|
| 1 | +(* STALLING *) |
| 2 | + |
| 3 | +let () = Random.self_init () |
| 4 | + |
| 5 | +let to_s_ns f = |
| 6 | + let ns = int_of_float (f *. 1_000_000_000.) in |
| 7 | + (ns / 1_000_000_000 , ns mod 1_000_000_000) |
| 8 | + |
| 9 | +let rec stall d = |
| 10 | + let open Lwt.Syntax in |
| 11 | + let s, ns = to_s_ns d in |
| 12 | + Printf.printf "stalling for %d.%09d\n" s ns; |
| 13 | + flush stdout; |
| 14 | + Unix.sleepf d; |
| 15 | + let* () = Lwt.pause () in |
| 16 | + if d > 3. then exit 2 else stall (1.2*.(d+. Random.float 0.05)) |
| 17 | + |
| 18 | + |
| 19 | +(* DETECTING *) |
| 20 | + |
| 21 | +let () = Runtime_events.start () |
| 22 | + |
| 23 | +(* set to maxint to avoid the first hit being a false positive *) |
| 24 | +let last_lap = ref Int64.max_int |
| 25 | +let alarmed = ref false |
| 26 | + |
| 27 | +let detect ?ringopt () = |
| 28 | + begin match ringopt with |
| 29 | + | None -> Printf.printf "starting detection on self (%d)\n" (Unix.getpid ()) |
| 30 | + | Some (path, pid) -> Printf.printf "starting detection on %s/%d.events\n" path pid |
| 31 | + end; |
| 32 | + let cursor = Runtime_events.create_cursor ringopt in |
| 33 | + let is_stall t = |
| 34 | + let delta = Int64.sub (Runtime_events.Timestamp.to_int64 t) !last_lap in |
| 35 | + if delta > 1_000_000_000L (* 1 second *) then begin |
| 36 | + Printf.printf "ALARM: stall 1s+ CRASHING\n"; flush stdout; |
| 37 | + (match ringopt with |
| 38 | + | None -> exit 1 |
| 39 | + | Some (_, pid) -> Unix.kill pid Sys.sigkill) |
| 40 | + end; |
| 41 | + if not !alarmed && delta > 500_000_000L (* 0.5 second *) then begin |
| 42 | + alarmed := true; |
| 43 | + Printf.printf "ALARM: stall .5s+\n"; flush stdout |
| 44 | + end |
| 45 | + in |
| 46 | + let cb = |
| 47 | + Runtime_events.Callbacks.create () |
| 48 | + |> Runtime_events.Callbacks.add_user_event |
| 49 | + Runtime_events.Type.unit |
| 50 | + (fun _ t e () -> |
| 51 | + match Runtime_events.User.tag e with |
| 52 | + | Lwt_main.Scheduler_lap -> |
| 53 | + alarmed := false; |
| 54 | + last_lap := Runtime_events.Timestamp.to_int64 t |
| 55 | + | _ -> ()) |
| 56 | + in |
| 57 | + let rec detect () = |
| 58 | + Unix.sleepf 0.01; |
| 59 | + let _ : int = Runtime_events.read_poll cursor cb None in |
| 60 | + is_stall (Runtime_events.Timestamp.get_current ()); |
| 61 | + detect () |
| 62 | + in |
| 63 | + detect () |
0 commit comments