2 * Copyright (C) 2012 Red Hat Inc.
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
27 external _exit : int -> 'a = "whenjobs__exit"
33 let state = ref Whenstate.empty
35 (* Jobs that are running: a map of PID -> (job, tmpdir, serial, start_time).
36 * Note that the job may no longer exist *OR* it may have been renamed,
37 * eg. if the jobs file was reloaded.
39 let runningmap = ref IntMap.empty
41 (* Serial numbers of running jobs. Map of serial -> PID (in runningmap). *)
42 let serialmap = ref BigIntMap.empty
44 (* Was debugging requested on the command line? *)
50 let esys = Unixqueue.standard_event_system ()
52 (* The timer. It's convenient to have this as a global variable
53 * because (a) there should only be one timer (which fires when the
54 * soonest every-job becomes ready), and (b) it's complicated to track
55 * that timer and avoid it getting double-scheduled (eg. when we
56 * reload the jobs file) without having a global variable.
58 let timer_group = ref None
64 Whenlock.create_lock !jobsdir;
66 (* Remove old socket if it exists. *)
67 let addr = sprintf "%s/socket" !jobsdir in
68 (try unlink addr with Unix_error _ -> ());
70 (* Create the Unix domain socket server. *)
72 Whenproto_srv.When.V1.create_server
76 ~proc_get_variable_names
86 ~proc_whisper_variables
87 (Rpc_server.Unix addr)
88 Rpc.Tcp (* not TCP, this is the same as SOCK_STREAM *)
93 (* Handle SIGCHLD to clean up jobs. *)
94 Sys.set_signal Sys.sigchld (Sys.Signal_handle handle_sigchld);
96 (* Initialize the variables. *)
97 state := Whenstate.set_variable !state "JOBSERIAL" (T_int zero_big_int)
99 and proc_reload_file () =
100 if !debug then Syslog.notice "remote call: reload_file";
102 try reload_files (); `ok
103 with Failure err -> `error err
105 and proc_set_variable (name, value) =
106 if !debug then Syslog.notice "remote call: set_variable %s" name;
109 check_valid_variable_name name;
111 let value = variable_of_rpc value in
112 state := Whenstate.set_variable !state name value;
114 (* Which jobs need to be re-evaluated? *)
115 let jobs = Whenstate.get_dependencies !state [name] in
116 let jobnames, state' = reevaluate_whenjobs !state jobs in
117 let state' = run_whenjobs state' jobnames in
122 Failure msg -> `error msg
124 and proc_get_variable name =
125 if !debug then Syslog.notice "remote call: get_variable %s" name;
127 rpc_of_variable (Whenstate.get_variable !state name)
129 and proc_get_variable_names () =
130 if !debug then Syslog.notice "remote call: get_variable_names";
132 let vars = Whenstate.get_variable_names !state in
134 (* Return variable names as a sorted array. *)
135 let vars = Array.of_list vars in
136 Array.sort compare vars;
139 and proc_exit_daemon () =
140 if !debug then Syslog.notice "remote call: exit_daemon";
144 `error "exit_daemon: no server handle"
146 Rpc_server.stop_server ~graceful:true s;
150 and proc_get_jobs () =
151 let running = Array.of_list (IntMap.values !runningmap) in
153 fun (job, dir, serial, start_time) ->
154 { Whenproto_aux.job_name = job.job_name;
155 job_serial = string_of_big_int serial;
156 job_tmpdir = dir; job_start_time = Int64.of_float start_time }
159 and proc_cancel_job serial =
161 let serial = big_int_of_string serial in
162 let pid = BigIntMap.find serial !serialmap in
166 | Not_found -> `error "job not found"
167 | exn -> `error (Printexc.to_string exn)
169 and proc_start_job jobname =
171 let job = Whenstate.get_job !state jobname in
172 let state' = run_job !state job in
176 | Not_found -> `error "job not found"
177 | exn -> `error (Printexc.to_string exn)
179 and proc_get_job serial =
181 let serial = big_int_of_string serial in
182 let pid = BigIntMap.find serial !serialmap in
183 let job, dir, serial, start_time = IntMap.find pid !runningmap in
184 { Whenproto_aux.job_name = job.job_name;
185 job_serial = string_of_big_int serial;
186 job_tmpdir = dir; job_start_time = Int64.of_float start_time }
188 | Not_found -> failwith "job not found"
189 | exn -> failwith (Printexc.to_string exn)
191 and proc_set_variables vars =
193 let vars = Array.map (
194 fun { Whenproto_aux.sv_name = name; sv_value = value } ->
195 name, variable_of_rpc value
197 let vars = Array.to_list vars in
200 Syslog.notice "remote call: set_variables (%s)"
204 sprintf "%s=%s" name (string_of_variable value)
207 List.iter (fun (name, _) -> check_valid_variable_name name) vars;
209 (* Update all the variables atomically. *)
210 let s = List.fold_left (
211 fun s (name, value) -> Whenstate.set_variable s name value
215 (* Which jobs need to be re-evaluated? *)
216 let jobs = Whenstate.get_dependencies !state (List.map fst vars) in
217 let jobnames, state' = reevaluate_whenjobs !state jobs in
218 let state' = run_whenjobs state' jobnames in
223 Failure msg -> `error msg
225 and proc_get_job_names () =
226 Array.of_list (Whenstate.get_job_names !state)
228 and proc_test_variables vars =
229 (* This is the same as proc_set_variables, except that it doesn't
230 * update the state, it just returns the jobs that *would* run if
231 * these variables were set to these values.
233 let vars = Array.map (
234 fun { Whenproto_aux.sv_name = name; sv_value = value } ->
235 name, variable_of_rpc value
237 let vars = Array.to_list vars in
240 Syslog.notice "remote call: test_variables (%s)"
244 sprintf "%s=%s" name (string_of_variable value)
247 List.iter (fun (name, _) -> check_valid_variable_name name) vars;
249 (* Update all the variables atomically. *)
250 let state = List.fold_left (
251 fun s (name, value) -> Whenstate.set_variable s name value
254 (* Which jobs WOULD be re-evaluated? *)
255 let jobs = Whenstate.get_dependencies state (List.map fst vars) in
256 let jobnames, _ = reevaluate_whenjobs state jobs in
258 (* Return the names. *)
259 Array.of_list jobnames
261 and proc_ping_daemon () = `ok
263 and proc_whisper_variables vars =
265 let vars = Array.map (
266 fun { Whenproto_aux.sv_name = name; sv_value = value } ->
267 name, variable_of_rpc value
269 let vars = Array.to_list vars in
272 Syslog.notice "remote call: whisper_variables (%s)"
276 sprintf "%s=%s" name (string_of_variable value)
279 List.iter (fun (name, _) -> check_valid_variable_name name) vars;
281 (* Update all the variables atomically. *)
282 let s = List.fold_left (
283 fun s (name, value) -> Whenstate.set_variable s name value
287 (* .. but don't reevaluate or run jobs. *)
291 Failure msg -> `error msg
293 (* Reload the jobs file(s). *)
294 and reload_files () =
295 (* Get the highest numbered dir/jobs__*.cmo (bytecode) or
296 * dir/jobs__*.cmxs (native code) file and load it. Delete
297 * lower-numbered (== older) files.
301 if not Dynlink.is_native then ".cmo", 4 else ".cmxs", 5 in
302 let dir = !jobsdir in
303 let files = Array.to_list (Sys.readdir dir) in
304 let times = filter_map (
306 if not (string_startswith file "jobs__") ||
307 not (string_endswith file suffix) then
310 let len = String.length file in
311 let t = String.sub file 6 (len-slen-6) in
312 try Some (int_of_string t) with Failure "int_of_string" -> None
315 let times = List.rev (List.sort compare times) in
319 (* Unlink the older files. *)
322 try unlink (dir // sprintf "jobs__%d%s" t suffix)
323 with Unix_error _ -> ()
325 (* Return the newest (highest numbered) file. *)
326 Some (dir // sprintf "jobs__%d%s" x suffix) in
328 (* As we are reloading the file, we want to create a new state
329 * that has no jobs, but has all the variables from the previous
332 let s = Whenstate.copy_variables !state Whenstate.empty in
338 (* no jobs file, return the same state *)
339 Syslog.notice "no jobs file found";
343 Dynlink.loadfile filename;
344 let s = Whenfile.get_state () in
345 Syslog.notice "loaded %d job(s)" (Whenstate.nr_jobs s);
348 | Dynlink.Error err ->
349 let err = Dynlink.error_message err in
350 Syslog.error "error loading jobs: %s" err;
353 failwith (Printexc.to_string exn) in
355 let s = Whenstate.copy_prev_state !state s in
358 (* Re-evaluate all when jobs. *)
359 let jobs = Whenstate.get_whenjobs !state in
360 let jobnames, state' = reevaluate_whenjobs ~onload:true !state jobs in
361 let state' = run_whenjobs state' jobnames in
364 (* Schedule the next every job to run. *)
365 schedule_next_everyjob ()
367 (* Re-evaluate each when-statement job, in a loop until we reach
368 * a fixpoint. Return the list of job names that should run and
371 and reevaluate_whenjobs ?onload state jobs =
372 let rec loop (set, state) jobs =
375 fun (set, state) job ->
377 try Whenstate.evaluate_whenjob ?onload state job
378 with Invalid_argument err | Failure err ->
379 Syslog.error "error evaluating job %s (at %s): %s"
380 job.job_name (Camlp4.PreCast.Ast.Loc.to_string job.job_loc) err;
384 Syslog.notice "evaluate %s -> %b\n" job.job_name r;
386 (if r then StringSet.add job.job_name set else set), state'
387 ) (set, state) jobs in
388 (* reached a fixpoint? *)
389 if StringSet.compare set set' <> 0 then
390 loop (set', state') jobs
394 let set, state = loop (StringSet.empty, state) jobs in
395 let jobnames = StringSet.elements set in
397 (* Ensure the jobs always run in predictable (name) order. *)
398 let jobnames = List.sort compare_jobnames jobnames in
401 and run_whenjobs state jobnames =
403 let jobs = List.map (Whenstate.get_job state) jobnames in
404 List.fold_left run_job state jobs
406 (* Schedule the next every-statement job to run, if there is one. We
407 * look at the every jobs, work out the time that each must run at,
408 * pick the job(s) which must run soonest, and schedule a timer to run
409 * them. When the timer fires, it runs those jobs, then calls this
412 and schedule_next_everyjob () =
415 (* Get only everyjobs. *)
416 let jobs = Whenstate.get_everyjobs !state in
417 let jobs = List.map (
419 | { job_cond = Every_job period } as job -> (job, period)
420 | { job_cond = When_job _ } -> assert false
423 (* Map everyjob to next time it must run. *)
424 let jobs = List.map (
426 let t' = next_periodexpr t period in
427 assert (t' > t); (* serious bug in next_periodexpr if false *)
431 (* Sort, soonest first. *)
432 let jobs = List.sort (fun (_,a) (_,b) -> compare a b) jobs in
437 Syslog.notice "%s: next scheduled run at %s"
438 job.job_name (string_of_time_t t)
442 (* Pick the job(s) which run soonest. *)
443 let rec pick = function
446 | (j1, t) :: (j2, t') :: _ when t < t' -> t, [j1]
447 | (j1, t) :: (((j2, t') :: _) as rest) -> t, (j1 :: snd (pick rest))
449 let t, jobs = pick jobs in
453 (* Ensure the jobs always run in predictable (name) order. *)
455 List.sort (fun {job_name = a} {job_name = b} -> compare_jobnames a b)
459 Syslog.notice "scheduling job(s) %s to run at %s"
460 (String.concat ", " (List.map (fun { job_name = name } -> name) jobs))
461 (string_of_time_t t);
463 (* Schedule them to run at time t. *)
464 let g = new_timer_group () in
465 let t_diff = t -. Unix.time () in
466 let t_diff = if t_diff < 0. then 0. else t_diff in
468 delete_timer_group (); (* Delete the timer. *)
469 let state' = List.fold_left run_job !state jobs in
471 schedule_next_everyjob ()
473 Unixqueue.weak_once esys g t_diff run_jobs;
477 and new_timer_group () =
478 delete_timer_group ();
479 let g = Unixqueue.new_group esys in
480 timer_group := Some g;
483 and delete_timer_group () =
484 match !timer_group with
487 Unixqueue.clear esys g;
490 and run_job state job =
491 (* Increment JOBSERIAL. *)
493 match Whenstate.get_variable state "JOBSERIAL" with
495 let serial = succ_big_int serial in
496 let state' = Whenstate.set_variable state "JOBSERIAL" (T_int serial) in
498 | _ -> assert false in
500 (* Call the pre-condition script. Note this may decide not to run
501 * the job by returning false.
503 let pre_condition () =
504 match job.job_pre with
509 fun pid (job, _, serial, start_time) ->
510 let r = { pirun_job_name = job.job_name;
511 pirun_serial = serial;
512 pirun_start_time = start_time;
517 pi_job_name = job.job_name;
519 pi_variables = Whenstate.get_variables state;
524 if pre_condition () then (
525 Syslog.notice "running %s (JOBSERIAL=%s)"
526 job.job_name (string_of_big_int serial);
528 (* Create a temporary directory. The current directory of the job
529 * will be in this directory. The directory is removed when the
530 * child process exits.
532 let dir = tmpdir () in
535 if pid = 0 then ( (* child process running the job *)
538 (* Set environment variables corresponding to each variable. *)
540 (fun (name, value) -> putenv name (string_of_variable value))
541 (Whenstate.get_variables state);
543 (* Set the $JOBNAME environment variable. *)
544 putenv "JOBNAME" job.job_name;
546 (* Create a temporary file containing the shell script fragment. *)
547 let script = dir // "script.sh" in
548 let chan = open_out script in
549 fprintf chan "set -e\n"; (* So that jobs exit on error. *)
550 output_string chan job.job_script.sh_script;
554 let shell = try getenv "SHELL" with Not_found -> "/bin/sh" in
556 (* Set output to file. *)
557 let output = dir // "output.txt" in
558 let fd = openfile output [O_WRONLY; O_CREAT; O_TRUNC; O_NOCTTY] 0o600 in
563 (* Execute the shell script. *)
564 (try execvp shell [| shell; "-c"; script |];
565 with Unix_error (err, fn, _) ->
566 Syslog.error "%s failed: %s: %s" fn script (error_message err)
571 (* Remember this PID, the job and the temporary directory, so we
572 * can clean up when the child exits.
574 runningmap := IntMap.add pid (job, dir, serial, time ()) !runningmap;
575 serialmap := BigIntMap.add serial pid !serialmap;
580 Syslog.notice "not running %s (JOBSERIAL=%s) because pre() condition returned false"
581 job.job_name (string_of_big_int serial);
587 let chan = open_in "/dev/urandom" in
588 let data = String.create 16 in
589 really_input chan data 0 (String.length data);
591 let data = Digest.to_hex (Digest.string data) in
592 let dir = Filename.temp_dir_name // sprintf "whenjobs%s" data in
596 (* This is called when a job (child process) exits. *)
597 and handle_sigchld _ =
599 let pid, status = waitpid [WNOHANG] 0 in
601 (* Look up the PID in the running jobs map. *)
602 let job, dir, serial, time = IntMap.find pid !runningmap in
603 runningmap := IntMap.remove pid !runningmap;
604 serialmap := BigIntMap.remove serial !serialmap;
605 post_job job dir serial time status
607 with Unix_error _ | Not_found -> ()
609 and post_job job dir serial time status =
610 (* If there is a post function, run it. *)
611 (match job.job_post with
617 | WSIGNALED s | WSTOPPED s -> 1 in
619 res_job_name = job.job_name;
623 res_output = dir // "output.txt";
624 res_start_time = time
629 Syslog.error "job %s post function failed: %s" job.job_name msg
631 Syslog.error "job %s post function exception: %s"
632 job.job_name (Printexc.to_string exn)
635 (* This should be safe because the path cannot contain shell metachars. *)
636 let cmd = sprintf "rm -rf '%s'" dir in
637 ignore (Sys.command cmd)
639 (* Intelligent comparison of job names. *)
640 and compare_jobnames name1 name2 =
642 let len1 = String.length name1
643 and len2 = String.length name2 in
644 if len1 > 4 && len2 > 4 &&
645 String.sub name1 0 4 = "job$" && String.sub name2 0 4 = "job$"
647 let i1 = int_of_string (String.sub name1 4 (len1-4)) in
648 let i2 = int_of_string (String.sub name2 4 (len2-4)) in