Files
upstream-mattermost/server/channels/jobs/jobs.go
upstream-mirror 240db4e2a7
Some checks failed
Server CI / Postgres (shard 0) (push) Has been skipped
Server CI / Postgres (shard 3) (push) Has been skipped
BuildEnv Docker Image / build-image (push) Failing after 4s
BuildEnv Docker Image / build-image-fips (push) Failing after 4s
Migration Review & Release Notes / Generate migration review & release notes (push) Failing after 3s
Server CI / Build Environment (push) Failing after 3s
Server CI / Postgres (shard 1) (push) Has been skipped
Tools CI / check-style (mattermost-govet) (push) Failing after 4s
Server CI / Postgres (shard 2) (push) Has been skipped
Tools CI / check-style (mmgotool) (push) Failing after 3s
Server CI / Elasticsearch v8 Compatibility (push) Has been skipped
Server CI / OpenSearch v2 Compatibility (push) Has been skipped
Server CI / Postgres FIPS (shard 0) (push) Has been skipped
Tools CI / Test (mattermost-govet) (push) Failing after 3s
Server CI / Postgres FIPS (shard 1) (push) Has been skipped
Server CI / Postgres FIPS (shard 2) (push) Has been skipped
Tools CI / Test (mmgotool) (push) Failing after 3s
Server CI / Postgres FIPS (shard 3) (push) Has been skipped
Server CI / Run mmctl tests (push) Has been skipped
Server CI / Run mmctl tests (FIPS) (push) Has been skipped
Server CI / Merge Postgres Test Results (push) Has been skipped
Server CI / Merge Postgres FIPS Test Results (push) Has been skipped
Server CI / Check backport migrations (push) Has been skipped
Server CI / Check generated files (push) Has been skipped
Server CI / check-style (push) Has been skipped
Server CI / Vet API (push) Has been skipped
Server CI / Build mattermost server app (push) Has been skipped
Server CI / Server CI Complete (push) Failing after 2s
Server CI / Server CI Artifacts (push) Has been skipped
Server CI / Sentry Upload (push) Has been skipped
Server CI / Server CI Report (push) Failing after 5m1s
Server CI Nightly Race / Race Detector (push) Has been skipped
CodeQL / Analyze (go) (push) Has been cancelled
CodeQL / Analyze (javascript) (push) Has been cancelled
Webapp CI Cache Warm / Warm Webapp node_modules Cache (push) Has been cancelled
E2E CI Cache Warm / Warm E2E npm Registry Cache (push) Has been cancelled
API / build (push) Has been cancelled
Web App CI / check-types (push) Has been cancelled
Web App CI / check-lint (push) Has been cancelled
Web App CI / check-i18n (push) Has been cancelled
Web App CI / check-external-links (push) Has been cancelled
Web App CI / test (platform) (push) Has been cancelled
Web App CI / test (mattermost-redux) (push) Has been cancelled
Web App CI / test (channels shard 1/4) (push) Has been cancelled
Web App CI / test (channels shard 2/4) (push) Has been cancelled
Web App CI / test (channels shard 3/4) (push) Has been cancelled
Web App CI / test (channels shard 4/4) (push) Has been cancelled
Web App CI / upload-coverage (push) Has been cancelled
Web App CI / build (push) Has been cancelled
YAML Lint / yamllint (push) Has been cancelled
Push mirrored docker images / cd/Push mirrored docker images (push) Has been cancelled
docs-cd / Build and deploy docs (push) Has been cancelled
Docs CI / build (push) Has been cancelled
E2E Cache Warm: Cypress deps / Warm Cypress deps cache (push) Has been cancelled
Scorecards supply-chain security / Scorecard analysis (push) Has been cancelled
Server CI Weekly / Postgres FIPS (push) Has been skipped
Server CI Weekly / Run mmctl tests (FIPS) (push) Has been skipped
Server CI Weekly / Postgres with binary parameters (push) Has been skipped
Server CI Weekly / Build Environment (push) Failing after 1s
Server CI Cache Warm / Warm Go Module and Build Cache (push) Failing after 1s
Server CI Nightly Race / Build Environment (push) Failing after 1s
snapshot of https://github.com/mattermost/mattermost.git (master, 2026-09-16; blobs > 131072B omitted)
2026-09-16 00:29:55 +05:00

430 lines
14 KiB
Go

// Copyright (c) 2015-present Mattermost, Inc. All Rights Reserved.
// See LICENSE.txt for license information.
package jobs
import (
"encoding/json"
"errors"
"fmt"
"net/http"
"runtime/pprof"
"strings"
"time"
"github.com/mattermost/mattermost/server/public/model"
"github.com/mattermost/mattermost/server/public/shared/mlog"
"github.com/mattermost/mattermost/server/public/shared/request"
"github.com/mattermost/mattermost/server/v8/channels/store"
)
const (
CancelWatcherPollingInterval = 5000
)
// JobLoggerFields returns the logger annotations reflecting the given job metadata.
func JobLoggerFields(job *model.Job) []mlog.Field {
if job == nil {
return nil
}
return []mlog.Field{
mlog.String("job_id", job.Id),
mlog.String("job_type", job.Type),
mlog.Millis("job_create_at", job.CreateAt),
}
}
func (srv *JobServer) CreateJob(rctx request.CTX, jobType string, jobData map[string]string) (*model.Job, *model.AppError) {
job, appErr := srv._createJob(rctx, jobType, jobData)
if appErr != nil {
return nil, appErr
}
if _, err := srv.Store.Job().Save(job); err != nil {
return nil, model.NewAppError("CreateJob", "app.job.save.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
return job, nil
}
func (srv *JobServer) CreateJobOnce(rctx request.CTX, jobType string, jobData map[string]string) (*model.Job, *model.AppError) {
job, appErr := srv._createJob(rctx, jobType, jobData)
if appErr != nil {
return nil, appErr
}
if _, err := srv.Store.Job().SaveOnce(job); err != nil {
return nil, model.NewAppError("CreateJob", "app.job.save.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
return job, nil
}
func (srv *JobServer) CreateJobOnceByTypeAndData(rctx request.CTX, jobType string, jobData map[string]string, data map[string]string) (*model.Job, *model.AppError) {
job, appErr := srv._createJob(rctx, jobType, jobData)
if appErr != nil {
return nil, appErr
}
savedJob, err := srv.Store.Job().SaveOnceByTypeAndData(job, data)
if err != nil {
return nil, model.NewAppError("CreateJob", "app.job.save.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
return savedJob, nil
}
func (srv *JobServer) _createJob(rctx request.CTX, jobType string, jobData map[string]string) (*model.Job, *model.AppError) {
job := model.Job{
Id: model.NewId(),
Type: jobType,
CreateAt: model.GetMillis(),
Status: model.JobStatusPending,
Data: jobData,
}
if err := job.IsValid(); err != nil {
return nil, err
}
if srv.workers.Get(job.Type) == nil {
return nil, model.NewAppError("Job.IsValid", "model.job.is_valid.type.app_error", nil, "id="+job.Id, http.StatusBadRequest)
}
return &job, nil
}
func (srv *JobServer) GetJob(rctx request.CTX, id string) (*model.Job, *model.AppError) {
job, err := srv.Store.Job().Get(rctx, id)
if err != nil {
var nfErr *store.ErrNotFound
switch {
case errors.As(err, &nfErr):
return nil, model.NewAppError("GetJob", "app.job.get.app_error", nil, "", http.StatusNotFound).Wrap(err)
default:
return nil, model.NewAppError("GetJob", "app.job.get.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
}
return job, nil
}
func (srv *JobServer) ClaimJob(job *model.Job) (*model.Job, *model.AppError) {
newJob, err := srv.Store.Job().UpdateStatusOptimistically(job.Id, model.JobStatusPending, model.JobStatusInProgress)
if err != nil {
return nil, model.NewAppError("ClaimJob", "app.job.update.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
if newJob != nil {
if srv.metrics != nil {
srv.metrics.IncrementJobActive(newJob.Type)
}
srv.publishJobStatus(newJob, model.JobStatusInProgress)
}
return newJob, nil
}
func (srv *JobServer) publishJobStatus(job *model.Job, status string) {
if srv.publish == nil {
return
}
jobCopy := *job
jobCopy.Status = status
jobJSON, err := json.Marshal(&jobCopy)
if err != nil {
srv.logger.Warn("Failed to marshal job for WebSocket event", mlog.Err(err))
return
}
message := model.NewWebSocketEvent(model.WebsocketEventJobUpdated, "", "", "", nil, "")
message.Add("job", string(jobJSON))
var permission *model.Permission
switch job.Type {
case model.JobTypeDataRetention:
permission = model.PermissionReadDataRetentionJob
case model.JobTypeMessageExport:
permission = model.PermissionReadComplianceExportJob
case model.JobTypeElasticsearchPostIndexing:
permission = model.PermissionReadElasticsearchPostIndexingJob
case model.JobTypeElasticsearchPostAggregation:
permission = model.PermissionReadElasticsearchPostAggregationJob
case model.JobTypeLdapSync:
permission = model.PermissionReadLdapSyncJob
case
model.JobTypeMigrations,
model.JobTypePlugins,
model.JobTypeProductNotices,
model.JobTypeExpiryNotify,
model.JobTypeActiveUsers,
model.JobTypeImportProcess,
model.JobTypeImportDelete,
model.JobTypeExportProcess,
model.JobTypeExportDelete,
model.JobTypeCloud,
model.JobTypeMobileSessionMetadata,
model.JobTypeExtractContent,
model.JobTypeCleanupExpiredAccessTokens:
permission = model.PermissionReadJobs
case model.JobTypeAccessControlSync:
permission = model.PermissionManageSystem
case model.JobTypeAccessControlTeamSync:
permission = model.PermissionManageTeamAccessRules
}
// TODO: ContainsSensitiveData is set unconditionally as a temporary workaround for mixed-version
// cluster rollouts. Older nodes don't know about RequiredPermissions and silently drop it when
// decoding the cluster message, so this makes them fail closed (sysadmin-only) instead of
// broadcasting the unfiltered job to everyone. ShouldSendEvent on upgraded nodes ignores
// ContainsSensitiveData whenever RequiredPermissions is set. Remove once permission-scoped
// WebSocket delivery no longer needs to interoperate with pre-MM-69403 nodes.
message.GetBroadcast().ContainsSensitiveData = true
if permission != nil {
message.GetBroadcast().RequiredPermissions = []string{permission.Id}
}
srv.publish(message)
}
func (srv *JobServer) SetJobProgress(job *model.Job, progress int64) *model.AppError {
job.Status = model.JobStatusInProgress
job.Progress = progress
ret, err := srv.Store.Job().UpdateOptimistically(job, model.JobStatusInProgress)
if err != nil {
return model.NewAppError("SetJobProgress", "app.job.update.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
if ret != nil {
srv.publishJobStatus(ret, model.JobStatusInProgress)
}
return nil
}
func (srv *JobServer) SetJobWarning(job *model.Job) *model.AppError {
ret, err := srv.Store.Job().UpdateStatus(job.Id, model.JobStatusWarning)
if err != nil {
return model.NewAppError("SetJobWarning", "app.job.update.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
if srv.metrics != nil {
srv.metrics.DecrementJobActive(job.Type)
}
srv.publishJobStatus(ret, model.JobStatusWarning)
return nil
}
func (srv *JobServer) SetJobSuccess(job *model.Job) *model.AppError {
ret, err := srv.Store.Job().UpdateStatus(job.Id, model.JobStatusSuccess)
if err != nil {
return model.NewAppError("SetJobSuccess", "app.job.update.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
if srv.metrics != nil {
srv.metrics.DecrementJobActive(ret.Type)
}
srv.publishJobStatus(ret, model.JobStatusSuccess)
return nil
}
func (srv *JobServer) SetJobError(job *model.Job, jobError *model.AppError) *model.AppError {
if jobError == nil {
ret, err := srv.Store.Job().UpdateStatus(job.Id, model.JobStatusError)
if err != nil {
return model.NewAppError("SetJobError", "app.job.update.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
if srv.metrics != nil {
srv.metrics.DecrementJobActive(ret.Type)
}
srv.publishJobStatus(ret, model.JobStatusError)
return nil
}
job.Status = model.JobStatusError
job.Progress = -1
if job.Data == nil {
job.Data = make(map[string]string)
}
job.Data["error"] = jobError.Message
if jobError.DetailedError != "" {
job.Data["error"] += " — " + jobError.DetailedError
}
if wrapped := jobError.Unwrap(); wrapped != nil {
job.Data["error"] += " — " + wrapped.Error()
}
ret, err := srv.Store.Job().UpdateOptimistically(job, model.JobStatusInProgress)
if err != nil {
return model.NewAppError("SetJobError", "app.job.update.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
if ret != nil && srv.metrics != nil {
srv.metrics.DecrementJobActive(job.Type)
}
if ret == nil {
ret, err = srv.Store.Job().UpdateOptimistically(job, model.JobStatusCancelRequested)
if err != nil {
return model.NewAppError("SetJobError", "app.job.update.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
if ret == nil {
return model.NewAppError("SetJobError", "jobs.set_job_error.update.error", nil, "id="+job.Id, http.StatusInternalServerError)
}
}
srv.publishJobStatus(ret, ret.Status)
return nil
}
func (srv *JobServer) SetJobCanceled(job *model.Job) *model.AppError {
ret, err := srv.Store.Job().UpdateStatus(job.Id, model.JobStatusCanceled)
if err != nil {
return model.NewAppError("SetJobCanceled", "app.job.update.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
if srv.metrics != nil {
srv.metrics.DecrementJobActive(ret.Type)
}
srv.publishJobStatus(ret, model.JobStatusCanceled)
return nil
}
func (srv *JobServer) SetJobPending(job *model.Job) *model.AppError {
ret, err := srv.Store.Job().UpdateStatus(job.Id, model.JobStatusPending)
if err != nil {
return model.NewAppError("SetJobPending", "app.job.update.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
if srv.metrics != nil {
srv.metrics.DecrementJobActive(ret.Type)
}
srv.publishJobStatus(ret, model.JobStatusPending)
return nil
}
func (srv *JobServer) UpdateInProgressJobData(job *model.Job) *model.AppError {
job.Status = model.JobStatusInProgress
job.LastActivityAt = model.GetMillis()
if _, err := srv.Store.Job().UpdateOptimistically(job, model.JobStatusInProgress); err != nil {
return model.NewAppError("UpdateInProgressJobData", "app.job.update.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
return nil
}
// HandleJobPanic is used to handle panics during the execution of a job. It logs the panic and sets the status for the job.
// After handling, the method repanics! This method is supposed to be `defer`'d at the start of the job.
func (srv *JobServer) HandleJobPanic(logger mlog.LoggerIFace, job *model.Job) {
r := recover()
if r == nil {
return
}
sb := &strings.Builder{}
err := pprof.Lookup("goroutine").WriteTo(sb, 2)
if err != nil {
logger.Error("Error fetching goroutine stack", mlog.Err(err))
}
logger.Error("Unhandled panic in job", mlog.Any("panic", r), mlog.Any("job", job), mlog.String("stack", sb.String()))
rerr, ok := r.(error)
if !ok {
rerr = fmt.Errorf("job panic: %v", r)
}
appErr := srv.SetJobError(job, model.NewAppError("HandleJobPanic", "app.job.update.app_error", nil, "", http.StatusInternalServerError))
if appErr != nil {
appErr = appErr.Wrap(rerr)
logger.Error("Failed to set the job status to 'failed'", mlog.Err(appErr), mlog.Any("job", job))
}
panic(r)
}
func (srv *JobServer) RequestCancellation(rctx request.CTX, jobId string) *model.AppError {
newJob, err := srv.Store.Job().UpdateStatusOptimistically(jobId, model.JobStatusPending, model.JobStatusCanceled)
if err != nil {
return model.NewAppError("RequestCancellation", "app.job.update.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
if newJob != nil {
if srv.metrics != nil {
srv.metrics.DecrementJobActive(newJob.Type)
}
srv.publishJobStatus(newJob, model.JobStatusCanceled)
return nil
}
newJob, err = srv.Store.Job().UpdateStatusOptimistically(jobId, model.JobStatusInProgress, model.JobStatusCancelRequested)
if err != nil {
return model.NewAppError("RequestCancellation", "app.job.update.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
if newJob != nil {
srv.publishJobStatus(newJob, model.JobStatusCancelRequested)
return nil
}
return model.NewAppError("RequestCancellation", "jobs.request_cancellation.status.error", nil, "id="+jobId, http.StatusInternalServerError)
}
func (srv *JobServer) CancellationWatcher(rctx request.CTX, jobId string, cancelChan chan struct{}) {
for {
select {
case <-rctx.Context().Done():
rctx.Logger().Debug("CancellationWatcher for Job Aborting as job has finished.", mlog.String("job_id", jobId))
return
case <-time.After(CancelWatcherPollingInterval * time.Millisecond):
rctx.Logger().Debug("CancellationWatcher for Job started polling.", mlog.String("job_id", jobId))
jobStatus, err := srv.Store.Job().Get(rctx, jobId)
if err != nil {
rctx.Logger().Warn("Error getting job", mlog.String("job_id", jobId), mlog.Err(err))
continue
}
if jobStatus.Status == model.JobStatusCancelRequested {
close(cancelChan)
return
}
}
}
}
func GenerateNextStartDateTime(now time.Time, nextStartTime time.Time) *time.Time {
nextTime := time.Date(now.Year(), now.Month(), now.Day(), nextStartTime.Hour(), nextStartTime.Minute(), 0, 0, time.Local)
if !now.Before(nextTime) {
nextTime = nextTime.AddDate(0, 0, 1)
}
return &nextTime
}
func (srv *JobServer) CheckForPendingJobsByType(jobType string) (bool, *model.AppError) {
count, err := srv.Store.Job().GetCountByStatusAndType(model.JobStatusPending, jobType)
if err != nil {
return false, model.NewAppError("CheckForPendingJobsByType", "app.job.get_count_by_status_and_type.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
return count > 0, nil
}
func (srv *JobServer) GetJobsByTypeAndStatus(rctx request.CTX, jobType string, status string) ([]*model.Job, *model.AppError) {
jobs, err := srv.Store.Job().GetAllByTypeAndStatus(rctx, jobType, status)
if err != nil {
return nil, model.NewAppError("GetJobsByTypeAndStatus", "app.job.get_all_jobs_by_type_and_status.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
return jobs, nil
}
func (srv *JobServer) GetLastSuccessfulJobByType(jobType string) (*model.Job, *model.AppError) {
statuses := []string{model.JobStatusSuccess}
if jobType == model.JobTypeMessageExport {
statuses = []string{model.JobStatusWarning, model.JobStatusSuccess}
}
job, err := srv.Store.Job().GetNewestJobByStatusesAndType(statuses, jobType)
var nfErr *store.ErrNotFound
if err != nil && !errors.As(err, &nfErr) {
return nil, model.NewAppError("GetLastSuccessfulJobByType", "app.job.get_newest_job_by_status_and_type.app_error", nil, "", http.StatusInternalServerError).Wrap(err)
}
return job, nil
}