: unit -> t
val report
: t
- -> time_all:float
- -> time_group_by_size:float
- -> time_group_by_head:float
- -> time_group_by_digest:float
+ -> wall_time_all:float
+ -> wall_time_group_by_size:float
+ -> wall_time_group_by_head:float
+ -> wall_time_group_by_digest:float
+ -> proc_time_all:float
+ -> proc_time_group_by_size:float
+ -> proc_time_group_by_head:float
+ -> proc_time_group_by_digest:float
-> unit
val file_considered
: t -> size:int -> unit
val digest
: t -> unit
+ val redundant_data
+ : t -> size:int -> unit
end = struct
type t =
{ considered_files : int ref
; hashed_files : int ref
; hashed_bytes : int ref
; digests : int ref
+ ; redundant_data : int ref
}
let init () =
; unique_sample_files = ref 0
; unique_sample_bytes = ref 0
; digests = ref 0
+ ; redundant_data = ref 0
}
let add sum addend =
let digest t =
incr t.digests
+ let redundant_data t ~size =
+ add t.redundant_data size
+
let report
t
- ~time_all
- ~time_group_by_size
- ~time_group_by_head
- ~time_group_by_digest
+ ~wall_time_all
+ ~wall_time_group_by_size
+ ~wall_time_group_by_head
+ ~wall_time_group_by_digest
+ ~proc_time_all
+ ~proc_time_group_by_size
+ ~proc_time_group_by_head
+ ~proc_time_group_by_digest
=
let b_to_mb b = (float_of_int b) /. 1024. /. 1024. in
let b_to_gb b = (b_to_mb b) /. 1024. in
- eprintf "Time : %8.2f seconds\n%!"
- time_all;
+ eprintf "Total time : %.2f wall sec %.2f proc sec\n%!"
+ wall_time_all
+ proc_time_all;
eprintf "Considered : %8d files %6.2f Gb\n%!"
!(t.considered_files)
(b_to_gb !(t.considered_bytes));
eprintf "Sampled : %8d files %6.2f Gb\n%!"
!(t.sampled_files)
(b_to_gb !(t.sampled_bytes));
- eprintf "Hashed : %8d files %6.2f Gb %6.2f seconds\n%!"
+ eprintf "Hashed : %8d files %6.2f Gb %6.2f wall sec %6.2f proc sec\n%!"
!(t.hashed_files)
(b_to_gb !(t.hashed_bytes))
- time_group_by_digest;
+ wall_time_group_by_digest
+ proc_time_group_by_digest;
eprintf "Digests : %8d\n%!"
!(t.digests);
- eprintf "Duplicates (Hashed - Digests): %8d\n%!"
- (!(t.hashed_files) - !(t.digests));
+ eprintf "Duplicates (Hashed - Digests): %8d files %6.2f Gb\n%!"
+ (!(t.hashed_files) - !(t.digests))
+ (b_to_gb !(t.redundant_data));
eprintf "Skipped due to 0 size : %8d files\n%!" !(t.empty);
- eprintf "Skipped due to unique size : %8d files %6.2f Gb %6.2f seconds\n%!"
+ eprintf "Skipped due to unique size : %8d files %6.2f Gb %6.2f wall sec %6.2f proc sec\n%!"
!(t.unique_size_files)
(b_to_gb !(t.unique_size_bytes))
- time_group_by_size;
- eprintf "Skipped due to unique sample : %8d files %6.2f Gb %6.2f seconds\n%!"
+ wall_time_group_by_size
+ proc_time_group_by_size;
+ eprintf "Skipped due to unique sample : %8d files %6.2f Gb %6.2f wall sec %6.2f proc sec\n%!"
!(t.unique_sample_files)
(b_to_gb !(t.unique_sample_bytes))
- time_group_by_head;
+ wall_time_group_by_head
+ proc_time_group_by_head;
eprintf "Ignored due to regex match : %8d files %6.2f Gb\n%!"
!(t.ignored_files)
(b_to_gb !(t.ignored_bytes))
val iter : 'a t -> f:('a -> unit) -> unit
+ val bag_map : 'a t -> njobs:int -> f:('a -> 'b) -> ('a * 'b) t
+ (** Parallel map with arbitrarily-reordered elements. *)
+
val map : 'a t -> f:('a -> 'b) -> 'b t
val filter : 'a t -> f:('a -> bool) -> 'a t
type 'a t =
{mutable streams : ('a S.t) list}
+ type ('input, 'output) msg_from_vassal =
+ | Ready of int
+ | Result of (int * ('input * 'output))
+ | Exiting of int
+
+ type 'input msg_from_lord =
+ | Job of 'input option
+
let create f =
{streams = [S.from (fun _ -> f ())]}
(fun name (length, members) -> Queue.add (name, length, members) groups)
groups_tbl;
of_queue groups
+
+ module Ipc : sig
+ val send : out_channel -> 'a -> unit
+ val recv : in_channel -> 'a
+ end = struct
+ let send oc msg =
+ Marshal.to_channel oc msg [];
+ flush oc
+
+ let recv ic =
+ Marshal.from_channel ic
+ end
+
+ let lord t ~njobs ~vassals ~ic ~ocs =
+ eprintf "[debug] [lord] started\n%!";
+ let active_vassals = ref njobs in
+ let results = Queue.create () in
+ let rec dispatch () =
+ match Ipc.recv ic with
+ | ((Exiting i) : ('input, 'output) msg_from_vassal) ->
+ close_out ocs.(i);
+ decr active_vassals;
+ if !active_vassals = 0 then
+ ()
+ else
+ dispatch ()
+ | ((Ready i) : ('input, 'output) msg_from_vassal) ->
+ Ipc.send ocs.(i) (Job (next t));
+ dispatch ()
+ | ((Result (i, result)) : ('input, 'output) msg_from_vassal) ->
+ Queue.add result results;
+ Ipc.send ocs.(i) (Job (next t));
+ dispatch ()
+ in
+ let rec wait = function
+ | [] -> ()
+ | vassals ->
+ let pid, _process_status = Unix.wait () in
+ (* TODO: handle process_status *)
+ wait (List.filter vassals ~f:(fun p -> p <> pid))
+ in
+ dispatch ();
+ close_in ic;
+ wait vassals;
+ of_queue results
+
+ let vassal i ~f ~vassal_pipe_r ~lord_pipe_w =
+ eprintf "[debug] [vassal %d] started\n%!" i;
+ let ic = Unix.in_channel_of_descr vassal_pipe_r in
+ let oc = Unix.out_channel_of_descr lord_pipe_w in
+ let rec work msg =
+ Ipc.send oc msg;
+ match Ipc.recv ic with
+ | (Job (Some x) : 'input msg_from_lord) ->
+ work (Result (i, (x, f x)))
+ | (Job None : 'input msg_from_lord) ->
+ Ipc.send oc (Exiting i)
+ in
+ work (Ready i);
+ close_in ic;
+ close_out oc;
+ exit 0
+
+ let bag_map t ~njobs ~f =
+ let lord_pipe_r, lord_pipe_w = Unix.pipe () in
+ let vassal_pipes = Array.init njobs ~f:(fun _ -> Unix.pipe ()) in
+ let vassal_pipes_r = Array.map vassal_pipes ~f:(fun (r, _) -> r) in
+ let vassal_pipes_w = Array.map vassal_pipes ~f:(fun (_, w) -> w) in
+ let vassals = ref [] in
+ for i=0 to (njobs - 1) do
+ begin match Unix.fork () with
+ | 0 ->
+ Unix.close lord_pipe_r;
+ vassal i ~f ~lord_pipe_w ~vassal_pipe_r:vassal_pipes_r.(i)
+ | pid ->
+ vassals := pid :: !vassals
+ end
+ done;
+ Unix.close lord_pipe_w;
+ lord
+ t
+ ~njobs
+ ~vassals:!vassals
+ ~ic:(Unix.in_channel_of_descr lord_pipe_r)
+ ~ocs:(Array.map vassal_pipes_w ~f:Unix.out_channel_of_descr)
end
module In_channel : sig
val lookup : string Stream.t -> t Stream.t
(** Lookup file info for given paths *)
+ val head : t -> len:int -> metrics:M.t -> string
+
val filter_out_unique_sizes : t Stream.t -> metrics:M.t -> t Stream.t
val filter_out_unique_heads : t Stream.t -> len:int -> metrics:M.t -> t Stream.t
end = struct
~group:(fun {size; _} -> size)
~handle_singleton:(fun (size, _, _) -> M.file_unique_size metrics ~size)
- let head path ~len ~metrics =
+ let head {path; _} ~len ~metrics =
+ M.file_sampled metrics;
let buf = Bytes.make len ' ' in
let ic = open_in_bin path in
let rec read pos len =
let filter_out_unique_heads files ~len ~metrics =
filter_out_singletons
files
- ~group:(fun {path; _} ->
- M.file_sampled metrics;
- head path ~len ~metrics
- )
+ ~group:(head ~len ~metrics)
~handle_singleton:(fun (_, _, files) ->
let {size; _} = List.hd files in (* Guaranteed non-empty *)
M.file_unique_sample metrics ~size
type opt =
{ input : input
; output : output
- ; ignore : Str.regexp option
+ ; ignore : string -> bool
; sample : int
+ ; njobs : int
}
let make_input_stream input ignore ~metrics =
Stream.filter input ~f:(fun {File.path; size} ->
M.file_considered metrics ~size;
let empty = size = 0 in
+ let ignored = ignore path in
if empty then M.file_empty metrics;
- let ignored =
- match ignore with
- | Some regexp when (Str.string_match regexp path 0) ->
- M.file_ignored metrics ~size;
- true
- | Some _ | None ->
- false
- in
+ if ignored then M.file_ignored metrics ~size;
(not empty) && (not ignored)
)
);
close_out oc
-let main {input; output; ignore; sample = sample_len} =
- let t0_all = Sys.time () in
+let time_wall () =
+ Unix.gettimeofday ()
+
+let time_proc () =
+ Sys.time ()
+
+let main {input; output; ignore; sample = sample_len; njobs} =
+ let wt0_all = time_wall () in
+ let pt0_all = time_proc () in
let metrics = M.init () in
let output = make_output_fun output in
let input = make_input_stream input ignore ~metrics in
let files = input in
- let t0_group_by_size = Sys.time () in
+ let wt0_group_by_size = time_wall () in
+ let pt0_group_by_size = time_proc () in
+ eprintf "[debug] filtering-out files with unique size\n%!";
let files = File.filter_out_unique_sizes files ~metrics in
- let t1_group_by_size = Sys.time () in
-
- let t0_group_by_sample = t1_group_by_size in
- let files = File.filter_out_unique_heads files ~len:sample_len ~metrics in
- let t1_group_by_sample = Sys.time () in
+ let pt1_group_by_size = time_proc () in
+ let wt1_group_by_size = time_wall () in
+
+ let wt0_group_by_sample = wt1_group_by_size in
+ let pt0_group_by_sample = pt1_group_by_size in
+ eprintf "[debug] filtering-out files with unique heads\n%!";
+ let files =
+ if njobs > 1 then begin
+ let q = Queue.create () in
+ files
+ |> Stream.bag_map ~njobs ~f:(File.head ~len:sample_len ~metrics)
+ |> Stream.group_by ~f:snd
+ |> Stream.map ~f:(fun (d, n, pairs) -> (d, n, List.map pairs ~f:fst))
+ |> Stream.filter ~f:(fun (_, n, _) -> n > 1)
+ |> Stream.iter ~f:(fun (_, _, fs) -> List.iter fs ~f:(fun f -> Queue.add f q))
+ ;
+ Stream.of_queue q
+ end else
+ File.filter_out_unique_heads files ~len:sample_len ~metrics
+ in
+ let pt1_group_by_sample = time_proc () in
+ let wt1_group_by_sample = time_wall () in
- let t0_group_by_digest = t1_group_by_sample in
+ let wt0_group_by_digest = wt1_group_by_sample in
+ let pt0_group_by_digest = pt1_group_by_sample in
+ eprintf "[debug] hashing\n%!";
let groups =
- Stream.group_by files ~f:(fun {File.path; size} ->
- M.file_hashed metrics ~size;
- Digest.file path
- )
+ if njobs > 1 then
+ let with_digests =
+ Stream.bag_map files ~njobs ~f:(fun {File.path; _} -> Digest.file path)
+ in
+ Stream.map (Stream.group_by with_digests ~f:snd) ~f:(
+ fun (digest, n, file_digest_pairs) ->
+ let files =
+ List.map file_digest_pairs ~f:(fun (file, _) ->
+ M.file_hashed metrics ~size:file.File.size;
+ file
+ )
+ in
+ (digest, n, files)
+ )
+ else
+ Stream.group_by files ~f:(fun {File.path; size} ->
+ M.file_hashed metrics ~size;
+ Digest.file path
+ )
in
- let t1_group_by_digest = Sys.time () in
+ let pt1_group_by_digest = time_proc () in
+ let wt1_group_by_digest = time_wall () in
+ eprintf "[debug] reporting\n%!";
Stream.iter groups ~f:(fun (d, n, files) ->
M.digest metrics;
- if n > 1 then output d n files
+ if n > 1 then
+ M.redundant_data metrics ~size:(n * (List.hd files).File.size);
+ output d n files
);
- let t1_all = Sys.time () in
+ let pt1_all = time_proc () in
+ let wt1_all = time_wall () in
M.report metrics
- ~time_all: (t1_all -. t0_all)
- ~time_group_by_size: (t1_group_by_size -. t0_group_by_size)
- ~time_group_by_head: (t1_group_by_sample -. t0_group_by_sample)
- ~time_group_by_digest:(t1_group_by_digest -. t0_group_by_digest)
+ ~wall_time_all: (wt1_all -. wt0_all)
+ ~wall_time_group_by_size: (wt1_group_by_size -. wt0_group_by_size)
+ ~wall_time_group_by_head: (wt1_group_by_sample -. wt0_group_by_sample)
+ ~wall_time_group_by_digest:(wt1_group_by_digest -. wt0_group_by_digest)
+ ~proc_time_all: (pt1_all -. pt0_all)
+ ~proc_time_group_by_size: (pt1_group_by_size -. pt0_group_by_size)
+ ~proc_time_group_by_head: (pt1_group_by_sample -. pt0_group_by_sample)
+ ~proc_time_group_by_digest:(pt1_group_by_digest -. pt0_group_by_digest)
let get_opt () : opt =
let assert_ test x msg =
in
let input = ref Stdin in
let output = ref Stdout in
- let ignore = ref None in
- let sample = ref 256 in
+ let ignore = ref (fun _ -> false) in
+ let sample = ref 512 in
+ let njobs = ref 6 in
let spec =
[ ( "-out"
, Arg.String (fun path ->
, " Output to this directory instead of stdout."
)
; ( "-ignore"
- , Arg.String (fun regexp -> ignore := Some (Str.regexp regexp))
+ , Arg.String (fun regexp ->
+ let regexp = Str.regexp regexp in
+ ignore := fun string -> Str.string_match regexp string 0)
, " Ignore file paths which match this regexp pattern (see Str module)."
)
; ( "-sample"
, Arg.Set_int sample
, (sprintf " Byte size of file samples to use. Default: %d" !sample)
)
+ ; ( "-j"
+ , Arg.Set_int njobs
+ , (sprintf " Number of parallel jobs. Default: %d" !njobs)
+ )
]
in
Arg.parse
; output = !output
; ignore = !ignore
; sample = !sample
+ ; njobs = !njobs
}
let () =