X-Git-Url: http://git.annexia.org/?a=blobdiff_plain;f=daemon%2Fdaemon.ml;h=6e6a5fd9634631ef433eaeeadc6e5fc97e1c6e60;hb=50a4457bd64045abfdc88237a934fadff0bb7f84;hp=6c3799ccd9d0142d720337bd1c6b1f10adf3a06f;hpb=108dd86b36e82df2a2029dbd12700f9c83e501c1;p=whenjobs.git diff --git a/daemon/daemon.ml b/daemon/daemon.ml index 6c3799c..6e6a5fd 100644 --- a/daemon/daemon.ml +++ b/daemon/daemon.ml @@ -32,11 +32,14 @@ let jobsdir = ref "" (* The state. *) let state = ref Whenstate.empty -(* Jobs that are running; map of PID -> (job, other data). Note that - * the job may no longer exist *OR* it may have been renamed, +(* Jobs that are running: a map of PID -> (job, tmpdir, serial, start_time). + * Note that the job may no longer exist *OR* it may have been renamed, * eg. if the jobs file was reloaded. *) -let running = ref IntMap.empty +let runningmap = ref IntMap.empty + +(* Serial numbers of running jobs. Map of serial -> PID (in runningmap). *) +let serialmap = ref BigIntMap.empty (* Was debugging requested on the command line? *) let debug = ref false @@ -72,6 +75,10 @@ let rec init j d = ~proc_get_variable ~proc_get_variable_names ~proc_exit_daemon + ~proc_get_jobs + ~proc_cancel_job + ~proc_start_job + ~proc_get_job (Rpc_server.Unix addr) Rpc.Tcp (* not TCP, this is the same as SOCK_STREAM *) Rpc.Socket @@ -94,23 +101,7 @@ and proc_set_variable (name, value) = if !debug then Syslog.notice "remote call: set_variable %s" name; try - (* Don't permit certain names. *) - if name = "JOBSERIAL" then - failwith "JOBSERIAL variable cannot be set"; - - let len = String.length name in - if len = 0 then - failwith "variable name is an empty string"; - if name.[0] <> '_' && not (isalpha name.[0]) then - failwith "variable name must start with alphabetic character or underscore"; - - let rec loop i = - if i >= len then () - else if name.[i] <> '_' && not (isalnum name.[i]) then - failwith "variable name contains non-alphanumeric non-underscore character" - else loop (i+1) - in - loop 1; + check_valid_variable_name name; let value = variable_of_rpc value in state := Whenstate.set_variable !state name value; @@ -149,6 +140,46 @@ and proc_exit_daemon () = server := None; `ok +and proc_get_jobs () = + let running = Array.of_list (IntMap.values !runningmap) in + Array.map ( + fun (job, dir, serial, start_time) -> + { Whenproto_aux.job_name = job.job_name; + job_serial = string_of_big_int serial; + job_tmpdir = dir; job_start_time = Int64.of_float start_time } + ) running + +and proc_cancel_job serial = + try + let serial = big_int_of_string serial in + let pid = BigIntMap.find serial !serialmap in + kill pid 15; + `ok + with + | Not_found -> `error "job not found" + | exn -> `error (Printexc.to_string exn) + +and proc_start_job jobname = + try + let job = Whenstate.get_job !state jobname in + run_job job; + `ok + with + | Not_found -> `error "job not found" + | exn -> `error (Printexc.to_string exn) + +and proc_get_job serial = + try + let serial = big_int_of_string serial in + let pid = BigIntMap.find serial !serialmap in + let job, dir, serial, start_time = IntMap.find pid !runningmap in + { Whenproto_aux.job_name = job.job_name; + job_serial = string_of_big_int serial; + job_tmpdir = dir; job_start_time = Int64.of_float start_time } + with + | Not_found -> failwith "job not found" + | exn -> failwith (Printexc.to_string exn) + (* Reload the jobs file. *) and reload_file () = let file = sprintf "%s/jobs.cmo" !jobsdir in @@ -174,6 +205,7 @@ and reload_file () = | exn -> failwith (Printexc.to_string exn) in + let s = Whenstate.copy_prev_state !state s in state := s; (* Re-evaluate all when jobs. *) @@ -301,66 +333,97 @@ and delete_timer_group () = Unixqueue.clear esys g; timer_group := None -and string_of_time_t t = - let tm = gmtime t in - sprintf "%04d-%02d-%02d %02d:%02d:%02d UTC" - (1900+tm.tm_year) (1+tm.tm_mon) tm.tm_mday - tm.tm_hour tm.tm_min tm.tm_sec - and run_job job = - let () = - (* Increment JOBSERIAL. *) - let serial = - match Whenstate.get_variable !state "JOBSERIAL" with - | T_int serial -> - let serial = succ_big_int serial in - state := Whenstate.set_variable !state "JOBSERIAL" (T_int serial); - serial - | _ -> assert false in - - Syslog.notice "running %s (JOBSERIAL=%s)" - job.job_name (string_of_big_int serial) in - - (* Create a temporary directory. The current directory of the job - * will be in this directory. The directory is removed when the - * child process exits. + (* Increment JOBSERIAL. *) + let serial = + match Whenstate.get_variable !state "JOBSERIAL" with + | T_int serial -> + let serial = succ_big_int serial in + state := Whenstate.set_variable !state "JOBSERIAL" (T_int serial); + serial + | _ -> assert false in + + (* Call the pre-condition script. Note this may decide not to run + * the job by returning false. *) - let dir = tmpdir () in - - let pid = fork () in - if pid = 0 then ( (* child process running the job *) - chdir dir; - - (* Set environment variables corresponding to each variable. *) - List.iter - (fun (name, value) -> putenv name (string_of_variable value)) - (Whenstate.get_variables !state); - - (* Set the $JOBNAME environment variable. *) - putenv "JOBNAME" job.job_name; - - (* Create a temporary file containing the shell script fragment. *) - let script = dir // "script" in - let chan = open_out script in - fprintf chan "set -e\n"; (* So that jobs exit on error. *) - output_string chan job.job_script.sh_script; - close_out chan; - chmod script 0o700; - - let shell = try getenv "SHELL" with Not_found -> "/bin/sh" in - - (* Execute the shell script. *) - (try execvp shell [| shell; "-c"; script |]; - with Unix_error (err, fn, _) -> - Syslog.error "%s failed: %s: %s" fn script (error_message err) + let pre_condition () = + match job.job_pre with + | None -> true + | Some pre -> + let rs = ref [] in + IntMap.iter ( + fun pid (job, _, serial, start_time) -> + let r = { pirun_job_name = job.job_name; + pirun_serial = serial; + pirun_start_time = start_time; + pirun_pid = pid } in + rs := r :: !rs + ) !runningmap; + let preinfo = { + pi_job_name = job.job_name; + pi_serial = serial; + pi_variables = Whenstate.get_variables !state; + pi_running = !rs; + } in + pre preinfo + in + if pre_condition () then ( + Syslog.notice "running %s (JOBSERIAL=%s)" + job.job_name (string_of_big_int serial); + + (* Create a temporary directory. The current directory of the job + * will be in this directory. The directory is removed when the + * child process exits. + *) + let dir = tmpdir () in + + let pid = fork () in + if pid = 0 then ( (* child process running the job *) + chdir dir; + + (* Set environment variables corresponding to each variable. *) + List.iter + (fun (name, value) -> putenv name (string_of_variable value)) + (Whenstate.get_variables !state); + + (* Set the $JOBNAME environment variable. *) + putenv "JOBNAME" job.job_name; + + (* Create a temporary file containing the shell script fragment. *) + let script = dir // "script.sh" in + let chan = open_out script in + fprintf chan "set -e\n"; (* So that jobs exit on error. *) + output_string chan job.job_script.sh_script; + close_out chan; + chmod script 0o700; + + let shell = try getenv "SHELL" with Not_found -> "/bin/sh" in + + (* Set output to file. *) + let output = dir // "output.txt" in + let fd = openfile output [O_WRONLY; O_CREAT; O_TRUNC; O_NOCTTY] 0o600 in + dup2 fd stdout; + dup2 fd stderr; + close fd; + + (* Execute the shell script. *) + (try execvp shell [| shell; "-c"; script |]; + with Unix_error (err, fn, _) -> + Syslog.error "%s failed: %s: %s" fn script (error_message err) + ); + _exit 1 ); - _exit 1 - ); - (* Remember this PID, the job and the temporary directory, so we - * can clean up when the child exits. - *) - running := IntMap.add pid (job, dir) !running + (* Remember this PID, the job and the temporary directory, so we + * can clean up when the child exits. + *) + runningmap := IntMap.add pid (job, dir, serial, time ()) !runningmap; + serialmap := BigIntMap.add serial pid !serialmap + ) + else ( + Syslog.notice "not running %s (JOBSERIAL=%s) because pre() condition returned false" + job.job_name (string_of_big_int serial); + ) and tmpdir () = let chan = open_in "/dev/urandom" in @@ -378,13 +441,39 @@ and handle_sigchld _ = let pid, status = waitpid [WNOHANG] 0 in if pid > 0 then ( (* Look up the PID in the running jobs map. *) - let job, dir = IntMap.find pid !running in - running := IntMap.remove pid !running; - cleanup_job job dir + let job, dir, serial, time = IntMap.find pid !runningmap in + runningmap := IntMap.remove pid !runningmap; + serialmap := BigIntMap.remove serial !serialmap; + post_job job dir serial time status ) with Unix_error _ | Not_found -> () -and cleanup_job job dir = +and post_job job dir serial time status = + (* If there is a post function, run it. *) + (match job.job_post with + | None -> () + | Some post -> + let code = + match status with + | WEXITED c -> c + | WSIGNALED s | WSTOPPED s -> 1 in + let result = { + res_job_name = job.job_name; + res_serial = serial; + res_code = code; + res_tmpdir = dir; + res_output = dir // "output.txt"; + res_start_time = time + } in + try post result + with + | Failure msg -> + Syslog.error "job %s post function failed: %s" job.job_name msg + | exn -> + Syslog.error "job %s post function exception: %s" + job.job_name (Printexc.to_string exn) + ); + (* This should be safe because the path cannot contain shell metachars. *) let cmd = sprintf "rm -rf '%s'" dir in ignore (Sys.command cmd)