4:47 a.m.
On 12/08/21 16:18, Richard W.M. Jones wrote:
On Wed, Dec 08, 2021 at 01:20:48PM +0100, Laszlo Ersek wrote:
> Connecting to an NBD server temporarily, for a "one-shot" operation, is
> quite similar to "Std_utils.with_open_in" and
"Std_utils.with_open_out",
> as there are cleanup operations regardless of whether the "one-shot"
> operation completes successfully or throws an exception.
>
> Introduce the "Nbdkit.with_connect_unix" function, which takes a Unix
> domain socket pathname, a list of metadata contexts to request from the
> NBD server, and calls a function with the live NBD server connection.
>
> Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=2027598
> Signed-off-by: Laszlo Ersek <lersek(a)redhat.com>
> ---
> lib/Makefile.am | 2 +-
> lib/nbdkit.mli | 10 ++++++++++
> lib/nbdkit.ml | 12 ++++++++++++
> 3 files changed, 23 insertions(+), 1 deletion(-)
>
> diff --git a/lib/Makefile.am b/lib/Makefile.am
> index c274b9ecf6c7..1fab25b326f5 100644
> --- a/lib/Makefile.am
> +++ b/lib/Makefile.am
> @@ -80,7 +80,7 @@ BOBJECTS = config.cmo $(SOURCES_ML:.ml=.cmo)
> XOBJECTS = $(BOBJECTS:.cmo=.cmx)
>
> OCAMLPACKAGES = \
> - -package str,unix \
> + -package str,unix,nbd \
> -I $(builddir) \
> -I $(top_builddir)/common/mlgettext \
> -I $(top_builddir)/common/mlpcre \
> diff --git a/lib/nbdkit.mli b/lib/nbdkit.mli
> index ae19295186ed..825755e61dbe 100644
> --- a/lib/nbdkit.mli
> +++ b/lib/nbdkit.mli
> @@ -107,3 +107,13 @@ val run_unix : ?socket:string -> cmd -> string * int
> The --exit-with-parent, --foreground, --pidfile, --newstyle and
> --unix flags are added automatically. Other flags are set as
> in the {!cmd} struct. *)
> +
> +val with_connect_unix : socket:string ->
> + meta_contexts:string list ->
> + f:(NBD.t -> 'a) ->
> + 'a
> +(** [with_connect_unix socket meta_contexts f] calls function [f] with the NBD
> + server at Unix domain socket [socket] connected, and the metadata contexts
> + in [meta_contexts] requested (each of which is not necessarily supported by
> + the server though). The connection is torn down either on normal return or
> + if the function [f] throws an exception. *)
Interesting choice of module for this. nbdkit is a server. I would
have put it into lib/utils.mli or created a new module.
Yes, I was certainly lost as to what module to pick :) I found
"run_unix" in this one, and thought that a wrapper for "connect_unix"
(from the OCaml binding of libnbd) belonged here.
I can add it to "utils" if you think that's a good fit. (I feel that
creating a new module just for this is overkill, but I could be wrong.)
The function itself looks fine although personally I might have
abstracted the meta_context function to be a "do anything between
creation and connection" (let's call that "preparation"). What do
you
think about:
val with_connect_unix : socket:string ->
?prepare:(NBD.t -> unit) ->
f:(NBD.t -> 'a) ->
'a
I think you would have to replace the List.iter line:
...
> + ~f:(fun () ->
(match prepare with Some pf -> pf nbd | None -> ());
> + NBD.connect_unix nbd socket;
> + protect
> + ~f:(fun () -> f nbd)
> + ~finally:(fun () -> NBD.shutdown nbd)
> + )
> + ~finally:(fun () -> NBD.close nbd)
It would be called like this:
let prepare nbd = NBD.add_meta_context nbd "base:allocation" in
with_connect_unix socket ~prepare (
fun () ->
(* the code *)
)
Hmm, maybe this is getting more complicated :-(
Well my concern is the following; let me paste the full function again:
let with_connect_unix ~socket ~meta_contexts ~f =
let nbd = NBD.create () in
protect
~f:(fun () ->
List.iter (NBD.add_meta_context nbd) meta_contexts;
NBD.connect_unix nbd socket;
protect
~f:(fun () -> f nbd)
~finally:(fun () -> NBD.shutdown nbd)
)
~finally:(fun () -> NBD.close nbd)
"NBD.close", which mirrors "NBD.create", suffices for undoing
"NBD.add_meta_context" as well.
But, if we turn "NBD.add_meta_context" into a generic "do anything
between NBD.create and NBD.connect_unix", I don't know if "NBD.close"
can still satisfy the teardown role. In that case, we might have to
accept another "rollback" function, and *that* I do find too complicated.
FWIW, the current code does make a similar assumption about "f", and
"NBD.shutdown". However, I thought that that was OK (and didn't need
spelling out): namely, once "NBD.connect_unix" completes, any valid
operation on the connection is considered... well, valid (or
"expected"), and "NBD.shutdown" is considered appropriate to finish
*any* such sequence of operations. I don't have the same confidence in a
generic "prepare" being undone by just NBD.close.
Thanks
Laszlo