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
81 (Rpc_server.Unix addr)
82 Rpc.Tcp (* not TCP, this is the same as SOCK_STREAM *)
87 (* Handle SIGCHLD to clean up jobs. *)
88 Sys.set_signal Sys.sigchld (Sys.Signal_handle handle_sigchld);
90 (* Initialize the variables. *)
91 state := Whenstate.set_variable !state "JOBSERIAL" (T_int zero_big_int)
93 and proc_reload_file () =
94 if !debug then Syslog.notice "remote call: reload_file";
96 try reload_file (); `ok
97 with Failure err -> `error err
99 and proc_set_variable (name, value) =
100 if !debug then Syslog.notice "remote call: set_variable %s" name;
103 check_valid_variable_name name;
105 let value = variable_of_rpc value in
106 state := Whenstate.set_variable !state name value;
108 (* Which jobs need to be re-evaluated? *)
109 let jobs = Whenstate.get_dependencies !state name in
110 reevaluate_whenjobs jobs;
114 Failure msg -> `error msg
116 and proc_get_variable name =
117 if !debug then Syslog.notice "remote call: get_variable %s" name;
119 rpc_of_variable (Whenstate.get_variable !state name)
121 and proc_get_variable_names () =
122 if !debug then Syslog.notice "remote call: get_variable_names";
124 let vars = Whenstate.get_variable_names !state in
126 (* Return variable names as a sorted array. *)
127 let vars = Array.of_list vars in
128 Array.sort compare vars;
131 and proc_exit_daemon () =
132 if !debug then Syslog.notice "remote call: exit_daemon";
136 `error "exit_daemon: no server handle"
138 Rpc_server.stop_server ~graceful:true s;
142 and proc_get_jobs () =
143 let running = Array.of_list (IntMap.values !runningmap) in
145 fun (job, dir, serial, start_time) ->
146 { Whenproto_aux.job_name = job.job_name;
147 job_serial = string_of_big_int serial;
148 job_tmpdir = dir; job_start_time = Int64.of_float start_time }
151 and proc_cancel_job serial =
153 let serial = big_int_of_string serial in
154 let pid = BigIntMap.find serial !serialmap in
158 | Not_found -> `error "job not found"
159 | exn -> `error (Printexc.to_string exn)
161 and proc_start_job jobname =
163 let job = Whenstate.get_job !state jobname in
167 | Not_found -> `error "job not found"
168 | exn -> `error (Printexc.to_string exn)
170 (* Reload the jobs file. *)
172 let file = sprintf "%s/jobs.cmo" !jobsdir in
174 (* As we are reloading the file, we want to create a new state
175 * that has no jobs, but has all the variables from the previous
178 let s = Whenstate.copy_variables !state Whenstate.empty in
183 Dynlink.loadfile file;
184 let s = Whenfile.get_state () in
185 Syslog.notice "loaded %d job(s) from %s" (Whenstate.nr_jobs s) file;
188 | Dynlink.Error err ->
189 let err = Dynlink.error_message err in
190 Syslog.error "error loading jobs: %s" err;
193 failwith (Printexc.to_string exn) in
197 (* Re-evaluate all when jobs. *)
198 reevaluate_whenjobs ~onload:true (Whenstate.get_whenjobs !state);
200 (* Schedule the next every job to run. *)
201 schedule_next_everyjob ()
203 (* Re-evaluate each when-statement job, in a loop until we reach
204 * a fixpoint. Run those that need to be run.
206 and reevaluate_whenjobs ?onload jobs =
207 let rec loop set jobs =
212 try Whenstate.evaluate_whenjob ?onload !state job
213 with Invalid_argument err | Failure err ->
214 Syslog.error "error evaluating job %s (at %s): %s"
215 job.job_name (Camlp4.PreCast.Ast.Loc.to_string job.job_loc) err;
221 Syslog.notice "evaluate %s -> %b\n" job.job_name r;
223 if r then StringSet.add job.job_name set else set
225 if StringSet.compare set set' <> 0 then
230 let set = loop StringSet.empty jobs in
231 let jobnames = StringSet.elements set in
233 (* Ensure the jobs always run in predictable (name) order. *)
234 let jobnames = List.sort compare_jobnames jobnames in
237 List.iter run_job (List.map (Whenstate.get_job !state) jobnames)
239 (* Schedule the next every-statement job to run, if there is one. We
240 * look at the every jobs, work out the time that each must run at,
241 * pick the job(s) which must run soonest, and schedule a timer to run
242 * them. When the timer fires, it runs those jobs, then calls this
245 and schedule_next_everyjob () =
248 (* Get only everyjobs. *)
249 let jobs = Whenstate.get_everyjobs !state in
250 let jobs = List.map (
252 | { job_cond = Every_job period } as job -> (job, period)
253 | { job_cond = When_job _ } -> assert false
256 (* Map everyjob to next time it must run. *)
257 let jobs = List.map (
259 let t' = next_periodexpr t period in
260 assert (t' > t); (* serious bug in next_periodexpr if false *)
264 (* Sort, soonest first. *)
265 let jobs = List.sort (fun (_,a) (_,b) -> compare a b) jobs in
270 Syslog.notice "%s: next scheduled run at %s"
271 job.job_name (string_of_time_t t)
275 (* Pick the job(s) which run soonest. *)
276 let rec pick = function
279 | (j1, t) :: (j2, t') :: _ when t < t' -> t, [j1]
280 | (j1, t) :: (((j2, t') :: _) as rest) -> t, (j1 :: snd (pick rest))
282 let t, jobs = pick jobs in
286 (* Ensure the jobs always run in predictable (name) order. *)
288 List.sort (fun {job_name = a} {job_name = b} -> compare_jobnames a b)
292 Syslog.notice "scheduling job(s) %s to run at %s"
293 (String.concat ", " (List.map (fun { job_name = name } -> name) jobs))
294 (string_of_time_t t);
296 (* Schedule them to run at time t. *)
297 let g = new_timer_group () in
298 let t_diff = t -. Unix.time () in
299 let t_diff = if t_diff < 0. then 0. else t_diff in
301 delete_timer_group (); (* Delete the timer. *)
302 List.iter run_job jobs;
303 schedule_next_everyjob ()
305 Unixqueue.weak_once esys g t_diff run_jobs;
309 and new_timer_group () =
310 delete_timer_group ();
311 let g = Unixqueue.new_group esys in
312 timer_group := Some g;
315 and delete_timer_group () =
316 match !timer_group with
319 Unixqueue.clear esys g;
323 (* Increment JOBSERIAL. *)
325 match Whenstate.get_variable !state "JOBSERIAL" with
327 let serial = succ_big_int serial in
328 state := Whenstate.set_variable !state "JOBSERIAL" (T_int serial);
330 | _ -> assert false in
332 (* Call the pre-condition script. Note this may decide not to run
333 * the job by returning false.
335 let pre_condition () =
336 match job.job_pre with
341 fun pid (job, _, serial, start_time) ->
342 let r = { pirun_job_name = job.job_name;
343 pirun_serial = serial;
344 pirun_start_time = start_time;
349 pi_job_name = job.job_name;
351 pi_variables = Whenstate.get_variables !state;
356 if pre_condition () then (
357 Syslog.notice "running %s (JOBSERIAL=%s)"
358 job.job_name (string_of_big_int serial);
360 (* Create a temporary directory. The current directory of the job
361 * will be in this directory. The directory is removed when the
362 * child process exits.
364 let dir = tmpdir () in
367 if pid = 0 then ( (* child process running the job *)
370 (* Set environment variables corresponding to each variable. *)
372 (fun (name, value) -> putenv name (string_of_variable value))
373 (Whenstate.get_variables !state);
375 (* Set the $JOBNAME environment variable. *)
376 putenv "JOBNAME" job.job_name;
378 (* Create a temporary file containing the shell script fragment. *)
379 let script = dir // "script.sh" in
380 let chan = open_out script in
381 fprintf chan "set -e\n"; (* So that jobs exit on error. *)
382 output_string chan job.job_script.sh_script;
386 let shell = try getenv "SHELL" with Not_found -> "/bin/sh" in
388 (* Set output to file. *)
389 let output = dir // "output.txt" in
390 let fd = openfile output [O_WRONLY; O_CREAT; O_TRUNC; O_NOCTTY] 0o600 in
395 (* Execute the shell script. *)
396 (try execvp shell [| shell; "-c"; script |];
397 with Unix_error (err, fn, _) ->
398 Syslog.error "%s failed: %s: %s" fn script (error_message err)
403 (* Remember this PID, the job and the temporary directory, so we
404 * can clean up when the child exits.
406 runningmap := IntMap.add pid (job, dir, serial, time ()) !runningmap;
407 serialmap := BigIntMap.add serial pid !serialmap
410 Syslog.notice "not running %s (JOBSERIAL=%s) because pre() condition returned false"
411 job.job_name (string_of_big_int serial);
415 let chan = open_in "/dev/urandom" in
416 let data = String.create 16 in
417 really_input chan data 0 (String.length data);
419 let data = Digest.to_hex (Digest.string data) in
420 let dir = Filename.temp_dir_name // sprintf "whenjobs%s" data in
424 (* This is called when a job (child process) exits. *)
425 and handle_sigchld _ =
427 let pid, status = waitpid [WNOHANG] 0 in
429 (* Look up the PID in the running jobs map. *)
430 let job, dir, serial, time = IntMap.find pid !runningmap in
431 runningmap := IntMap.remove pid !runningmap;
432 serialmap := BigIntMap.remove serial !serialmap;
433 post_job job dir serial time status
435 with Unix_error _ | Not_found -> ()
437 and post_job job dir serial time status =
438 (* If there is a post function, run it. *)
439 (match job.job_post with
445 | WSIGNALED s | WSTOPPED s -> 1 in
447 res_job_name = job.job_name;
451 res_output = dir // "output.txt";
452 res_start_time = time
457 Syslog.error "job %s post function failed: %s" job.job_name msg
459 Syslog.error "job %s post function exception: %s"
460 job.job_name (Printexc.to_string exn)
463 (* This should be safe because the path cannot contain shell metachars. *)
464 let cmd = sprintf "rm -rf '%s'" dir in
465 ignore (Sys.command cmd)
467 (* Intelligent comparison of job names. *)
468 and compare_jobnames name1 name2 =
470 let len1 = String.length name1
471 and len2 = String.length name2 in
472 if len1 > 4 && len2 > 4 &&
473 String.sub name1 0 4 = "job$" && String.sub name2 0 4 = "job$"
475 let i1 = int_of_string (String.sub name1 4 (len1-4)) in
476 let i2 = int_of_string (String.sub name2 4 (len2-4)) in