argoproj/argo-workflows · error

failed to create temp file: %w

Error message

failed to create temp file: %w

What it means

BufferReaderToTempFile creates a temp file with os.CreateTemp to buffer an artifact stream before saving. This error wraps the os.CreateTemp failure, meaning the process could not create a file in the system temp dir (os.TempDir, typically /tmp).

Source

Thrown at workflow/artifacts/common/streaming.go:33

	if err != nil {
		return err
	}
	defer cleanup()
	return save(path)
}

// BufferReaderToTempFile buffers reader into a new temp file (named per the os.CreateTemp
// pattern, e.g. "s3-upload-*") so its content can be re-read multiple times, which most
// storage SDKs require for retry/backoff. It returns the temp file's path and a cleanup
// function that removes it; cleanup is safe to call more than once. On error, any partially
// written temp file is removed before returning, so callers only need to defer cleanup
// after a nil error.
func BufferReaderToTempFile(reader io.Reader, pattern string) (path string, cleanup func(), err error) {
	noop := func() {}

	tmpFile, err := os.CreateTemp("", pattern)
	if err != nil {
		return "", noop, fmt.Errorf("failed to create temp file: %w", err)
	}
	name := tmpFile.Name()
	cleanup = func() {
		_ = os.Remove(name)
	}

	if _, err := io.Copy(tmpFile, reader); err != nil {
		_ = tmpFile.Close()
		cleanup()
		return "", noop, fmt.Errorf("failed to buffer stream to temp file: %w", err)
	}
	if err := tmpFile.Close(); err != nil {
		cleanup()
		return "", noop, fmt.Errorf("failed to close temp file: %w", err)
	}

	return name, cleanup, nil
}

View on GitHub (pinned to 35bff19146)

Solutions

1. Free space on the temp filesystem (df -h /tmp) or expand the node/pod disk.
2. Mount a writable tmpfs/emptyDir at /tmp or set TMPDIR to a writable volume.
3. If read-only rootfs, add an emptyDir volume and mount it at /tmp (or set TMPDIR to a mounted path).
4. Check open-file limits (lsof count, ulimit) for fd leaks; restart the leaking process.
5. Set the TMPDIR env var to a larger persistent volume if streams are very large.

Example fix

# before: read-only container, no tmp
spec:
  containers: [ ... ]
# after
spec:
  containers:
  - name: main
    volumeMounts: [{name: tmp, mountPath: /tmp}]
  volumes:
  - name: tmp
    emptyDir: {}

// verify the temp dir is writable and has space before streaming
if info, err := os.Stat(os.TempDir()); err != nil || !info.IsDir() {
    return fmt.Errorf("temp dir %s unusable", os.TempDir())
}
if err := unix.Access(os.TempDir(), unix.W_OK); err != nil {
    return fmt.Errorf("temp dir not writable: %w", err)
}

if err := artifactscommon.SaveStreamViaTempFile(reader, "upload-*", save); err != nil {
    if strings.Contains(err.Error(), "failed to create temp file") && errors.Is(err, syscall.ENOSPC) {
        return fmt.Errorf("disk full on temp volume — expand /tmp or set TMPDIR")
    }
    return err
}

When it happens

Trigger: os.CreateTemp returns an error: temp filesystem full (ENOSPC), no read-write /tmp in the container, permission denied on the temp dir, or too many open file descriptors (EMFILE/ENFILE).

Common situations: Pods with a tiny read-only emptyDir at /tmp; node disk pressure; containers run with read-only root filesystem and no tmpfs mount; ulimit -n exhaustion from leaked fds in a long-running controller.

Related errors

• failed to buffer stream to temp file: %w
• failed to copy %s to %s: %w
• failed to close temp file: %w
• failed to create folder path: %w
• failed to create ctr directory: %w

AI-assisted analysis of argoproj/argo-workflows@35bff19146 (2026-09-03). Data as JSON: /api/errors/f06bc1e0dcdf7666. Report an issue: GitHub.