_ interface{}, ttl time.Duration, size int) {
var jd schema.JobData
var err error
if job.State == schema.JobStateRunning ||
job.MonitoringStatus == schema.MonitoringStatusRunningOrArchiving ||
!config.Keys.DisableArchive {
repo, ok := metricdata.GetMetricDataRepo(job.Cluster)
if !ok {
return fmt.Errorf("METRICDATA/METRICDATA > no metric data repository configured for '%s'", job.Cluster), 0, 0
}
if scopes == nil {
scopes = append(scopes, schema.MetricScopeNode)
}
if metrics == nil {
cluster := archive.GetCluster(job.Cluster)
for _, mc := range cluster.MetricConfig {
metrics = append(metrics, mc.Name)
}
}
jd, err = repo.LoadData(job, metrics, scopes, ctx)
if err != nil {
if len(jd) != 0 {
log.Warnf("partial error: %s", err.Error())
// return err, 0, 0 // Reactivating will block archiving on one partial error
} else {
log.Error("Error while loading job data from metric repository")
return err, 0, 0
}
}
size = jd.Size()
} else {
jd, err = archive.GetHandle().LoadJobData(job)
if err != nil {
log.Error("Error while loading job data from archive")
return err, 0, 0
}
// Avoid sending unrequested data to the client:
if metrics != nil || scopes != nil {
if metrics == nil {
metrics = make([]string, 0, len(jd))
for k := range jd {
metrics = append(metrics, k)
}
}
res := schema.JobData{}
for _, metric := range metrics {
if perscope, ok := jd[metric]; ok {
if len(perscope) > 1 {
subset := make(map[schema.MetricScope]*schema.JobMetric)
for _, scope := range scopes {
if jm, ok := perscope[scope]; ok {
subset[scope] = jm
}
}
if len(subset) > 0 {
perscope = subset
}
}
res[metric] = perscope
}
}
jd = res
}
size = jd.Size()
}
ttl = 5 * time.Hour
if job.State == schema.JobStateRunning {
ttl = 2 * time.Minute
}
prepareJobData(jd, scopes)
return jd, ttl, size
})
if err, ok := data.(error); ok {
log.Error("Error in returned dataset")
return nil, err
}
return data.(schema.JobData), nil
}

View File

@ -13,7 +13,6 @@ import (
"github.com/ClusterCockpit/cc-backend/internal/config" "github.com/ClusterCockpit/cc-backend/internal/config"
"github.com/ClusterCockpit/cc-backend/pkg/archive" "github.com/ClusterCockpit/cc-backend/pkg/archive"
"github.com/ClusterCockpit/cc-backend/pkg/log" "github.com/ClusterCockpit/cc-backend/pkg/log"
"github.com/ClusterCockpit/cc-backend/pkg/lrucache"
"github.com/ClusterCockpit/cc-backend/pkg/schema" "github.com/ClusterCockpit/cc-backend/pkg/schema"
) )
@ -34,10 +33,7 @@ type MetricDataRepository interface {
var metricDataRepos map[string]MetricDataRepository = map[string]MetricDataRepository{} var metricDataRepos map[string]MetricDataRepository = map[string]MetricDataRepository{}
var useArchive bool func Init() error {
func Init(disableArchive bool) error {
useArchive = !disableArchive
for _, cluster := range config.Keys.Clusters { for _, cluster := range config.Keys.Clusters {
if cluster.MetricDataRepository != nil { if cluster.MetricDataRepository != nil {
var kind struct { var kind struct {
@ -72,106 +68,14 @@ func Init(disableArchive bool) error {
return nil return nil
} }
var cache *lrucache.Cache = lrucache.New(128 * 1024 * 1024) func GetMetricDataRepo(cluster string) MetricDataRepository {
repo, ok := metricDataRepos[cluster]
// Fetches the metric data for a job.
func LoadData(job *schema.Job,
metrics []string,
scopes []schema.MetricScope,
ctx context.Context,
) (schema.JobData, error) {
data := cache.Get(cacheKey(job, metrics, scopes), func() (_ interface{}, ttl time.Duration, size int) {
var jd schema.JobData
var err error
if job.State == schema.JobStateRunning ||
job.MonitoringStatus == schema.MonitoringStatusRunningOrArchiving ||
!useArchive {
repo, ok := metricDataRepos[job.Cluster]
if !ok { if !ok {
return fmt.Errorf("METRICDATA/METRICDATA > no metric data repository configured for '%s'", job.Cluster), 0, 0 return fmt.Errorf("METRICDATA/METRICDATA > no metric data repository configured for '%s'", job.Cluster), 0, 0
} }
if scopes == nil { return repo
scopes = append(scopes, schema.MetricScopeNode)
}
if metrics == nil {
cluster := archive.GetCluster(job.Cluster)
for _, mc := range cluster.MetricConfig {
metrics = append(metrics, mc.Name)
}
}
jd, err = repo.LoadData(job, metrics, scopes, ctx)
if err != nil {
if len(jd) != 0 {
log.Warnf("partial error: %s", err.Error())
// return err, 0, 0 // Reactivating will block archiving on one partial error
} else {
log.Error("Error while loading job data from metric repository")
return err, 0, 0
}
}
size = jd.Size()
} else {
jd, err = archive.GetHandle().LoadJobData(job)
if err != nil {
log.Error("Error while loading job data from archive")
return err, 0, 0
}
// Avoid sending unrequested data to the client:
if metrics != nil || scopes != nil {
if metrics == nil {
metrics = make([]string, 0, len(jd))
for k := range jd {
metrics = append(metrics, k)
}
}
res := schema.JobData{}
for _, metric := range metrics {
if perscope, ok := jd[metric]; ok {
if len(perscope) > 1 {
subset := make(map[schema.MetricScope]*schema.JobMetric)
for _, scope := range scopes {
if jm, ok := perscope[scope]; ok {
subset[scope] = jm
}
}
if len(subset) > 0 {
perscope = subset
}
}
res[metric] = perscope
}
}
jd = res
}
size = jd.Size()
}
ttl = 5 * time.Hour
if job.State == schema.JobStateRunning {
ttl = 2 * time.Minute
}
prepareJobData(jd, scopes)
return jd, ttl, size
})
if err, ok := data.(error); ok {
log.Error("Error in returned dataset")
return nil, err
}
return data.(schema.JobData), nil
} }
// Used for the jobsFootprint GraphQL-Query. TODO: Rename/Generalize. // Used for the jobsFootprint GraphQL-Query. TODO: Rename/Generalize.
@ -249,17 +153,6 @@ func LoadNodeData(
return data, nil return data, nil
} }
func cacheKey(
job *schema.Job,
metrics []string,
scopes []schema.MetricScope,
) string {
// Duration and StartTime do not need to be in the cache key as StartTime is less unique than
// job.ID and the TTL of the cache entry makes sure it does not stay there forever.
return fmt.Sprintf("%d(%s):[%v],[%v]",
job.ID, job.State, metrics, scopes)
}
// For /monitoring/job/<job> and some other places, flops_any and mem_bw need // For /monitoring/job/<job> and some other places, flops_any and mem_bw need
// to be available at the scope 'node'. If a job has a lot of nodes, // to be available at the scope 'node'. If a job has a lot of nodes,
// statisticsSeries should be available so that a min/median/max Graph can be // statisticsSeries should be available so that a min/median/max Graph can be