Add async archiving option; Move REST-API to new package

This commit is contained in:
Lou Knauer 2021-12-16 09:35:03 +01:00
parent 9c5c8a05e2
commit 7fcc39a144
6 changed files with 358 additions and 174 deletions

View File

@ -12,7 +12,7 @@ git clone --recursive git@github.com:ClusterCockpit/cc-jobarchive.git
cd ./cc-jobarchive/frontend
yarn install
export CCFRONTEND_ROLLUP_INTRO='
const JOBVIEW_URL = job => `/monitoring/job/${job.jobId}`;
const JOBVIEW_URL = job => `/monitoring/job/${job.id}`;
const USERVIEW_URL = userId => `/monitoring/user/${userId}`;
const TAG_URL = tag => `/monitoring/jobs/?tag=${tag.id}`;
'
@ -52,4 +52,3 @@ This project uses [gqlgen](https://github.com/99designs/gqlgen) for the GraphQL
- [ ] Write more TODOs
- [ ] Caching
- [ ] Generate JWTs based on the provided keys

276
api/rest.go Normal file
View File

@ -0,0 +1,276 @@
package api
import (
"context"
"database/sql"
"encoding/json"
"fmt"
"log"
"net/http"
"strings"
"github.com/ClusterCockpit/cc-jobarchive/config"
"github.com/ClusterCockpit/cc-jobarchive/graph"
"github.com/ClusterCockpit/cc-jobarchive/graph/model"
"github.com/ClusterCockpit/cc-jobarchive/metricdata"
sq "github.com/Masterminds/squirrel"
"github.com/gorilla/mux"
"github.com/jmoiron/sqlx"
)
type RestApi struct {
DB *sqlx.DB
Resolver *graph.Resolver
AsyncArchiving bool
}
func (api *RestApi) MountRoutes(r *mux.Router) {
r.HandleFunc("/api/jobs/start_job/", api.startJob).Methods(http.MethodPost, http.MethodPut)
r.HandleFunc("/api/jobs/stop_job/", api.stopJob).Methods(http.MethodPost, http.MethodPut)
r.HandleFunc("/api/jobs/stop_job/{id}", api.stopJob).Methods(http.MethodPost, http.MethodPut)
r.HandleFunc("/api/jobs/{id}", api.getJob).Methods(http.MethodGet)
r.HandleFunc("/api/jobs/tag_job/{id}", api.tagJob).Methods(http.MethodPost, http.MethodPatch)
}
type StartJobApiRequest struct {
JobId int64 `json:"jobId"`
UserId string `json:"userId"`
ClusterId string `json:"clusterId"`
StartTime int64 `json:"startTime"`
MetaData string `json:"metaData"`
ProjectId string `json:"projectId"`
Nodes []string `json:"nodes"`
NodeList string `json:"nodeList"`
}
type StartJobApiRespone struct {
DBID int64 `json:"id"`
}
type StopJobApiRequest struct {
// JobId, ClusterId and StartTime are optional.
// They are only used if no database id was provided.
JobId *string `json:"jobId"`
ClusterId *string `json:"clusterId"`
StartTime *int64 `json:"startTime"`
// Payload
StopTime int64 `json:"stopTime"`
}
type StopJobApiRespone struct {
DBID string `json:"id"`
}
type TagJobApiRequest []*struct {
Name string `json:"name"`
Type string `json:"type"`
}
func (api *RestApi) getJob(rw http.ResponseWriter, r *http.Request) {
id := mux.Vars(r)["id"]
job, err := api.Resolver.Query().Job(r.Context(), id)
if err != nil {
http.Error(rw, err.Error(), http.StatusNotFound)
return
}
job.Tags, err = api.Resolver.Job().Tags(r.Context(), job)
if err != nil {
http.Error(rw, err.Error(), http.StatusInternalServerError)
return
}
rw.Header().Add("Content-Type", "application/json")
rw.WriteHeader(http.StatusOK)
json.NewEncoder(rw).Encode(job)
}
func (api *RestApi) tagJob(rw http.ResponseWriter, r *http.Request) {
id := mux.Vars(r)["id"]
job, err := api.Resolver.Query().Job(r.Context(), id)
if err != nil {
http.Error(rw, err.Error(), http.StatusNotFound)
return
}
job.Tags, err = api.Resolver.Job().Tags(r.Context(), job)
if err != nil {
http.Error(rw, err.Error(), http.StatusInternalServerError)
return
}
var req TagJobApiRequest
if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
http.Error(rw, err.Error(), http.StatusBadRequest)
return
}
for _, tag := range req {
var tagId string
if err := sq.Select("id").From("tag").
Where("tag.tag_type = ?", tag.Type).Where("tag.tag_name = ?", tag.Name).
RunWith(api.DB).QueryRow().Scan(&tagId); err != nil {
http.Error(rw, fmt.Sprintf("the tag '%s:%s' does not exist", tag.Type, tag.Name), http.StatusNotFound)
return
}
if _, err := api.DB.Exec(`INSERT INTO jobtag (job_id, tag_id) VALUES (?, ?)`, job.ID, tagId); err != nil {
http.Error(rw, err.Error(), http.StatusInternalServerError)
return
}
job.Tags = append(job.Tags, &model.JobTag{
ID: tagId,
TagType: tag.Type,
TagName: tag.Name,
})
}
rw.Header().Add("Content-Type", "application/json")
rw.WriteHeader(http.StatusOK)
json.NewEncoder(rw).Encode(job)
}
func (api *RestApi) startJob(rw http.ResponseWriter, r *http.Request) {
req := StartJobApiRequest{}
if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
http.Error(rw, err.Error(), http.StatusBadRequest)
return
}
if config.GetClusterConfig(req.ClusterId) == nil {
http.Error(rw, fmt.Sprintf("cluster '%s' does not exist", req.ClusterId), http.StatusBadRequest)
return
}
if req.Nodes == nil {
req.Nodes = strings.Split(req.NodeList, "|")
if len(req.Nodes) == 1 {
req.Nodes = strings.Split(req.NodeList, ",")
}
}
if len(req.Nodes) == 0 || len(req.Nodes[0]) == 0 || len(req.UserId) == 0 {
http.Error(rw, "required fields are missing", http.StatusBadRequest)
return
}
// Check if combination of (job_id, cluster_id, start_time) already exists:
rows, err := api.DB.Query(`SELECT job.id FROM job WHERE job.job_id = ? AND job.cluster_id = ? AND job.start_time = ?`,
req.JobId, req.ClusterId, req.StartTime)
if err != nil {
http.Error(rw, err.Error(), http.StatusInternalServerError)
return
}
if rows.Next() {
var id int64 = -1
rows.Scan(&id)
http.Error(rw, fmt.Sprintf("a job with that job_id, cluster_id and start_time already exists (database id: %d)", id), http.StatusUnprocessableEntity)
return
}
res, err := api.DB.Exec(
`INSERT INTO job (job_id, user_id, project_id, cluster_id, start_time, duration, job_state, num_nodes, node_list, metadata) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?);`,
req.JobId, req.UserId, req.ProjectId, req.ClusterId, req.StartTime, 0, model.JobStateRunning, len(req.Nodes), strings.Join(req.Nodes, ","), req.MetaData)
if err != nil {
http.Error(rw, err.Error(), http.StatusInternalServerError)
return
}
id, err := res.LastInsertId()
if err != nil {
http.Error(rw, err.Error(), http.StatusInternalServerError)
return
}
log.Printf("new job (id: %d): clusterId=%s, jobId=%d, userId=%s, startTime=%d, nodes=%v\n", id, req.ClusterId, req.JobId, req.UserId, req.StartTime, req.Nodes)
rw.Header().Add("Content-Type", "application/json")
rw.WriteHeader(http.StatusCreated)
json.NewEncoder(rw).Encode(StartJobApiRespone{
DBID: id,
})
}
func (api *RestApi) stopJob(rw http.ResponseWriter, r *http.Request) {
req := StopJobApiRequest{}
if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
http.Error(rw, err.Error(), http.StatusBadRequest)
return
}
var err error
var job *model.Job
id, ok := mux.Vars(r)["id"]
if ok {
job, err = graph.ScanJob(sq.Select(graph.JobTableCols...).From("job").Where("job.id = ?", id).RunWith(api.DB).QueryRow())
} else {
job, err = graph.ScanJob(sq.Select(graph.JobTableCols...).From("job").
Where("job.job_id = ?", req.JobId).
Where("job.cluster_id = ?", req.ClusterId).
Where("job.start_time = ?", req.StartTime).
RunWith(api.DB).QueryRow())
}
if err != nil {
http.Error(rw, err.Error(), http.StatusNotFound)
return
}
if job == nil || job.StartTime.Unix() >= req.StopTime || job.State != model.JobStateRunning {
http.Error(rw, "stop_time must be larger than start_time and only running jobs can be stopped", http.StatusBadRequest)
return
}
doArchiving := func(job *model.Job, ctx context.Context) error {
job.Duration = int(req.StopTime - job.StartTime.Unix())
jobMeta, err := metricdata.ArchiveJob(job, ctx)
if err != nil {
log.Printf("archiving job (id: %s) failed: %s\n", job.ID, err.Error())
return err
}
getAvg := func(metric string) sql.NullFloat64 {
stats, ok := jobMeta.Statistics[metric]
if !ok {
return sql.NullFloat64{Valid: false}
}
return sql.NullFloat64{Valid: true, Float64: stats.Avg}
}
if _, err := api.DB.Exec(
`UPDATE job SET
job_state = ?, duration = ?,
flops_any_avg = ?, mem_bw_avg = ?, net_bw_avg = ?, file_bw_avg = ?, load_avg = ?
WHERE job.id = ?`,
model.JobStateCompleted, job.Duration,
getAvg("flops_any"), getAvg("mem_bw"), getAvg("net_bw"), getAvg("file_bw"), getAvg("load"),
job.ID); err != nil {
log.Printf("archiving job (id: %s) failed: %s\n", job.ID, err.Error())
return err
}
log.Printf("job stopped and archived (id: %s)\n", job.ID)
return nil
}
log.Printf("archiving job... (id: %s): clusterId=%s, jobId=%s, userId=%s, startTime=%s, nodes=%v\n", job.ID, job.ClusterID, job.JobID, job.UserID, job.StartTime, job.Nodes)
if api.AsyncArchiving {
rw.Header().Add("Content-Type", "application/json")
rw.WriteHeader(http.StatusOK)
json.NewEncoder(rw).Encode(StopJobApiRespone{
DBID: job.ID,
})
go doArchiving(job, context.Background())
} else {
err := doArchiving(job, r.Context())
if err != nil {
http.Error(rw, err.Error(), http.StatusInternalServerError)
} else {
rw.Header().Add("Content-Type", "application/json")
rw.WriteHeader(http.StatusOK)
json.NewEncoder(rw).Encode(job)
}
}
}

View File

@ -137,9 +137,9 @@ func loadAveragesFromArchive(job *model.Job, metrics []string, data [][]schema.F
}
// Writes a running job to the job-archive
func ArchiveJob(job *model.Job, ctx context.Context) error {
func ArchiveJob(job *model.Job, ctx context.Context) (*schema.JobMeta, error) {
if job.State != model.JobStateRunning {
return errors.New("cannot archive job that is not running")
return nil, errors.New("cannot archive job that is not running")
}
allMetrics := make([]string, 0)
@ -149,7 +149,7 @@ func ArchiveJob(job *model.Job, ctx context.Context) error {
}
jobData, err := LoadData(job, allMetrics, ctx)
if err != nil {
return err
return nil, err
}
tags := []struct {
@ -195,39 +195,46 @@ func ArchiveJob(job *model.Job, ctx context.Context) error {
}
}
// If the file based archive is disabled,
// only return the JobMeta structure as the
// statistics in there are needed.
if !useArchive {
return metaData, nil
}
dirPath, err := getPath(job, "", false)
if err != nil {
return err
return nil, err
}
if err := os.MkdirAll(dirPath, 0777); err != nil {
return err
return nil, err
}
f, err := os.Create(path.Join(dirPath, "meta.json"))
if err != nil {
return err
return nil, err
}
defer f.Close()
writer := bufio.NewWriter(f)
if err := json.NewEncoder(writer).Encode(metaData); err != nil {
return err
return nil, err
}
if err := writer.Flush(); err != nil {
return err
return nil, err
}
f, err = os.Create(path.Join(dirPath, "data.json"))
if err != nil {
return err
return nil, err
}
writer = bufio.NewWriter(f)
if err := json.NewEncoder(writer).Encode(jobData); err != nil {
return err
return nil, err
}
if err := writer.Flush(); err != nil {
return err
return nil, err
}
return f.Close()
return metaData, f.Close()
}

View File

@ -28,7 +28,10 @@ var metricDataRepos map[string]MetricDataRepository = map[string]MetricDataRepos
var JobArchivePath string
func Init(jobArchivePath string) error {
var useArchive bool
func Init(jobArchivePath string, disableArchive bool) error {
useArchive = !disableArchive
JobArchivePath = jobArchivePath
for _, cluster := range config.Clusters {
if cluster.MetricDataRepository != nil {
@ -55,7 +58,7 @@ func Init(jobArchivePath string) error {
// Fetches the metric data for a job.
func LoadData(job *model.Job, metrics []string, ctx context.Context) (schema.JobData, error) {
if job.State == model.JobStateRunning {
if job.State == model.JobStateRunning || !useArchive {
repo, ok := metricDataRepos[job.ClusterID]
if !ok {
return nil, fmt.Errorf("no metric data repository configured for '%s'", job.ClusterID)
@ -83,7 +86,7 @@ func LoadData(job *model.Job, metrics []string, ctx context.Context) (schema.Job
// Used for the jobsFootprint GraphQL-Query. TODO: Rename/Generalize.
func LoadAverages(job *model.Job, metrics []string, data [][]schema.Float, ctx context.Context) error {
if job.State != model.JobStateRunning {
if job.State != model.JobStateRunning && useArchive {
return loadAveragesFromArchive(job, metrics, data)
}
@ -120,6 +123,12 @@ func LoadNodeData(clusterId string, metrics, nodes []string, from, to int64, ctx
return nil, fmt.Errorf("no metric data repository configured for '%s'", clusterId)
}
if metrics == nil {
for _, m := range config.GetClusterConfig(clusterId).MetricConfig {
metrics = append(metrics, m.Name)
}
}
data, err := repo.LoadNodeData(clusterId, metrics, nodes, from, to, ctx)
if err != nil {
return nil, err

View File

@ -1,142 +0,0 @@
package main
import (
"encoding/json"
"fmt"
"log"
"net/http"
"strings"
"github.com/ClusterCockpit/cc-jobarchive/config"
"github.com/ClusterCockpit/cc-jobarchive/graph"
"github.com/ClusterCockpit/cc-jobarchive/graph/model"
"github.com/ClusterCockpit/cc-jobarchive/metricdata"
sq "github.com/Masterminds/squirrel"
"github.com/gorilla/mux"
)
type StartJobApiRequest struct {
JobId int64 `json:"jobId"`
UserId string `json:"userId"`
ClusterId string `json:"clusterId"`
StartTime int64 `json:"startTime"`
MetaData string `json:"metaData"`
ProjectId string `json:"projectId"`
Nodes []string `json:"nodes"`
NodeList string `json:"nodeList"`
}
type StartJobApiRespone struct {
DBID int64 `json:"id"`
}
type StopJobApiRequest struct {
// JobId, ClusterId and StartTime are optional.
// They are only used if no database id was provided.
JobId *string `json:"jobId"`
ClusterId *string `json:"clusterId"`
StartTime *int64 `json:"startTime"`
StopTime int64 `json:"stopTime"`
}
type StopJobApiRespone struct {
DBID string `json:"id"`
}
func startJob(rw http.ResponseWriter, r *http.Request) {
req := StartJobApiRequest{}
if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
http.Error(rw, err.Error(), http.StatusBadRequest)
return
}
if config.GetClusterConfig(req.ClusterId) == nil {
http.Error(rw, fmt.Sprintf("cluster '%s' does not exist", req.ClusterId), http.StatusBadRequest)
return
}
if req.Nodes == nil {
req.Nodes = strings.Split(req.NodeList, "|")
if len(req.Nodes) == 1 {
req.Nodes = strings.Split(req.NodeList, ",")
}
}
if len(req.Nodes) == 0 || len(req.Nodes[0]) == 0 || len(req.UserId) == 0 {
http.Error(rw, "required fields are missing", http.StatusBadRequest)
return
}
res, err := db.Exec(
`INSERT INTO job (job_id, user_id, project_id, cluster_id, start_time, duration, job_state, num_nodes, node_list, metadata) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?);`,
req.JobId, req.UserId, req.ProjectId, req.ClusterId, req.StartTime, 0, model.JobStateRunning, len(req.Nodes), strings.Join(req.Nodes, ","), req.MetaData)
if err != nil {
http.Error(rw, err.Error(), http.StatusInternalServerError)
return
}
id, err := res.LastInsertId()
if err != nil {
http.Error(rw, err.Error(), http.StatusInternalServerError)
return
}
log.Printf("new job (id: %d): clusterId=%s, jobId=%d, userId=%s, startTime=%d, nodes=%v\n", id, req.ClusterId, req.JobId, req.UserId, req.StartTime, req.Nodes)
rw.Header().Add("Content-Type", "application/json")
rw.WriteHeader(http.StatusCreated)
json.NewEncoder(rw).Encode(StartJobApiRespone{
DBID: id,
})
}
func stopJob(rw http.ResponseWriter, r *http.Request) {
req := StopJobApiRequest{}
if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
http.Error(rw, err.Error(), http.StatusBadRequest)
return
}
var err error
var job *model.Job
id, ok := mux.Vars(r)["id"]
if ok {
job, err = graph.ScanJob(sq.Select(graph.JobTableCols...).From("job").Where("job.id = ?", id).RunWith(db).QueryRow())
} else {
job, err = graph.ScanJob(sq.Select(graph.JobTableCols...).From("job").
Where("job.job_id = ?", req.JobId).
Where("job.cluster_id = ?", req.ClusterId).
Where("job.start_time = ?", req.StartTime).
RunWith(db).QueryRow())
}
if err != nil {
http.Error(rw, err.Error(), http.StatusBadRequest)
return
}
if job == nil || job.StartTime.Unix() >= req.StopTime || job.State != model.JobStateRunning {
http.Error(rw, "stop_time must be larger than start_time and only running jobs can be stopped", http.StatusBadRequest)
return
}
job.Duration = int(req.StopTime - job.StartTime.Unix())
if err := metricdata.ArchiveJob(job, r.Context()); err != nil {
log.Printf("archiving job (id: %s) failed: %s\n", job.ID, err.Error())
http.Error(rw, err.Error(), http.StatusInternalServerError)
return
}
if _, err := db.Exec(
`UPDATE job SET job_state = ?, duration = ? WHERE job.id = ?`,
model.JobStateCompleted, job.Duration, job.ID); err != nil {
http.Error(rw, err.Error(), http.StatusInternalServerError)
return
}
log.Printf("job stoped and archived (id: %s): clusterId=%s, jobId=%s, userId=%s, startTime=%s, nodes=%v\n", job.ID, job.ClusterID, job.JobID, job.UserID, job.StartTime, job.Nodes)
rw.Header().Add("Content-Type", "application/json")
rw.WriteHeader(http.StatusCreated)
json.NewEncoder(rw).Encode(StopJobApiRespone{
DBID: job.ID,
})
}

View File

@ -10,6 +10,7 @@ import (
"github.com/99designs/gqlgen/graphql/handler"
"github.com/99designs/gqlgen/graphql/playground"
"github.com/ClusterCockpit/cc-jobarchive/api"
"github.com/ClusterCockpit/cc-jobarchive/auth"
"github.com/ClusterCockpit/cc-jobarchive/config"
"github.com/ClusterCockpit/cc-jobarchive/graph"
@ -24,15 +25,39 @@ import (
var db *sqlx.DB
// Format of the configurartion (file). See below for the defaults.
type ProgramConfig struct {
// Address where the http (or https) server will listen on (for example: 'localhost:80').
Addr string `json:"addr"`
// Disable authentication (for everything: API, Web-UI, ...)
DisableAuthentication bool `json:"disable-authentication"`
// Folder where static assets can be found, will be served directly
StaticFiles string `json:"static-files"`
// Currently only SQLite3 ist supported, so this should be a filename
DB string `json:"db"`
// Path to the job-archive
JobArchive string `json:"job-archive"`
// Make the /api/jobs/stop_job endpoint do the heavy work in the background.
AsyncArchiving bool `json:"async-archive"`
// Keep all metric data in the metric data repositories,
// do not write to the job-archive.
DisableArchive bool `json:"disable-archive"`
// For LDAP Authentication and user syncronisation.
LdapConfig *auth.LdapConfig `json:"ldap"`
// If both those options are not empty, use HTTPS using those certificates.
HttpsCertFile string `json:"https-cert-file"`
HttpsKeyFile string `json:"https-key-file"`
// If overwriten, at least all the options in the defaults below must
// be provided! Most options here can be overwritten by the user.
UiDefaults map[string]interface{} `json:"ui-defaults"`
}
@ -42,6 +67,8 @@ var programConfig ProgramConfig = ProgramConfig{
StaticFiles: "./frontend/public",
DB: "./var/job.db",
JobArchive: "./var/job-archive",
AsyncArchiving: true,
DisableArchive: false,
LdapConfig: &auth.LdapConfig{
Url: "ldap://localhost",
UserBase: "ou=hpc,dc=rrze,dc=uni-erlangen,dc=de",
@ -92,11 +119,15 @@ func main() {
}
var err error
db, err = sqlx.Open("sqlite3", programConfig.DB)
// This might need to change for other databases:
db, err = sqlx.Open("sqlite3", fmt.Sprintf("%s?_foreign_keys=on", programConfig.DB))
if err != nil {
log.Fatal(err)
}
// Only for sqlite, not needed for any other database:
db.SetMaxOpenConns(1)
// Initialize sub-modules...
if !programConfig.DisableAuthentication {
@ -126,7 +157,7 @@ func main() {
log.Fatal(err)
}
if err := metricdata.Init(programConfig.JobArchive); err != nil {
if err := metricdata.Init(programConfig.JobArchive, programConfig.DisableArchive); err != nil {
log.Fatal(err)
}
@ -145,6 +176,11 @@ func main() {
resolver := &graph.Resolver{DB: db}
graphQLEndpoint := handler.NewDefaultServer(generated.NewExecutableSchema(generated.Config{Resolvers: resolver}))
graphQLPlayground := playground.Handler("GraphQL playground", "/query")
restApi := &api.RestApi{
DB: db,
Resolver: resolver,
AsyncArchiving: programConfig.AsyncArchiving,
}
handleGetLogin := func(rw http.ResponseWriter, r *http.Request) {
templates.Render(rw, r, "login", &templates.Page{
@ -170,9 +206,7 @@ func main() {
secured.Use(auth.Auth)
}
secured.Handle("/query", graphQLEndpoint)
secured.HandleFunc("/api/jobs/start_job/", startJob).Methods(http.MethodPost)
secured.HandleFunc("/api/jobs/stop_job/", stopJob).Methods(http.MethodPost, http.MethodPut)
secured.HandleFunc("/api/jobs/stop_job/{id:[0-9]+}", stopJob).Methods(http.MethodPost, http.MethodPut)
secured.HandleFunc("/config.json", config.ServeConfig).Methods(http.MethodGet)
secured.HandleFunc("/", func(rw http.ResponseWriter, r *http.Request) {
@ -201,12 +235,13 @@ func main() {
})
monitoringRoutes(secured, resolver)
restApi.MountRoutes(secured)
r.PathPrefix("/").Handler(http.FileServer(http.Dir(programConfig.StaticFiles)))
handler := handlers.CORS(
handlers.AllowedHeaders([]string{"X-Requested-With", "Content-Type", "Authorization"}),
handlers.AllowedMethods([]string{"GET", "POST", "HEAD", "OPTIONS"}),
handlers.AllowedOrigins([]string{"*"}))(handlers.LoggingHandler(os.Stdout, r))
handlers.AllowedOrigins([]string{"*"}))(handlers.LoggingHandler(os.Stdout, handlers.CompressHandler(r)))
// Start http or https server
if programConfig.HttpsCertFile != "" && programConfig.HttpsKeyFile != "" {