rclone/fs/rc/jobs/job.go
Michał Matczuk 3247e69cf5 fs/rc/jobs: ExecuteJob propagate the error returned by function
Without this patch the resulting error is first converted to string and then recreated.
This makes it impossible to use the defined error types to figure out the cause of the error,
and may result in invalid HTTP status codes.

This patch adds a test TestExecuteJobErrorPropagation to validate that the errors are
properly propagated.
2019-08-22 16:10:48 +01:00

323 lines
7.5 KiB
Go

// Manage background jobs that the rc is running
package jobs
import (
"context"
"fmt"
"runtime/debug"
"sync"
"sync/atomic"
"time"
"github.com/pkg/errors"
"github.com/rclone/rclone/fs"
"github.com/rclone/rclone/fs/accounting"
"github.com/rclone/rclone/fs/rc"
)
// Job describes a asynchronous task started via the rc package
type Job struct {
mu sync.Mutex
ID int64 `json:"id"`
Group string `json:"group"`
StartTime time.Time `json:"startTime"`
EndTime time.Time `json:"endTime"`
Error string `json:"error"`
Finished bool `json:"finished"`
Success bool `json:"success"`
Duration float64 `json:"duration"`
Output rc.Params `json:"output"`
Stop func() `json:"-"`
// realErr is the Error before printing it as a string, it's used to return
// the real error to the upper application layers while still printing the
// string error message.
realErr error
}
// Jobs describes a collection of running tasks
type Jobs struct {
mu sync.RWMutex
jobs map[int64]*Job
opt *rc.Options
expireRunning bool
}
var (
running = newJobs()
jobID = int64(0)
)
// newJobs makes a new Jobs structure
func newJobs() *Jobs {
return &Jobs{
jobs: map[int64]*Job{},
opt: &rc.DefaultOpt,
}
}
// SetOpt sets the options when they are known
func SetOpt(opt *rc.Options) {
running.opt = opt
}
// SetInitialJobID allows for setting jobID before starting any jobs.
func SetInitialJobID(id int64) {
if !atomic.CompareAndSwapInt64(&jobID, 0, id) {
panic("Setting jobID is only possible before starting any jobs")
}
}
// kickExpire makes sure Expire is running
func (jobs *Jobs) kickExpire() {
jobs.mu.Lock()
defer jobs.mu.Unlock()
if !jobs.expireRunning {
time.AfterFunc(jobs.opt.JobExpireInterval, jobs.Expire)
jobs.expireRunning = true
}
}
// Expire expires any jobs that haven't been collected
func (jobs *Jobs) Expire() {
jobs.mu.Lock()
defer jobs.mu.Unlock()
now := time.Now()
for ID, job := range jobs.jobs {
job.mu.Lock()
if job.Finished && now.Sub(job.EndTime) > jobs.opt.JobExpireDuration {
delete(jobs.jobs, ID)
}
job.mu.Unlock()
}
if len(jobs.jobs) != 0 {
time.AfterFunc(jobs.opt.JobExpireInterval, jobs.Expire)
jobs.expireRunning = true
} else {
jobs.expireRunning = false
}
}
// IDs returns the IDs of the running jobs
func (jobs *Jobs) IDs() (IDs []int64) {
jobs.mu.RLock()
defer jobs.mu.RUnlock()
IDs = []int64{}
for ID := range jobs.jobs {
IDs = append(IDs, ID)
}
return IDs
}
// Get a job with a given ID or nil if it doesn't exist
func (jobs *Jobs) Get(ID int64) *Job {
jobs.mu.RLock()
defer jobs.mu.RUnlock()
return jobs.jobs[ID]
}
// mark the job as finished
func (job *Job) finish(out rc.Params, err error) {
job.mu.Lock()
job.EndTime = time.Now()
if out == nil {
out = make(rc.Params)
}
job.Output = out
job.Duration = job.EndTime.Sub(job.StartTime).Seconds()
if err != nil {
job.realErr = err
job.Error = err.Error()
job.Success = false
} else {
job.realErr = nil
job.Error = ""
job.Success = true
}
job.Finished = true
job.mu.Unlock()
running.kickExpire() // make sure this job gets expired
}
// run the job until completion writing the return status
func (job *Job) run(ctx context.Context, fn rc.Func, in rc.Params) {
defer func() {
if r := recover(); r != nil {
job.finish(nil, errors.Errorf("panic received: %v \n%s", r, string(debug.Stack())))
}
}()
job.finish(fn(ctx, in))
}
func getGroup(in rc.Params) string {
// Check to see if the group is set
group, err := in.GetString("_group")
if rc.NotErrParamNotFound(err) {
fs.Errorf(nil, "Can't get _group param %+v", err)
}
delete(in, "_group")
return group
}
// NewAsyncJob start a new asynchronous Job off
func (jobs *Jobs) NewAsyncJob(fn rc.Func, in rc.Params) *Job {
id := atomic.AddInt64(&jobID, 1)
group := getGroup(in)
if group == "" {
group = fmt.Sprintf("job/%d", id)
}
ctx := accounting.WithStatsGroup(context.Background(), group)
ctx, cancel := context.WithCancel(ctx)
stop := func() {
cancel()
// Wait for cancel to propagate before returning.
<-ctx.Done()
}
job := &Job{
ID: id,
Group: group,
StartTime: time.Now(),
Stop: stop,
}
jobs.mu.Lock()
jobs.jobs[job.ID] = job
jobs.mu.Unlock()
go job.run(ctx, fn, in)
return job
}
// NewSyncJob start a new synchronous Job off
func (jobs *Jobs) NewSyncJob(ctx context.Context, in rc.Params) (*Job, context.Context) {
id := atomic.AddInt64(&jobID, 1)
group := getGroup(in)
if group == "" {
group = fmt.Sprintf("job/%d", id)
}
ctxG := accounting.WithStatsGroup(ctx, fmt.Sprintf("job/%d", id))
ctx, cancel := context.WithCancel(ctxG)
stop := func() {
cancel()
// Wait for cancel to propagate before returning.
<-ctx.Done()
}
job := &Job{
ID: id,
Group: group,
StartTime: time.Now(),
Stop: stop,
}
jobs.mu.Lock()
jobs.jobs[job.ID] = job
jobs.mu.Unlock()
return job, ctx
}
// StartAsyncJob starts a new job asynchronously and returns a Param suitable
// for output.
func StartAsyncJob(fn rc.Func, in rc.Params) (rc.Params, error) {
job := running.NewAsyncJob(fn, in)
out := make(rc.Params)
out["jobid"] = job.ID
return out, nil
}
// ExecuteJob executes new job synchronously and returns a Param suitable for
// output.
func ExecuteJob(ctx context.Context, fn rc.Func, in rc.Params) (rc.Params, int64, error) {
job, ctx := running.NewSyncJob(ctx, in)
job.run(ctx, fn, in)
return job.Output, job.ID, job.realErr
}
func init() {
rc.Add(rc.Call{
Path: "job/status",
Fn: rcJobStatus,
Title: "Reads the status of the job ID",
Help: `Parameters
- jobid - id of the job (integer)
Results
- finished - boolean
- duration - time in seconds that the job ran for
- endTime - time the job finished (eg "2018-10-26T18:50:20.528746884+01:00")
- error - error from the job or empty string for no error
- finished - boolean whether the job has finished or not
- id - as passed in above
- startTime - time the job started (eg "2018-10-26T18:50:20.528336039+01:00")
- success - boolean - true for success false otherwise
- output - output of the job as would have been returned if called synchronously
- progress - output of the progress related to the underlying job
`,
})
}
// Returns the status of a job
func rcJobStatus(ctx context.Context, in rc.Params) (out rc.Params, err error) {
jobID, err := in.GetInt64("jobid")
if err != nil {
return nil, err
}
job := running.Get(jobID)
if job == nil {
return nil, errors.New("job not found")
}
job.mu.Lock()
defer job.mu.Unlock()
out = make(rc.Params)
err = rc.Reshape(&out, job)
if err != nil {
return nil, errors.Wrap(err, "reshape failed in job status")
}
return out, nil
}
func init() {
rc.Add(rc.Call{
Path: "job/list",
Fn: rcJobList,
Title: "Lists the IDs of the running jobs",
Help: `Parameters - None
Results
- jobids - array of integer job ids
`,
})
}
// Returns list of job ids.
func rcJobList(ctx context.Context, in rc.Params) (out rc.Params, err error) {
out = make(rc.Params)
out["jobids"] = running.IDs()
return out, nil
}
func init() {
rc.Add(rc.Call{
Path: "job/stop",
Fn: rcJobStop,
Title: "Stop the running job",
Help: `Parameters
- jobid - id of the job (integer)
`,
})
}
// Stops the running job.
func rcJobStop(ctx context.Context, in rc.Params) (out rc.Params, err error) {
jobID, err := in.GetInt64("jobid")
if err != nil {
return nil, err
}
job := running.Get(jobID)
if job == nil {
return nil, errors.New("job not found")
}
job.mu.Lock()
defer job.mu.Unlock()
out = make(rc.Params)
job.Stop()
return out, nil
}