Michael Gaffney
9529491fa6
This changes the periodic job that deletes terminated sessions to delete the those sessions in batches. Prior to this, a single SQL DELETE command was used to delete all qualifying terminated sessions. In systems under heavy load, the statement could timeout leading to a situation where no qualifying terminated sessions were ever being deleted.
106 lines
3.0 KiB
Go
106 lines
3.0 KiB
Go
// Copyright (c) HashiCorp, Inc.
|
||
// SPDX-License-Identifier: BUSL-1.1
|
||
|
||
package session
|
||
|
||
import (
|
||
"context"
|
||
"sync"
|
||
"time"
|
||
|
||
"github.com/hashicorp/boundary/internal/errors"
|
||
"github.com/hashicorp/boundary/internal/scheduler"
|
||
"github.com/hashicorp/boundary/internal/scheduler/batch"
|
||
)
|
||
|
||
type deleteTerminatedJob struct {
|
||
repo *Repository
|
||
|
||
// the amount of time that a session must be in the terminated
|
||
// state for it to be deleted.
|
||
threshold time.Duration
|
||
|
||
mu sync.Mutex
|
||
batch *batch.Batch
|
||
}
|
||
|
||
func newDeleteTerminatedJob(ctx context.Context, repo *Repository, threshold time.Duration) (*deleteTerminatedJob, error) {
|
||
const op = "session.newDeleteTerminatedJob"
|
||
switch {
|
||
case repo == nil:
|
||
return nil, errors.New(ctx, errors.InvalidParameter, op, "missing repository")
|
||
}
|
||
|
||
return &deleteTerminatedJob{
|
||
repo: repo,
|
||
threshold: threshold,
|
||
}, nil
|
||
}
|
||
|
||
// Status reports the job’s current status. The status is periodically persisted by
|
||
// the scheduler when a job is running, and will be used to verify a job is making progress.
|
||
func (d *deleteTerminatedJob) Status() scheduler.JobStatus {
|
||
d.mu.Lock()
|
||
defer d.mu.Unlock()
|
||
if d.batch != nil {
|
||
return d.batch.Status()
|
||
}
|
||
return scheduler.JobStatus{}
|
||
}
|
||
|
||
// Run performs the required work depending on the implementation.
|
||
// The context is used to notify the job that it should exit early.
|
||
func (d *deleteTerminatedJob) Run(ctx context.Context, statusThreshold time.Duration) error {
|
||
const op = "session.(deleteTerminatedJob).Run"
|
||
|
||
params, err := d.repo.getDeleteJobParams(ctx, d.threshold)
|
||
switch {
|
||
case err != nil:
|
||
return errors.Wrap(ctx, err, op)
|
||
case params.TotalToDelete == 0:
|
||
return nil
|
||
}
|
||
|
||
exec := func() batch.Exec {
|
||
return func(ctx context.Context, batchSize int) (int, error) {
|
||
return d.repo.deleteTerminatedSessionsBatch(ctx, params.WindowStartTime, batchSize)
|
||
}
|
||
}
|
||
|
||
config := &batch.Config{
|
||
Size: params.BatchSize,
|
||
TotalToComplete: params.TotalToDelete,
|
||
StatusThreshold: statusThreshold,
|
||
Exec: exec(),
|
||
Store: d.repo.setDeleteJobBatchSize,
|
||
}
|
||
|
||
batch, err := batch.New(ctx, config)
|
||
if err != nil {
|
||
return errors.Wrap(ctx, err, op)
|
||
}
|
||
d.mu.Lock()
|
||
d.batch = batch
|
||
d.mu.Unlock()
|
||
return batch.Run(ctx)
|
||
}
|
||
|
||
// NextRunIn returns the duration until the next job run should be scheduled. This
|
||
// method is invoked after a run has successfully completed and the next run time
|
||
// is being persisted by the scheduler. If an error is returned, the error will be logged
|
||
// but the duration returned will still be used in scheduling. If a zero duration is returned
|
||
// the job will be scheduled to run again immediately.
|
||
func (d *deleteTerminatedJob) NextRunIn(_ context.Context) (time.Duration, error) {
|
||
return time.Minute * 30, nil
|
||
}
|
||
|
||
// Name is the unique name of the job.
|
||
func (d *deleteTerminatedJob) Name() string {
|
||
return "delete_terminated_sessions"
|
||
}
|
||
|
||
// Description is the human readable description of the job.
|
||
func (d *deleteTerminatedJob) Description() string {
|
||
return "Delete terminated sessions that were terminated "
|
||
}
|