| 1 | #!/usr/bin/env bash
 | 
| 2 | #
 | 
| 3 | # Do a quick test of virtual memory.
 | 
| 4 | #
 | 
| 5 | # Note: This is probably very similar to max RSS of
 | 
| 6 | # testdata/osh-runtime/hello-world.sh, so it could be retired.
 | 
| 7 | #
 | 
| 8 | # Usage:
 | 
| 9 | #   benchmarks/vm-baseline.sh <function name>
 | 
| 10 | 
 | 
| 11 | set -o nounset
 | 
| 12 | set -o pipefail
 | 
| 13 | set -o errexit
 | 
| 14 | 
 | 
| 15 | REPO_ROOT=$(cd "$(dirname $0)/.."; pwd)
 | 
| 16 | 
 | 
| 17 | source test/common.sh  # log
 | 
| 18 | source benchmarks/common.sh
 | 
| 19 | source test/tsv-lib.sh  # tsv2html
 | 
| 20 | 
 | 
| 21 | readonly BASE_DIR=_tmp/vm-baseline
 | 
| 22 | 
 | 
| 23 | measure() {
 | 
| 24 |   local provenance=$1
 | 
| 25 |   local host_job_id=$2
 | 
| 26 |   local base_dir=${3:-_tmp/vm-baseline}
 | 
| 27 | 
 | 
| 28 |   local out_dir="$base_dir/$host_job_id"
 | 
| 29 |   mkdir -p $out_dir
 | 
| 30 | 
 | 
| 31 |   # TODO:
 | 
| 32 |   # print-tasks should:
 | 
| 33 |   # - use the whole shell path like _bin/osh
 | 
| 34 |   # - the host name should be a column
 | 
| 35 |   # - the join ID can be a file, and construct the task name from that
 | 
| 36 |   # - Then maybe use tsv_columns_from_files.py like we do with cachegrind
 | 
| 37 | 
 | 
| 38 |   # - should not
 | 
| 39 |   #   - get shell name from the filename
 | 
| 40 |   #   - get host name from the filename
 | 
| 41 |   # - should use TSV files
 | 
| 42 | 
 | 
| 43 |   # Fourth column is the shell.
 | 
| 44 |   cat $provenance | filter-provenance "${SHELLS[@]}" "$OSH_CPP_REGEX" |
 | 
| 45 |   while read _ _ _ sh_path shell_hash; do
 | 
| 46 | 
 | 
| 47 |     local sh_name
 | 
| 48 |     sh_name=$(basename $sh_path)
 | 
| 49 | 
 | 
| 50 |     local out="$out_dir/${sh_name}-${shell_hash}.txt"
 | 
| 51 | 
 | 
| 52 |     # There is a race condition on the status but sleep helps.
 | 
| 53 |     # Bug fix: ALIVE to prevent exec optimization in OSH and zsh.
 | 
| 54 |     $sh_path -c 'sleep 0.001; cat /proc/$$/status; echo ALIVE' > $out
 | 
| 55 |   done
 | 
| 56 | 
 | 
| 57 |   echo
 | 
| 58 |   echo "$out_dir:"
 | 
| 59 |   ls -l $out_dir
 | 
| 60 | }
 | 
| 61 | 
 | 
| 62 | # Run a single file through stage 1 and report.
 | 
| 63 | demo() {
 | 
| 64 |   local -a job_dirs=($BASE_DIR/lisa.2017-*)
 | 
| 65 |   local dir1=$BASE_DIR/stage1
 | 
| 66 |   local dir2=$BASE_DIR/stage2
 | 
| 67 | 
 | 
| 68 |   mkdir -p $dir1 $dir2
 | 
| 69 |   
 | 
| 70 |   benchmarks/virtual_memory.py baseline ${job_dirs[-1]} \
 | 
| 71 |     > $dir1/vm-baseline.tsv
 | 
| 72 | 
 | 
| 73 |   benchmarks/report.R vm-baseline $dir1 $dir2
 | 
| 74 | }
 | 
| 75 | 
 | 
| 76 | # Combine CSV files.
 | 
| 77 | stage1() {
 | 
| 78 |   local raw_dir=${1:-$BASE_DIR/raw}
 | 
| 79 |   local single_machine=${2:-}
 | 
| 80 | 
 | 
| 81 |   local out=$BASE_DIR/stage1
 | 
| 82 |   mkdir -p $out
 | 
| 83 | 
 | 
| 84 |   local base_dir=
 | 
| 85 | 
 | 
| 86 |   local -a raw=()
 | 
| 87 | 
 | 
| 88 |   if test -n "$single_machine"; then
 | 
| 89 |     base_dir=_tmp/vm-baseline
 | 
| 90 |     local -a m1=( $base_dir/$single_machine.* )
 | 
| 91 |     raw+=( ${m1[-1]} )
 | 
| 92 |   else
 | 
| 93 |     base_dir=../benchmark-data/vm-baseline
 | 
| 94 |     # Globs are in lexicographical order, which works for our dates.
 | 
| 95 |     local -a m1=( $base_dir/$MACHINE1.* )
 | 
| 96 |     local -a m2=( $base_dir/$MACHINE2.* )
 | 
| 97 | 
 | 
| 98 |     raw+=( ${m1[-1]} ${m2[-1]} )
 | 
| 99 |   fi
 | 
| 100 | 
 | 
| 101 |   benchmarks/virtual_memory.py baseline "${raw[@]}" \
 | 
| 102 |     | tee $out/vm-baseline.tsv
 | 
| 103 | }
 | 
| 104 | 
 | 
| 105 | print-report() {
 | 
| 106 |   local in_dir=$1
 | 
| 107 | 
 | 
| 108 |   benchmark-html-head 'Virtual Memory Baseline'
 | 
| 109 | 
 | 
| 110 |   cat <<EOF
 | 
| 111 |   <body class="width60">
 | 
| 112 |     <p id="home-link">
 | 
| 113 |       <a href="/">oilshell.org</a>
 | 
| 114 |     </p>
 | 
| 115 | EOF
 | 
| 116 | 
 | 
| 117 |   cmark << 'EOF'
 | 
| 118 | ## Virtual Memory Baseline
 | 
| 119 | 
 | 
| 120 | Source code: [oil/benchmarks/vm-baseline.sh](https://github.com/oilshell/oil/tree/master/benchmarks/vm-baseline.sh)
 | 
| 121 | 
 | 
| 122 | ### Memory Used at Startup (MB)
 | 
| 123 | 
 | 
| 124 | Memory usage is measured in MB (powers of 10), not MiB (powers of 2).
 | 
| 125 | 
 | 
| 126 | EOF
 | 
| 127 |   # highlight OSH lines
 | 
| 128 |   tsv2html --css-class-pattern 'special ^osh' $in_dir/vm-baseline.tsv
 | 
| 129 | 
 | 
| 130 |   # R code doesn't generate this
 | 
| 131 |   if false; then
 | 
| 132 |     cmark <<< '### Shell and Host Details'
 | 
| 133 | 
 | 
| 134 |     tsv2html $in_dir/shells.tsv
 | 
| 135 |     tsv2html $in_dir/hosts.tsv
 | 
| 136 |   fi
 | 
| 137 | 
 | 
| 138 |   cat <<EOF
 | 
| 139 |   </body>
 | 
| 140 | </html>
 | 
| 141 | EOF
 | 
| 142 | }
 | 
| 143 | 
 | 
| 144 | 
 | 
| 145 | #
 | 
| 146 | # Other
 | 
| 147 | #
 | 
| 148 | 
 | 
| 149 | soil-run() {
 | 
| 150 |   ### Run it on just this machine, and make a report
 | 
| 151 | 
 | 
| 152 |   rm -r -f $BASE_DIR
 | 
| 153 |   mkdir -p $BASE_DIR
 | 
| 154 | 
 | 
| 155 |   local -a osh_bin=( $OSH_CPP_NINJA_BUILD )
 | 
| 156 |   ninja "${osh_bin[@]}"
 | 
| 157 | 
 | 
| 158 |   local single_machine='no-host'
 | 
| 159 | 
 | 
| 160 |   local job_id
 | 
| 161 |   job_id=$(benchmarks/id.sh print-job-id)
 | 
| 162 | 
 | 
| 163 |   benchmarks/id.sh shell-provenance-2 \
 | 
| 164 |     $single_machine $job_id _tmp \
 | 
| 165 |     bash dash bin/osh "${osh_bin[@]}"
 | 
| 166 | 
 | 
| 167 |   # TODO: measure* should use print-tasks | run-tasks
 | 
| 168 |   local provenance=_tmp/provenance.txt
 | 
| 169 |   local host_job_id="$single_machine.$job_id"
 | 
| 170 | 
 | 
| 171 |   measure $provenance $host_job_id
 | 
| 172 | 
 | 
| 173 |   # Make it run on one machine
 | 
| 174 |   stage1 '' $single_machine 
 | 
| 175 | 
 | 
| 176 |   benchmarks/report.sh stage2 $BASE_DIR
 | 
| 177 |   benchmarks/report.sh stage3 $BASE_DIR
 | 
| 178 | }
 | 
| 179 | 
 | 
| 180 | "$@"
 |