Method CleanStaleSessions

provisionersdk/tfpath/tfpath.go:217–259

CleanStaleSessions browses the work directory searching for stale session directories. Coder provisioner is supposed to remove them once after finishing the provisioning, but there is a risk of keeping them in case of a failure.

(ctx context.Context, logger slog.Logger, fs afero.Fs, now time.Time)

// directories. Coder provisioner is supposed to remove them once after finishing the provisioning,
// but there is a risk of keeping them in case of a failure.
func (l Layout) CleanStaleSessions(ctx context.Context, logger slog.Logger, fs afero.Fs, now time.Time) error {
	parent := filepath.Dir(l.WorkDirectory())
	entries, err := afero.ReadDir(fs, filepath.Dir(l.WorkDirectory()))
	if err != nil {
		return xerrors.Errorf("can't read %q directory", parent)
	}

	for _, fi := range entries {
		dirName := fi.Name()

		if fi.IsDir() && isValidSessionDir(dirName) {
			sessionDirPath := filepath.Join(parent, dirName)

			modTime := fi.ModTime() // fallback to modTime if modTime is not available (afero)

			if modTime.Add(staleSessionRetention).After(now) {
				continue
			}

			logger.Info(ctx, "remove stale session directory", slog.F("session_path", sessionDirPath))
			err = fs.RemoveAll(sessionDirPath)
			if err != nil {
				// This should not be a fatal error. If it is, the provisioner would be rendered
				// non-functional until this directory is cleaned up. Ideally there would be a
				// way to escalate this to an operator alert in Coder. Until then, the best we
				// can do is log it on every cleanup attempt (every build). Eventually the disk
				// usage will be noticeable, and hopefully these logs are noticed.
				logger.Error(ctx, "failed to remove stale session directory",
					slog.F("directory", sessionDirPath),
					slog.Error(err),
				)

				if l.WorkDirectory() == sessionDirPath {
					// This should never happen because sessions are uuid's. But if that logic ever
					// changes, this would be a bad state to be in. The directory that the
					// provisioner is going to use cannot be stale.
					return xerrors.Errorf("remove %q directory, will not work inside a stale directory: %w", sessionDirPath, err)
				}
			}
		}
	}
	return nil
}