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 reevaluate_whenjobs jobs;
117 Failure msg -> `error msg
119 and proc_get_variable name =
120 if !debug then Syslog.notice "remote call: get_variable %s" name;
122 rpc_of_variable (Whenstate.get_variable !state name)
124 and proc_get_variable_names () =
125 if !debug then Syslog.notice "remote call: get_variable_names";
127 let vars = Whenstate.get_variable_names !state in
129 (* Return variable names as a sorted array. *)
130 let vars = Array.of_list vars in
131 Array.sort compare vars;
134 and proc_exit_daemon () =
135 if !debug then Syslog.notice "remote call: exit_daemon";
139 `error "exit_daemon: no server handle"
141 Rpc_server.stop_server ~graceful:true s;
145 and proc_get_jobs () =
146 let running = Array.of_list (IntMap.values !runningmap) in
148 fun (job, dir, serial, start_time) ->
149 { Whenproto_aux.job_name = job.job_name;
150 job_serial = string_of_big_int serial;
151 job_tmpdir = dir; job_start_time = Int64.of_float start_time }
154 and proc_cancel_job serial =
156 let serial = big_int_of_string serial in
157 let pid = BigIntMap.find serial !serialmap in
161 | Not_found -> `error "job not found"
162 | exn -> `error (Printexc.to_string exn)
164 and proc_start_job jobname =
166 let job = Whenstate.get_job !state jobname in
170 | Not_found -> `error "job not found"
171 | exn -> `error (Printexc.to_string exn)
173 and proc_get_job serial =
175 let serial = big_int_of_string serial in
176 let pid = BigIntMap.find serial !serialmap in
177 let job, dir, serial, start_time = IntMap.find pid !runningmap in
178 { Whenproto_aux.job_name = job.job_name;
179 job_serial = string_of_big_int serial;
180 job_tmpdir = dir; job_start_time = Int64.of_float start_time }
182 | Not_found -> failwith "job not found"
183 | exn -> failwith (Printexc.to_string exn)
185 and proc_set_variables vars =
187 let vars = Array.map (
188 fun { Whenproto_aux.sv_name = name; sv_value = value } ->
189 name, variable_of_rpc value
191 let vars = Array.to_list vars in
194 Syslog.notice "remote call: set_variables (%s)"
198 sprintf "%s=%s" name (string_of_variable value)
201 List.iter (fun (name, _) -> check_valid_variable_name name) vars;
203 (* Update all the variables atomically. *)
204 let s = List.fold_left (
205 fun s (name, value) -> Whenstate.set_variable s name value
209 (* Which jobs need to be re-evaluated? *)
210 let jobs = Whenstate.get_dependencies !state (List.map fst vars) in
211 reevaluate_whenjobs jobs;
215 Failure msg -> `error msg
217 and proc_get_job_names () =
218 Array.of_list (Whenstate.get_job_names !state)
220 (* Reload the jobs file. *)
222 let file = sprintf "%s/jobs.cmo" !jobsdir in
224 (* As we are reloading the file, we want to create a new state
225 * that has no jobs, but has all the variables from the previous
228 let s = Whenstate.copy_variables !state Whenstate.empty in
233 Dynlink.loadfile file;
234 let s = Whenfile.get_state () in
235 Syslog.notice "loaded %d job(s) from %s" (Whenstate.nr_jobs s) file;
238 | Dynlink.Error err ->
239 let err = Dynlink.error_message err in
240 Syslog.error "error loading jobs: %s" err;
243 failwith (Printexc.to_string exn) in
245 let s = Whenstate.copy_prev_state !state s in
248 (* Re-evaluate all when jobs. *)
249 reevaluate_whenjobs ~onload:true (Whenstate.get_whenjobs !state);
251 (* Schedule the next every job to run. *)
252 schedule_next_everyjob ()
254 (* Re-evaluate each when-statement job, in a loop until we reach
255 * a fixpoint. Run those that need to be run.
257 and reevaluate_whenjobs ?onload jobs =
258 let rec loop set jobs =
263 try Whenstate.evaluate_whenjob ?onload !state job
264 with Invalid_argument err | Failure err ->
265 Syslog.error "error evaluating job %s (at %s): %s"
266 job.job_name (Camlp4.PreCast.Ast.Loc.to_string job.job_loc) err;
272 Syslog.notice "evaluate %s -> %b\n" job.job_name r;
274 if r then StringSet.add job.job_name set else set
276 if StringSet.compare set set' <> 0 then
281 let set = loop StringSet.empty jobs in
282 let jobnames = StringSet.elements set in
284 (* Ensure the jobs always run in predictable (name) order. *)
285 let jobnames = List.sort compare_jobnames jobnames in
288 List.iter run_job (List.map (Whenstate.get_job !state) jobnames)
290 (* Schedule the next every-statement job to run, if there is one. We
291 * look at the every jobs, work out the time that each must run at,
292 * pick the job(s) which must run soonest, and schedule a timer to run
293 * them. When the timer fires, it runs those jobs, then calls this
296 and schedule_next_everyjob () =
299 (* Get only everyjobs. *)
300 let jobs = Whenstate.get_everyjobs !state in
301 let jobs = List.map (
303 | { job_cond = Every_job period } as job -> (job, period)
304 | { job_cond = When_job _ } -> assert false
307 (* Map everyjob to next time it must run. *)
308 let jobs = List.map (
310 let t' = next_periodexpr t period in
311 assert (t' > t); (* serious bug in next_periodexpr if false *)
315 (* Sort, soonest first. *)
316 let jobs = List.sort (fun (_,a) (_,b) -> compare a b) jobs in
321 Syslog.notice "%s: next scheduled run at %s"
322 job.job_name (string_of_time_t t)
326 (* Pick the job(s) which run soonest. *)
327 let rec pick = function
330 | (j1, t) :: (j2, t') :: _ when t < t' -> t, [j1]
331 | (j1, t) :: (((j2, t') :: _) as rest) -> t, (j1 :: snd (pick rest))
333 let t, jobs = pick jobs in
337 (* Ensure the jobs always run in predictable (name) order. *)
339 List.sort (fun {job_name = a} {job_name = b} -> compare_jobnames a b)
343 Syslog.notice "scheduling job(s) %s to run at %s"
344 (String.concat ", " (List.map (fun { job_name = name } -> name) jobs))
345 (string_of_time_t t);
347 (* Schedule them to run at time t. *)
348 let g = new_timer_group () in
349 let t_diff = t -. Unix.time () in
350 let t_diff = if t_diff < 0. then 0. else t_diff in
352 delete_timer_group (); (* Delete the timer. *)
353 List.iter run_job jobs;
354 schedule_next_everyjob ()
356 Unixqueue.weak_once esys g t_diff run_jobs;
360 and new_timer_group () =
361 delete_timer_group ();
362 let g = Unixqueue.new_group esys in
363 timer_group := Some g;
366 and delete_timer_group () =
367 match !timer_group with
370 Unixqueue.clear esys g;
374 (* Increment JOBSERIAL. *)
376 match Whenstate.get_variable !state "JOBSERIAL" with
378 let serial = succ_big_int serial in
379 state := Whenstate.set_variable !state "JOBSERIAL" (T_int serial);
381 | _ -> assert false in
383 (* Call the pre-condition script. Note this may decide not to run
384 * the job by returning false.
386 let pre_condition () =
387 match job.job_pre with
392 fun pid (job, _, serial, start_time) ->
393 let r = { pirun_job_name = job.job_name;
394 pirun_serial = serial;
395 pirun_start_time = start_time;
400 pi_job_name = job.job_name;
402 pi_variables = Whenstate.get_variables !state;
407 if pre_condition () then (
408 Syslog.notice "running %s (JOBSERIAL=%s)"
409 job.job_name (string_of_big_int serial);
411 (* Create a temporary directory. The current directory of the job
412 * will be in this directory. The directory is removed when the
413 * child process exits.
415 let dir = tmpdir () in
418 if pid = 0 then ( (* child process running the job *)
421 (* Set environment variables corresponding to each variable. *)
423 (fun (name, value) -> putenv name (string_of_variable value))
424 (Whenstate.get_variables !state);
426 (* Set the $JOBNAME environment variable. *)
427 putenv "JOBNAME" job.job_name;
429 (* Create a temporary file containing the shell script fragment. *)
430 let script = dir // "script.sh" in
431 let chan = open_out script in
432 fprintf chan "set -e\n"; (* So that jobs exit on error. *)
433 output_string chan job.job_script.sh_script;
437 let shell = try getenv "SHELL" with Not_found -> "/bin/sh" in
439 (* Set output to file. *)
440 let output = dir // "output.txt" in
441 let fd = openfile output [O_WRONLY; O_CREAT; O_TRUNC; O_NOCTTY] 0o600 in
446 (* Execute the shell script. *)
447 (try execvp shell [| shell; "-c"; script |];
448 with Unix_error (err, fn, _) ->
449 Syslog.error "%s failed: %s: %s" fn script (error_message err)
454 (* Remember this PID, the job and the temporary directory, so we
455 * can clean up when the child exits.
457 runningmap := IntMap.add pid (job, dir, serial, time ()) !runningmap;
458 serialmap := BigIntMap.add serial pid !serialmap
461 Syslog.notice "not running %s (JOBSERIAL=%s) because pre() condition returned false"
462 job.job_name (string_of_big_int serial);
466 let chan = open_in "/dev/urandom" in
467 let data = String.create 16 in
468 really_input chan data 0 (String.length data);
470 let data = Digest.to_hex (Digest.string data) in
471 let dir = Filename.temp_dir_name // sprintf "whenjobs%s" data in
475 (* This is called when a job (child process) exits. *)
476 and handle_sigchld _ =
478 let pid, status = waitpid [WNOHANG] 0 in
480 (* Look up the PID in the running jobs map. *)
481 let job, dir, serial, time = IntMap.find pid !runningmap in
482 runningmap := IntMap.remove pid !runningmap;
483 serialmap := BigIntMap.remove serial !serialmap;
484 post_job job dir serial time status
486 with Unix_error _ | Not_found -> ()
488 and post_job job dir serial time status =
489 (* If there is a post function, run it. *)
490 (match job.job_post with
496 | WSIGNALED s | WSTOPPED s -> 1 in
498 res_job_name = job.job_name;
502 res_output = dir // "output.txt";
503 res_start_time = time
508 Syslog.error "job %s post function failed: %s" job.job_name msg
510 Syslog.error "job %s post function exception: %s"
511 job.job_name (Printexc.to_string exn)
514 (* This should be safe because the path cannot contain shell metachars. *)
515 let cmd = sprintf "rm -rf '%s'" dir in
516 ignore (Sys.command cmd)
518 (* Intelligent comparison of job names. *)
519 and compare_jobnames name1 name2 =
521 let len1 = String.length name1
522 and len2 = String.length name2 in
523 if len1 > 4 && len2 > 4 &&
524 String.sub name1 0 4 = "job$" && String.sub name2 0 4 = "job$"
526 let i1 = int_of_string (String.sub name1 4 (len1-4)) in
527 let i2 = int_of_string (String.sub name2 4 (len2-4)) in