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
85 (Rpc_server.Unix addr)
86 Rpc.Tcp (* not TCP, this is the same as SOCK_STREAM *)
91 (* Handle SIGCHLD to clean up jobs. *)
92 Sys.set_signal Sys.sigchld (Sys.Signal_handle handle_sigchld);
94 (* Initialize the variables. *)
95 state := Whenstate.set_variable !state "JOBSERIAL" (T_int zero_big_int)
97 and proc_reload_file () =
98 if !debug then Syslog.notice "remote call: reload_file";
100 try reload_file (); `ok
101 with Failure err -> `error err
103 and proc_set_variable (name, value) =
104 if !debug then Syslog.notice "remote call: set_variable %s" name;
107 check_valid_variable_name name;
109 let value = variable_of_rpc value in
110 state := Whenstate.set_variable !state name value;
112 (* Which jobs need to be re-evaluated? *)
113 let jobs = Whenstate.get_dependencies !state [name] in
114 let jobnames, state' = reevaluate_whenjobs !state jobs in
115 let state' = run_whenjobs state' jobnames in
120 Failure msg -> `error msg
122 and proc_get_variable name =
123 if !debug then Syslog.notice "remote call: get_variable %s" name;
125 rpc_of_variable (Whenstate.get_variable !state name)
127 and proc_get_variable_names () =
128 if !debug then Syslog.notice "remote call: get_variable_names";
130 let vars = Whenstate.get_variable_names !state in
132 (* Return variable names as a sorted array. *)
133 let vars = Array.of_list vars in
134 Array.sort compare vars;
137 and proc_exit_daemon () =
138 if !debug then Syslog.notice "remote call: exit_daemon";
142 `error "exit_daemon: no server handle"
144 Rpc_server.stop_server ~graceful:true s;
148 and proc_get_jobs () =
149 let running = Array.of_list (IntMap.values !runningmap) in
151 fun (job, dir, serial, start_time) ->
152 { Whenproto_aux.job_name = job.job_name;
153 job_serial = string_of_big_int serial;
154 job_tmpdir = dir; job_start_time = Int64.of_float start_time }
157 and proc_cancel_job serial =
159 let serial = big_int_of_string serial in
160 let pid = BigIntMap.find serial !serialmap in
164 | Not_found -> `error "job not found"
165 | exn -> `error (Printexc.to_string exn)
167 and proc_start_job jobname =
169 let job = Whenstate.get_job !state jobname in
170 let state' = run_job !state job in
174 | Not_found -> `error "job not found"
175 | exn -> `error (Printexc.to_string exn)
177 and proc_get_job serial =
179 let serial = big_int_of_string serial in
180 let pid = BigIntMap.find serial !serialmap in
181 let job, dir, serial, start_time = IntMap.find pid !runningmap in
182 { Whenproto_aux.job_name = job.job_name;
183 job_serial = string_of_big_int serial;
184 job_tmpdir = dir; job_start_time = Int64.of_float start_time }
186 | Not_found -> failwith "job not found"
187 | exn -> failwith (Printexc.to_string exn)
189 and proc_set_variables vars =
191 let vars = Array.map (
192 fun { Whenproto_aux.sv_name = name; sv_value = value } ->
193 name, variable_of_rpc value
195 let vars = Array.to_list vars in
198 Syslog.notice "remote call: set_variables (%s)"
202 sprintf "%s=%s" name (string_of_variable value)
205 List.iter (fun (name, _) -> check_valid_variable_name name) vars;
207 (* Update all the variables atomically. *)
208 let s = List.fold_left (
209 fun s (name, value) -> Whenstate.set_variable s name value
213 (* Which jobs need to be re-evaluated? *)
214 let jobs = Whenstate.get_dependencies !state (List.map fst vars) in
215 let jobnames, state' = reevaluate_whenjobs !state jobs in
216 let state' = run_whenjobs state' jobnames in
221 Failure msg -> `error msg
223 and proc_get_job_names () =
224 Array.of_list (Whenstate.get_job_names !state)
226 and proc_test_variables vars =
227 (* This is the same as proc_set_variables, except that it doesn't
228 * update the state, it just returns the jobs that *would* run if
229 * these variables were set to these values.
231 let vars = Array.map (
232 fun { Whenproto_aux.sv_name = name; sv_value = value } ->
233 name, variable_of_rpc value
235 let vars = Array.to_list vars in
238 Syslog.notice "remote call: test_variables (%s)"
242 sprintf "%s=%s" name (string_of_variable value)
245 List.iter (fun (name, _) -> check_valid_variable_name name) vars;
247 (* Update all the variables atomically. *)
248 let state = List.fold_left (
249 fun s (name, value) -> Whenstate.set_variable s name value
252 (* Which jobs WOULD be re-evaluated? *)
253 let jobs = Whenstate.get_dependencies state (List.map fst vars) in
254 let jobnames, _ = reevaluate_whenjobs state jobs in
256 (* Return the names. *)
257 Array.of_list jobnames
259 (* Reload the jobs file. *)
261 let file = sprintf "%s/jobs.cmo" !jobsdir in
263 (* As we are reloading the file, we want to create a new state
264 * that has no jobs, but has all the variables from the previous
267 let s = Whenstate.copy_variables !state Whenstate.empty in
272 Dynlink.loadfile file;
273 let s = Whenfile.get_state () in
274 Syslog.notice "loaded %d job(s) from %s" (Whenstate.nr_jobs s) file;
277 | Dynlink.Error err ->
278 let err = Dynlink.error_message err in
279 Syslog.error "error loading jobs: %s" err;
282 failwith (Printexc.to_string exn) in
284 let s = Whenstate.copy_prev_state !state s in
287 (* Re-evaluate all when jobs. *)
288 let jobs = Whenstate.get_whenjobs !state in
289 let jobnames, state' = reevaluate_whenjobs ~onload:true !state jobs in
290 let state' = run_whenjobs state' jobnames in
293 (* Schedule the next every job to run. *)
294 schedule_next_everyjob ()
296 (* Re-evaluate each when-statement job, in a loop until we reach
297 * a fixpoint. Return the list of job names that should run and
300 and reevaluate_whenjobs ?onload state jobs =
301 let rec loop (set, state) jobs =
304 fun (set, state) job ->
306 try Whenstate.evaluate_whenjob ?onload state job
307 with Invalid_argument err | Failure err ->
308 Syslog.error "error evaluating job %s (at %s): %s"
309 job.job_name (Camlp4.PreCast.Ast.Loc.to_string job.job_loc) err;
313 Syslog.notice "evaluate %s -> %b\n" job.job_name r;
315 (if r then StringSet.add job.job_name set else set), state'
316 ) (set, state) jobs in
317 (* reached a fixpoint? *)
318 if StringSet.compare set set' <> 0 then
319 loop (set', state') jobs
323 let set, state = loop (StringSet.empty, state) jobs in
324 let jobnames = StringSet.elements set in
326 (* Ensure the jobs always run in predictable (name) order. *)
327 let jobnames = List.sort compare_jobnames jobnames in
330 and run_whenjobs state jobnames =
332 let jobs = List.map (Whenstate.get_job state) jobnames in
333 List.fold_left run_job state jobs
335 (* Schedule the next every-statement job to run, if there is one. We
336 * look at the every jobs, work out the time that each must run at,
337 * pick the job(s) which must run soonest, and schedule a timer to run
338 * them. When the timer fires, it runs those jobs, then calls this
341 and schedule_next_everyjob () =
344 (* Get only everyjobs. *)
345 let jobs = Whenstate.get_everyjobs !state in
346 let jobs = List.map (
348 | { job_cond = Every_job period } as job -> (job, period)
349 | { job_cond = When_job _ } -> assert false
352 (* Map everyjob to next time it must run. *)
353 let jobs = List.map (
355 let t' = next_periodexpr t period in
356 assert (t' > t); (* serious bug in next_periodexpr if false *)
360 (* Sort, soonest first. *)
361 let jobs = List.sort (fun (_,a) (_,b) -> compare a b) jobs in
366 Syslog.notice "%s: next scheduled run at %s"
367 job.job_name (string_of_time_t t)
371 (* Pick the job(s) which run soonest. *)
372 let rec pick = function
375 | (j1, t) :: (j2, t') :: _ when t < t' -> t, [j1]
376 | (j1, t) :: (((j2, t') :: _) as rest) -> t, (j1 :: snd (pick rest))
378 let t, jobs = pick jobs in
382 (* Ensure the jobs always run in predictable (name) order. *)
384 List.sort (fun {job_name = a} {job_name = b} -> compare_jobnames a b)
388 Syslog.notice "scheduling job(s) %s to run at %s"
389 (String.concat ", " (List.map (fun { job_name = name } -> name) jobs))
390 (string_of_time_t t);
392 (* Schedule them to run at time t. *)
393 let g = new_timer_group () in
394 let t_diff = t -. Unix.time () in
395 let t_diff = if t_diff < 0. then 0. else t_diff in
397 delete_timer_group (); (* Delete the timer. *)
398 let state' = List.fold_left run_job !state jobs in
400 schedule_next_everyjob ()
402 Unixqueue.weak_once esys g t_diff run_jobs;
406 and new_timer_group () =
407 delete_timer_group ();
408 let g = Unixqueue.new_group esys in
409 timer_group := Some g;
412 and delete_timer_group () =
413 match !timer_group with
416 Unixqueue.clear esys g;
419 and run_job state job =
420 (* Increment JOBSERIAL. *)
422 match Whenstate.get_variable state "JOBSERIAL" with
424 let serial = succ_big_int serial in
425 let state' = Whenstate.set_variable state "JOBSERIAL" (T_int serial) in
427 | _ -> assert false in
429 (* Call the pre-condition script. Note this may decide not to run
430 * the job by returning false.
432 let pre_condition () =
433 match job.job_pre with
438 fun pid (job, _, serial, start_time) ->
439 let r = { pirun_job_name = job.job_name;
440 pirun_serial = serial;
441 pirun_start_time = start_time;
446 pi_job_name = job.job_name;
448 pi_variables = Whenstate.get_variables state;
453 if pre_condition () then (
454 Syslog.notice "running %s (JOBSERIAL=%s)"
455 job.job_name (string_of_big_int serial);
457 (* Create a temporary directory. The current directory of the job
458 * will be in this directory. The directory is removed when the
459 * child process exits.
461 let dir = tmpdir () in
464 if pid = 0 then ( (* child process running the job *)
467 (* Set environment variables corresponding to each variable. *)
469 (fun (name, value) -> putenv name (string_of_variable value))
470 (Whenstate.get_variables state);
472 (* Set the $JOBNAME environment variable. *)
473 putenv "JOBNAME" job.job_name;
475 (* Create a temporary file containing the shell script fragment. *)
476 let script = dir // "script.sh" in
477 let chan = open_out script in
478 fprintf chan "set -e\n"; (* So that jobs exit on error. *)
479 output_string chan job.job_script.sh_script;
483 let shell = try getenv "SHELL" with Not_found -> "/bin/sh" in
485 (* Set output to file. *)
486 let output = dir // "output.txt" in
487 let fd = openfile output [O_WRONLY; O_CREAT; O_TRUNC; O_NOCTTY] 0o600 in
492 (* Execute the shell script. *)
493 (try execvp shell [| shell; "-c"; script |];
494 with Unix_error (err, fn, _) ->
495 Syslog.error "%s failed: %s: %s" fn script (error_message err)
500 (* Remember this PID, the job and the temporary directory, so we
501 * can clean up when the child exits.
503 runningmap := IntMap.add pid (job, dir, serial, time ()) !runningmap;
504 serialmap := BigIntMap.add serial pid !serialmap;
509 Syslog.notice "not running %s (JOBSERIAL=%s) because pre() condition returned false"
510 job.job_name (string_of_big_int serial);
516 let chan = open_in "/dev/urandom" in
517 let data = String.create 16 in
518 really_input chan data 0 (String.length data);
520 let data = Digest.to_hex (Digest.string data) in
521 let dir = Filename.temp_dir_name // sprintf "whenjobs%s" data in
525 (* This is called when a job (child process) exits. *)
526 and handle_sigchld _ =
528 let pid, status = waitpid [WNOHANG] 0 in
530 (* Look up the PID in the running jobs map. *)
531 let job, dir, serial, time = IntMap.find pid !runningmap in
532 runningmap := IntMap.remove pid !runningmap;
533 serialmap := BigIntMap.remove serial !serialmap;
534 post_job job dir serial time status
536 with Unix_error _ | Not_found -> ()
538 and post_job job dir serial time status =
539 (* If there is a post function, run it. *)
540 (match job.job_post with
546 | WSIGNALED s | WSTOPPED s -> 1 in
548 res_job_name = job.job_name;
552 res_output = dir // "output.txt";
553 res_start_time = time
558 Syslog.error "job %s post function failed: %s" job.job_name msg
560 Syslog.error "job %s post function exception: %s"
561 job.job_name (Printexc.to_string exn)
564 (* This should be safe because the path cannot contain shell metachars. *)
565 let cmd = sprintf "rm -rf '%s'" dir in
566 ignore (Sys.command cmd)
568 (* Intelligent comparison of job names. *)
569 and compare_jobnames name1 name2 =
571 let len1 = String.length name1
572 and len2 = String.length name2 in
573 if len1 > 4 && len2 > 4 &&
574 String.sub name1 0 4 = "job$" && String.sub name2 0 4 = "job$"
576 let i1 = int_of_string (String.sub name1 4 (len1-4)) in
577 let i2 = int_of_string (String.sub name2 4 (len2-4)) in