Relocate JobStatistics from GraphQL resolver to JobRepository

This commit is contained in:
Jan Eitzinger 2023-02-13 13:53:24 +01:00
parent aef821fa7e
commit 51b2a7f30e
3 changed files with 202 additions and 204 deletions

View File

@ -16,8 +16,8 @@ import (
"github.com/ClusterCockpit/cc-backend/internal/metricdata"
"github.com/ClusterCockpit/cc-backend/internal/repository"
"github.com/ClusterCockpit/cc-backend/pkg/archive"
"github.com/ClusterCockpit/cc-backend/pkg/schema"
"github.com/ClusterCockpit/cc-backend/pkg/log"
"github.com/ClusterCockpit/cc-backend/pkg/schema"
)
// Partitions is the resolver for the partitions field.
@ -238,7 +238,7 @@ func (r *queryResolver) Jobs(ctx context.Context, filter []*model.JobFilter, pag
// JobsStatistics is the resolver for the jobsStatistics field.
func (r *queryResolver) JobsStatistics(ctx context.Context, filter []*model.JobFilter, groupBy *model.Aggregate) ([]*model.JobsStatistics, error) {
return r.jobsStatistics(ctx, filter, groupBy)
return r.Repo.JobsStatistics(ctx, filter, groupBy)
}
// JobsCount is the resolver for the jobsCount field.

View File

@ -6,207 +6,16 @@ package graph
import (
"context"
"database/sql"
"errors"
"fmt"
"math"
"time"
"github.com/99designs/gqlgen/graphql"
"github.com/ClusterCockpit/cc-backend/internal/graph/model"
"github.com/ClusterCockpit/cc-backend/internal/metricdata"
"github.com/ClusterCockpit/cc-backend/internal/repository"
"github.com/ClusterCockpit/cc-backend/pkg/archive"
"github.com/ClusterCockpit/cc-backend/pkg/schema"
"github.com/ClusterCockpit/cc-backend/pkg/log"
sq "github.com/Masterminds/squirrel"
"github.com/ClusterCockpit/cc-backend/pkg/schema"
)
// GraphQL validation should make sure that no unkown values can be specified.
var groupBy2column = map[model.Aggregate]string{
model.AggregateUser: "job.user",
model.AggregateProject: "job.project",
model.AggregateCluster: "job.cluster",
}
const ShortJobDuration int = 5 * 60
// Helper function for the jobsStatistics GraphQL query placed here so that schema.resolvers.go is not too full.
func (r *queryResolver) jobsStatistics(ctx context.Context, filter []*model.JobFilter, groupBy *model.Aggregate) ([]*model.JobsStatistics, error) {
// In case `groupBy` is nil (not used), the model.JobsStatistics used is at the key '' (empty string)
stats := map[string]*model.JobsStatistics{}
// `socketsPerNode` and `coresPerSocket` can differ from cluster to cluster, so we need to explicitly loop over those.
for _, cluster := range archive.Clusters {
for _, subcluster := range cluster.SubClusters {
corehoursCol := fmt.Sprintf("CAST(ROUND(SUM(job.duration * job.num_nodes * %d * %d) / 3600) as int)", subcluster.SocketsPerNode, subcluster.CoresPerSocket)
var query sq.SelectBuilder
if groupBy == nil {
query = sq.Select(
"''",
"COUNT(job.id)",
"CAST(ROUND(SUM(job.duration) / 3600) as int)",
corehoursCol,
).From("job")
} else {
col := groupBy2column[*groupBy]
query = sq.Select(
col,
"COUNT(job.id)",
"CAST(ROUND(SUM(job.duration) / 3600) as int)",
corehoursCol,
).From("job").GroupBy(col)
}
query = query.
Where("job.cluster = ?", cluster.Name).
Where("job.subcluster = ?", subcluster.Name)
query = repository.SecurityCheck(ctx, query)
for _, f := range filter {
query = repository.BuildWhereClause(f, query)
}
rows, err := query.RunWith(r.DB).Query()
if err != nil {
log.Warn("Error while querying DB for job statistics")
return nil, err
}
for rows.Next() {
var id sql.NullString
var jobs, walltime, corehours sql.NullInt64
if err := rows.Scan(&id, &jobs, &walltime, &corehours); err != nil {
log.Warn("Error while scanning rows")
return nil, err
}
if id.Valid {
if s, ok := stats[id.String]; ok {
s.TotalJobs += int(jobs.Int64)
s.TotalWalltime += int(walltime.Int64)
s.TotalCoreHours += int(corehours.Int64)
} else {
stats[id.String] = &model.JobsStatistics{
ID: id.String,
TotalJobs: int(jobs.Int64),
TotalWalltime: int(walltime.Int64),
TotalCoreHours: int(corehours.Int64),
}
}
}
}
}
}
if groupBy == nil {
query := sq.Select("COUNT(job.id)").From("job").Where("job.duration < ?", ShortJobDuration)
query = repository.SecurityCheck(ctx, query)
for _, f := range filter {
query = repository.BuildWhereClause(f, query)
}
if err := query.RunWith(r.DB).QueryRow().Scan(&(stats[""].ShortJobs)); err != nil {
log.Warn("Error while scanning rows for short job stats")
return nil, err
}
} else {
col := groupBy2column[*groupBy]
query := sq.Select(col, "COUNT(job.id)").From("job").Where("job.duration < ?", ShortJobDuration)
query = repository.SecurityCheck(ctx, query)
for _, f := range filter {
query = repository.BuildWhereClause(f, query)
}
rows, err := query.RunWith(r.DB).Query()
if err != nil {
log.Warn("Error while querying jobs for short jobs")
return nil, err
}
for rows.Next() {
var id sql.NullString
var shortJobs sql.NullInt64
if err := rows.Scan(&id, &shortJobs); err != nil {
log.Warn("Error while scanning rows for short jobs")
return nil, err
}
if id.Valid {
stats[id.String].ShortJobs = int(shortJobs.Int64)
}
}
}
// Calculating the histogram data is expensive, so only do it if needed.
// An explicit resolver can not be used because we need to know the filters.
histogramsNeeded := false
fields := graphql.CollectFieldsCtx(ctx, nil)
for _, col := range fields {
if col.Name == "histDuration" || col.Name == "histNumNodes" {
histogramsNeeded = true
}
}
res := make([]*model.JobsStatistics, 0, len(stats))
for _, stat := range stats {
res = append(res, stat)
id, col := "", ""
if groupBy != nil {
id = stat.ID
col = groupBy2column[*groupBy]
}
if histogramsNeeded {
var err error
value := fmt.Sprintf(`CAST(ROUND((CASE WHEN job.job_state = "running" THEN %d - job.start_time ELSE job.duration END) / 3600) as int) as value`, time.Now().Unix())
stat.HistDuration, err = r.jobsStatisticsHistogram(ctx, value, filter, id, col)
if err != nil {
log.Warn("Error while loading job statistics histogram: running jobs")
return nil, err
}
stat.HistNumNodes, err = r.jobsStatisticsHistogram(ctx, "job.num_nodes as value", filter, id, col)
if err != nil {
log.Warn("Error while loading job statistics histogram: num nodes")
return nil, err
}
}
}
return res, nil
}
// `value` must be the column grouped by, but renamed to "value". `id` and `col` can optionally be used
// to add a condition to the query of the kind "<col> = <id>".
func (r *queryResolver) jobsStatisticsHistogram(ctx context.Context, value string, filters []*model.JobFilter, id, col string) ([]*model.HistoPoint, error) {
query := sq.Select(value, "COUNT(job.id) AS count").From("job")
query = repository.SecurityCheck(ctx, query)
for _, f := range filters {
query = repository.BuildWhereClause(f, query)
}
if len(id) != 0 && len(col) != 0 {
query = query.Where(col+" = ?", id)
}
rows, err := query.GroupBy("value").RunWith(r.DB).Query()
if err != nil {
log.Error("Error while running query")
return nil, err
}
points := make([]*model.HistoPoint, 0)
for rows.Next() {
point := model.HistoPoint{}
if err := rows.Scan(&point.Value, &point.Count); err != nil {
log.Warn("Error while scanning rows")
return nil, err
}
points = append(points, &point)
}
return points, nil
}
const MAX_JOBS_FOR_ANALYSIS = 500
// Helper function for the rooflineHeatmap GraphQL query placed here so that schema.resolvers.go is not too full.

View File

@ -14,9 +14,11 @@ import (
"sync"
"time"
"github.com/99designs/gqlgen/graphql"
"github.com/ClusterCockpit/cc-backend/internal/auth"
"github.com/ClusterCockpit/cc-backend/internal/graph/model"
"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"
@ -67,7 +69,7 @@ func scanJob(row interface{ Scan(...interface{}) error }) (*schema.Job, error) {
if err := row.Scan(
&job.ID, &job.JobID, &job.User, &job.Project, &job.Cluster, &job.SubCluster, &job.StartTimeUnix, &job.Partition, &job.ArrayJobId,
&job.NumNodes, &job.NumHWThreads, &job.NumAcc, &job.Exclusive, &job.MonitoringStatus, &job.SMT, &job.State,
&job.Duration, &job.Walltime, &job.RawResources, /*&job.RawMetaData*/); err != nil {
&job.Duration, &job.Walltime, &job.RawResources /*&job.RawMetaData*/); err != nil {
log.Warn("Error while scanning rows")
return nil, err
}
@ -635,3 +637,190 @@ func (r *JobRepository) StopJobsExceedingWalltimeBy(seconds int) error {
}
return nil
}
// TODO: Move to config
const ShortJobDuration int = 5 * 60
// GraphQL validation should make sure that no unkown values can be specified.
var groupBy2column = map[model.Aggregate]string{
model.AggregateUser: "job.user",
model.AggregateProject: "job.project",
model.AggregateCluster: "job.cluster",
}
// Helper function for the jobsStatistics GraphQL query placed here so that schema.resolvers.go is not too full.
func (r *JobRepository) JobsStatistics(ctx context.Context, filter []*model.JobFilter, groupBy *model.Aggregate) ([]*model.JobsStatistics, error) {
// In case `groupBy` is nil (not used), the model.JobsStatistics used is at the key '' (empty string)
stats := map[string]*model.JobsStatistics{}
// `socketsPerNode` and `coresPerSocket` can differ from cluster to cluster, so we need to explicitly loop over those.
for _, cluster := range archive.Clusters {
for _, subcluster := range cluster.SubClusters {
corehoursCol := fmt.Sprintf("CAST(ROUND(SUM(job.duration * job.num_nodes * %d * %d) / 3600) as int)", subcluster.SocketsPerNode, subcluster.CoresPerSocket)
var query sq.SelectBuilder
if groupBy == nil {
query = sq.Select(
"''",
"COUNT(job.id)",
"CAST(ROUND(SUM(job.duration) / 3600) as int)",
corehoursCol,
).From("job")
} else {
col := groupBy2column[*groupBy]
query = sq.Select(
col,
"COUNT(job.id)",
"CAST(ROUND(SUM(job.duration) / 3600) as int)",
corehoursCol,
).From("job").GroupBy(col)
}
query = query.
Where("job.cluster = ?", cluster.Name).
Where("job.subcluster = ?", subcluster.Name)
query = SecurityCheck(ctx, query)
for _, f := range filter {
query = BuildWhereClause(f, query)
}
rows, err := query.RunWith(r.DB).Query()
if err != nil {
log.Warn("Error while querying DB for job statistics")
return nil, err
}
for rows.Next() {
var id sql.NullString
var jobs, walltime, corehours sql.NullInt64
if err := rows.Scan(&id, &jobs, &walltime, &corehours); err != nil {
log.Warn("Error while scanning rows")
return nil, err
}
if id.Valid {
if s, ok := stats[id.String]; ok {
s.TotalJobs += int(jobs.Int64)
s.TotalWalltime += int(walltime.Int64)
s.TotalCoreHours += int(corehours.Int64)
} else {
stats[id.String] = &model.JobsStatistics{
ID: id.String,
TotalJobs: int(jobs.Int64),
TotalWalltime: int(walltime.Int64),
TotalCoreHours: int(corehours.Int64),
}
}
}
}
}
}
if groupBy == nil {
query := sq.Select("COUNT(job.id)").From("job").Where("job.duration < ?", ShortJobDuration)
query = SecurityCheck(ctx, query)
for _, f := range filter {
query = BuildWhereClause(f, query)
}
if err := query.RunWith(r.DB).QueryRow().Scan(&(stats[""].ShortJobs)); err != nil {
log.Warn("Error while scanning rows for short job stats")
return nil, err
}
} else {
col := groupBy2column[*groupBy]
query := sq.Select(col, "COUNT(job.id)").From("job").Where("job.duration < ?", ShortJobDuration)
query = SecurityCheck(ctx, query)
for _, f := range filter {
query = BuildWhereClause(f, query)
}
rows, err := query.RunWith(r.DB).Query()
if err != nil {
log.Warn("Error while querying jobs for short jobs")
return nil, err
}
for rows.Next() {
var id sql.NullString
var shortJobs sql.NullInt64
if err := rows.Scan(&id, &shortJobs); err != nil {
log.Warn("Error while scanning rows for short jobs")
return nil, err
}
if id.Valid {
stats[id.String].ShortJobs = int(shortJobs.Int64)
}
}
}
// Calculating the histogram data is expensive, so only do it if needed.
// An explicit resolver can not be used because we need to know the filters.
histogramsNeeded := false
fields := graphql.CollectFieldsCtx(ctx, nil)
for _, col := range fields {
if col.Name == "histDuration" || col.Name == "histNumNodes" {
histogramsNeeded = true
}
}
res := make([]*model.JobsStatistics, 0, len(stats))
for _, stat := range stats {
res = append(res, stat)
id, col := "", ""
if groupBy != nil {
id = stat.ID
col = groupBy2column[*groupBy]
}
if histogramsNeeded {
var err error
value := fmt.Sprintf(`CAST(ROUND((CASE WHEN job.job_state = "running" THEN %d - job.start_time ELSE job.duration END) / 3600) as int) as value`, time.Now().Unix())
stat.HistDuration, err = r.jobsStatisticsHistogram(ctx, value, filter, id, col)
if err != nil {
log.Warn("Error while loading job statistics histogram: running jobs")
return nil, err
}
stat.HistNumNodes, err = r.jobsStatisticsHistogram(ctx, "job.num_nodes as value", filter, id, col)
if err != nil {
log.Warn("Error while loading job statistics histogram: num nodes")
return nil, err
}
}
}
return res, nil
}
// `value` must be the column grouped by, but renamed to "value". `id` and `col` can optionally be used
// to add a condition to the query of the kind "<col> = <id>".
func (r *JobRepository) jobsStatisticsHistogram(ctx context.Context, value string, filters []*model.JobFilter, id, col string) ([]*model.HistoPoint, error) {
query := sq.Select(value, "COUNT(job.id) AS count").From("job")
query = SecurityCheck(ctx, query)
for _, f := range filters {
query = BuildWhereClause(f, query)
}
if len(id) != 0 && len(col) != 0 {
query = query.Where(col+" = ?", id)
}
rows, err := query.GroupBy("value").RunWith(r.DB).Query()
if err != nil {
log.Error("Error while running query")
return nil, err
}
points := make([]*model.HistoPoint, 0)
for rows.Next() {
point := model.HistoPoint{}
if err := rows.Scan(&point.Value, &point.Count); err != nil {
log.Warn("Error while scanning rows")
return nil, err
}
points = append(points, &point)
}
return points, nil
}