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
83 (Rpc_server.Unix addr)
84 Rpc.Tcp (* not TCP, this is the same as SOCK_STREAM *)
89 (* Handle SIGCHLD to clean up jobs. *)
90 Sys.set_signal Sys.sigchld (Sys.Signal_handle handle_sigchld);
92 (* Initialize the variables. *)
93 state := Whenstate.set_variable !state "JOBSERIAL" (T_int zero_big_int)
95 and proc_reload_file () =
96 if !debug then Syslog.notice "remote call: reload_file";
98 try reload_file (); `ok
99 with Failure err -> `error err
101 and proc_set_variable (name, value) =
102 if !debug then Syslog.notice "remote call: set_variable %s" name;
105 check_valid_variable_name name;
107 let value = variable_of_rpc value in
108 state := Whenstate.set_variable !state name value;
110 (* Which jobs need to be re-evaluated? *)
111 let jobs = Whenstate.get_dependencies !state [name] in
112 reevaluate_whenjobs jobs;
116 Failure msg -> `error msg
118 and proc_get_variable name =
119 if !debug then Syslog.notice "remote call: get_variable %s" name;
121 rpc_of_variable (Whenstate.get_variable !state name)
123 and proc_get_variable_names () =
124 if !debug then Syslog.notice "remote call: get_variable_names";
126 let vars = Whenstate.get_variable_names !state in
128 (* Return variable names as a sorted array. *)
129 let vars = Array.of_list vars in
130 Array.sort compare vars;
133 and proc_exit_daemon () =
134 if !debug then Syslog.notice "remote call: exit_daemon";
138 `error "exit_daemon: no server handle"
140 Rpc_server.stop_server ~graceful:true s;
144 and proc_get_jobs () =
145 let running = Array.of_list (IntMap.values !runningmap) in
147 fun (job, dir, serial, start_time) ->
148 { Whenproto_aux.job_name = job.job_name;
149 job_serial = string_of_big_int serial;
150 job_tmpdir = dir; job_start_time = Int64.of_float start_time }
153 and proc_cancel_job serial =
155 let serial = big_int_of_string serial in
156 let pid = BigIntMap.find serial !serialmap in
160 | Not_found -> `error "job not found"
161 | exn -> `error (Printexc.to_string exn)
163 and proc_start_job jobname =
165 let job = Whenstate.get_job !state jobname in
169 | Not_found -> `error "job not found"
170 | exn -> `error (Printexc.to_string exn)
172 and proc_get_job serial =
174 let serial = big_int_of_string serial in
175 let pid = BigIntMap.find serial !serialmap in
176 let job, dir, serial, start_time = IntMap.find pid !runningmap in
177 { Whenproto_aux.job_name = job.job_name;
178 job_serial = string_of_big_int serial;
179 job_tmpdir = dir; job_start_time = Int64.of_float start_time }
181 | Not_found -> failwith "job not found"
182 | exn -> failwith (Printexc.to_string exn)
184 and proc_set_variables vars =
186 let vars = Array.map (
187 fun { Whenproto_aux.sv_name = name; sv_value = value } ->
188 name, variable_of_rpc value
190 let vars = Array.to_list vars in
193 Syslog.notice "remote call: set_variables (%s)"
197 sprintf "%s=%s" name (string_of_variable value)
200 List.iter (fun (name, _) -> check_valid_variable_name name) vars;
202 (* Update all the variables atomically. *)
203 let s = List.fold_left (
204 fun s (name, value) -> Whenstate.set_variable s name value
208 (* Which jobs need to be re-evaluated? *)
209 let jobs = Whenstate.get_dependencies !state (List.map fst vars) in
210 reevaluate_whenjobs jobs;
214 Failure msg -> `error msg
216 (* Reload the jobs file. *)
218 let file = sprintf "%s/jobs.cmo" !jobsdir in
220 (* As we are reloading the file, we want to create a new state
221 * that has no jobs, but has all the variables from the previous
224 let s = Whenstate.copy_variables !state Whenstate.empty in
229 Dynlink.loadfile file;
230 let s = Whenfile.get_state () in
231 Syslog.notice "loaded %d job(s) from %s" (Whenstate.nr_jobs s) file;
234 | Dynlink.Error err ->
235 let err = Dynlink.error_message err in
236 Syslog.error "error loading jobs: %s" err;
239 failwith (Printexc.to_string exn) in
241 let s = Whenstate.copy_prev_state !state s in
244 (* Re-evaluate all when jobs. *)
245 reevaluate_whenjobs ~onload:true (Whenstate.get_whenjobs !state);
247 (* Schedule the next every job to run. *)
248 schedule_next_everyjob ()
250 (* Re-evaluate each when-statement job, in a loop until we reach
251 * a fixpoint. Run those that need to be run.
253 and reevaluate_whenjobs ?onload jobs =
254 let rec loop set jobs =
259 try Whenstate.evaluate_whenjob ?onload !state job
260 with Invalid_argument err | Failure err ->
261 Syslog.error "error evaluating job %s (at %s): %s"
262 job.job_name (Camlp4.PreCast.Ast.Loc.to_string job.job_loc) err;
268 Syslog.notice "evaluate %s -> %b\n" job.job_name r;
270 if r then StringSet.add job.job_name set else set
272 if StringSet.compare set set' <> 0 then
277 let set = loop StringSet.empty jobs in
278 let jobnames = StringSet.elements set in
280 (* Ensure the jobs always run in predictable (name) order. *)
281 let jobnames = List.sort compare_jobnames jobnames in
284 List.iter run_job (List.map (Whenstate.get_job !state) jobnames)
286 (* Schedule the next every-statement job to run, if there is one. We
287 * look at the every jobs, work out the time that each must run at,
288 * pick the job(s) which must run soonest, and schedule a timer to run
289 * them. When the timer fires, it runs those jobs, then calls this
292 and schedule_next_everyjob () =
295 (* Get only everyjobs. *)
296 let jobs = Whenstate.get_everyjobs !state in
297 let jobs = List.map (
299 | { job_cond = Every_job period } as job -> (job, period)
300 | { job_cond = When_job _ } -> assert false
303 (* Map everyjob to next time it must run. *)
304 let jobs = List.map (
306 let t' = next_periodexpr t period in
307 assert (t' > t); (* serious bug in next_periodexpr if false *)
311 (* Sort, soonest first. *)
312 let jobs = List.sort (fun (_,a) (_,b) -> compare a b) jobs in
317 Syslog.notice "%s: next scheduled run at %s"
318 job.job_name (string_of_time_t t)
322 (* Pick the job(s) which run soonest. *)
323 let rec pick = function
326 | (j1, t) :: (j2, t') :: _ when t < t' -> t, [j1]
327 | (j1, t) :: (((j2, t') :: _) as rest) -> t, (j1 :: snd (pick rest))
329 let t, jobs = pick jobs in
333 (* Ensure the jobs always run in predictable (name) order. *)
335 List.sort (fun {job_name = a} {job_name = b} -> compare_jobnames a b)
339 Syslog.notice "scheduling job(s) %s to run at %s"
340 (String.concat ", " (List.map (fun { job_name = name } -> name) jobs))
341 (string_of_time_t t);
343 (* Schedule them to run at time t. *)
344 let g = new_timer_group () in
345 let t_diff = t -. Unix.time () in
346 let t_diff = if t_diff < 0. then 0. else t_diff in
348 delete_timer_group (); (* Delete the timer. *)
349 List.iter run_job jobs;
350 schedule_next_everyjob ()
352 Unixqueue.weak_once esys g t_diff run_jobs;
356 and new_timer_group () =
357 delete_timer_group ();
358 let g = Unixqueue.new_group esys in
359 timer_group := Some g;
362 and delete_timer_group () =
363 match !timer_group with
366 Unixqueue.clear esys g;
370 (* Increment JOBSERIAL. *)
372 match Whenstate.get_variable !state "JOBSERIAL" with
374 let serial = succ_big_int serial in
375 state := Whenstate.set_variable !state "JOBSERIAL" (T_int serial);
377 | _ -> assert false in
379 (* Call the pre-condition script. Note this may decide not to run
380 * the job by returning false.
382 let pre_condition () =
383 match job.job_pre with
388 fun pid (job, _, serial, start_time) ->
389 let r = { pirun_job_name = job.job_name;
390 pirun_serial = serial;
391 pirun_start_time = start_time;
396 pi_job_name = job.job_name;
398 pi_variables = Whenstate.get_variables !state;
403 if pre_condition () then (
404 Syslog.notice "running %s (JOBSERIAL=%s)"
405 job.job_name (string_of_big_int serial);
407 (* Create a temporary directory. The current directory of the job
408 * will be in this directory. The directory is removed when the
409 * child process exits.
411 let dir = tmpdir () in
414 if pid = 0 then ( (* child process running the job *)
417 (* Set environment variables corresponding to each variable. *)
419 (fun (name, value) -> putenv name (string_of_variable value))
420 (Whenstate.get_variables !state);
422 (* Set the $JOBNAME environment variable. *)
423 putenv "JOBNAME" job.job_name;
425 (* Create a temporary file containing the shell script fragment. *)
426 let script = dir // "script.sh" in
427 let chan = open_out script in
428 fprintf chan "set -e\n"; (* So that jobs exit on error. *)
429 output_string chan job.job_script.sh_script;
433 let shell = try getenv "SHELL" with Not_found -> "/bin/sh" in
435 (* Set output to file. *)
436 let output = dir // "output.txt" in
437 let fd = openfile output [O_WRONLY; O_CREAT; O_TRUNC; O_NOCTTY] 0o600 in
442 (* Execute the shell script. *)
443 (try execvp shell [| shell; "-c"; script |];
444 with Unix_error (err, fn, _) ->
445 Syslog.error "%s failed: %s: %s" fn script (error_message err)
450 (* Remember this PID, the job and the temporary directory, so we
451 * can clean up when the child exits.
453 runningmap := IntMap.add pid (job, dir, serial, time ()) !runningmap;
454 serialmap := BigIntMap.add serial pid !serialmap
457 Syslog.notice "not running %s (JOBSERIAL=%s) because pre() condition returned false"
458 job.job_name (string_of_big_int serial);
462 let chan = open_in "/dev/urandom" in
463 let data = String.create 16 in
464 really_input chan data 0 (String.length data);
466 let data = Digest.to_hex (Digest.string data) in
467 let dir = Filename.temp_dir_name // sprintf "whenjobs%s" data in
471 (* This is called when a job (child process) exits. *)
472 and handle_sigchld _ =
474 let pid, status = waitpid [WNOHANG] 0 in
476 (* Look up the PID in the running jobs map. *)
477 let job, dir, serial, time = IntMap.find pid !runningmap in
478 runningmap := IntMap.remove pid !runningmap;
479 serialmap := BigIntMap.remove serial !serialmap;
480 post_job job dir serial time status
482 with Unix_error _ | Not_found -> ()
484 and post_job job dir serial time status =
485 (* If there is a post function, run it. *)
486 (match job.job_post with
492 | WSIGNALED s | WSTOPPED s -> 1 in
494 res_job_name = job.job_name;
498 res_output = dir // "output.txt";
499 res_start_time = time
504 Syslog.error "job %s post function failed: %s" job.job_name msg
506 Syslog.error "job %s post function exception: %s"
507 job.job_name (Printexc.to_string exn)
510 (* This should be safe because the path cannot contain shell metachars. *)
511 let cmd = sprintf "rm -rf '%s'" dir in
512 ignore (Sys.command cmd)
514 (* Intelligent comparison of job names. *)
515 and compare_jobnames name1 name2 =
517 let len1 = String.length name1
518 and len2 = String.length name2 in
519 if len1 > 4 && len2 > 4 &&
520 String.sub name1 0 4 = "job$" && String.sub name2 0 4 = "job$"
522 let i1 = int_of_string (String.sub name1 4 (len1-4)) in
523 let i2 = int_of_string (String.sub name2 4 (len2-4)) in