ClusterCockpit/cc-backend
1
0
Fork 0
mirror of https://github.com/ClusterCockpit/cc-backend synced 2025-07-23 12:51:40 +02:00
Code Issues Packages Releases Activity

Merge pull request #143 from ClusterCockpit/102-add-rest-api-endpoint-to-query-full-job-data-including-all-metric-timeseries

Browse Source 
102 add rest api endpoint to query full job data including all metric timeseries
This commit is contained in:
Jan Eitzinger
2023-06-15 07:06:45 +02:00
committed by GitHub
parent 817efde844 54aa940d3e
commit 113cb863e0
5 changed files with 877 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

290
internal/api/docs.go
View File

@@ -628,6 +628,91 @@ const docTemplate = `{
}
}
}
},
"/jobs/{id}": {
"post": {
"security": [
{
"ApiKeyAuth": []
}
],
"description": "Job to get is specified by database ID\nReturns full job resource information according to 'JobMeta' scheme and all metrics according to 'JobData'.",
"consumes": [
"application/json"
],
"produces": [
"application/json"
],
"tags": [
"query"
],
"summary": "Get complete job meta and metric data",
"parameters": [
{
"type": "integer",
"description": "Database ID of Job",
"name": "id",
"in": "path",
"required": true
},
{
"description": "Array of metric names",
"name": "request",
"in": "body",
"required": true,
"schema": {
"type": "array",
"items": {
"type": "string"
}
}
}
],
"responses": {
"200": {
"description": "Job resource",
"schema": {
"$ref": "#/definitions/api.GetJobApiResponse"
}
},
"400": {
"description": "Bad Request",
"schema": {
"$ref": "#/definitions/api.ErrorResponse"
}
},
"401": {
"description": "Unauthorized",
"schema": {
"$ref": "#/definitions/api.ErrorResponse"
}
},
"403": {
"description": "Forbidden",
"schema": {
"$ref": "#/definitions/api.ErrorResponse"
}
},
"404": {
"description": "Resource not found",
"schema": {
"$ref": "#/definitions/api.ErrorResponse"
}
},
"422": {
"description": "Unprocessable Entity: finding job failed: sql: no rows in result set",
"schema": {
"$ref": "#/definitions/api.ErrorResponse"
}
},
"500": {
"description": "Internal Server Error",
"schema": {
"$ref": "#/definitions/api.ErrorResponse"
}
}
}
}
}
},
"definitions": {
@@ -690,6 +775,20 @@ const docTemplate = `{
}
}
},
"api.GetJobApiResponse": {
"type": "object",
"properties": {
"data": {
"type": "array",
"items": {
"$ref": "#/definitions/api.JobMetricWithName"
}
},
"meta": {
"$ref": "#/definitions/schema.Job"
}
}
},
"api.GetJobsApiResponse": {
"type": "object",
"properties": {
@@ -710,6 +809,20 @@ const docTemplate = `{
}
}
},
"api.JobMetricWithName": {
"type": "object",
"properties": {
"metric": {
"$ref": "#/definitions/schema.JobMetric"
},
"name": {
"type": "string"
},
"scope": {
"$ref": "#/definitions/schema.MetricScope"
}
}
},
"api.StartJobApiResponse": {
"type": "object",
"properties": {
@@ -771,6 +884,9 @@ const docTemplate = `{
"type": "string",
"example": "fritz"
},
"concurrentJobs": {
"$ref": "#/definitions/schema.JobLinkResultList"
},
"duration": {
"description": "Duration of job in seconds (Min \u003e 0)",
"type": "integer",
@@ -795,6 +911,14 @@ const docTemplate = `{
},
"jobState": {
"description": "Final state of job",
"enum": [
"completed",
"failed",
"cancelled",
"stopped",
"timeout",
"out_of_memory"
],
"allOf": [
{
"$ref": "#/definitions/schema.JobState"
@@ -823,7 +947,7 @@ const docTemplate = `{
"example": 2
},
"numHwthreads": {
"description": "Number of HWThreads used (Min \u003e 0)",
"description": "NumCores int32 ` + "`" + `json:\"numCores\" db:\"num_cores\" example:\"20\" minimum:\"1\"` + "`" + ` // Number of HWThreads used (Min \u003e 0)",
"type": "integer",
"minimum": 1,
"example": 20
@@ -885,6 +1009,31 @@ const docTemplate = `{
}
}
},
"schema.JobLink": {
"type": "object",
"properties": {
"id": {
"type": "integer"
},
"jobId": {
"type": "integer"
}
}
},
"schema.JobLinkResultList": {
"type": "object",
"properties": {
"count": {
"type": "integer"
},
"items": {
"type": "array",
"items": {
"$ref": "#/definitions/schema.JobLink"
}
}
}
},
"schema.JobMeta": {
"description": "Meta data information of a HPC job.",
"type": "object",
@@ -899,6 +1048,9 @@ const docTemplate = `{
"type": "string",
"example": "fritz"
},
"concurrentJobs": {
"$ref": "#/definitions/schema.JobLinkResultList"
},
"duration": {
"description": "Duration of job in seconds (Min \u003e 0)",
"type": "integer",
@@ -923,6 +1075,14 @@ const docTemplate = `{
},
"jobState": {
"description": "Final state of job",
"enum": [
"completed",
"failed",
"cancelled",
"stopped",
"timeout",
"out_of_memory"
],
"allOf": [
{
"$ref": "#/definitions/schema.JobState"
@@ -951,7 +1111,7 @@ const docTemplate = `{
"example": 2
},
"numHwthreads": {
"description": "Number of HWThreads used (Min \u003e 0)",
"description": "NumCores int32 ` + "`" + `json:\"numCores\" db:\"num_cores\" example:\"20\" minimum:\"1\"` + "`" + ` // Number of HWThreads used (Min \u003e 0)",
"type": "integer",
"minimum": 1,
"example": 20
@@ -1022,6 +1182,26 @@ const docTemplate = `{
}
}
},
"schema.JobMetric": {
"type": "object",
"properties": {
"series": {
"type": "array",
"items": {
"$ref": "#/definitions/schema.Series"
}
},
"statisticsSeries": {
"$ref": "#/definitions/schema.StatsSeries"
},
"timestep": {
"type": "integer"
},
"unit": {
"$ref": "#/definitions/schema.Unit"
}
}
},
"schema.JobState": {
"type": "string",
"enum": [
@@ -1068,9 +1248,42 @@ const docTemplate = `{
"example": 2000
},
"unit": {
"description": "Metric unit (see schema/unit.schema.json)",
"type": "string",
"example": "GHz"
"$ref": "#/definitions/schema.Unit"
}
}
},
"schema.MetricScope": {
"type": "string",
"enum": [
"invalid_scope",
"node",
"socket",
"memoryDomain",
"core",
"hwthread",
"accelerator"
],
"x-enum-varnames": [
"MetricScopeInvalid",
"MetricScopeNode",
"MetricScopeSocket",
"MetricScopeMemoryDomain",
"MetricScopeCore",
"MetricScopeHWThread",
"MetricScopeAccelerator"
]
},
"schema.MetricStatistics": {
"type": "object",
"properties": {
"avg": {
"type": "number"
},
"max": {
"type": "number"
},
"min": {
"type": "number"
}
}
},
@@ -1102,12 +1315,64 @@ const docTemplate = `{
}
}
},
"schema.Series": {
"type": "object",
"properties": {
"data": {
"type": "array",
"items": {
"type": "number"
}
},
"hostname": {
"type": "string"
},
"id": {
"type": "string"
},
"statistics": {
"$ref": "#/definitions/schema.MetricStatistics"
}
}
},
"schema.StatsSeries": {
"type": "object",
"properties": {
"max": {
"type": "array",
"items": {
"type": "number"
}
},
"mean": {
"type": "array",
"items": {
"type": "number"
}
},
"min": {
"type": "array",
"items": {
"type": "number"
}
},
"percentiles": {
"type": "object",
"additionalProperties": {
"type": "array",
"items": {
"type": "number"
}
}
}
}
},
"schema.Tag": {
"description": "Defines a tag using name and type.",
"type": "object",
"properties": {
"id": {
"description": "The unique DB identifier of a tag",
"description": "The unique DB identifier of a tag\nThe unique DB identifier of a tag",
"type": "integer"
},
"name": {
@@ -1121,6 +1386,17 @@ const docTemplate = `{
"example": "Debug"
}
}
},
"schema.Unit": {
"type": "object",
"properties": {
"base": {
"type": "string"
},
"prefix": {
"type": "string"
}
}
}
},
"securityDefinitions": {
@@ -1139,7 +1415,7 @@ const docTemplate = `{
// SwaggerInfo holds exported Swagger Info so clients can modify it
var SwaggerInfo = &swag.Spec{
Version: "0.2.0",
Version: "1",
Host: "localhost:8080",
BasePath: "/api",
Schemes: []string{},

109
internal/api/rest.go
View File

@@ -23,6 +23,7 @@ import (
"github.com/ClusterCockpit/cc-backend/internal/graph"
"github.com/ClusterCockpit/cc-backend/internal/graph/model"
"github.com/ClusterCockpit/cc-backend/internal/importer"
"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/log"
@@ -68,7 +69,7 @@ func (api *RestApi) MountRoutes(r *mux.Router) {
// r.HandleFunc("/jobs/import/", api.importJob).Methods(http.MethodPost, http.MethodPut)
r.HandleFunc("/jobs/", api.getJobs).Methods(http.MethodGet)
// r.HandleFunc("/jobs/{id}", api.getJob).Methods(http.MethodGet)
r.HandleFunc("/jobs/{id}", api.getJobById).Methods(http.MethodPost)
r.HandleFunc("/jobs/tag_job/{id}", api.tagJob).Methods(http.MethodPost, http.MethodPatch)
r.HandleFunc("/jobs/metrics/{id}", api.getJobMetrics).Methods(http.MethodGet)
r.HandleFunc("/jobs/delete_job/", api.deleteJobByRequest).Methods(http.MethodDelete)
@@ -142,6 +143,19 @@ type ApiTag struct {
type TagJobApiRequest []*ApiTag
type GetJobApiRequest []string
type GetJobApiResponse struct {
Meta *schema.Job
Data []*JobMetricWithName
}
type JobMetricWithName struct {
Name string `json:"name"`
Scope schema.MetricScope `json:"scope"`
Metric *schema.JobMetric `json:"metric"`
}
func handleError(err error, statusCode int, rw http.ResponseWriter) {
log.Warnf("REST ERROR : %s", err.Error())
rw.Header().Add("Content-Type", "application/json")
@@ -301,6 +315,99 @@ func (api *RestApi) getJobs(rw http.ResponseWriter, r *http.Request) {
}
}
// getJobById godoc
// @summary Get complete job meta and metric data
// @tags query
// @description Job to get is specified by database ID
// @description Returns full job resource information according to 'JobMeta' scheme and all metrics according to 'JobData'.
// @accept json
// @produce json
// @param id path int true "Database ID of Job"
// @param request body api.GetJobApiRequest true "Array of metric names"
// @success 200 {object} api.GetJobApiResponse "Job resource"
// @failure 400 {object} api.ErrorResponse "Bad Request"
// @failure 401 {object} api.ErrorResponse "Unauthorized"
// @failure 403 {object} api.ErrorResponse "Forbidden"
// @failure 404 {object} api.ErrorResponse "Resource not found"
// @failure 422 {object} api.ErrorResponse "Unprocessable Entity: finding job failed: sql: no rows in result set"
// @failure 500 {object} api.ErrorResponse "Internal Server Error"
// @security ApiKeyAuth
// @router /jobs/{id} [post]
func (api *RestApi) getJobById(rw http.ResponseWriter, r *http.Request) {
if user := auth.GetUser(r.Context()); user != nil && !user.HasRole(auth.RoleApi) {
handleError(fmt.Errorf("missing role: %v",
auth.GetRoleString(auth.RoleApi)), http.StatusForbidden, rw)
return
}
// Fetch job from db
id, ok := mux.Vars(r)["id"]
var job *schema.Job
var err error
if ok {
id, e := strconv.ParseInt(id, 10, 64)
if e != nil {
handleError(fmt.Errorf("integer expected in path for id: %w", e), http.StatusBadRequest, rw)
return
}
job, err = api.JobRepository.FindById(id)
} else {
handleError(errors.New("the parameter 'id' is required"), http.StatusBadRequest, rw)
return
}
if err != nil {
handleError(fmt.Errorf("finding job failed: %w", err), http.StatusUnprocessableEntity, rw)
return
}
var metrics GetJobApiRequest
if err = decode(r.Body, &metrics); err != nil {
http.Error(rw, err.Error(), http.StatusBadRequest)
return
}
var scopes []schema.MetricScope
if job.NumNodes == 1 {
scopes = []schema.MetricScope{"core"}
} else {
scopes = []schema.MetricScope{"node"}
}
data, err := metricdata.LoadData(job, metrics, scopes, r.Context())
if err != nil {
log.Warn("Error while loading job data")
return
}
res := []*JobMetricWithName{}
for name, md := range data {
for scope, metric := range md {
res = append(res, &JobMetricWithName{
Name: name,
Scope: scope,
Metric: metric,
})
}
}
log.Debugf("/api/job/%s: get job %d", id, job.JobID)
rw.Header().Add("Content-Type", "application/json")
bw := bufio.NewWriter(rw)
defer bw.Flush()
payload := GetJobApiResponse{
Meta: job,
Data: res,
}
if err := json.NewEncoder(bw).Encode(payload); err != nil {
handleError(err, http.StatusInternalServerError, rw)
return
}
}
// tagJob godoc
// @summary Adds one or more tags to a job
// @tags add and modify