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
84 (Rpc_server.Unix addr)
85 Rpc.Tcp (* not TCP, this is the same as SOCK_STREAM *)
90 (* Handle SIGCHLD to clean up jobs. *)
91 Sys.set_signal Sys.sigchld (Sys.Signal_handle handle_sigchld);
93 (* Initialize the variables. *)
94 state := Whenstate.set_variable !state "JOBSERIAL" (T_int zero_big_int)
96 and proc_reload_file () =
97 if !debug then Syslog.notice "remote call: reload_file";
99 try reload_file (); `ok
100 with Failure err -> `error err
102 and proc_set_variable (name, value) =
103 if !debug then Syslog.notice "remote call: set_variable %s" name;
106 check_valid_variable_name name;
108 let value = variable_of_rpc value in
109 state := Whenstate.set_variable !state name value;
111 (* Which jobs need to be re-evaluated? *)
112 let jobs = Whenstate.get_dependencies !state [name] in
113 let state' = reevaluate_whenjobs !state jobs in
118 Failure msg -> `error msg
120 and proc_get_variable name =
121 if !debug then Syslog.notice "remote call: get_variable %s" name;
123 rpc_of_variable (Whenstate.get_variable !state name)
125 and proc_get_variable_names () =
126 if !debug then Syslog.notice "remote call: get_variable_names";
128 let vars = Whenstate.get_variable_names !state in
130 (* Return variable names as a sorted array. *)
131 let vars = Array.of_list vars in
132 Array.sort compare vars;
135 and proc_exit_daemon () =
136 if !debug then Syslog.notice "remote call: exit_daemon";
140 `error "exit_daemon: no server handle"
142 Rpc_server.stop_server ~graceful:true s;
146 and proc_get_jobs () =
147 let running = Array.of_list (IntMap.values !runningmap) in
149 fun (job, dir, serial, start_time) ->
150 { Whenproto_aux.job_name = job.job_name;
151 job_serial = string_of_big_int serial;
152 job_tmpdir = dir; job_start_time = Int64.of_float start_time }
155 and proc_cancel_job serial =
157 let serial = big_int_of_string serial in
158 let pid = BigIntMap.find serial !serialmap in
162 | Not_found -> `error "job not found"
163 | exn -> `error (Printexc.to_string exn)
165 and proc_start_job jobname =
167 let job = Whenstate.get_job !state jobname in
168 let state' = run_job !state job in
172 | Not_found -> `error "job not found"
173 | exn -> `error (Printexc.to_string exn)
175 and proc_get_job serial =
177 let serial = big_int_of_string serial in
178 let pid = BigIntMap.find serial !serialmap in
179 let job, dir, serial, start_time = IntMap.find pid !runningmap in
180 { Whenproto_aux.job_name = job.job_name;
181 job_serial = string_of_big_int serial;
182 job_tmpdir = dir; job_start_time = Int64.of_float start_time }
184 | Not_found -> failwith "job not found"
185 | exn -> failwith (Printexc.to_string exn)
187 and proc_set_variables vars =
189 let vars = Array.map (
190 fun { Whenproto_aux.sv_name = name; sv_value = value } ->
191 name, variable_of_rpc value
193 let vars = Array.to_list vars in
196 Syslog.notice "remote call: set_variables (%s)"
200 sprintf "%s=%s" name (string_of_variable value)
203 List.iter (fun (name, _) -> check_valid_variable_name name) vars;
205 (* Update all the variables atomically. *)
206 let s = List.fold_left (
207 fun s (name, value) -> Whenstate.set_variable s name value
211 (* Which jobs need to be re-evaluated? *)
212 let jobs = Whenstate.get_dependencies !state (List.map fst vars) in
213 let state' = reevaluate_whenjobs !state jobs in
218 Failure msg -> `error msg
220 and proc_get_job_names () =
221 Array.of_list (Whenstate.get_job_names !state)
223 (* Reload the jobs file. *)
225 let file = sprintf "%s/jobs.cmo" !jobsdir in
227 (* As we are reloading the file, we want to create a new state
228 * that has no jobs, but has all the variables from the previous
231 let s = Whenstate.copy_variables !state Whenstate.empty in
236 Dynlink.loadfile file;
237 let s = Whenfile.get_state () in
238 Syslog.notice "loaded %d job(s) from %s" (Whenstate.nr_jobs s) file;
241 | Dynlink.Error err ->
242 let err = Dynlink.error_message err in
243 Syslog.error "error loading jobs: %s" err;
246 failwith (Printexc.to_string exn) in
248 let s = Whenstate.copy_prev_state !state s in
251 (* Re-evaluate all when jobs. *)
252 let jobs = Whenstate.get_whenjobs !state in
253 let state' = reevaluate_whenjobs ~onload:true !state jobs in
256 (* Schedule the next every job to run. *)
257 schedule_next_everyjob ()
259 (* Re-evaluate each when-statement job, in a loop until we reach
260 * a fixpoint. Run the jobs and return the updated state.
262 and reevaluate_whenjobs ?onload state jobs =
263 let rec loop (set, state) jobs =
266 fun (set, state) job ->
268 try Whenstate.evaluate_whenjob ?onload state job
269 with Invalid_argument err | Failure err ->
270 Syslog.error "error evaluating job %s (at %s): %s"
271 job.job_name (Camlp4.PreCast.Ast.Loc.to_string job.job_loc) err;
275 Syslog.notice "evaluate %s -> %b\n" job.job_name r;
277 (if r then StringSet.add job.job_name set else set), state'
278 ) (set, state) jobs in
279 (* reached a fixpoint? *)
280 if StringSet.compare set set' <> 0 then
281 loop (set', state') jobs
285 let set, state = loop (StringSet.empty, state) jobs in
286 let jobnames = StringSet.elements set in
288 (* Ensure the jobs always run in predictable (name) order. *)
289 let jobnames = List.sort compare_jobnames jobnames in
292 let jobs = List.map (Whenstate.get_job state) jobnames in
293 List.fold_left run_job state jobs
295 (* Schedule the next every-statement job to run, if there is one. We
296 * look at the every jobs, work out the time that each must run at,
297 * pick the job(s) which must run soonest, and schedule a timer to run
298 * them. When the timer fires, it runs those jobs, then calls this
301 and schedule_next_everyjob () =
304 (* Get only everyjobs. *)
305 let jobs = Whenstate.get_everyjobs !state in
306 let jobs = List.map (
308 | { job_cond = Every_job period } as job -> (job, period)
309 | { job_cond = When_job _ } -> assert false
312 (* Map everyjob to next time it must run. *)
313 let jobs = List.map (
315 let t' = next_periodexpr t period in
316 assert (t' > t); (* serious bug in next_periodexpr if false *)
320 (* Sort, soonest first. *)
321 let jobs = List.sort (fun (_,a) (_,b) -> compare a b) jobs in
326 Syslog.notice "%s: next scheduled run at %s"
327 job.job_name (string_of_time_t t)
331 (* Pick the job(s) which run soonest. *)
332 let rec pick = function
335 | (j1, t) :: (j2, t') :: _ when t < t' -> t, [j1]
336 | (j1, t) :: (((j2, t') :: _) as rest) -> t, (j1 :: snd (pick rest))
338 let t, jobs = pick jobs in
342 (* Ensure the jobs always run in predictable (name) order. *)
344 List.sort (fun {job_name = a} {job_name = b} -> compare_jobnames a b)
348 Syslog.notice "scheduling job(s) %s to run at %s"
349 (String.concat ", " (List.map (fun { job_name = name } -> name) jobs))
350 (string_of_time_t t);
352 (* Schedule them to run at time t. *)
353 let g = new_timer_group () in
354 let t_diff = t -. Unix.time () in
355 let t_diff = if t_diff < 0. then 0. else t_diff in
357 delete_timer_group (); (* Delete the timer. *)
358 let state' = List.fold_left run_job !state jobs in
360 schedule_next_everyjob ()
362 Unixqueue.weak_once esys g t_diff run_jobs;
366 and new_timer_group () =
367 delete_timer_group ();
368 let g = Unixqueue.new_group esys in
369 timer_group := Some g;
372 and delete_timer_group () =
373 match !timer_group with
376 Unixqueue.clear esys g;
379 and run_job state job =
380 (* Increment JOBSERIAL. *)
382 match Whenstate.get_variable state "JOBSERIAL" with
384 let serial = succ_big_int serial in
385 let state' = Whenstate.set_variable state "JOBSERIAL" (T_int serial) in
387 | _ -> assert false in
389 (* Call the pre-condition script. Note this may decide not to run
390 * the job by returning false.
392 let pre_condition () =
393 match job.job_pre with
398 fun pid (job, _, serial, start_time) ->
399 let r = { pirun_job_name = job.job_name;
400 pirun_serial = serial;
401 pirun_start_time = start_time;
406 pi_job_name = job.job_name;
408 pi_variables = Whenstate.get_variables state;
413 if pre_condition () then (
414 Syslog.notice "running %s (JOBSERIAL=%s)"
415 job.job_name (string_of_big_int serial);
417 (* Create a temporary directory. The current directory of the job
418 * will be in this directory. The directory is removed when the
419 * child process exits.
421 let dir = tmpdir () in
424 if pid = 0 then ( (* child process running the job *)
427 (* Set environment variables corresponding to each variable. *)
429 (fun (name, value) -> putenv name (string_of_variable value))
430 (Whenstate.get_variables state);
432 (* Set the $JOBNAME environment variable. *)
433 putenv "JOBNAME" job.job_name;
435 (* Create a temporary file containing the shell script fragment. *)
436 let script = dir // "script.sh" in
437 let chan = open_out script in
438 fprintf chan "set -e\n"; (* So that jobs exit on error. *)
439 output_string chan job.job_script.sh_script;
443 let shell = try getenv "SHELL" with Not_found -> "/bin/sh" in
445 (* Set output to file. *)
446 let output = dir // "output.txt" in
447 let fd = openfile output [O_WRONLY; O_CREAT; O_TRUNC; O_NOCTTY] 0o600 in
452 (* Execute the shell script. *)
453 (try execvp shell [| shell; "-c"; script |];
454 with Unix_error (err, fn, _) ->
455 Syslog.error "%s failed: %s: %s" fn script (error_message err)
460 (* Remember this PID, the job and the temporary directory, so we
461 * can clean up when the child exits.
463 runningmap := IntMap.add pid (job, dir, serial, time ()) !runningmap;
464 serialmap := BigIntMap.add serial pid !serialmap;
469 Syslog.notice "not running %s (JOBSERIAL=%s) because pre() condition returned false"
470 job.job_name (string_of_big_int serial);
476 let chan = open_in "/dev/urandom" in
477 let data = String.create 16 in
478 really_input chan data 0 (String.length data);
480 let data = Digest.to_hex (Digest.string data) in
481 let dir = Filename.temp_dir_name // sprintf "whenjobs%s" data in
485 (* This is called when a job (child process) exits. *)
486 and handle_sigchld _ =
488 let pid, status = waitpid [WNOHANG] 0 in
490 (* Look up the PID in the running jobs map. *)
491 let job, dir, serial, time = IntMap.find pid !runningmap in
492 runningmap := IntMap.remove pid !runningmap;
493 serialmap := BigIntMap.remove serial !serialmap;
494 post_job job dir serial time status
496 with Unix_error _ | Not_found -> ()
498 and post_job job dir serial time status =
499 (* If there is a post function, run it. *)
500 (match job.job_post with
506 | WSIGNALED s | WSTOPPED s -> 1 in
508 res_job_name = job.job_name;
512 res_output = dir // "output.txt";
513 res_start_time = time
518 Syslog.error "job %s post function failed: %s" job.job_name msg
520 Syslog.error "job %s post function exception: %s"
521 job.job_name (Printexc.to_string exn)
524 (* This should be safe because the path cannot contain shell metachars. *)
525 let cmd = sprintf "rm -rf '%s'" dir in
526 ignore (Sys.command cmd)
528 (* Intelligent comparison of job names. *)
529 and compare_jobnames name1 name2 =
531 let len1 = String.length name1
532 and len2 = String.length name2 in
533 if len1 > 4 && len2 > 4 &&
534 String.sub name1 0 4 = "job$" && String.sub name2 0 4 = "job$"
536 let i1 = int_of_string (String.sub name1 4 (len1-4)) in
537 let i2 = int_of_string (String.sub name2 4 (len2-4)) in