~ubuntu-branches/ubuntu/precise/torque/precise-updates

« back to all changes in this revision

Viewing changes to src/resmom/linux/mom_mach.c

Committer: Bazaar Package Importer
Author(s): Dominique Belhachemi
Date: 2010-05-17 20:56:46 UTC
mto: This revision was merged to the branch mainline in revision 8.
Revision ID: james.westby@ubuntu.com-20100517205646-yjsoqs5r1s9xpnu9

Tags: upstream-2.4.8+dfsg

Import upstream version 2.4.8+dfsg

files added:
contrib/blcr

contrib/blcr/checkpoint_script

contrib/blcr/restart_script

doc/man1/qchkpt.1.in

src/cmds/qchkpt.c

src/include/assertions.h

src/include/csv.h

src/include/pbs_batchreqtype_db.h

src/include/pbs_error_db.h

src/include/pbs_job.h

src/include/utils.h

src/lib/Libcsv

src/lib/Libcsv/Makefile.am

src/lib/Libcsv/Makefile.in

src/lib/Libcsv/csv.c

src/lib/Libifl/dec_ReturnFile.c

src/lib/Libifl/enc_ReturnFile.c

src/lib/Libifl/pbsD_chkptjob.c

src/lib/Libutils

src/lib/Libutils/Makefile.am

src/lib/Libutils/Makefile.in

src/lib/Libutils/u_MXML.c

src/lib/Libutils/u_groups.c

src/lib/Libutils/u_mu.c

src/lib/Libutils/u_tree.c

src/resmom/checkpoint.c

src/server/req_holdarray.c

src/server/svr_format_job.c

files removed:
src/include/job.h

files modified:
CHANGELOG

DEVELOPMENT

INSTALL

Makefile.am

Makefile.in

README.configure

README.torque

aclocal.m4

buildutils/ltmain.sh

buildutils/pbs_mkdirs.in

buildutils/torque.spec.in

configure

configure.ac

contrib/init.d/pbs_mom

contrib/init.d/pbs_server

contrib/init.d/suse.pbs_mom

contrib/init.d/suse.pbs_sched

contrib/init.d/suse.pbs_server

doc/Makefile.am

doc/Makefile.in

doc/man1/Makefile.in

doc/man1/qdel.1.in

doc/man1/qstat.1.in

doc/man1/qsub.1.in

doc/man3/Makefile.in

doc/man7/Makefile.in

doc/man7/pbs_job_attributes.7.in

doc/man7/pbs_queue_attributes.7.in

doc/man7/pbs_server_attributes.7.in

doc/man8/Makefile.in

doc/man8/pbs_mom.8.in

doc/man8/pbs_server.8.in

doc/man8/pbsnodes.8.in

doc/man8/qrun.8.in

doc/man8/qterm.8.in

src/Makefile.in

src/cmds/MXML.c

src/cmds/Makefile.am

src/cmds/Makefile.in

src/cmds/pbs_track.c

src/cmds/pbsdsh.c

src/cmds/pbsnodes.c

src/cmds/pbspd.c

src/cmds/pbspoe.c

src/cmds/qalter.c

src/cmds/qdel.c

src/cmds/qmgr.c

src/cmds/qrerun.c

src/cmds/qsig.c

src/cmds/qstat.c

src/cmds/qstop.c

src/cmds/qsub.c

src/cmds/qterm.c

src/drmaa/Makefile.in

src/drmaa/src/Makefile.in

src/drmaa/src/lookup3.c

src/drmaa/test/Makefile.in

src/gui/Ccode/Makefile.in

src/gui/Ccode/xpbs_datadump.c

src/gui/Ccode/xpbs_scriptload.c

src/gui/Makefile.in

src/iff/Makefile.in

src/iff/iff2.c

src/include/Long.h

src/include/Long_.h

src/include/Makefile.am

src/include/Makefile.in

src/include/acct.h

src/include/array.h

src/include/attribute.h

src/include/batch_request.h

src/include/cmds.h

src/include/dis.h

src/include/dis_init.h

src/include/libpbs.h

src/include/list_link.h

src/include/log.h

src/include/md5.h

src/include/mom_func.h

src/include/net_connect.h

src/include/pbs_assert.h

src/include/pbs_config.h.in

src/include/pbs_cpa.h

src/include/pbs_cpuset.h

src/include/pbs_error.h

src/include/pbs_ifl.h

src/include/pbs_nodes.h

src/include/portability.h

src/include/qmgr.h

src/include/qmgr_node_public.h

src/include/qmgr_node_readonly.h

src/include/qmgr_que_public.h

src/include/qmgr_que_readonly.h

src/include/qmgr_svr_public.h

src/include/qmgr_svr_readonly.h

src/include/queue.h

src/include/resmon.h

src/include/resource.h

src/include/rpp.h

src/include/sched_cmds.h

src/include/server.h

src/include/server_limits.h

src/include/site_job_attr_enum.ht

src/include/svrfunc.h

src/include/tm.h

src/include/tm_.h

src/include/tracking.h

src/include/work_task.h

src/lib/Libattr/LTostr.c

src/lib/Libattr/Makefile.in

src/lib/Libattr/attr_atomic.c

src/lib/Libattr/attr_fn_arst.c

src/lib/Libattr/attr_fn_b.c

src/lib/Libattr/attr_fn_c.c

src/lib/Libattr/attr_fn_hold.c

src/lib/Libattr/attr_fn_intr.c

src/lib/Libattr/attr_fn_l.c

src/lib/Libattr/attr_fn_ll.c

src/lib/Libattr/attr_fn_resc.c

src/lib/Libattr/attr_fn_str.c

src/lib/Libattr/attr_fn_unkn.c

src/lib/Libattr/attr_func.c

src/lib/Libattr/attr_node_func.c

src/lib/Libcmds/Makefile.in

src/lib/Libcmds/cnt2server.c

src/lib/Libcmds/cvtdate.c

src/lib/Libcmds/get_server.c

src/lib/Libcmds/parse_at.c

src/lib/Libcmds/parse_depend.c

src/lib/Libcmds/parse_stage.c

src/lib/Libcmds/prepare_path.c

src/lib/Libdis/Makefile.in

src/lib/Libdis/dis.c

src/lib/Libdis/dis_.h

src/lib/Libdis/disiui_.c

src/lib/Libdis/disrf.c

src/lib/Libdis/disrl_.c

src/lib/Libdis/disrsi_.c

src/lib/Libdis/disrsl_.c

src/lib/Libdis/disrui.c

src/lib/Libdis/diswf.c

src/lib/Libdis/diswl_.c

src/lib/Libdis/diswsi.c

src/lib/Libdis/diswsl.c

src/lib/Libdis/diswui_.c

src/lib/Libdis/diswul.c

src/lib/Libifl/Makefile.in

src/lib/Libifl/PBSD_msg2.c

src/lib/Libifl/PBSD_sig2.c

src/lib/Libifl/PBS_attr.c

src/lib/Libifl/dec_Authen.c

src/lib/Libifl/dec_CpyFil.c

src/lib/Libifl/dec_JobCred.c

src/lib/Libifl/dec_JobFile.c

src/lib/Libifl/dec_JobId.c

src/lib/Libifl/dec_JobObit.c

src/lib/Libifl/dec_Manage.c

src/lib/Libifl/dec_MoveJob.c

src/lib/Libifl/dec_MsgJob.c

src/lib/Libifl/dec_QueueJob.c

src/lib/Libifl/dec_Reg.c

src/lib/Libifl/dec_ReqExt.c

src/lib/Libifl/dec_ReqHdr.c

src/lib/Libifl/dec_Resc.c

src/lib/Libifl/dec_RunJob.c

src/lib/Libifl/dec_Shut.c

src/lib/Libifl/dec_Sig.c

src/lib/Libifl/dec_Status.c

src/lib/Libifl/dec_Track.c

src/lib/Libifl/dec_attrl.c

src/lib/Libifl/dec_attropl.c

src/lib/Libifl/dec_rpyc.c

src/lib/Libifl/dec_rpys.c

src/lib/Libifl/dec_svrattrl.c

src/lib/Libifl/enc_CpyFil.c

src/lib/Libifl/enc_JobFile.c

src/lib/Libifl/enc_JobId.c

src/lib/Libifl/enc_Manage.c

src/lib/Libifl/enc_MsgJob.c

src/lib/Libifl/enc_QueueJob.c

src/lib/Libifl/enc_Reg.c

src/lib/Libifl/enc_ReqExt.c

src/lib/Libifl/enc_ReqHdr.c

src/lib/Libifl/enc_RunJob.c

src/lib/Libifl/enc_Shut.c

src/lib/Libifl/enc_Sig.c

src/lib/Libifl/enc_Status.c

src/lib/Libifl/enc_Track.c

src/lib/Libifl/enc_attrl.c

src/lib/Libifl/enc_attropl.c

src/lib/Libifl/enc_reply.c

src/lib/Libifl/enc_svrattrl.c

src/lib/Libifl/list_link.c

src/lib/Libifl/pbsD_alterjo.c

src/lib/Libifl/pbsD_connect.c

src/lib/Libifl/pbsD_deljob.c

src/lib/Libifl/pbsD_holdjob.c

src/lib/Libifl/pbsD_locjob.c

src/lib/Libifl/pbsD_movejob.c

src/lib/Libifl/pbsD_msgjob.c

src/lib/Libifl/pbsD_orderjo.c

src/lib/Libifl/pbsD_rerunjo.c

src/lib/Libifl/pbsD_resc.c

src/lib/Libifl/pbsD_rlsjob.c

src/lib/Libifl/pbsD_runjob.c

src/lib/Libifl/pbsD_sigjob.c

src/lib/Libifl/pbsD_stagein.c

src/lib/Libifl/rpp.c

src/lib/Libifl/tcp_dis.c

src/lib/Libifl/tm.c

src/lib/Liblog/Makefile.in

src/lib/Liblog/pbs_log.c

src/lib/Liblog/pbs_messages.c

src/lib/Libnet/Makefile.in

src/lib/Libnet/md5.c

src/lib/Libnet/net_client.c

src/lib/Libnet/net_server.c

src/lib/Libnet/rm.c

src/lib/Libpbs/Makefile.am

src/lib/Libpbs/Makefile.in

src/lib/Libsite/Makefile.in

src/lib/Libsite/site_alt_rte.c

src/lib/Libsite/site_check_u.c

src/lib/Libsite/site_mom_chu.c

src/lib/Libsite/site_mom_ckp.c

src/lib/Libsite/site_mom_jst.c

src/lib/Makefile.am

src/lib/Makefile.in

src/mom_rcp/Makefile.in

src/mom_rcp/util.c

src/momctl/Makefile.in

src/momctl/momctl.c

src/pam/Makefile.in

src/pam/pam_pbssimpleauth.c

src/resmom/Makefile.am

src/resmom/Makefile.in

src/resmom/aix4/Makefile.in

src/resmom/aix4/mom_mach.c

src/resmom/aix4/mom_mach.h

src/resmom/aix4/mom_start.c

src/resmom/aix4/pe_input.c

src/resmom/aix5/Makefile.in

src/resmom/aix5/mom_mach.c

src/resmom/aix5/mom_mach.h

src/resmom/aix5/mom_start.c

src/resmom/aix5/pe_input.c

src/resmom/catch_child.c

src/resmom/cray_cpa.c

src/resmom/darwin/Makefile.in

src/resmom/darwin/mom_mach.c

src/resmom/darwin/mom_mach.h

src/resmom/darwin/mom_start.c

src/resmom/digitalunix/Makefile.in

src/resmom/digitalunix/mom_mach.c

src/resmom/digitalunix/mom_mach.h

src/resmom/digitalunix/mom_start.c

src/resmom/digitalunix/pe_input.c

src/resmom/freebsd/Makefile.in

src/resmom/freebsd/mom_mach.c

src/resmom/freebsd/mom_mach.h

src/resmom/freebsd/mom_start.c

src/resmom/freebsd/pe_input.c

src/resmom/freebsd5/Makefile.in

src/resmom/freebsd5/mom_mach.c

src/resmom/freebsd5/mom_mach.h

src/resmom/freebsd5/mom_start.c

src/resmom/fujitsu/Makefile.in

src/resmom/fujitsu/mom_mach.c

src/resmom/fujitsu/mom_mach.h

src/resmom/fujitsu/mom_start.c

src/resmom/fujitsu/pe_input.c

src/resmom/hpux10/Makefile.in

src/resmom/hpux10/mom_mach.c

src/resmom/hpux10/mom_mach.h

src/resmom/hpux10/mom_start.c

src/resmom/hpux10/pe_input.c

src/resmom/hpux11/Makefile.in

src/resmom/hpux11/mom_mach.c

src/resmom/hpux11/mom_mach.h

src/resmom/hpux11/mom_start.c

src/resmom/hpux11/pe_input.c

src/resmom/irix5/Makefile.in

src/resmom/irix5/mom_mach.c

src/resmom/irix5/mom_mach.h

src/resmom/irix5/mom_start.c

src/resmom/irix5/pe_input.c

src/resmom/irix6/Makefile.in

src/resmom/irix6/mom_mach.c

src/resmom/irix6/mom_mach.h

src/resmom/irix6/mom_start.c

src/resmom/irix6/pe_input.c

src/resmom/irix6array/Makefile.in

src/resmom/irix6array/mom_mach.c

src/resmom/irix6array/mom_mach.h

src/resmom/irix6array/mom_start.c

src/resmom/irix6array/pe_input.c

src/resmom/job_func.c

src/resmom/linux/Makefile.in

src/resmom/linux/cpuset.c

src/resmom/linux/mom_mach.c

src/resmom/linux/mom_mach.h

src/resmom/linux/mom_start.c

src/resmom/mom_comm.c

src/resmom/mom_inter.c

src/resmom/mom_main.c

src/resmom/mom_server.c

src/resmom/netbsd/Makefile.in

src/resmom/netbsd/mom_mach.c

src/resmom/netbsd/mom_mach.h

src/resmom/netbsd/mom_start.c

src/resmom/netbsd/pe_input.c

src/resmom/pbs_demux.c

src/resmom/prolog.c

src/resmom/req_quejob.c

src/resmom/requests.c

src/resmom/solaris5/Makefile.in

src/resmom/solaris5/mom_mach.c

src/resmom/solaris5/mom_mach.h

src/resmom/solaris5/mom_start.c

src/resmom/solaris5/pe_input.c

src/resmom/solaris7/Makefile.in

src/resmom/solaris7/mom_mach.c

src/resmom/solaris7/mom_mach.h

src/resmom/solaris7/mom_start.c

src/resmom/solaris7/pe_input.c

src/resmom/start_exec.c

src/resmom/sunos4/Makefile.in

src/resmom/sunos4/mom_mach.c

src/resmom/sunos4/mom_mach.h

src/resmom/sunos4/mom_start.c

src/resmom/sunos4/pe_input.c

src/resmom/tmsock_recov.c

src/resmom/unicos8/Makefile.in

src/resmom/unicos8/mom_mach.c

src/resmom/unicos8/mom_mach.h

src/resmom/unicos8/mom_start.c

src/resmom/unicos8/pe_input.c

src/resmom/unicosmk2/Makefile.in

src/resmom/unicosmk2/mom_mach.c

src/resmom/unicosmk2/mom_mach.h

src/resmom/unicosmk2/mom_start.c

src/resmom/unicosmk2/pe_input.c

src/scheduler.basl/Basl2c.c

src/scheduler.basl/CodeGen.c

src/scheduler.basl/List.c

src/scheduler.basl/Makefile.in

src/scheduler.basl/Node.c

src/scheduler.basl/Semantic.c

src/scheduler.basl/SymTab.c

src/scheduler.basl/af.c

src/scheduler.basl/af_cnode.c

src/scheduler.basl/af_cnodemap.c

src/scheduler.basl/af_config.c

src/scheduler.basl/af_job.c

src/scheduler.basl/af_que.c

src/scheduler.basl/af_server.c

src/scheduler.basl/af_system.c

src/scheduler.cc/Makefile.in

src/scheduler.cc/pbs_sched.c

src/scheduler.cc/samples/Makefile.in

src/scheduler.cc/samples/cray_t3e/Makefile.in

src/scheduler.cc/samples/cray_t3e/allocations.c

src/scheduler.cc/samples/cray_t3e/bitfield.h

src/scheduler.cc/samples/cray_t3e/byte2val.c

src/scheduler.cc/samples/cray_t3e/cleanup.c

src/scheduler.cc/samples/cray_t3e/comment.c

src/scheduler.cc/samples/cray_t3e/dedicated.c

src/scheduler.cc/samples/cray_t3e/dyn_nodemask.c

src/scheduler.cc/samples/cray_t3e/evaluate_system.c

src/scheduler.cc/samples/cray_t3e/file_chgs.c

src/scheduler.cc/samples/cray_t3e/fragments.c

src/scheduler.cc/samples/cray_t3e/gblxvars.h

src/scheduler.cc/samples/cray_t3e/getat.c

src/scheduler.cc/samples/cray_t3e/getconfig.c

src/scheduler.cc/samples/cray_t3e/getjobs.c

src/scheduler.cc/samples/cray_t3e/getqueues.c

src/scheduler.cc/samples/cray_t3e/getrsrcs.c

src/scheduler.cc/samples/cray_t3e/how_many.c

src/scheduler.cc/samples/cray_t3e/jobinfo.c

src/scheduler.cc/samples/cray_t3e/misc.c

src/scheduler.cc/samples/cray_t3e/movejob.c

src/scheduler.cc/samples/cray_t3e/msgs.c

src/scheduler.cc/samples/cray_t3e/msgs.h

src/scheduler.cc/samples/cray_t3e/overlaps.c

src/scheduler.cc/samples/cray_t3e/pack_queues.c

src/scheduler.cc/samples/cray_t3e/pnp.c

src/scheduler.cc/samples/cray_t3e/queue_limits.c

src/scheduler.cc/samples/cray_t3e/rejectjob.c

src/scheduler.cc/samples/cray_t3e/resource_limits.c

src/scheduler.cc/samples/cray_t3e/runjob.c

src/scheduler.cc/samples/cray_t3e/schedinit.c

src/scheduler.cc/samples/cray_t3e/schedule.c

src/scheduler.cc/samples/cray_t3e/sec2val.c

src/scheduler.cc/samples/cray_t3e/time_limits.c

src/scheduler.cc/samples/cray_t3e/toolkit.h

src/scheduler.cc/samples/cray_t3e/user_limits.c

src/scheduler.cc/samples/cray_t3e/usersort.c

src/scheduler.cc/samples/cray_t3e/val2bool.c

src/scheduler.cc/samples/cray_t3e/val2byte.c

src/scheduler.cc/samples/cray_t3e/val2sec.c

src/scheduler.cc/samples/dec_cluster/Makefile.in

src/scheduler.cc/samples/dec_cluster/cleanup.c

src/scheduler.cc/samples/dec_cluster/comment.c

src/scheduler.cc/samples/dec_cluster/evaluate_system.c

src/scheduler.cc/samples/dec_cluster/fair_access.c

src/scheduler.cc/samples/dec_cluster/gblxvars.h

src/scheduler.cc/samples/dec_cluster/getconfig.c

src/scheduler.cc/samples/dec_cluster/getjobs.c

src/scheduler.cc/samples/dec_cluster/getqueues.c

src/scheduler.cc/samples/dec_cluster/getrsrcs.c

src/scheduler.cc/samples/dec_cluster/jobinfo.c

src/scheduler.cc/samples/dec_cluster/msgs.c

src/scheduler.cc/samples/dec_cluster/msgs.h

src/scheduler.cc/samples/dec_cluster/pack_queues.c

src/scheduler.cc/samples/dec_cluster/queue_limits.c

src/scheduler.cc/samples/dec_cluster/resource_limits.c

src/scheduler.cc/samples/dec_cluster/runjob.c

src/scheduler.cc/samples/dec_cluster/schedinit.c

src/scheduler.cc/samples/dec_cluster/schedule.c

src/scheduler.cc/samples/dec_cluster/time_limits.c

src/scheduler.cc/samples/dec_cluster/toolkit.c

src/scheduler.cc/samples/dec_cluster/toolkit.h

src/scheduler.cc/samples/dec_cluster/user_limits.c

src/scheduler.cc/samples/dec_cluster/usersort.c

src/scheduler.cc/samples/fifo/Makefile.in

src/scheduler.cc/samples/fifo/check.c

src/scheduler.cc/samples/fifo/check.h

src/scheduler.cc/samples/fifo/config.h

src/scheduler.cc/samples/fifo/constant.h

src/scheduler.cc/samples/fifo/data_types.h

src/scheduler.cc/samples/fifo/fairshare.c

src/scheduler.cc/samples/fifo/fifo.c

src/scheduler.cc/samples/fifo/parse.c

src/scheduler.cc/samples/fifo/prime.c

src/scheduler.cc/samples/fifo/queue_info.c

src/scheduler.cc/samples/fifo/sched_config

src/scheduler.cc/samples/fifo/token_accounting.c

src/scheduler.cc/samples/msic_cluster/Makefile.in

src/scheduler.cc/samples/msic_cluster/acl_support.c

src/scheduler.cc/samples/msic_cluster/cleanup.c

src/scheduler.cc/samples/msic_cluster/comment.c

src/scheduler.cc/samples/msic_cluster/evaluate_system.c

src/scheduler.cc/samples/msic_cluster/fair_access.c

src/scheduler.cc/samples/msic_cluster/gblxvars.h

src/scheduler.cc/samples/msic_cluster/getconfig.c

src/scheduler.cc/samples/msic_cluster/getjobs.c

src/scheduler.cc/samples/msic_cluster/getqueues.c

src/scheduler.cc/samples/msic_cluster/getrsrcs.c

src/scheduler.cc/samples/msic_cluster/jobinfo.c

src/scheduler.cc/samples/msic_cluster/msgs.c

src/scheduler.cc/samples/msic_cluster/msgs.h

src/scheduler.cc/samples/msic_cluster/pack_queues.c

src/scheduler.cc/samples/msic_cluster/queue_limits.c

src/scheduler.cc/samples/msic_cluster/resource_limits.c

src/scheduler.cc/samples/msic_cluster/runjob.c

src/scheduler.cc/samples/msic_cluster/schedinit.c

src/scheduler.cc/samples/msic_cluster/schedule.c

src/scheduler.cc/samples/msic_cluster/time_limits.c

src/scheduler.cc/samples/msic_cluster/toolkit.h

src/scheduler.cc/samples/msic_cluster/user_limits.c

src/scheduler.cc/samples/msic_cluster/usersort.c

src/scheduler.cc/samples/sgi_origin/Makefile.in

src/scheduler.cc/samples/sgi_origin/allocations.c

src/scheduler.cc/samples/sgi_origin/bitfield.h

src/scheduler.cc/samples/sgi_origin/byte2val.c

src/scheduler.cc/samples/sgi_origin/cleanup.c

src/scheduler.cc/samples/sgi_origin/comment.c

src/scheduler.cc/samples/sgi_origin/dedicated.c

src/scheduler.cc/samples/sgi_origin/dyn_nodemask.c

src/scheduler.cc/samples/sgi_origin/evaluate_system.c

src/scheduler.cc/samples/sgi_origin/file_chgs.c

src/scheduler.cc/samples/sgi_origin/fragments.c

src/scheduler.cc/samples/sgi_origin/gblxvars.h

src/scheduler.cc/samples/sgi_origin/getat.c

src/scheduler.cc/samples/sgi_origin/getconfig.c

src/scheduler.cc/samples/sgi_origin/getjobs.c

src/scheduler.cc/samples/sgi_origin/getqueues.c

src/scheduler.cc/samples/sgi_origin/getrsrcs.c

src/scheduler.cc/samples/sgi_origin/how_many.c

src/scheduler.cc/samples/sgi_origin/hpm.c

src/scheduler.cc/samples/sgi_origin/hpm.h

src/scheduler.cc/samples/sgi_origin/jobinfo.c

src/scheduler.cc/samples/sgi_origin/misc.c

src/scheduler.cc/samples/sgi_origin/movejob.c

src/scheduler.cc/samples/sgi_origin/msgs.c

src/scheduler.cc/samples/sgi_origin/msgs.h

src/scheduler.cc/samples/sgi_origin/overlaps.c

src/scheduler.cc/samples/sgi_origin/pack_queues.c

src/scheduler.cc/samples/sgi_origin/pnp.c

src/scheduler.cc/samples/sgi_origin/queue_limits.c

src/scheduler.cc/samples/sgi_origin/rejectjob.c

src/scheduler.cc/samples/sgi_origin/resource_limits.c

src/scheduler.cc/samples/sgi_origin/runjob.c

src/scheduler.cc/samples/sgi_origin/schedinit.c

src/scheduler.cc/samples/sgi_origin/schedule.c

src/scheduler.cc/samples/sgi_origin/sec2val.c

src/scheduler.cc/samples/sgi_origin/time_limits.c

src/scheduler.cc/samples/sgi_origin/toolkit.h

src/scheduler.cc/samples/sgi_origin/user_limits.c

src/scheduler.cc/samples/sgi_origin/usersort.c

src/scheduler.cc/samples/sgi_origin/val2bool.c

src/scheduler.cc/samples/sgi_origin/val2byte.c

src/scheduler.cc/samples/sgi_origin/val2sec.c

src/scheduler.cc/samples/umn_cluster/Makefile.in

src/scheduler.cc/samples/umn_cluster/acl_support.c

src/scheduler.cc/samples/umn_cluster/cleanup.c

src/scheduler.cc/samples/umn_cluster/comment.c

src/scheduler.cc/samples/umn_cluster/evaluate_system.c

src/scheduler.cc/samples/umn_cluster/fair_access.c

src/scheduler.cc/samples/umn_cluster/gblxvars.h

src/scheduler.cc/samples/umn_cluster/getconfig.c

src/scheduler.cc/samples/umn_cluster/getjobs.c

src/scheduler.cc/samples/umn_cluster/getqueues.c

src/scheduler.cc/samples/umn_cluster/getrsrcs.c

src/scheduler.cc/samples/umn_cluster/jobinfo.c

src/scheduler.cc/samples/umn_cluster/msgs.c

src/scheduler.cc/samples/umn_cluster/msgs.h

src/scheduler.cc/samples/umn_cluster/pack_queues.c

src/scheduler.cc/samples/umn_cluster/qschedule.c

src/scheduler.cc/samples/umn_cluster/queue_limits.c

src/scheduler.cc/samples/umn_cluster/resource_limits.c

src/scheduler.cc/samples/umn_cluster/runjob.c

src/scheduler.cc/samples/umn_cluster/schedinit.c

src/scheduler.cc/samples/umn_cluster/schedule.c

src/scheduler.cc/samples/umn_cluster/time_limits.c

src/scheduler.cc/samples/umn_cluster/toolkit.c

src/scheduler.cc/samples/umn_cluster/toolkit.h

src/scheduler.cc/samples/umn_cluster/user_limits.c

src/scheduler.cc/samples/umn_cluster/usersort.c

src/scheduler.tcl/Makefile.in

src/scheduler.tcl/site_tclWrap.c

src/server/Makefile.am

src/server/Makefile.in

src/server/accounting.c

src/server/array_func.c

src/server/attr_recov.c

src/server/dis_read.c

src/server/geteusernam.c

src/server/issue_request.c

src/server/job_attr_def.c

src/server/job_func.c

src/server/job_qs_upgrade.c

src/server/job_recov.c

src/server/job_route.c

src/server/node_func.c

src/server/node_manager.c

src/server/pbsd_init.c

src/server/pbsd_main.c

src/server/process_request.c

src/server/queue_attr_def.c

src/server/queue_recov.c

src/server/reply_send.c

src/server/req_delete.c

src/server/req_deletearray.c

src/server/req_holdjob.c

src/server/req_jobobit.c

src/server/req_locate.c

src/server/req_manager.c

src/server/req_message.c

src/server/req_modify.c

src/server/req_movejob.c

src/server/req_quejob.c

src/server/req_register.c

src/server/req_rerun.c

src/server/req_runjob.c

src/server/req_select.c

src/server/req_shutdown.c

src/server/req_signal.c

src/server/req_stat.c

src/server/req_tokens.c

src/server/req_track.c

src/server/resc_def_all.c

src/server/run_sched.c

src/server/stat_job.c

src/server/svr_attr_def.c

src/server/svr_chk_owner.c

src/server/svr_connect.c

src/server/svr_func.c

src/server/svr_jobfunc.c

src/server/svr_mail.c

src/server/svr_movejob.c

src/server/svr_recov.c

src/server/svr_resccost.c

src/server/svr_task.c

src/tools/Makefile.in

src/tools/hostn.c

src/tools/printjob.c

src/tools/tracejob.c

src/tools/tracejob.h

src/tools/xpbsmon/Makefile.in

torque.setup

torque.spec

Show diffs side-by-side

added added

removed removed

src/resmom/linux/mom_mach.c

* are permitted provided that all of the following conditions are met.

* After December 31, 2001, only conditions 3-6 must be met:

* 1. Commercial and/or non-commercial use of the Software is permitted

* provided a current software registration is on file at www.OpenPBS.org.

* If use of this software contributes to a publication, product, or

* service, proper attribution must be given; see www.OpenPBS.org/credit.html

* 2. Redistribution in any form is only permitted for non-commercial,

* non-profit purposes. There can be no charge for the Software or any

* software incorporating the Software. Further, there can be no

* expectation of revenue generated as a consequence of redistributing

* the Software.

* 3. Any Redistribution of source code must retain the above copyright notice

* and the acknowledgment contained in paragraph 6, this list of conditions

* and the disclaimer contained in paragraph 7.

105

#include <syscall.h>

106

#include <ctype.h>

107

#include <string.h>

#include <csv.h>

108

109

#ifdef Q_6_5_QUOTAON

110

100

/* remap dqblk for SUSE 9.0 */

124

114

#include "server_limits.h"

125

115

#include "attribute.h"

126

116

#include "resource.h"

127

#include "job.h"

117

#include "pbs_job.h"

128

118

#include "log.h"

129

119

#include "mom_mach.h"

130

120

#include "mom_func.h"

169

159

extern char extra_parm[];

170

160

extern char no_parm[];

171

161

172

extern struct pbs_err_to_txt pbs_err_to_txt[];

173

162

extern time_t time_now;

174

163

175

164

extern int LOGLEVEL;

176

extern char checkpoint_script_name[1024];

177

extern char restart_script_name[1024];

178

165

extern char PBSNodeMsgBuf[1024];

179

166

180

167

#define TBL_INC 200 /* initial proc table */

186

173

187

174

** external functions and data

188

175

189

190

extern struct config *search A_((struct config *, char *));

191

192

extern struct rm_attribute *momgetattr A_((char *));

176

extern struct config *search A_((struct config *,char *));

177

extern struct rm_attribute *momgetattr A_((char *));

193

178

extern int rm_errno;

194

extern double cputfactor;

195

extern double wallfactor;

196

extern long system_ncpus;

197

extern int ignwalltime;

198

extern int ignvmem;

179

extern double cputfactor;

180

extern double wallfactor;

181

extern long system_ncpus;

182

extern int ignwalltime;

183

extern int igncput;

184

extern int ignvmem;

185

extern int ignmem;

199

186

200

187

201

188

** local functions and data

202

189

203

static char *resi A_((struct rm_attribute *));

204

static char *totmem A_((struct rm_attribute *));

190

static char *resi A_((struct rm_attribute *));

191

static char *totmem A_((struct rm_attribute *));

205

192

static char *availmem A_((struct rm_attribute *));

206

static char *physmem A_((struct rm_attribute *));

207

static char *ncpus A_((struct rm_attribute *));

193

static char *physmem A_((struct rm_attribute *));

194

static char *ncpus A_((struct rm_attribute *));

208

195

static char *walltime A_((struct rm_attribute *));

209

static char *quota A_((struct rm_attribute *));

210

static char *netload A_((struct rm_attribute *));

196

static char *quota A_((struct rm_attribute *));

197

static char *netload A_((struct rm_attribute *));

198

199

#ifndef mbool_t

200

#define mbool_t char

201

#endif /* mbool_t */

202

203

mbool_t ProcIsChild(char *,char *,char *);

211

204

212

205

extern char *loadave A_((struct rm_attribute *));

213

206

extern char *nullproc A_((struct rm_attribute *));

389

382

390

383

lastbracket++;

391

384

392

if (sscanf(readbuf, "%d (%[^\n]", &ps.pid, path) != 2)

385

if (sscanf(readbuf,"%d (%[^\n]",&ps.pid,path) != 2)

393

386

{

394

387

/* FAILURE */

395

388

399

392

}

400

393

401

394

/* see stat_str[] value for mapping 'stat' format */

402

if (sscanf(lastbracket, stat_str,

403

&ps.state, /* state (one of RSDZTW) */

404

&ps.ppid, /* ppid */

405

&ps.pgrp, /* pgrp */

406

&ps.session, /* session id */

407

&ps.flags, /* flags - kernel flags of the process, see the PF_* in <linux/sched.h> */

408

&ps.utime, /* utime - jiffies that this process has been scheduled in user mode */

409

&ps.stime, /* stime - jiffies that this process has been scheduled in kernel mode */

410

&ps.cutime, /* cutime - jiffies that this process’s waited-for children have been scheduled in user mode */

411

&ps.cstime, /* cstime - jiffies that this process’s waited-for children have been scheduled in kernel mode */

412

&jstarttime, /* starttime */

413

&ps.vsize, /* vsize */

414

&ps.rss) != 12) /* rss */

395

396

if (sscanf(lastbracket,stat_str,

397

&ps.state, /* state (one of RSDZTW) */

398

&ps.ppid, /* ppid */

399

&ps.pgrp, /* pgrp */

400

&ps.session, /* session id */

401

&ps.flags, /* flags - kernel flags of the process, see the PF_* in <linux/sched.h> */

402

&ps.utime, /* utime - jiffies that this process has been scheduled in user mode */

403

&ps.stime, /* stime - jiffies that this process has been scheduled in kernel mode */

404

&ps.cutime, /* cutime - jiffies that this process’s waited-for children have been scheduled in user mode */

405

&ps.cstime, /* cstime - jiffies that this process’s waited-for children have been scheduled in kernel mode */

406

&jstarttime, /* starttime */

407

&ps.vsize, /* vsize */

408

&ps.rss) != 12) /* rss */

415

409

{

416

410

/* FAILURE */

417

411

449

443

450

444

451

445

452

proc_mem_t *

453

get_proc_mem(void)

446

447

proc_mem_t *get_proc_mem(void)

454

448

455

449

{

456

450

static proc_mem_t mm;

458

452

char str[32];

459

453

unsigned long long bfsz, casz;

460

454

461

if ((fp = fopen("/proc/meminfo", "r")) == NULL)

462

{

463

return(NULL);

464

}

465

466

if (fscanf(fp, "%30s", str) != 1)

467

{

468

return(NULL);

469

}

470

471

if (!strncmp(str, "total:", sizeof(str)))

455

if ((fp = fopen("/proc/meminfo","r")) == NULL)

456

{

457

return(NULL);

458

}

459

460

if (fscanf(fp,"%30s",str) != 1)

461

{

462

return(NULL);

463

}

464

465

if (!strncmp(str,"total:",sizeof(str)))

472

466

{

473

467

/* old format */

474

468

475

if (fscanf(fp, "%*[^\n]%*c") != 0) /* remove text header */

469

if (fscanf(fp,"%*[^\n]%*c") != 0) /* remove text header */

476

470

{

477

471

return(NULL);

478

472

}

628

622

629

623

630

624

631

void

632

dep_initialize(void)

625

void dep_initialize(void)

633

626

634

627

{

635

628

char *id = "dep_initialize";

981

974

982

975

983

976

977

984

978

985

979

* Internal session workingset size function.

986

980

1078

1072

int value)

1079

1073

1080

1074

{

1081

int i = 0;

1082

1075

char *message;

1083

1076

1084

1077

assert(string != NULL);

1085

1078

assert(*string != '\0');

1086

assert(value > PBSE_); /* minimum PBS error number */

1087

assert(value <= PBSE_NOSYNCMSTR); /* maximum PBS error number */

1088

assert(pbs_err_to_txt[i].err_no != 0);

1089

1090

1091

{

1092

if (pbs_err_to_txt[i].err_no == value)

1093

break;

1094

}

1095

while (pbs_err_to_txt[++i].err_no != 0);

1096

1097

assert(pbs_err_to_txt[i].err_txt != NULL);

1098

1099

message = *pbs_err_to_txt[i].err_txt;

1079

1080

message = pbse_to_txt(value);

1100

1081

1101

1082

assert(message != NULL);

1102

1103

1083

assert(*message != '\0');

1104

1084

1105

1085

fprintf(stderr, msg_momsetlim, string, message);

1204

1184

1205

1185

if (!strcmp(pname, "cput"))

1206

1186

{

1207

/* cpu time - check, if less than pcput use it */

1208

1209

retval = gettime(pres, &value);

1210

1211

if (retval != PBSE_NONE)

1187

if (igncput == FALSE)

1212

1188

{

1213

sprintf(log_buffer, "cput gettime failed in %s",

1214

id);

1215

1216

return(error(pname, retval));

1189

/* cpu time - check, if less than pcput use it */

1190

1191

retval = gettime(pres, &value);

1192

1193

if (retval != PBSE_NONE)

1194

{

1195

sprintf(log_buffer, "cput gettime failed in %s",id);

1196

1197

return(error(pname, retval));

1198

}

1217

1199

}

1218

1200

}

1219

1201

else if (!strcmp(pname, "pcput"))

1220

1202

{

1221

if (set_mode == SET_LIMIT_SET)

1203

if (igncput == FALSE)

1222

1204

{

1223

/* process cpu time - set */

1224

1225

retval = gettime(pres, &value);

1226

1227

if (retval != PBSE_NONE)

1228

{

1229

sprintf(log_buffer, "pcput gettime failed in %s",

1230

id);

1231

1232

return(error(pname, retval));

1233

}

1234

1235

reslim.rlim_cur = reslim.rlim_max = (unsigned long)((double)value / cputfactor);

1236

1237

if (LOGLEVEL >= 2)

1238

{

1239

sprintf(log_buffer, "setting cpu time limit to %ld for job %s",

1240

(long int)reslim.rlim_cur,

1241

pjob->ji_qs.ji_jobid);

1242

1243

log_record(PBSEVENT_SYSTEM, 0, id, log_buffer);

1244

1245

log_buffer[0] = '\0';

1246

}

1247

1248

/* NOTE: some versions of linux have a bug which causes the parent

1249

process to receive a SIGKILL if the child's cpu limit is exceeded */

1250

1251

if (setrlimit(RLIMIT_CPU, &reslim) < 0)

1252

{

1253

sprintf(log_buffer, "setrlimit for RLIMIT_CPU failed in %s, errno=%d (%s)",

1254

id,

1255

errno, strerror(errno));

1256

1257

return(error("RLIMIT_CPU", PBSE_SYSTEM));

1258

}

1259

} /* END if (set_mode == SET_LIMIT_SET) */

1205

if (set_mode == SET_LIMIT_SET)

1206

{

1207

/* process cpu time - set */

1208

1209

retval = gettime(pres, &value);

1210

1211

if (retval != PBSE_NONE)

1212

{

1213

sprintf(log_buffer, "pcput gettime failed in %s",id);

1214

1215

return(error(pname, retval));

1216

}

1217

1218

reslim.rlim_cur = reslim.rlim_max =

1219

(unsigned long)((double)value / cputfactor);

1220

1221

if (LOGLEVEL >= 2)

1222

{

1223

sprintf(log_buffer, "setting cpu time limit to %ld for job %s",

1224

(long int)reslim.rlim_cur,

1225

pjob->ji_qs.ji_jobid);

1226

1227

log_record(PBSEVENT_SYSTEM, 0, id, log_buffer);

1228

1229

log_buffer[0] = '\0';

1230

}

1231

1232

/* NOTE: some versions of linux have a bug which causes the parent

1233

process to receive a SIGKILL if the child's cpu limit is exceeded */

1234

1235

if (setrlimit(RLIMIT_CPU, &reslim) < 0)

1236

{

1237

sprintf(log_buffer, "setrlimit for RLIMIT_CPU failed in %s, errno=%d (%s)",

1238

id,

1239

errno, strerror(errno));

1240

1241

return(error("RLIMIT_CPU", PBSE_SYSTEM));

1242

}

1243

} /* END if (set_mode == SET_LIMIT_SET) */

1244

}

1260

1245

}

1261

1246

else if (!strcmp(pname, "file"))

1262

1247

{

1308

1293

}

1309

1294

else if (!strcmp(pname, "vmem"))

1310

1295

{

1311

/* check */

1312

1313

retval = getsize(pres, &value);

1314

1315

if (retval != PBSE_NONE)

1316

{

1317

sprintf(log_buffer, "getsize() failed for vmem in %s",

1318

id);

1319

1320

return(error(pname, retval));

1321

}

1322

1323

if ((vmem_limit == 0) || (value < vmem_limit))

1324

vmem_limit = value;

1325

}

1326

else if (!strcmp(pname, "pvmem"))

1327

{

1328

/* set */

1329

1330

if (set_mode == SET_LIMIT_SET)

1331

{

1296

if (ignvmem == FALSE)

1297

{

1298

/* check */

1299

1332

1300

retval = getsize(pres, &value);

1333

1301

1334

1302

if (retval != PBSE_NONE)

1335

1303

{

1336

sprintf(log_buffer, "getsize() failed for pvmem in %s",

1337

id);

1304

sprintf(log_buffer, "getsize() failed for vmem in %s",id);

1338

1305

1339

1306

return(error(pname, retval));

1340

1307

}

1341

1308

1342

if (value > ULONG_MAX)

1343

{

1344

log_buffer[0] = '\0';

1345

1346

sprintf(log_buffer, "invalid value returned by getsize() for pvmem in %s",

1347

id);

1348

1349

return(error(pname, PBSE_BADATVAL));

1350

}

1351

1352

1309

if ((vmem_limit == 0) || (value < vmem_limit))

1353

1310

vmem_limit = value;

1354

1311

}

1355

1312

}

1356

else if ((!strcmp(pname, "mem") && (pjob->ji_numnodes != 1)) ||

1357

!strcmp(pname, "mppmem"))

1313

else if (!strcmp(pname, "pvmem"))

1314

{

1315

if (ignvmem == FALSE)

1316

{

1317

/* set */

1318

1319

if (set_mode == SET_LIMIT_SET)

1320

{

1321

retval = getsize(pres, &value);

1322

1323

if (retval != PBSE_NONE)

1324

{

1325

sprintf(log_buffer, "getsize() failed for pvmem in %s",

1326

id);

1327

1328

return(error(pname, retval));

1329

}

1330

1331

if (value > ULONG_MAX)

1332

{

1333

log_buffer[0] = '\0';

1334

1335

sprintf(log_buffer, "invalid value returned by getsize() for pvmem in %s",

1336

id);

1337

1338

return(error(pname, PBSE_BADATVAL));

1339

}

1340

1341

if ((vmem_limit == 0) || (value < vmem_limit))

1342

vmem_limit = value;

1343

}

1344

}

1345

}

1346

else if ((!strcmp(pname,"mem") && (pjob->ji_numnodes != 1)) ||

1347

!strcmp(pname,"mppmem"))

1358

1348

{

1359

1349

/* ignore. If we ever get rid of support for the UNICOS OS then we can

1360

1350

remove the ATR_DFLAG_MOM | ATR_DFLAG_ALTRUN flags from mppmem */

1362

1352

else if ((!strcmp(pname, "mem") && (pjob->ji_numnodes == 1)) ||

1363

1353

!strcmp(pname, "pmem"))

1364

1354

{

1365

/* set */

1366

1367

if (set_mode == SET_LIMIT_SET)

1355

if (ignmem == FALSE)

1368

1356

{

1369

retval = getsize(pres, &value);

1370

1371

if (retval != PBSE_NONE)

1372

{

1373

sprintf(log_buffer, "getsize() failed for mem/pmem in %s",

1374

id);

1375

1376

return(error(pname, retval));

1377

}

1378

1379

reslim.rlim_cur = reslim.rlim_max = value;

1380

1381

if (setrlimit(RLIMIT_DATA, &reslim) < 0)

1382

{

1383

sprintf(log_buffer, "cannot set data limit to %ld for job %s (setrlimit failed w/errno=%d (%s) - check default user limits)",

1384

reslim.rlim_max,

1385

pjob->ji_qs.ji_jobid,

1386

errno,

1387

strerror(errno));

1388

1389

return(error("RLIMIT_DATA", PBSE_SYSTEM));

1390

}

1391

1392

if (setrlimit(RLIMIT_RSS, &reslim) < 0)

1393

{

1394

sprintf(log_buffer, "cannot set RSS limit to %ld for job %s (setrlimit failed w/errno=%d (%s) - check default user limits)",

1395

reslim.rlim_max,

1396

pjob->ji_qs.ji_jobid,

1397

errno,

1398

strerror(errno));

1399

1400

return(error("RLIMIT_RSS", PBSE_SYSTEM));

1401

}

1357

/* set */

1358

1359

if (set_mode == SET_LIMIT_SET)

1360

{

1361

retval = getsize(pres, &value);

1362

1363

if (retval != PBSE_NONE)

1364

{

1365

sprintf(log_buffer, "getsize() failed for mem/pmem in %s",

1366

id);

1367

1368

return(error(pname, retval));

1369

}

1370

1371

reslim.rlim_cur = reslim.rlim_max = value;

1372

1373

if (setrlimit(RLIMIT_DATA, &reslim) < 0)

1374

{

1375

sprintf(log_buffer, "cannot set data limit to %ld for job %s (setrlimit failed w/errno=%d (%s) - check default user limits)",

1376

(long int)reslim.rlim_max,

1377

pjob->ji_qs.ji_jobid,

1378

errno,

1379

strerror(errno));

1380

1381

return(error("RLIMIT_DATA", PBSE_SYSTEM));

1382

}

1383

1384

if (setrlimit(RLIMIT_RSS, &reslim) < 0)

1385

{

1386

sprintf(log_buffer, "cannot set RSS limit to %ld for job %s (setrlimit failed w/errno=%d (%s) - check default user limits)",

1387

(long int)reslim.rlim_max,

1388

pjob->ji_qs.ji_jobid,

1389

errno,

1390

strerror(errno));

1391

1392

return(error("RLIMIT_RSS", PBSE_SYSTEM));

1393

}

1402

1394

1403

1395

#ifdef __GATECH

1404

/* NOTE: best patch may be to change to 'vmem_limit = value;' */

1405

1406

if (setrlimit(RLIMIT_STACK, &reslim) < 0)

1407

{

1408

sprintf(log_buffer, "cannot set stack limit to %ld for job %s (setrlimit failed w/errno=%d (%s) - check default user limits)",

1409

reslim.rlim_max,

1410

pjob->ji_qs.ji_jobid,

1411

errno,

1412

strerror(errno));

1413

1414

return(error("RLIMIT_STACK", PBSE_SYSTEM));

1415

}

1416

1417

/* set address space */

1418

1419

if (setrlimit(RLIMIT_AS, &reslim) < 0)

1420

{

1421

sprintf(log_buffer, "cannot set AS limit to %ld for job %s (setrlimit failed w/errno=%d (%s) - check default user limits)",

1422

reslim.rlim_max,

1423

pjob->ji_qs.ji_jobid,

1424

errno,

1425

strerror(errno));

1426

1427

return(error("RLIMIT_AS", PBSE_SYSTEM));

1428

}

1396

/* NOTE: best patch may be to change to 'vmem_limit = value;' */

1397

1398

if (setrlimit(RLIMIT_STACK, &reslim) < 0)

1399

{

1400

sprintf(log_buffer, "cannot set stack limit to %ld for job %s (setrlimit failed w/errno=%d (%s) - check default user limits)",

1401

(long int)reslim.rlim_max,

1402

pjob->ji_qs.ji_jobid,

1403

errno,

1404

strerror(errno));

1405

1406

return(error("RLIMIT_STACK", PBSE_SYSTEM));

1407

}

1408

1409

/* set address space */

1410

1411

if (setrlimit(RLIMIT_AS, &reslim) < 0)

1412

{

1413

sprintf(log_buffer, "cannot set AS limit to %ld for job %s (setrlimit failed w/errno=%d (%s) - check default user limits)",

1414

(long int)reslim.rlim_max,

1415

pjob->ji_qs.ji_jobid,

1416

errno,

1417

strerror(errno));

1418

1419

return(error("RLIMIT_AS", PBSE_SYSTEM));

1420

}

1429

1421

1430

1422

#endif /* __GATECH */

1431

1423

1432

mem_limit = value;

1433

1434

if (getrlimit(RLIMIT_STACK, &reslim) >= 0)

1435

{

1436

/* NOTE: mem_limit no longer used with UMU patch in place */

1437

1438

mem_limit = value + reslim.rlim_cur;

1424

mem_limit = value;

1425

1426

if (getrlimit(RLIMIT_STACK, &reslim) >= 0)

1427

{

1428

/* NOTE: mem_limit no longer used with UMU patch in place */

1429

1430

mem_limit = value + reslim.rlim_cur;

1431

}

1439

1432

}

1440

1433

}

1441

1434

} /* END else if (!strcmp(pname,"mem") && ... */

1478

1471

1479

1472

/* NO-OP */

1480

1473

}

1474

else if(!strcmp(pname, "prologue"))

1475

{

1476

}

1477

else if(!strcmp(pname, "epilogue"))

1478

{

1479

}

1481

1480

else if ((pres->rs_defin->rs_flags & ATR_DFLAG_RMOMIG) == 0)

1482

1481

{

1483

1482

/* don't recognize and not marked as ignore by mom */

1658

1657

1659

1658

1660

1659

1661

1662

1660

1663

1661

* Declare start of polling loop.

1664

1662

1672

1670

* This function is called from the main MOM loop once every "check_poll_interval"

1673

1671

* seconds.

1674

1672

1675

* @see get_proc_stat()

1673

* @see get_proc_stat() - child

1676

1674

* @see mom_set_use() - Aggregates data collected here

1677

1675

1678

1676

* NOTE: populates global 'proc_array[]' variable.

1807

1805

assert(pname != NULL);

1808

1806

assert(*pname != '\0');

1809

1807

1810

if (strcmp(pname, "cput") == 0)

1808

if ((igncput == FALSE) && (strcmp(pname, "cput") == 0))

1811

1809

{

1812

1810

retval = gettime(pres, &value);

1813

1811

1823

1821

return(TRUE);

1824

1822

}

1825

1823

}

1826

else if (strcmp(pname, "pcput") == 0)

1824

else if ((igncput == FALSE) && (strcmp(pname, "pcput") == 0))

1827

1825

{

1828

1826

retval = gettime(pres, &value);

1829

1827

1873

1871

return(TRUE);

1874

1872

}

1875

1873

}

1876

else if (strcmp(pname, "walltime") == 0)

1874

else if (ignwalltime == 0 && strcmp(pname, "walltime") == 0)

1877

1875

{

1876

1878

1877

/* no need to check walltime on sisters, MS will get it */

1879

1880

1878

if ((pjob->ji_qs.ji_svrflags & JOB_SVFLG_HERE) == 0)

1881

1879

continue;

1882

1880

1894

1892

num,

1895

1893

value);

1896

1894

1897

if (ignwalltime == 0)

1898

{

1899

return(TRUE);

1900

}

1895

return(TRUE);

1901

1896

}

1902

1897

}

1903

1898

} /* END for (pres) */

1921

1916

* stats gathered by the mom_get_sample() function. This function

1922

1917

* is often called by "im_request()" as a result of POLL_JOB query

1923

1918

* from the mother superior.

1924

1925

* @return An error code if something goes wrong.

1926

1919

1927

1920

* @see im_request() - parent - respond to poll_job request from mother superior

1928

1921

* @see examine_all_running_jobs() - parent - update local use on mother superior

1929

1922

* @see TMomFinalizeJob1() - parent - update serial job immediately at job start

1923

1924

* @return An error code if something goes wrong.

1930

1925

1931

1926

1932

1927

int mom_set_use(

1943

1938

at = &pjob->ji_wattr[(int)JOB_ATR_resc_used];

1944

1939

assert(at->at_type == ATR_TYPE_RESC);

1945

1940

1941

#ifdef USESAVEDRESOURCES

1942

/* don't update jobs that are marked as recovery */

1943

if (pjob->ji_flags & MOM_JOB_RECOVERY)

1944

{

1945

return(PBSE_NONE);

1946

}

1947

#endif /* USESAVEDRESOURCES */

1948

1946

1949

at->at_flags |= ATR_VFLAG_MODIFY;

1947

1950

1948

1951

if ((at->at_flags & ATR_VFLAG_SET) == 0)

2033

2036

assert(pres != NULL);

2034

2037

2035

2038

/* NOTE: starting jobs can come through here before stime is recorded */

2036

2037

2039

if (pjob->ji_qs.ji_stime == 0)

2038

2040

pres->rs_value.at_val.at_long = 0;

2039

2041

else

2063

2065

2064

2066

2065

2067

2066

2068

/**

2067

2069

* Kill a task session.

2068

2070

* Call with the task pointer and a signal number.

2069

* return number of tasks signalled (0 = failure)

2071

2072

* @return number of tasks signalled (0 = failure)

2073

2074

* @see kill_job() - parent

2075

2076

* NOTE: should support killpg() or killpidtree() - (NYI)

2077

* may be required for suspend/resume

2070

2078

2071

2079

2072

/* NOTE: should support killpg() or killpidtree() (NYI)

2073

may be required for suspend resume */

2074

2075

2080

int kill_task(

2076

2081

2077

2082

task *ptask, /* I */

2085

2090

int NumProcessesFound = 0; /* number of processes found with session ID */

2086

2091

2087

2092

struct dirent *dent;

2088

proc_stat_t *ps;

2093

proc_stat_t *ps;

2089

2094

int sesid;

2090

2095

pid_t mompid;

2091

2096

2095

2100

if (LOGLEVEL >= 5)

2096

2101

{

2097

2102

sprintf(log_buffer, "%s: sending signal %d to task %d, session %d",

2098

id,

2099

sig,

2100

ptask->ti_qs.ti_task,

2101

sesid);

2103

id,

2104

sig,

2105

ptask->ti_qs.ti_task,

2106

sesid);

2102

2107

2103

2108

log_record(

2104

2109

PBSEVENT_JOB,

2112

2117

if (LOGLEVEL >= 3)

2113

2118

{

2114

2119

sprintf(log_buffer, "cannot send signal %d to task (no session id)",

2115

sig);

2120

sig);

2116

2121

2117

2122

log_record(

2118

2123

PBSEVENT_ERROR,

2146

2151

sprintf(log_buffer, "%s: get_proc_stat",

2147

2152

dent->d_name);

2148

2153

2149

log_err(errno, id, log_buffer);

2154

log_err(errno,id,log_buffer);

2150

2155

}

2151

2156

2152

2157

continue;

2153

2158

}

2154

2159

2155

if (sesid == ps->session)

2160

if ((sesid == ps->session) ||

2161

(ProcIsChild(procfs,dent->d_name,ptask->ti_job->ji_qs.ji_jobid) == TRUE))

2162

2156

2163

{

2157

2164

NumProcessesFound++;

2158

2165

2201

2208

log_buffer);

2202

2209

}

2203

2210

2211

if (LOGLEVEL >= 3)

2212

{

2213

sprintf(log_buffer, "%s: not killing process %d. Avoid sending signal because child task still has MOM's session id", id, ps->pid);

2214

2215

log_record(

2216

PBSEVENT_JOB,

2217

PBS_EVENTCLASS_JOB,

2218

ptask->ti_job->ji_qs.ji_jobid,

2219

log_buffer);

2220

}

2221

2204

2222

continue;

2205

2223

} /* END if (ps->pid == mompid) */

2206

2224

2211

2229

req.tv_sec = 0;

2212

2230

req.tv_nsec = 250000000; /* .25 seconds */

2213

2231

2214

/* give the process some time to quit gracefully first (up to 5 seconds) */

2232

/* give the process some time to quit gracefully first (up to .25*20=5 seconds) */

2215

2233

2216

2234

sprintf(log_buffer, "%s: killing pid %d task %d gracefully with sig %d",

2217

2235

id,

2277

2295

{

2278

2296

/* kill process hard */

2279

2297

2298

/* why is this not killing with SIGKILL? */

2299

2280

2300

sprintf(log_buffer, "%s: killing pid %d task %d with sig %d",

2281

2301

id,

2282

2302

ps->pid,

2294

2314

else

2295

2315

killpg(ps->pid, sig);

2296

2316

}

2297

}

2298

} /* END if (i >= 20) */

2317

} /* END if ((ps = get_proc_stat(ps->pid)) != NULL) */

2318

} /* END if (i >= 20) */

2299

2319

2300

2320

++ct;

2301

2321

} /* END else ((ps->state == 'Z') || (ps->pid == 0)) */

2302

2322

} /* END if (sesid == ps->session) */

2303

2323

} /* END while ((dent = readdir(pdir)) != NULL) */

2304

2324

2325

/* NOTE: to fix bad state situations resulting from a hard crash, the logic

2326

below should be triggered any time no processes are found (NYI) */

2327

2305

2328

if (IS_ADOPTED_TASK(ptask->ti_qs.ti_task) && (NumProcessesFound == 0))

2306

2329

{

2307

2330

/* no process was found, but for an adopted task this is OK (we don't find

2311

2334

2312

2335

ct++;

2313

2336

2314

/* do code to mark task as finished (borrowed from Linux scan_for_terminating())... */

2337

/* do code to mark task as finished (borrowed from Linux scan_for_terminated())... */

2315

2338

2316

2339

ptask->ti_qs.ti_exitstat = 0; /* assume successful completion */

2317

2340

ptask->ti_qs.ti_status = TI_STATE_EXITED;

2338

2361

2339

2362

ct++;

2340

2363

2341

/* do code to mark task as finished (borrowed from Linux scan_for_terminating())... */

2364

/* do code to mark task as finished (borrowed from Linux scan_for_terminated())... */

2342

2365

2343

2366

ptask->ti_qs.ti_exitstat = 0; /* assume successful completion */

2344

2367

ptask->ti_qs.ti_status = TI_STATE_EXITED;

2412

2435

2413

2436

2414

2437

2415

2416

2438

2417

* mom_does_chkpnt - return 1 if mom supports checkpoint

2418

* 0 if not

2439

* mom_does_checkpoint

2440

2441

* @returns CST values as described in resmon.h.

2419

2442

2420

2443

2421

2444

int

2422

mom_does_chkpnt(void)

2423

2445

mom_does_checkpoint(void)

2424

2446

{

2425

if (checkpoint_script_name[0] != '\0')

2426

{

2427

return(1);

2428

}

2429

2430

return(0);

2447

return(CST_BLCR); /* Use the BLCR checkpointing system. */

2431

2448

}

2432

2449

2433

2450

2438

2455

2439

2456

int mach_checkpoint(

2440

2457

2441

task *ptask,

2442

char *file,

2458

task *ptask, /* I */

2459

char *file, /* I */

2443

2460

int abort) /* I */

2444

2461

2445

2462

{

2446

char *id = "mach_checkpoint";

2447

int pid;

2448

char sid[20];

2449

char *arg[20];

2450

char buf[1024];

2451

char **ap;

2452

2453

#define SET_ARG(x) (((x) == NULL) || (*(x) == 0))?"-":(x)

2454

2455

/* if a checkpoint script is defined launch it */

2456

2457

if (checkpoint_script_name[0] == '\0')

2458

{

2459

log_err(-1, id, "No checkpoint script defined");

2460

}

2461

else

2462

{

2463

/* launch the script and return success */

2464

2465

pid = fork();

2466

2467

if (pid > 0)

2468

{

2469

/* parent: pid = child's pid */

2470

2471

/* NO-OP, but may need to waitpid() the child (NYI) */

2472

2473

return(0);

2474

}

2475

else if (pid == 0)

2476

{

2477

/* child: execv the script */

2478

2479

sprintf(sid, "%ld",

2480

ptask->ti_job->ji_wattr[(int)JOB_ATR_session_id].at_val.at_long);

2481

2482

arg[0] = checkpoint_script_name;

2483

arg[1] = sid;

2484

arg[2] = SET_ARG(ptask->ti_job->ji_qs.ji_jobid);

2485

arg[3] = SET_ARG(ptask->ti_job->ji_wattr[(int)JOB_ATR_euser].at_val.at_str);

2486

arg[4] = SET_ARG(ptask->ti_job->ji_wattr[(int)JOB_ATR_chkptdir].at_val.at_str);

2487

arg[5] = SET_ARG(ptask->ti_job->ji_wattr[(int)JOB_ATR_chkptname].at_val.at_str);

2488

arg[6] = (abort) ? "15" /*abort*/ : "0" /*run/continue*/;

2489

arg[7] = NULL;

2490

2491

strcpy(buf, "checkpoint args:");

2492

2493

for (ap = arg; *ap; ap++)

2494

{

2495

strcat(buf, " ");

2496

strcat(buf, *ap);

2497

}

2498

2499

log_err(-1, id, buf);

2500

2501

execv(arg[0], arg);

2502

2503

/* NOTE: The right way to do this is to put the job into some sort

2504

* of waiting state before returning, and the monitor the

2505

* the child to see what it's exit status is. We are delaying

2506

* the full implementation for now. NYI

2507

2508

} /* END if (pid == 0) */

2509

}

2510

2511

2463

return(-1);

2512

2464

} /* END mach_checkpoint() */

2513

2465

2527

2479

char *file)

2528

2480

2529

2481

{

2530

char *id = "mach_restart";

2531

int pid;

2532

char sid[20];

2533

char *arg[20];

2534

2535

2536

/* if a restart script is defined launch it */

2537

2538

if (restart_script_name[0] == '\0')

2539

{

2540

log_record(PBSEVENT_DEBUG, 0, id, "No restart script defined");

2541

}

2542

else

2543

{

2544

/* launch the script and return success */

2545

2546

pid = fork();

2547

2548

if (pid > 0)

2549

{

2550

/* parent: pid = child's pid */

2551

2552

/* NO-OP, but may need to waitpid() the child (NYI) */

2553

2554

return(0);

2555

}

2556

2557

if (pid == 0)

2558

{

2559

/* child: execv the script */

2560

2561

sprintf(sid, "%ld",

2562

ptask->ti_job->ji_wattr[(int)JOB_ATR_session_id].at_val.at_long);

2563

2564

arg[0] = restart_script_name;

2565

arg[1] = sid;

2566

arg[2] = ptask->ti_job->ji_qs.ji_jobid;

2567

arg[3] = ptask->ti_job->ji_wattr[(int)JOB_ATR_euser].at_val.at_str;

2568

arg[4] = NULL;

2569

2570

execv(arg[0], arg);

2571

2572

/* NOTE: The right way to do this is to put the job into some sort

2573

* of waiting state before returning, and the monitor the

2574

* the child to see what it's exit status is. We are delaying

2575

* the full implementation for now. NYI

2576

2577

} /* END if (pid == 0) */

2578

}

2579

2580

2581

2482

return(-1);

2582

2483

}

2583

2484

2643

2544

}

2644

2545

2645

2546

sprintf(ret_string, "%.2f",

2646

cputime * cputfactor);

2547

2548

cputime * cputfactor);

2647

2549

2648

2550

return(ret_string);

2649

2551

} /* END cput_job() */

2944

2846

pid_t pid)

2945

2847

2946

2848

{

2947

char *id = "resi_proc";

2849

char *id = "resi_proc";

2948

2850

proc_stat_t *ps;

2949

2851

2950

2852

if ((ps = get_proc_stat(pid)) == NULL)

3074

2976

if ((jobid = ps->session) == 0)

3075

2977

continue;

3076

2978

3077

if (LOGLEVEL >= 6)

2979

if (LOGLEVEL >= 7)

3078

2980

{

3079

2981

sprintf(log_buffer, "%s[%d]: pid %d sid %d",

3080

2982

id,

3320

3222

if ((uid = ps->uid) == 0)

3321

3223

continue;

3322

3224

3323

if (LOGLEVEL >= 6)

3225

if (LOGLEVEL >= 7)

3324

3226

{

3325

3227

sprintf(log_buffer, "%s[%d]: pid %d uid %d",

3326

3228

id,

3746

3648

3747

3649

3748

3650

3749

3750

3751

3651

3752

3652

* For a recovering (-p) mom, look through existing tasks in existing

3753

3653

* jobs for things that have exited that are not owned by us through a

3762

3662

3763

3663

job *job;

3764

3664

extern tlist_head svr_alljobs;

3665

static int first_time = TRUE;

3765

3666

3766

3667

DIR *pdir; /* use local pdir to prevent race conditions associated w/global pdir (VPAC) */

3767

3668

3777

3678

struct dirent *dent;

3778

3679

int found;

3779

3680

3681

3682

* Check for tasks that were exiting when mom went down, set back to

3683

* running so we can reprocess them and send the obit

3684

3685

if ((first_time) && (task->ti_qs.ti_sid != 0) &&

3686

((task->ti_qs.ti_status == TI_STATE_EXITED) ||

3687

(task->ti_qs.ti_status == TI_STATE_DEAD)))

3688

{

3689

3690

if (LOGLEVEL >= 7)

3691

{

3692

sprintf(log_buffer, "marking task %d as TI_STATE_RUNNING was %d",

3693

task->ti_qs.ti_task,

3694

task->ti_qs.ti_status);

3695

3696

LOG_EVENT(

3697

PBSEVENT_DEBUG,

3698

PBS_EVENTCLASS_JOB,

3699

job->ji_qs.ji_jobid,

3700

log_buffer);

3701

}

3702

task->ti_qs.ti_status = TI_STATE_RUNNING;

3703

}

3704

3780

3705

/* only check on tasks that we think should still be around */

3781

3706

3782

3707

if (task->ti_qs.ti_status != TI_STATE_RUNNING)

3841

3766

3842

3767

task_save(task);

3843

3768

3769

#ifdef USESAVEDRESOURCES

3770

if (first_time)

3771

{

3772

job->ji_flags |= MOM_JOB_RECOVERY;

3773

if (LOGLEVEL >= 7)

3774

{

3775

sprintf(buf, "marking job as MOM_JOB_RECOVERY for task %d",

3776

task->ti_qs.ti_task);

3777

3778

LOG_EVENT(

3779

PBSEVENT_DEBUG,

3780

PBS_EVENTCLASS_JOB,

3781

job->ji_qs.ji_jobid,

3782

buf);

3783

}

3784

}

3785

#endif /* USESAVEDRESOURCES */

3786

3844

3787

exiting_tasks = 1;

3845

3788

}

3846

3789

}

3847

3790

} /* END for (job = GET_NEXT(svr_alljobs)) */

3848

3791

3849

3792

closedir(pdir);

3793

3794

first_time = FALSE;

3850

3795

3851

3796

return;

3852

3797

} /* END scan_non_child_tasks() */

4468

4413

4469

4414

if (strncmp(interfaceName[interface],"eth",3) == 0)

4470

4415

{

4471

rc = sscanf(interfaceName[interface],"eth%d",

4472

&ethNum);

4416

rc = sscanf(interfaceName[interface],"eth%d",

4417

&ethNum);

4473

4418

}

4474

4419

4475

4420

4490

4435

return(ret_string);

4491

4436

} /* END netload() */

4492

4437

4438

4439

4440

4441

4442

mbool_t ProcIsChild(

4443

4444

char *Dir, /* I */

4445

char *PID, /* I */

4446

char *JobID) /* I */

4447

4448

{

4449

return(FALSE);

4450

} /* END ProcIsChild() */

4451

4493

4452

/* END mom_mach.c */

4494

4453

4495

4454

Older »