#! /bin/bash ##************************************************************** ## ## Copyright (C) 1990-2013, Condor Team, Computer Sciences Department, ## University of Wisconsin-Madison, WI. ## ## Licensed under the Apache License, Version 2.0 (the "License"); you ## may not use this file except in compliance with the License. You may ## obtain a copy of the License at ## ## http://www.apache.org/licenses/LICENSE-2.0 ## ## Unless required by applicable law or agreed to in writing, software ## distributed under the License is distributed on an "AS IS" BASIS, ## WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. ## See the License for the specific language governing permissions and ## limitations under the License. ## ##************************************************************** # Original Author Michael Hanke (v0.1) # Updated by Samuel Friedman (03/13) # # To generate a manpage, simply run through help2man. # play safe set -u set -e # define variable defaults scriptname="condor_qsub" version="0.2" error_file="" accounting_group="" binary_flag=0 checkpoint_interval="" cluster_id=0 cwd_flag=0 deferral_date_time="" environment="" hold_jid="" hold_flag=0 init_dir="" input_file="" interactive_option="" mail_config="" mail_user="" merge_streams="" job_name="" no_output_batch_id_flag=0 number_specs="" output_file="" print_flag=0 priority=0 requirements_string="" rerun_flag="" resource_list="" shell=${SHELL:-/bin/sh} sync_flag=0 taskid_first="" taskid_last="" taskid_step="" test_qsub_flag=0 total_procs=1 transfer_files="" qsub_cmd="" qsub_args="" qsub_file="" queue_name="" export_env_flag=0 condor_keep_files=0 print_description() { cat << EOT Minimalistic emulation of SGE's qsub for Condor. EOT } print_version() { cat << EOT $scriptname $version This script is released under the Apache V2.0 License. It was written by Michael Hanke and Samuel Friedman EOT } print_help() { cat << EOT Usage: $scriptname [OPTIONS] [ []] The primary purpose of this emulation is to allow PBS/Torque-style and SGE-style submission of dependent jobs without the need to specify the full dependency graph at submission time. This implementation is neither as efficient as HTCondor's DAGMan, nor as functional as SGE's qsub/qalter. It merely serves as a minimal adaptor to be able to use software original written to interact with PBS/Torque/SGE in an HTCondor pool. In Version 0.2, many additional features have been added. In general $scriptname behaves just like qsub. However, only a fraction of the original functionality is available (~1/3). The following list of options only describes the differences in the behavior of PBS/Torque/SGE's qsub and this emulation. Qsub options not listed here are not supported. For a full listing of options, please see POSIX: http://pubs.opengroup.org/onlinepubs/9699919799/utilities/qsub.html SGE: http://gridscheduler.sourceforge.net/htmlman/htmlman1/qsub.html PBS/Torque: http://docs.adaptivecomputing.com/torque/4-1-3/Content/topics/commands/qsub.htm Options: -a date_time We follow the PBS/Torque implementation of this option. The date_time argument is in the form: [[[[CC]YY]MM]DD]hhmm[.SS] where MM and DD are potentially optional. SGE does not permit MM or DD to be optional. -A account_string Creates an attribute specifying what accounting group should list the time associated with this job. Unlike SGE, we do not give a default group of "sge". -b If 'y', command and arguments given on the command line are wrapped into a shell script which is then submitted to HTCondor. SGE only. PBS/Torque version not supported. -c checkpoint_options There is some differences between POSIX, SGE, and PBS/Torque. This area could use some quick and helpful improvement. Currently, checkpointing can be turned off entirely or have periodic checkpointing turned off. Otherwise, the code assumes that the executable is standard universe and will do automatic checkpointing as HTCondor currently implements it. --condor-keep-files This is a non-SGE option. If given, it will prevent $scriptname from deleting temporary files (generated submit files, sentinel jobs). This is mostly useful for debugging. -cwd If given, this option will cause the 'initialdir' value in the Condor submit file to be set to the current directory. -d path -wd working_dir For -d, the default working directory is not the home directory as speicifed by PBS/Torque. Otherwise, works as expected for SGE-style (-wd). -e Name of the file to contain the STDERR output of the job. By default this will be job_name.ejob_id[.task_id]. If an existing directory is specified, the file will be placed inside this directory using the default schema for the filename. -f qsub_file Non-PBS/Torque/SGE option. condor_qsub was originally intended to accept command line arguments. However, submissions to qsub are often done in a batch file with commands like #PBS or #SGE. condor_qsub will parse the batch file listed as qsub_file. Note, this may be similar to the SGE -@ option. -h Implemented as listed. --help Print usage summary and option list. As of Version 0.2, no longer will -h be an option for usage summary and option list as it is a feature. We may consider changing this option to -help as this is a valid SGE option. -hold_jid If given, the job will be submitted in 'hold' state. Along with the actual job a 'sentinel' job will be submitted to Condor's local universe. This sentinel watches the specified job and releases the submitted job whenever the job has completed. The sentinel observes SGE's behavior to detect job exiting with code 100 and not start depedent job in this case. If a cluster id of an array job is given the dependent job will only be released after all individual jobs of a cluster have completed. -l This option is currently under development. We support from PBS/Torque the following options: arch, file, host, mem, nodes (partial), opsys, pmem, procs. We do not support parallel jobs at the moment. -m <...> SGE's notification labels will be translated (approximately) into Condor's notifications states (Never, Error, Complete). -M Added as 'notify_user' to the submit file. -N Determines the default name of logfile (stdout, stderr). -o See -e option, but for a job's stdout. -p Added a 'priority' to the submit file. --print Non-PBS/Torque/SGE option. Print to STDOUT the contents of the HTCondor Submit File. -q This option is permanently ignored, as HTCondor doesn't have multiple queues. -r Defaults to 'y' as this is the default value for HTCondor, PBS/Torque, and the POSIX standard. HTCondor likes to try to rerun jobs, so we've done as best of a job as possible to ensure that reruns do not occur. -S Path to a shell binary for script execution. -shell This option is currently ignored. -t [-[:]] Task ID specification for array job submissions. --test This non-PBS/Torque/SGE option makes condor_qsub not submit the job. Useful for testing purposes, especially when used in conjunction with --print and/or --condor-keep-files. -v variable_list This option does not handle whitespace well. When used on the command line, single and double quotes need to be escaped by backslashes. Also, HTCondor does not accept instead of =. If just is used, =True is added. It also ingores the environment Variable_List attribute. -V If given, 'getenv = True' is added to the submit file. -W attr_name=attr_value[,attr_name=attr_value...] PBS/Torque supports a number of attributes. Only stagein and stageout are supported for attr_name. The format of attr_value for stagein and stageout is local_file@hostname:remote_file[,...] and we strip it to remote_file[,...]. HTCondor's file transfer mechanism is then used if needed. --version Print version information and exit. EOT } create_dep_sentinel() { cat << EOT > $1 #!/bin/sh hold_jids="$2" dep_job="$3" clean_error() { printf "\$1\n" condor_rm \$dep_job exit 1 } # as long as there are relevant job in the queue wait and try again counter=3 while [ \$(condor_q -long -attributes Owner \$hold_jids | grep "$USER" | wc -l) -ge 1 -o \$counter -ge 1 ]; do job_count=\$(condor_q -long -attributes Owner \$hold_jids | grep "$USER" | wc -l) if [ \$job_count -eq 0 ]; then counter=\$((counter-1)) else counter=3 fi sleep 5 done # now check whether all deps have exited properly (i.e. not with code 100) for jid in \$hold_jids; do job_status="\$(condor_history \$jid |tail -n 1 | awk '{print \$6}')" case "\$job_status" in C) if [ "\$(condor_history -long \$jid |grep ExitCode | cut -d= -f2,2)" = " 100" ]; \ then clean_error "Error: Job dependency \$jid completed but exited with code 100."; fi;; X) clean_error "Error: Job dependency \$jid has been removed from the queue" ;; *) clean_error "Error: Job dependency \$jid doesn't exist" ;; esac done # all good -- let the job go condor_release \$dep_job # in the end we can safely clean this sentinel script [ $condor_keep_files -eq 0 ] && rm \$0 || true EOT chmod +x $1 cat << EOT > $1.submit universe = local executable = $1 Queue EOT chmod +x $1.submit } # PBS/Torque: http://docs.adaptivecomputing.com/torque/4-1-3/help.htm#topics/commands/qsub.htm # SGE/Oracle Grid Engine: http://docs.oracle.com/cd/E24901_01/doc.62/e21978/configuration.htm#autoId60 parse_args() { ############################################################### # cmdline args ############################################################### # Parse commandline options # Note that we use `"$@"' to let each command-line parameter expand to a # separate word. The quotes around `$@' are essential! # We need CLOPTS as the `eval set --' would nuke the return value of getopt. command_line_flags="-o A:,a:,b:,c:,d:,e:,f:,h,i:,I,j:,l:,m:,M:,N:,o:,p:,q:,r:,S:,t:,v:,V,W:,z --long condor-keep-files,cwd,help,hold_jid:,print,shell:,test,verbose,version,wd:" CLOPTS=`getopt -a $command_line_flags -n "$scriptname" -- "$@"` if [ $? != 0 ] ; then echo "Terminating..." >&2 exit 1 fi # If we have #PBS commands or #SGE commands # Parse the file for PBS or SGE commands if [ "$1" = "-f" ]; then # Read through the file and "create" commands line options shift qsub_file=$1 qsub_cmd=$qsub_file qsub_new_command_line="" pbs_regex='^#PBS ' sge_regex='^#\$ ' qsub_new_command_line="" while read qsub_file_line; do # Check each line in the file to see if they start with #PBS or #SGE line_has_command=0 if [[ $qsub_file_line =~ $pbs_regex ]]; then line_qsub_option=${qsub_file_line#*#PBS} line_has_command=1 fi if [[ $qsub_file_line =~ $sge_regex ]]; then line_qsub_option=${qsub_file_line#*#\$} line_has_command=1 fi if [ "$line_has_command" = 1 ]; then qsub_new_command_line="$qsub_new_command_line $line_qsub_option" fi done < $qsub_file # echo New command line options: "$qsub_new_command_line" qsub_new_command_line="$@ $qsub_new_command_line" # echo New command line: $qsub_new_command_line set -- $qsub_new_command_line CLOPTS=`getopt -a $command_line_flags -n "$scriptname" -- "$@"` fi # Note the quotes around `$CLOPTS': they are essential! eval set -- "$CLOPTS" while true ; do case "$1" in -A) shift; accounting_group=$1; shift;; -a) shift; deferral_date_time=$1; shift;; -b) shift; if [ "$1" = "y" ]; then binary_flag=1; fi; shift;; -c) shift; checkpoint_interval=$1;; # if [ ]; then asdf; fi; shift;; #Figure out how to get optional argument --cwd) shift; cwd_flag=1;; --condor-keep-files) shift; condor_keep_files=1;; -d|--wd) shift; init_dir=$1; shift;; # --wd) shift; init_dir=$1; shift;; -e) shift; error_file=$1; shift;; -h) shift; hold_flag=1;; --help) print_description; print_help; exit 0;; --hold_jid) shift; hold_jid=$1; shift;; # we would be only interested in the arch spec -- right now it is ignored -i) shift; input_file=$1; shift;; -I) shift; interactive_option="-interactive";; -j) shift; merge_streams=$1; shift;; -l) shift if [ -z "$resource_list" ]; then resource_list=$1 else resource_list="$resource_list,$1" fi shift;; -m) shift; mail_config=$1; shift;; -M) shift; mail_user=$1; shift;; -N) shift; job_name=$1; shift;; -o) shift; output_file=$1; shift;; -p) shift; priority=$1; shift;; --print) shift; print_flag=1;; -q) shift; queue_name=$1; shift;; -r) shift; rerun_flag=$1; shift;; -S) shift; shell=$1; shift;; --shell) shift; shift;; # --sync) shift; sync_flag=1;; #Needs work. -t) shift taskid_first="$(echo "$1" | awk -F- '{print $1}')" taskid_last="$(echo "$1" | awk -F- '{print $2}' | awk -F: '{print $1}')" taskid_step="$(echo "$1" | awk -F- '{print $2}' | awk -F: '{print $2}')" shift;; # needs to handle SGE_TASK_ID, SGE_TASK_FIRST, SGE_TASK_LAST and SGE_TASK_STEPSIZE # log goes to: .['e'|'o']'.' --test) shift; test_qsub_flag=1;; -v) shift; environment=$1; shift;; -V) shift; export_env_flag=1;; --version) print_version; exit 0;; -W) shift; transfer_files="$transfer_files $1"; shift;; -z) shift; no_output_batch_id_flag=1;; --) shift ; break ;; *) echo "Internal error! ($1)"; exit 1;; esac done # the first arg is the command the rest are its arguments if [ $# -ge 1 ] && [ "$1" != "XXcondor_sub_scriptmodeXX" ]; then qsub_cmd="$1" shift qsub_args="$@" fi } # parse all commandline args parse_args $@ # no arguments -> need to read from stdin if [ -z "$qsub_cmd" ]; then # redirect STDIN into a file, place it into the current dir to increase the # likelihood of being available on the execute host too (NFS-mounted FS) # unfortunately the author cannot think of a way to clean this tempfile up # as it need to be available until the last job in the cluster actually # started running -- yet another sentinel could do it .... if [ -z "$job_name" ]; then stdin_script_file=$(mktemp --tmpdir=$(pwd) STDIN_qsub.XXXXXXXXXXXXX) else stdin_script_file=$(mktemp --tmpdir=$(pwd) ${job_name}_qsub.XXXXXXXXXXXXX) fi cat < /dev/stdin > $stdin_script_file chmod +x $stdin_script_file # use same default name as SGE if [ -z "$job_name" ]; then job_name="STDIN"; fi qsub_cmd="$stdin_script_file" qsub_args="" fi # if we're not in "binary" mode, we also need to parse the submitted script for # additional arguments -- in an ideal world this would be done before parsing # the actual commandline args to give them a higher priority if [ $binary_flag -ne 1 ]; then script_args=`grep '^\#\$ ' < $qsub_cmd |cut -d' ' -f 2- | tr "\n" " "` if [ $? = 0 ] ; then # found some args parse_args $script_args XXcondor_sub_scriptmodeXX fi fi # have a meaningful job name in any case if [ -z "$job_name" ]; then job_name="$(basename $qsub_cmd)" fi # handle job dependency # store effective job deps job_deps="" if [ -n "$hold_jid" ]; then # loop over potentially multiple job deps for jid in $(echo "$hold_jid" |tr "," " ") ; do if [ $(condor_q -long -attributes Owner $jid | grep "$USER" | wc -l) -ge 1 ]; then # job is currently in the queue and owned by $USER -> store job_deps="$job_deps $jid" else # job not owned by this user or not in the queue, but maybe already # done? job_status="$(condor_history $jid |tail -n 1 | awk '{print $6}')" echo "JOBSTATUS: $job_status" >&2 case "$job_status" in C) if [ "$(condor_history -long $jid |grep ExitCode | cut -d= -f2,2)" = " 100" ]; \ then printf "Error: Job dependency $jid completed but exited with code 100.\n"; exit 100; fi;; X) printf "Error: Job dependency $jid has been removed from the queue\n" ; exit 1;; *) printf "Error: Job dependency $jid doesn't exist\n" ; exit 1;; esac fi done fi # compose the names of logfiles log_file="${log_file:-$job_name.log\$(Cluster)}" # SGE also accepts directories for stdout and stderr. in this case it would # place a file with the default name in this directory. This is probably # evaluated at execution time. for condor we can only easily do this at # submission time -- this is not the same -- let's hope it works nevertheless if [ -d "$error_file" ]; then error_file="${error_file}/$job_name.e\$(Cluster)" else error_file="${error_file:-$job_name.e\$(Cluster)}" fi if [ -d "$output_file" ]; then output_file="${output_file}/$job_name.o\$(Cluster)" else output_file="${output_file:-$job_name.o\$(Cluster)}" fi # main submit file submit_file=$(mktemp --tmpdir condor_qsub.XXXXXXXXXXXXX) # write submit file header cat << EOT > $submit_file # condor_qsub call: $@ universe = vanilla #should_transfer_files = NO should_transfer_files = IF_NEEDED #log = $log_file EOT # handle 'binary mode' if [ $binary_flag = 1 ]; then ## go with current shell or POSIX if undefined cat << EOT >> $submit_file executable = $shell arguments = "-c '$qsub_cmd $qsub_args'" EOT qsub_cmd="$shell" else cat << EOT >> $submit_file executable = $shell arguments = $qsub_cmd $qsub_args EOT fi if [ $cwd_flag = 1 ]; then printf "initialdir = $(pwd)\n" >> $submit_file fi if [ -n "$init_dir" ]; then printf "initialdir = $init_dir\n" >> $submit_file fi if [ -n "$merge_streams" ]; then # Check to see if "e" is in the string # echo $merge_streams error_flag=0 if [ `echo ${merge_streams} | grep -c "e"` -gt 0 ]; then error_flag=1 fi # Check to see if "o" is in the string output_flag=0 if [ `echo ${merge_streams} | grep -c "o"` -gt 0 ]; then output_flag=1 fi # If we are to merge the two streams, merge them into output if [ "$output_flag" -eq 1 ] && [ "$error_flag" -eq 1 ]; then # Make sure we have a file name for the output if [ -n "$output_file" ]; then output_file="$job_name.o\$(Cluster)" fi error_file=$output_file fi fi if [ -n "$error_file" ]; then printf "error = $error_file\n" >> $submit_file fi if [ -n "$output_file" ]; then printf "output = $output_file\n" >> $submit_file fi if [ -n "$input_file" ]; then printf "input = $input_file\n" >> $submit_file fi if [ $export_env_flag = 1 ]; then printf "getenv = True\n" >> $submit_file fi if [ -n "$mail_user" ]; then printf "notify_user = $mail_user\n" >> $submit_file fi if [ -n "$mail_config" ]; then if [ "$mail_config" != "$(echo "$mail_config" | sed -e 's/n//')" ]; then printf "notification = Never\n" >> $submit_file elif [ "$mail_config" != "$(echo "$mail_config" | sed -e 's/e//')" ]; then printf "notification = Complete\n" >> $submit_file elif [ "$mail_config" != "$(echo "$mail_config" | sed -e 's/a//')" ]; then printf "notification = Error\n" >> $submit_file fi else printf "notification = Never\n" >> $submit_file fi if [ -n "$priority" ]; then printf "priority = $priority\n" >> $submit_file fi if [ -n "$job_deps" ]; then # in case of job deps, submit the job held and release it via a sentinel printf "hold = True\n" >> $submit_file fi if [ -n "$deferral_date_time" ]; then # Convert [[[[CC]YY]MM]DD]hhmm[.SS] into a format for UNIX's date # Calculate length of string before period pre_period=${deferral_date_time%%.*} original_pre_period_length="${#pre_period}" post_period=${deferral_date_time##*.} #First, check to make sure that lengths are valid if [ "${#pre_period}" -gt 12 ] || [ "${#pre_period}" -lt 4 ] || [ "$((${#pre_period} % 2))" -eq 1 ]; then echo "Invalid Deferral Date. Incorrect lengths. Ignoring..." else # We have a (semi-)valid date. We need to parse [[[[CC]YY]MM]DD]hhmm[.SS] so that the Unix date command can use it. # No need to add_year if [ "${#pre_period}" = 12 ]; then date_year=${pre_period:0:4} pre_period=${pre_period:4:((${#pre_period}-4))} # Chop off CCYY else # Need to add year current_year="$(date +%Y)" if [ "${#pre_period}" = 10 ]; then # Need to add century current_century=${current_year:0:2} date_year="$current_century${pre_period:0:2}" pre_period=${pre_period:2:((${#pre_period}-2))} #Chop off YY else date_year=$current_year fi fi # Need to add month current_month="" # We need this variable for testing to see if MM is present if [ "${#pre_period}" = 8 ]; then date_month=${pre_period:0:2} pre_period=${pre_period:2:((${#pre_period}-2))} # Chop off MM else current_month="$(date +%m)" date_month=$current_month fi # Need to add day current_day="" # We need this variable for testing to see if DD is present if [ "${#pre_period}" = 6 ]; then date_day=${pre_period:0:2} pre_period=${pre_period:2:((${#pre_period}-2))} # Chop off MM else current_day="$(date +%d)" date_day=$current_day fi # We will have a time and minute date_hour=${pre_period:0:2} date_minute=${pre_period:2:2} # Check to see if we have seconds with correct length if [ "${#post_period}" = 2 ]; then date_seconds=$post_period else date_seconds="00" fi current_date="$(date +%s)" date_raw_string="$date_year-$date_month-$date_day" time_raw_string="$date_hour:$date_minute:$date_seconds" # echo Date Raw String: $date_raw_string # echo Time Raw String: $time_raw_string date_from_command="$(date --date="$date_raw_string $time_raw_string" +%s)" if [ $? = 0 ]; then # echo Date from command: $date_from_command # We have an valid date update_day=0 update_month=0 if [ -n "$current_month" ]; then # If MM is not specified # Check to see if we need to run tomorrow if [ -n "$current_day" ]; then # If DD is not specified as well # Get the epoch for time specified if [ "$date_from_command" -lt "$current_date" ]; then update_day=1 fi else # DD is specified # Get the epoch for the day and time specified if [ "$date_from_command" -lt "$current_date" ]; then update_month=1 fi fi fi if [ "$update_day" = 1 ]; then # echo Plus one day "$(date --date="$date_raw_string + 1 day $time_raw_string")" deferral_unix_epoch="$(date --date="$date_raw_string + 1 day $time_raw_string" +%s)" elif [ "$update_month" = 1 ]; then # echo Plus one month "$(date --date="$date_raw_string + 1 month $time_raw_string")" deferral_unix_epoch="$(date --date="$date_raw_string + 1 month $time_raw_string" +%s)" else deferral_unix_epoch=$date_from_command fi ############################################################ # We have two paths we can take here. Either we can use the # Periodic Release Route (which has no maximum time) or we # can use the Deferral Time Route (which does not have to # deal with any sort of periodic release issues). We are # going with the Periodic Release Route because of the lack # of a maximum time. ############################################################ # The Periodic Release Route ############################################################ # printf "hold = True\n" >> $submit_file # printf "+HoldReason = \"Job held until job becomes eligable for execution.\"\n" >> $submit_file # printf "+HoldReasonCode = 15\n" >> $submit_file # # We want to make sure that the reason why the job would be released from being on hold is because we put the job # # on hold on submission and the current time is after the specified eligible execution time. # printf "periodic_release = ( ( time() >= $deferral_unix_epoch ) && ( HoldReasonCode == 15 ) )\n" >> $submit_file ############################################################ # The Deferral Time Route ############################################################ printf "deferral_time = $deferral_unix_epoch\n" >> $submit_file printf "deferral_window = 3153600000\n" >> $submit_file #100 years printf "deferral_prep_time = 0\n" >> $submit_file else echo "Invalid Deferral Date. Incorrect values for date and/or time. Ignoring..." fi fi fi if [ -n "$accounting_group" ]; then printf "+AccountingGroup = \"$accounting_group\"\n" >> $submit_file fi if [ -n "$environment" ]; then # The environment variables for SGE/PBS/Torque are comma separated IFS=',' read -ra environment_variables <<< "$environment" environment_string="" # echo "$environment" # echo "${#environment_variables[@]}" environment_variable="" for environment_variable in "${environment_variables[@]}"; do # echo $environment_variable # We want to escape spaces, single quotes, and double quotes # First, escape double quotes # Second, append =True if there is no equal sign # Third, escape single quotes # Fourth, place single quotes around string if there is whitespace or single quotes # First, escape double quotes environment_variable="$(echo $environment_variable | sed s/\"/\"\"/g)" # Second, append =True if there is no equal sign if [ "$environment_variable" = "${environment_variable/=/}" ]; then # We need to have = # Assume that the variable should be set to true. Not a great assumption, but it's something environment_variable="$environment_variable=True" fi # Third, escape single quotes environment_variable="$(echo $environment_variable | sed s/\'/\'\'/g)" # Fourth, place single quotes around string if there is whitespace or single quotes if [ "$environment_variable" != "${environment_variable/\'/}" ]; then environment_single_quote=1 else environment_single_quote=0 fi environment_no_space="$(echo $environment_variable | sed s/\\s//g)" if [ "$environment_variable" != "$environment_no_space" ]; then environment_whitespace=1 else environment_whitespace=0 fi if [[ "$environment_single_quote" -eq 1 || "$environment_whitespace" -eq 1 ]]; then # Split it up to pre and post equal sign envname=${environment_variable%=*} envvalue=${environment_variable#*=} environment_variable="$envname='$envvalue'" fi if [ -z "$environment_string" ]; then environment_string="$environment_variable" else environment_string="$environment_string $environment_variable" fi done printf "environment = \"$environment_string\"\n" >> $submit_file fi if [ -n "$rerun_flag" ]; then if [ "$(echo $rerun_flag | cut -c 1 | tr a-z A-Z)" = "N" ]; then # https://htcondor-wiki.cs.wisc.edu/index.cgi/wiki?p=HowToAvoidJobRestarts rerun_string="( NumJobStarts == 0 )" # TJ says that JobCurrenStartExecutingDate might be better. Sam is inclind to agree, but needs to do more investigation # rerun_string="( ( CompletionDate =?= 0 ) && ( JobCurrentStartExecutingDate =?= UNDEFINED ) )" if [ -z "$requirements_string" ]; then requirements_string="$rerun_string" else requirements_string="$requirements_string && $rerun_string" fi # on_exit_remove only evaluates after a job has started, which means that NumShadowStarts must be already defined. printf "periodic_remove = ( JobStatus == 1 && NumJobStarts > 0)\n" >> $submit_file # printf "periodic_remove = ( ( CompletionDate =?= 0 ) && ( JobCurrentStartExecutingDate =!= 0 ) )\n" >> $submit_file fi fi if [ -n "$checkpoint_interval" ]; then if [ "$(echo $checkpoint_interval | cut -c 1 | tr A-Z a-z)" = "n" ]; then printf "+WantCheckpoint = False\n" >> $submit_file fi if [ "$(echo $checkpoint_interval | cut -c 1 | tr A-Z a-z)" = "s" ]; then printf "PERIODIC_CHECKPOINT = False\n" >> $submit_file fi # Try to support better checkpointing. # if [ "$checkpoint_interval" = "periodic" ]; then # printf "PERIODIC_CHECKPOINT = 10\n" >> $submit_file ## http://research.cs.wisc.edu/htcondor/manual/v7.9/7_2Setting_up.html#SECTION00827000000000000000 # fi # number_regex='^[0-9]+$' # number_interval_regex='^interval=[0-9]+$' # if [ "$checkpoint_interval" =~ $number_regex ] || [ "$checkpoint_interval" =~ $number_interval_regex ]; then # number_of_checkpoint_minutes=${checkpoint_interval#*=} # printf "PERIODIC_CHECKPOINT = $number_of_checkpoint_minutes\n" >> $submit_file # fi # If we have any checkpointing, we need to be in Standard Universe if [ "$(echo $checkpoint_interval | cut -c 1 | tr A-Z a-z)" != "n" ]; then printf "UNIVERSE = standard\n" >> $submit_file fi fi if [ $sync_flag = 1 ]; then printf "log = $log_file\n" >> $submit_file fi if [ $hold_flag = 1 ]; then printf "hold = True\n" >> $submit_file printf "+HoldReason = \"Job held because job was submitted on hold.\"\n" >> $submit_file printf "+HoldReasonCode = 15\n" >> $submit_file fi if [ -n "$resource_list" ]; then #Parse list. IFS=',' read -ra resources <<< "$resource_list" requirements_string="" num_nodes=0 for resource_type in "${resources[@]}"; do case "$resource_type" in # PBS/Torque arch=*) arch_value=${resource_type#*=} arch_string="\"Arch == $arch_value\"" if [ -z "$requirements_string" ]; then requirements_string="$arch_string" else requirements_string="$requirements_string && $arch_string" fi;; # cput=*) shift;; #ignore # epilogue=*) shift;; #NEED DAGMan or PostCmd file=*) file_value=${resource_type#*=} file_value=`echo $file_value | tr a-z A-Z` #Uppercase printf "request_disk = $file_value\n" >> $submit_file;; host=*) host_value=${resource_type#*=} host_string="\"Machine == $host_value\"" if [ -z "$requirements_string" ]; then requirements_string="$host_string" else requirements_string="$requirements_string && $host_string" fi;; # printf "requirements = \"Machine == $host_value\"\n" >> $submit_file;; mem=*) mem_value=${resource_type#*=} mem_value=`echo $mem_value | tr a-z A-Z` printf "request_memory = $mem_value\n" >> $submit_file;; # nice=*) nice_value=(((${resource_type#*=})*(-1))) #issues with *-1 # printf "JobPriority = $nice_value\n" >> $submit_file;; nodes=*) IFS='+' read -ra node_specs <<< "${resource_type#*=}" number_specs=$(( number_specs + ${#node_specs[@]} )) node_spec_index=0 total_procs=0 for node_spec in "${node_specs[@]}"; do IFS=':' read -ra node_properties <<< "${node_spec}" # echo Node spec: $node_spec # Check to see if we have a number of nodes or a name #node_number=${node_info%%*[!0-9]*} node_num_regex='^[0-9]+$' if [[ ${node_properties[0]} =~ $node_num_regex ]]; then num_nodes[$node_spec_index]=${node_properties[0]} node_name_string[$node_spec_index]="" else node_type_name=${node_properties[0]} num_nodes[$node_spec_index]=1 #Assume that the number of nodes on a named node is 1. node_name_string[$node_spec_index]="( Machine == $node_type_name )" fi ppn_value="" gpu_value="" ppn_array[$node_spec_index]=1 # echo Length of node properties: "${#node_properties[@]}" if [ "${#node_properties[@]}" \> 1 ]; then # echo Entering Node Loop for node_info in "${node_properties[@]:1}"; do # echo "Node Info: ${node_info}" case "$node_info" in #*[!0-9]*) node_number=${node_info##*[!0-9]*};; ppn=*) ppn_value=${node_info#*=} ppn_array[$node_spec_index]=$ppn_value;; gpus=*) gpu_value=${node_info#*=} gpu_array[$node_spec_index]=$gpu_value;; *) if [ -z "${node_name_string[$node_spec_index]}" ]; then node_name_string[$node_spec_index]=" ( Machine == $node_info )" else node_name_string[$node_spec_index]="${node_name_string[$node_spec_index]} && ( Machine == $node_info )" fi;; # Ignore all property cases for now *) other_value=${node_info#*=} other_array[$node_spec_index]=$gpu_value;; esac done fi echo PPN Array: "${ppn_array[$node_spec_index]}" # If there is no value for ppn given, assume that 1 cpu is needed on each node if [ -z "$ppn_value" ]; then ppn_array[$node_spec_index]=1 fi # If there is no value for gpu given, assume that no gpus are needed on a node if [ -n "$gpu_value" ]; then gpu_array[$node_spec_index]=0 fi total_procs=$(( total_procs + num_nodes[$node_spec_index]*${ppn_array[$node_spec_index]} )) if [ -n "${node_name_string[$node_spec_index]}" ]; then node_name_string[$node_spec_index]="\"${node_name_string[$node_spec_index]}\"" fi # echo Total Procs: $total_procs node_spec_index=$node_spec_index+1 done if [ "$total_procs" -gt 1 ]; then printf "request_cpus = $total_procs\n" >> $submit_file # The next few lines are for the start of trying to support MPI commands/parallel universe. # printf "machine_count = $total_procs\n" >> $submit_file # printf "Universe = parallel" >> $submit_file #Easy way to detect vanilla vs. parallel universe fi;; opsys=*) opsys_value=${resource_type#*=} opsys_string="\"OpSys == $opsys_value\"" if [ -z "$requirements_string" ]; then requirements_string="$opsys_string" else requirements_string="$requirements_string && $opsys_string" fi;; # other=*) shift;; #ignore # pcput=*) shift;; #ignore # pmem=*) pmem_value=${resource_type#*=} # printf "RequestMemory = $pmem_value\n" >> $submit_file;; procs=*) procs_value=${resource_type#*=} printf "request_cpus = $procs_value\n" >> $submit_file;; # More potential support for MPI commands/parallel universe. # printf "machine_count = $procs_value\n" >> $submit_file # if [ "$procs_value" -gt 1 ]; then # printf "Universe = parallel" >> $submit_file #Easy way to detect vanilla vs. parallel universe # fi;; # probs_bitmap=*) shift;; #ignore # prologue=*) shift;; #NEED DAGMan or PreCmd # pvmem=*) pvmem_value=${resource_type#*=} #ImageSize? # printf "RequestMemory = $pvmem_value\n" >> $submit_file;; # size=*) shift;;# # software=*) shift;; # vmem=*) shift;; # walltime=*) walltime_value=${resource_type#*=} #ignore # print "maxjobretirementtime = $walltime_value" >> $submit_file;; # SGE # Some of these SGE options may overlap with PBS/Torque options. Further investigation is needed. # We have added comments about what each option is supposed to do. the "shift" commands here are just place holders. # qname=*) shift;; # name of the queue. The default value is Template. # hostlist=*) shift;; # A list of host names or host group names on which the queue runs. The default value is NONE. # hostname=*) shift;; #??? # notify=*) shift;; # ??? # calendar=*) shift;; # Specifies the calendar associated with this queue, if there is one. The default value for this parameter is NONE. # min_cpu_interval=*) shift;; # time between two automatic checkpoints in the case of transparent checkpointing jobs. # ckpt_list=*) shift;; # The list of administrator-defined checkpointing interface names. The default is NONE. # tmpdir=*) shift;; # temporary directory path. The default value is /tmp. # seq_no=*) shift;; # sequence number of the queue. The default value is 0. # slots=*) shift;; # maximum number of concurrently executing jobs allowed in the queue. ## http://docs.oracle.com/cd/E24901_01/doc.62/e21978/configuration.htm#autoId30 # s_rt=*) shift;; # soft runtime limit # h_rt=*) shift;; # hard runtime limit # s_cpu=*) shift;; # per-process CPU time limit in seconds # h_cpu=*) shift;; # per-job CPU time limit in seconds # s_data=*) shift;; # per-process maximum memory limit in bytes # h_data=*) shift;; # per-job maximum memory limit in bytes # s_stack=*) shift;; # soft limit for the stack size # h_stack=*) shift;; # hard limit for the stack size # s_core=*) shift;; # soft limit for the per-process maximum core file size in bytes. # h_core=*) shift;; # hard limit for the per-process maximum core file size in bytes. # s_rss=*) shift;; # soft resident set size limit. # h_rss=*) shift;; # hard resident set size limit. # s_vmem=*) shift;; # per-process maximum memory limit in bytes. The same as s_data. If both are set, the minimum is used. # h_vmem=*) shift;; # per-job maximum memory limit in bytes. The same as h_data. If both are set, the minimum is used. # s_fsize=*) shift;; # soft limit for the number of disk blocks that this job can create. # h_fsize=*) shift;; # hard limit for the number of disk blocks that this job can create. esac done fi # Make printing the requirements string one of the last commands before entering the queue command if [ -n "$requirements_string" ]; then printf "requirements = $requirements_string\n" >> $submit_file fi # Transfer files if [ -n "$transfer_files" ]; then read -ra stages <<< "$transfer_files" set_should_transfer_files=0 for particular_staging in "${stages[@]}"; do case "$particular_staging" in stagein=*) #Transfer files in. First remove quotes if present input_file_transfer_list=${particular_staging#*stagein=} input_file_transfer_list="$(echo $input_file_transfer_list | sed s/\'//g)" input_file_list="" # echo Input File Transfer List: $input_file_transfer_list IFS=',' read -ra input_file_array <<< "$input_file_transfer_list" # Iterate over each file for input_file_full_path in "${input_file_array[@]}"; do input_file_name=${input_file_full_path#*:} # Remove everything before the : if [ -z "$input_file_list" ]; then input_file_list=$input_file_name else input_file_list="$input_file_list,$input_file_name" fi done printf "transfer_input_files = $input_file_list\n" >> $submit_file;; stageout=*) #Transfer files out. First remove quotes if present output_file_transfer_list=${particular_staging#*stageout=} output_file_transfer_list="$(echo $output_file_transfer_list | sed s/\'//g)" output_file_list="" # echo Output File Transfer List: $output_file_transfer_list IFS=',' read -ra output_file_array <<< "$output_file_transfer_list" # Iterate over each file for output_file_full_path in "${output_file_array[@]}"; do output_file_name=${output_file_full_path#*:} # Remove everything before the : if [ -z "$output_file_list" ]; then output_file_list=$output_file_name else output_file_list="$output_file_list,$output_file_name" fi done printf "when_to_transfer_output = ON_EXIT \n" >> $submit_file printf "transfer_output_files = $output_file_list \n" >> $submit_file;; esac done fi # transfer_executable = False for not transfering executable # -k # handle array jobs if [ -n "$taskid_first" ]; then for taskid in $(seq $taskid_first $taskid_step $taskid_last); do printf "environment = \"$environment SGE_TASK_ID=$taskid SGE_TASK_FIRST=$taskid_first SGE_TASK_STEPSIZE=$taskid_step SGE_TASK_LAST=$taskid_last\"\n" >> $submit_file if [ -n "$error_file" ]; then printf "error = $error_file.$taskid\n" >> $submit_file fi if [ -n "$output_file" ]; then printf "output = $output_file.$taskid\n" >> $submit_file fi # queue this job printf "Queue\n" >> $submit_file done else if [ "$total_procs" -gt 1 ] && [ "$number_specs" -gt 1 ]; then number_specs_max=$(( number_specs - 1)) for node_spec_index in $(seq 0 $number_specs_max ); do echo Node Specification Index: $node_spec_index if [ -n "${node_name_string[$node_spec_index]}" ]; then #echo ${node_name_string[$node_spec_index]} #host_string="\"Machine == ${node_name_string[$node_spec_index]}\"" host_string=${node_name_string[$node_spec_index]} if [ -z "$requirements_string" ]; then queued_requirements_string="$host_string" else queued_requirements_string="$requirements_string && $host_string" fi # This will change the requirements for each collection of nodes printf "requirements = $queued_requirements_string\n" >> $submit_file fi num_queued_procs=$(( num_nodes[$node_spec_index] * ppn_array[$node_spec_index] )) printf "Queue $num_queued_procs\n" >> $submit_file done else # unconditional queuing printf "Queue\n" >> $submit_file fi fi # Done with creating to submission file if [ $test_qsub_flag = 0 ]; then # actually submit the job, printing only the cluster id #printf "Trying to submit job.\n" cluster_id_with_period="$(condor_submit $interactive_option < $submit_file | grep 'cluster' | awk '{print $6}')" cluster_id=${cluster_id_with_period%.} # Hanke's original (V0.1) submit line is below #cluster_id="$(condor_submit -verbose $interactive_option < $submit_file |grep '^** Proc' || true)" #printf "Post condor_submit\n" if [ -z "$cluster_id" ]; then printf "Job submission failed.\n" [ $condor_keep_files -eq 0 ] && rm $submit_file || true exit 1 fi cluster_id="$(echo "$cluster_id" | head -n1 | cut -d ' ' -f 3,3 | cut -d. -f1,1)" # if [ $sync_flag = 1 ]; then # echo "Attempting to sync" ## # condor_q needs to be called to make sure to wait that the job has been scheduled to run. # condor_wait $log_file $cluster_id # fi #cleanup [ $condor_keep_files -eq 0 ] && rm $submit_file || true # deal with local job sentinel if [ -n "$job_deps" ]; then sentinel_file="$(mktemp --tmpdir cluster${cluster_id}_sentinel.XXXXXXXXXXXXX)" create_dep_sentinel $sentinel_file "$job_deps" "$cluster_id" condor_submit < $sentinel_file.submit > /dev/null [ $condor_keep_files -eq 0 ] && rm $sentinel_file.submit || true fi else if [ $print_flag = 1 ]; then echo "$(cat $submit_file)" else echo "$submit_file" fi [ $condor_keep_files -eq 0 ] && rm $submit_file || true fi if [ $no_output_batch_id_flag = 0 ]; then printf "Your job $cluster_id (\"$job_name\") has been submitted\n" fi exit 0