What causes "create temp for %s: %w" in multica-ai/multica?

Check df -h and quota on the filesystem holding the destination path; free space or raise the quota.

create temp for %s: %w

Error message

create temp for %s: %w

What it means

writeFileAtomic could not create a temp file in the destination directory (os.CreateTemp with prefix .hermes-tmp-*). Every derived-file write (config.yaml, marker, memory store sidecars) goes through this atomic-write helper; failure here usually means the destination directory is missing, unwritable, or the filesystem is exhausted.

Source

Thrown at server/internal/daemon/execenv/hermes_home.go:917

// marshalYAMLToFile renders a YAML node to dst as a 0600 file (it can hold
// inline secrets) via atomic replace.
func marshalYAMLToFile(doc *yaml.Node, dst string) error {
	out, err := yaml.Marshal(doc)
	if err != nil {
		return fmt.Errorf("marshal hermes config: %w", err)
	}
	return writeFileAtomic(dst, out, 0o600)
}

// writeFileAtomic writes data to a temp file in the destination directory with
// the given perms, then renames it over dst — so readers never see a partial
// file and a prior file's looser permissions are replaced.
func writeFileAtomic(dst string, data []byte, perm os.FileMode) error {
	dir := filepath.Dir(dst)
	tmp, err := os.CreateTemp(dir, ".hermes-tmp-*")
	if err != nil {
		return fmt.Errorf("create temp for %s: %w", dst, err)
	}
	tmpName := tmp.Name()
	defer os.Remove(tmpName) // no-op once renamed
	if _, err := tmp.Write(data); err != nil {
		tmp.Close()
		return fmt.Errorf("write temp for %s: %w", dst, err)
	}
	if err := tmp.Chmod(perm); err != nil {
		tmp.Close()
		return fmt.Errorf("chmod temp for %s: %w", dst, err)
	}
	if err := tmp.Close(); err != nil {
		return fmt.Errorf("close temp for %s: %w", dst, err)
	}
	if err := os.Rename(tmpName, dst); err != nil {
		return fmt.Errorf("rename temp to %s: %w", dst, err)
	}
	return nil

View on GitHub (pinned to 2c0912b6ec)

Solutions

Check df -h and quota on the filesystem holding the destination path; free space or raise the quota.
Verify the destination directory exists and is writable by the daemon user; recreate/fix as needed.
Check the daemon's open-FD count (`ls /proc/<pid>/fd | wc -l`) and restart it if leaking.
If the FS remounted read-only after an error, address the disk problem and remount.

Defensive patterns

Strategy: validation

Validate before calling

if err := ensureWritableDir(filepath.Dir(dst)); err != nil {
	return fmt.Errorf("cannot write to %s: %w", filepath.Dir(dst), err)
}

func ensureWritableDir(dir string) error {
	fi, err := os.Stat(dir)
	if err != nil {
		return err
	}
	if !fi.IsDir() {
		return fmt.Errorf("not a directory: %s", dir)
	}
	probe, perr := os.CreateTemp(dir, ".probe-*")
	if perr != nil {
		return perr
	}
	probe.Close()
	os.Remove(probe.Name())
	return nil
}

Try / catch

if err := writeFileAtomic(dst, data, 0o600); err != nil {
	var pe *os.PathError
	if errors.As(err, &pe) && (errors.Is(pe.Err, syscall.ENOSPC) || errors.Is(pe.Err, syscall.EROFS)) {
		log.Printf("storage problem on %s: %v — free space / check mount", filepath.Dir(dst), pe.Err)
	}
	return err
}

Prevention

Alert on disk usage of the env/workspaces volume before it hits 100%.
Validate writability of destination dirs at daemon startup, not mid-task.

When it happens

Trigger: Destination dir was deleted concurrently; daemon lacks write permission on the overlay/config dir; ENOSPC (disk/quota full); EMFILE (too many open files); read-only filesystem.

Common situations: Disk-full CI boxes or containers where env RootDir lives on a small volume; overlay dirs wiped by GC mid-prepare; overlay on a read-only mount after a disk error; long-running daemon leaking file descriptors.

Related errors

AI-assisted analysis of multica-ai/multica@2c0912b6ec (2026-08-15). Data as JSON: /api/errors/9838f8777938fdbb. Report an issue: GitHub.