Index: /branches/eam_branches/20091201/Ohana/src/opihi/include/pcontrol.h =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/include/pcontrol.h (revision 26714) +++ /branches/eam_branches/20091201/Ohana/src/opihi/include/pcontrol.h (revision 26715) @@ -98,4 +98,14 @@ int requested; } JobOutput; + +/* A machine has a unique name and may have multiple Hosts (each of which can run a single job) + We use this to track aspects of the analysis per machine, to eg, limit the number of jobs + desired on a single machine */ +typedef struct { + char *name; + int Nhosts; // how many hosts are selected for this machine (whatever state) + int NjobsRealhost; + int NjobsWanthost; +} Machine; /* data to define a job */ @@ -140,4 +150,5 @@ } Host; +/* the Jobs and Hosts are managed in a set of Stacks which define their state */ typedef struct { void **object; @@ -280,4 +291,24 @@ void LinkJobAndHost (Job *job, Host *host); +/*** MachineOps.c ***/ +void InitMachines (); +void FreeMachines (); +Machine *FindMachineByName (char *name); +Machine *AddMachine (char *name); +int DelMachine (char *name); +int AddMachineHost (Host *host); +int DelMachineHost (Host *host); +int AddMachineJob (Host *host, Job *job); +int DelMachineJob (Host *host, Job *job); +int PrintMachines (); +int CheckMachineJobs (Host *host, Job *job); +int GetMaxUnwantedHostJobs (void); +void SetMaxUnwantedHostJobs (int value); + +float GetMaxConnectTime (void); +void SetMaxConnectTime (float value); +float GetMaxWantHostWait (void); +void SetMaxWantHostWait (float value); + void pcontrol_exit (int n); Index: /branches/eam_branches/20091201/Ohana/src/opihi/lib.shell/ListOps.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/lib.shell/ListOps.c (revision 26714) +++ /branches/eam_branches/20091201/Ohana/src/opihi/lib.shell/ListOps.c (revision 26715) @@ -3,6 +3,6 @@ /*** local static variables used to track the command lists ***/ -static List *lists; /* variable to store the list of all lists */ -static int Nlists; /* number of currently available lists */ +static List *lists = NULL; /* variable to store the list of all lists */ +static int Nlists = 0; /* number of currently available lists */ void InitLists () { @@ -16,5 +16,7 @@ int i, j; - for (i = 0; i < Nlists; i++) { + // Nlists is a bit weird: it is the currently highest valid list, not the number of lists + // Nlists = 0 is never allocated + for (i = 1; i < Nlists + 1; i++) { for (j = 0; j < lists[i].Nlines; j++) { free (lists[i].line[j]); Index: /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/Makefile =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/Makefile (revision 26714) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/Makefile (revision 26715) @@ -73,4 +73,8 @@ $(SRC)/controller_status.$(ARCH).o \ $(SRC)/controller_jobstack.$(ARCH).o \ +$(SRC)/controller_hoststack.$(ARCH).o \ +$(SRC)/controller_machines.$(ARCH).o \ +$(SRC)/controller_parameters.$(ARCH).o \ +$(SRC)/controller_version.$(ARCH).o \ $(SRC)/controller_verbose.$(ARCH).o \ $(SRC)/controller_run.$(ARCH).o \ Index: /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller.c (revision 26714) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller.c (revision 26715) @@ -1,26 +1,36 @@ # include "pantasks.h" -int controller_host PROTO((int, char **)); -int controller_exit PROTO((int, char **)); -int controller_status PROTO((int, char **)); -int controller_jobstack PROTO((int, char **)); -int controller_verbose PROTO((int, char **)); -int controller_run PROTO((int, char **)); -int controller_stop PROTO((int, char **)); -int controller_check PROTO((int, char **)); -int controller_output PROTO((int, char **)); -int controller_pulse PROTO((int, char **)); +int controller_host PROTO((int, char **)); +int controller_exit PROTO((int, char **)); +int controller_status PROTO((int, char **)); +int controller_jobstack PROTO((int, char **)); +int controller_hoststack PROTO((int, char **)); +int controller_machines PROTO((int, char **)); +int controller_parameters PROTO((int, char **)); +int controller_verbose PROTO((int, char **)); +int controller_version PROTO((int, char **)); +int controller_run PROTO((int, char **)); +int controller_stop PROTO((int, char **)); +int controller_check PROTO((int, char **)); +int controller_output PROTO((int, char **)); +int controller_pulse PROTO((int, char **)); static Command controller_cmds[] = { - {1, "exit", controller_exit, "shutdown controller"}, - {1, "host", controller_host, "define host for controller"}, - {1, "check", controller_check, "check controller host/job"}, - {1, "run", controller_run, "start controller operation / set run levels"}, - {1, "stop", controller_run, "stop controller (no disconnect)"}, - {1, "status", controller_status, "check controller status"}, - {1, "jobstack", controller_jobstack, "check controller status"}, - {1, "verbose", controller_verbose, "set controller verbosity"}, - {1, "output", controller_output, "print controller output"}, - {1, "pulse", controller_pulse, "set controller pulse"}, + {1, "check", controller_check, "check controller host/job"}, + // {1, "delete", controller_delete, "delete job on controller"}, + {1, "exit", controller_exit, "shutdown controller"}, + {1, "host", controller_host, "define host for controller"}, + {1, "hoststack", controller_hoststack, "define host for controller"}, + // {1, "job", controller_job, "add jobs to controller"}, + {1, "jobstack", controller_jobstack, "check controller status"}, + {1, "machines", controller_machines, "print controller machine status"}, + {1, "parameters",controller_parameters, "modify controller parameters"}, + {1, "output", controller_output, "print controller output"}, + {1, "run", controller_run, "start controller operation / set run levels"}, + {1, "status", controller_status, "check controller status"}, + {1, "stop", controller_run, "stop controller (no disconnect)"}, + {1, "verbose", controller_verbose, "set controller verbosity"}, + {1, "version", controller_version, "show controller version"}, + {1, "pulse", controller_pulse, "set controller pulse"}, }; Index: /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller_hoststack.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller_hoststack.c (revision 26715) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller_hoststack.c (revision 26715) @@ -0,0 +1,35 @@ +# include "pantasks.h" + +int controller_hoststack (int argc, char **argv) { + + int status; + char command[1024]; + IOBuffer buffer; + + if (argc != 2) { + gprint (GP_ERR, "USAGE: controller hoststack (hoststack)\n"); + gprint (GP_ERR, " (hoststack) : pending, busy, exit, crash, hung, done\n"); + return (FALSE); + } + + /* check if controller is running */ + status = CheckControllerStatus (); + if (!status) { + gprint (GP_LOG, "controller is not active\n"); + return (TRUE); + } + + // XXX this has an error? test this out... + sprintf (command, "hoststack %s", argv[1]); + InitIOBuffer (&buffer, 0x100); + + status = ControllerCommand (command, CONTROLLER_PROMPT, &buffer); + + if (status) { + gwrite (buffer.buffer, 1, buffer.Nbuffer, GP_LOG); + } else { + gprint (GP_LOG, "controller is not communicating\n"); + } + FreeIOBuffer (&buffer); + return (TRUE); +} Index: /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller_machines.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller_machines.c (revision 26715) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller_machines.c (revision 26715) @@ -0,0 +1,35 @@ +# include "pantasks.h" + +int controller_machines (int argc, char **argv) { + + int status; + char command[1024]; + IOBuffer buffer; + + if (argc != 1) { + gprint (GP_ERR, "USAGE: controller machines\n"); + gprint (GP_ERR, " prints status of controller machines\n"); + return (FALSE); + } + + /* check if controller is running */ + status = CheckControllerStatus (); + if (!status) { + gprint (GP_LOG, "controller is not active\n"); + return (TRUE); + } + + // XXX this has an error? test this out... + sprintf (command, "machines"); + InitIOBuffer (&buffer, 0x100); + + status = ControllerCommand (command, CONTROLLER_PROMPT, &buffer); + + if (status) { + gwrite (buffer.buffer, 1, buffer.Nbuffer, GP_LOG); + } else { + gprint (GP_LOG, "controller is not communicating\n"); + } + FreeIOBuffer (&buffer); + return (TRUE); +} Index: /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller_parameters.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller_parameters.c (revision 26715) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller_parameters.c (revision 26715) @@ -0,0 +1,46 @@ +# include "pantasks.h" + +int controller_parameters (int argc, char **argv) { + + int status; + char command[1024]; + IOBuffer buffer; + + if (argc < 2) goto usage; + if (argc > 4) goto usage; + if (argc == 3) goto usage; + if ((argc == 4) && strcmp(argv[2], "=")) goto usage; + + /* check if controller is running */ + status = CheckControllerStatus (); + if (!status) { + gprint (GP_LOG, "controller is not active\n"); + return (TRUE); + } + + // XXX this has an error? test this out... + if (argc == 2) { + sprintf (command, "parameters %s", argv[1]); + } else { + sprintf (command, "parameters %s = %s", argv[1], argv[3]); + } + InitIOBuffer (&buffer, 0x100); + + status = ControllerCommand (command, CONTROLLER_PROMPT, &buffer); + + if (status) { + gwrite (buffer.buffer, 1, buffer.Nbuffer, GP_LOG); + } else { + gprint (GP_LOG, "controller is not communicating\n"); + } + FreeIOBuffer (&buffer); + return (TRUE); + + usage: + + gprint (GP_LOG, "USAGE: controller parameters (param) [= value])\n"); + gprint (GP_LOG, " valid parameters: connect_time, wanthost_wait, unwanted_host_jobs\n"); + gprint (GP_LOG, " (minimum matching word is allowed)\n"); + gprint (GP_LOG, " example: controller parameters connect = 2.0\n"); + return (FALSE); +} Index: /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller_version.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller_version.c (revision 26715) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pantasks/controller_version.c (revision 26715) @@ -0,0 +1,34 @@ +# include "pantasks.h" + +int controller_version (int argc, char **argv) { + + int status; + char command[1024]; + IOBuffer buffer; + + if (argc != 1) { + gprint (GP_ERR, "USAGE: controller version\n"); + return (FALSE); + } + + /* check if controller is running */ + status = CheckControllerStatus (); + if (!status) { + gprint (GP_LOG, "controller is not active\n"); + return (TRUE); + } + + + sprintf (command, "version"); + InitIOBuffer (&buffer, 0x100); + + status = ControllerCommand (command, CONTROLLER_PROMPT, &buffer); + + if (status) { + gwrite (buffer.buffer, 1, buffer.Nbuffer, GP_LOG); + } else { + gprint (GP_LOG, "controller is not communicating\n"); + } + FreeIOBuffer (&buffer); + return (TRUE); +} Index: /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/CheckDoneJob.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/CheckDoneJob.c (revision 26714) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/CheckDoneJob.c (revision 26715) @@ -28,4 +28,8 @@ if ((status1 == PCLIENT_DOWN) || (status2 == PCLIENT_DOWN)) { + + // decrement the machine job-host counters + DelMachineJob (host, job); + // unlink host & job if (DEBUG || VerboseMode()) gprint (GP_ERR, "host %s is down\n", host[0].hostname); @@ -53,4 +57,7 @@ } + // decrement the machine job-host counters + DelMachineJob (host, job); + /* job's state is either EXIT or CRASH (verify?) */ // unlink host & job Index: /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/CheckIdleHost.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/CheckIdleHost.c (revision 26714) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/CheckIdleHost.c (revision 26715) @@ -1,11 +1,11 @@ # include "pcontrol.h" -// The connection to the remote host is only allow to live for MAX_CONNECT_TIME seconds. We -// disconnect and reconnect if a remote host has been connected for too long. This is a -// (temporary?) work-around for the problem that the remote pclient job tends to grow too large -// over time. +// The connection to the remote host is only allowed to live for MAX_CONNECT_TIME seconds. +// We disconnect and reconnect if a remote host has been connected for too long. This is +// a (temporary?) work-around for the problem that the remote pclient job tends to grow +// too large over time. -# define MAX_CONNECT_TIME 36000.0 - +static float MAX_WANTHOST_WAIT = 10.0; +static float MAX_CONNECT_TIME = 36000.0; static FILE *logfile = NULL; @@ -70,4 +70,5 @@ // if (logfile) fprintf (logfile, "start needhost %s (job host %s) : %s\n", host[0].hostname, job[0].hostname, job[0].argv[0]); + AddMachineJob (host, job); /* take the job off the stack and unlock the stack */ @@ -90,4 +91,5 @@ // if (logfile) fprintf (logfile, "start wanthost %s (job host %s) : %s\n", host[0].hostname, job[0].hostname, job[0].argv[0]); + AddMachineJob (host, job); /* take the job off the stack and unlock the stack */ @@ -103,4 +105,6 @@ if (job[0].mode != PCONTROL_JOB_ANYHOST) continue; + if (!CheckMachineJobs (host, job)) continue; + /* we have found an appropriate job; link it to the host and send to StartJob */ job[0].host = (struct Host *) host; @@ -108,4 +112,5 @@ // if (logfile) fprintf (logfile, "start anyhost %s (job host %s) : %s\n", host[0].hostname, job[0].hostname, job[0].argv[0]); + AddMachineJob (host, job); /* take the job off the stack and unlock the stack */ @@ -124,7 +129,10 @@ gettimeofday (&now, (void *) NULL); dtime = DTIME (now, job[0].start); - if (dtime < 10.0) continue; + if (dtime < MAX_WANTHOST_WAIT) continue; + + if (!CheckMachineJobs (host, job)) continue; if (logfile) fprintf (logfile, "start wanthost(2) %s (job host %s) : %s\n", host[0].hostname, job[0].hostname, job[0].argv[0]); + AddMachineJob (host, job); /* we have found an appropriate job; link it to the host and send to StartJob */ @@ -145,4 +153,26 @@ } +void SetMaxWantHostWait (float value) { + + MAX_WANTHOST_WAIT = value; + return; +} + +float GetMaxWantHostWait (void) { + + return MAX_WANTHOST_WAIT; +} + +void SetMaxConnectTime (float value) { + + MAX_CONNECT_TIME = value; + return; +} + +float GetMaxConnectTime (void) { + + return MAX_CONNECT_TIME; +} + /** note : host and job popped off IDLE and PENDING stacks, unless no job is available **/ Index: /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/CheckSystem.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/CheckSystem.c (revision 26714) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/CheckSystem.c (revision 26715) @@ -122,8 +122,10 @@ if (!Njobchecks && !Nhostchecks && (RunLevel != PCONTROL_RUN_NONE)) { CheckLiveHosts(0.040); + // fprintf (stderr, "sleep a bit\n"); usleep (100000); // idle if no jobs are waiting } else { // if we only have busy jobs, pause a moment before trying again if (!Ndonejobs) { + // fprintf (stderr, "sleep a bit\n"); usleep (100000); } Index: /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/HostOps.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/HostOps.c (revision 26714) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/HostOps.c (revision 26715) @@ -214,4 +214,6 @@ host[0].job = NULL; + AddMachineHost (host); + PutHost (host, PCONTROL_HOST_ALLHOSTS, STACK_BOTTOM); PutHost (host, PCONTROL_HOST_DOWN, STACK_BOTTOM); @@ -225,4 +227,6 @@ copy = PullStackByID (HostPool_AllHosts, host[0].HostID); ASSERT (copy == host, "programming error: ALLHOSTS entry does not match"); + + DelMachineHost (host); FreeIOBuffer (&host[0].comms_buffer); Index: /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/KillJob.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/KillJob.c (revision 26714) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/KillJob.c (revision 26715) @@ -19,4 +19,8 @@ job[0].host = NULL; host[0].job = NULL; + + // decrement the machine job-host counters + DelMachineJob (host, job); + HarvestHost (host[0].pid); PutHost (host, PCONTROL_HOST_DOWN, STACK_BOTTOM); @@ -74,4 +78,8 @@ job[0].host = NULL; host[0].job = NULL; + + // decrement the machine job-host counters + DelMachineJob (host, job); + PutHost (host, PCONTROL_HOST_IDLE, STACK_BOTTOM); PutJob (job, PCONTROL_JOB_CRASH, STACK_BOTTOM); Index: /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/MachineOps.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/MachineOps.c (revision 26715) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/MachineOps.c (revision 26715) @@ -0,0 +1,207 @@ +# include "pcontrol.h" + +static int MAX_UNWANTED_HOST_JOBS = 5; + +static Machine **machines = NULL; +static int Nmachines = 0; +static int NMACHINES = 0; + +void InitMachines () { + + if (machines != NULL) return; + + NMACHINES = 10; + Nmachines = 0; + + ALLOCATE (machines, Machine *, NMACHINES); + memset (machines, 0, NMACHINES*sizeof(Machine *)); + + return; +} + +void FreeMachines () { + + int i; + + for (i = 0; i < NMACHINES; i++) { + FREE (machines[i]); + } + FREE (machines); + + return; +} + +// find a machine matching the given name (if not found, return NULL) +Machine *FindMachineByName (char *name) { + + int i; + + for (i = 0; i < Nmachines; i++) { + if (!strcmp (machines[i][0].name, name)) { + return machines[i]; + } + } + return NULL; +} + +// add a new machine (if not new, return TRUE) +Machine *AddMachine (char *name) { + + Machine *machine; + + machine = FindMachineByName (name); + if (machine) return machine; + + ALLOCATE (machines[Nmachines], Machine, 1); + machines[Nmachines][0].name = strcreate (name); + machines[Nmachines][0].Nhosts = 0; + machines[Nmachines][0].NjobsRealhost = 0; + machines[Nmachines][0].NjobsWanthost = 0; + + machine = machines[Nmachines]; + Nmachines ++; + + if (Nmachines >= NMACHINES - 1) { + NMACHINES += 10; + REALLOCATE (machines, Machine *, NMACHINES); + memset (&machines[Nmachines], 0, (NMACHINES - Nmachines)*sizeof(Machine *)); + } + + return machine; +} + +// delete a new machine (if not new, return TRUE) +int DelMachine (char *name) { + + return TRUE; +} + +int AddMachineHost (Host *host) { + + Machine *machine; + + // find or add machine matching this host name + machine = AddMachine (host[0].hostname); // Can this fail? + machine[0].Nhosts ++; + // XXX do we need this? host[0].machine = machine; + return (TRUE); +} + +int DelMachineHost (Host *host) { + + Machine *machine; + + // machine = host[0].machine; + machine = FindMachineByName (host[0].hostname); + if (machine == NULL) return (FALSE); + + machine[0].Nhosts --; + if (machine[0].Nhosts < 0) { + fprintf (stderr, "warning: mis-match in host count for machine %s\n", machine[0].name); + machine[0].Nhosts = 0; + } + + return (TRUE); +} + +int AddMachineJob (Host *host, Job *job) { + + Machine *machine; + + // find machine matching the real host name + machine = FindMachineByName (host[0].hostname); // Can this fail? + ASSERT (machine, "cannot find machine associated with host"); + machine[0].NjobsRealhost ++; + + // skip jobs that do not have a targeted host + if (job[0].hostname == NULL) { + return (TRUE); + } + + // do not double count jobs on the wanted host + if (!strcmp (job[0].hostname, host[0].hostname)) { + return (TRUE); + } + + // find machine matching the want host name (these are running on an unwanted host) + machine = FindMachineByName (job[0].hostname); // Can this fail? + if (machine == NULL) return (TRUE); + machine[0].NjobsWanthost ++; + + return (TRUE); +} + +int DelMachineJob (Host *host, Job *job) { + + Machine *machine; + + // find machine matching the real host name + machine = FindMachineByName (host[0].hostname); // Can this fail? + ASSERT (machine, "cannot find machine associated with host"); + machine[0].NjobsRealhost --; + + // skip jobs that do not have a targeted host + if (job[0].hostname == NULL) { + return (TRUE); + } + + // do not double count jobs on the wanted host + if (!strcmp (job[0].hostname, host[0].hostname)) { + return (TRUE); + } + + // find machine matching the want host name + machine = FindMachineByName (job[0].hostname); // Can this fail? + if (machine == NULL) return (TRUE); + machine[0].NjobsWanthost --; + + return (TRUE); +} + +int PrintMachines () { + + int i; + Machine *machine; + + gprint (GP_LOG, "Nmachines: %d\n", Nmachines); + for (i = 0; i < Nmachines; i++) { + machine = machines[i]; + gprint (GP_LOG, "%s : %d : %d : %d\n", machine[0].name, machine[0].Nhosts, machine[0].NjobsRealhost, machine[0].NjobsWanthost); + } + + return (TRUE); +} + +int CheckMachineJobs (Host *host, Job *job) { + + Machine *machine; + + machine = FindMachineByName (job[0].hostname); + if (machine == NULL) return (TRUE); + // fprintf (stderr, "wanthost: %s, Ntotal: %d, Nmax: %d\n", job[0].hostname, machine[0].NjobsWanthost + machine[0].NjobsRealhost, machine[0].Nhosts + MAX_UNWANTED_HOST_JOBS); + // if (machine[0].NjobsWanthost + machine[0].NjobsRealhost >= machine[0].Nhosts + MAX_UNWANTED_HOST_JOBS) { + if (machine[0].NjobsWanthost >= MAX_UNWANTED_HOST_JOBS) { + // fprintf (stderr, "too many outstanding jobs wanting host %s, delay job %s for now\n", job[0].hostname, job[0].argv[0]); + return (FALSE); + } + + machine = FindMachineByName (host[0].hostname); + // fprintf (stderr, "realhost: %s, Ntotal: %d, Nmax: %d\n", host[0].hostname, machine[0].NjobsWanthost + machine[0].NjobsRealhost, machine[0].Nhosts + MAX_UNWANTED_HOST_JOBS); + if (machine[0].NjobsWanthost >= MAX_UNWANTED_HOST_JOBS) { + // fprintf (stderr, "too many outstanding jobs wanting host %s, delay job %s for now\n", job[0].hostname, job[0].argv[0]); + return (FALSE); + } + return (TRUE); +} + +void SetMaxUnwantedHostJobs (int value) { + + MAX_UNWANTED_HOST_JOBS = value; + return; +} + +int GetMaxUnwantedHostJobs (void) { + + return MAX_UNWANTED_HOST_JOBS; +} + Index: /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/Makefile =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/Makefile (revision 26714) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/Makefile (revision 26715) @@ -37,4 +37,5 @@ $(SRC)/IDops.$(ARCH).o \ $(SRC)/JobOps.$(ARCH).o \ +$(SRC)/MachineOps.$(ARCH).o \ $(SRC)/StackOps.$(ARCH).o \ $(SRC)/PclientCommand.$(ARCH).o \ @@ -53,5 +54,7 @@ $(SRC)/kill.$(ARCH).o \ $(SRC)/pulse.$(ARCH).o \ +$(SRC)/parameters.$(ARCH).o \ $(SRC)/run.$(ARCH).o \ +$(SRC)/machines.$(ARCH).o \ $(SRC)/status.$(ARCH).o \ $(SRC)/stdout.$(ARCH).o \ Index: /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/host.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/host.c (revision 26714) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/host.c (revision 26715) @@ -96,6 +96,6 @@ usage: gprint (GP_LOG, "USAGE: host (command) (hostname)\n"); - gprint (GP_ERR, " valid commands: add, on, retry, check, off, delete\n"); - gprint (GP_ERR, " -threads Nthreads is optional for 'add'\n"); + gprint (GP_LOG, " valid commands: add, on, retry, check, off, delete\n"); + gprint (GP_LOG, " -threads Nthreads is optional for 'add'\n"); return (FALSE); } Index: /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/init.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/init.c (revision 26714) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/init.c (revision 26715) @@ -8,6 +8,8 @@ int jobstack PROTO((int, char **)); int kill_pc PROTO((int, char **)); +int machines PROTO((int, char **)); +int parameters PROTO((int, char **)); +int run PROTO((int, char **)); int status PROTO((int, char **)); -int run PROTO((int, char **)); int stderr_pc PROTO((int, char **)); int stdout_pc PROTO((int, char **)); @@ -19,20 +21,22 @@ static Command cmds[] = { - {1, "host", host, "add / delete / modify host"}, - {1, "hoststack", hoststack, "list hosts for a single stack"}, - {1, "status", status, "get system status"}, - {1, "stop", run, "stop controller processing"}, - {1, "run", run, "set controller runlevel"}, - {1, "verbose", verbose, "set the verbose mode for job"}, - {1, "version", version, "show version information"}, - {1, "job", job, "add job"}, - {1, "jobstack", jobstack, "list jobs for a single stack"}, - {1, "check", check, "get job or host status"}, - {1, "delete", delete, "delete job"}, - {1, "kill", kill_pc, "kill job"}, - {1, "stderr", stderr_pc, "get stderr buffer for job"}, - {1, "stdout", stdout_pc, "get stdout buffer for job"}, -# ifndef THREADED - {1, "pulse", pulse, "set system pulse"}, + {1, "check", check, "get job or host status"}, + {1, "delete", delete, "delete job"}, + {1, "host", host, "add / delete / modify host"}, + {1, "hoststack", hoststack, "list hosts for a single stack"}, + {1, "job", job, "add job"}, + {1, "jobstack", jobstack, "list jobs for a single stack"}, + {1, "kill", kill_pc, "kill job"}, + {1, "machines", machines, "list machines"}, + {1, "parameters", parameters, "get / set system parameters"}, + {1, "run", run, "set controller runlevel"}, + {1, "status", status, "get system status"}, + {1, "stderr", stderr_pc, "get stderr buffer for job"}, + {1, "stdout", stdout_pc, "get stdout buffer for job"}, + {1, "stop", run, "stop controller processing"}, + {1, "verbose", verbose, "set the verbose mode for job"}, + {1, "version", version, "show version information"}, +# ifndef THREADED + {1, "pulse", pulse, "set system pulse"}, # endif }; @@ -47,4 +51,5 @@ InitJobStacks (); InitHostStacks (); + InitMachines (); } @@ -52,3 +57,4 @@ FreeJobStacks (); FreeHostStacks (); + FreeMachines (); } Index: /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/machines.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/machines.c (revision 26715) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/machines.c (revision 26715) @@ -0,0 +1,9 @@ +# include "pcontrol.h" + +int machines (int argc, char **argv) { + + PrintMachines (); + + return (TRUE); +} + Index: /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/parameters.c =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/parameters.c (revision 26715) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/parameters.c (revision 26715) @@ -0,0 +1,57 @@ +# include "pcontrol.h" + +int parameters (int argc, char **argv) { + + int ivalue; + float value; + + if (argc < 2) goto usage; + if (argc > 4) goto usage; + if (argc == 3) goto usage; + if ((argc == 4) && strcmp(argv[2], "=")) goto usage; + + if (!strncasecmp (argv[1], "connect_time", strlen(argv[1]))) { + if (argc == 2) { + value = GetMaxConnectTime(); + gprint (GP_LOG, "Max Connect Time : %f\n", value); + return (TRUE); + } + + value = atof(argv[3]); + SetMaxConnectTime(value); + return (TRUE); + } + + if (!strncasecmp (argv[1], "wanthost_wait", strlen(argv[1]))) { + if (argc == 2) { + value = GetMaxWantHostWait(); + gprint (GP_LOG, "Max WantHost Wait : %f\n", value); + return (TRUE); + } + + value = atof(argv[3]); + SetMaxWantHostWait(value); + return (TRUE); + } + + if (!strncasecmp (argv[1], "unwanted_host_jobs", strlen(argv[1]))) { + if (argc == 2) { + ivalue = GetMaxUnwantedHostJobs(); + gprint (GP_LOG, "Max Unwanted Host Jobs : %d\n", ivalue); + return (TRUE); + } + + ivalue = atof(argv[3]); + SetMaxUnwantedHostJobs(ivalue); + return (TRUE); + } + + usage: + + gprint (GP_LOG, "USAGE: parameters (param) [= value])\n"); + gprint (GP_LOG, " valid parameters: connect_time, wanthost_wait, unwanted_host_jobs\n"); + gprint (GP_LOG, " (minimum matching word is allowed)\n"); + gprint (GP_LOG, " example: parameters connect = 2.0\n"); + return (FALSE); +} + Index: /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/test/machines.sh =================================================================== --- /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/test/machines.sh (revision 26715) +++ /branches/eam_branches/20091201/Ohana/src/opihi/pcontrol/test/machines.sh (revision 26715) @@ -0,0 +1,27 @@ + +macro load.hosts + parameters unwant = 1 + parameters want = 2.0 + + host add pikake + host add pikake + + host add ipp022 + host add ipp022 + host add ipp022 + host add ipp022 + + machines +end + +macro load.jobs + job -host pikake sleep 10 + job -host pikake sleep 10 + job -host pikake sleep 10 + job -host pikake sleep 10 + job -host pikake sleep 10 + job -host pikake sleep 10 + job -host pikake sleep 10 + job -host pikake sleep 10 +end +