t/perf/README

   1 Git performance tests
   2 =====================
   3
   4 This directory holds performance testing scripts for git tools.  The
   5 first part of this document describes the various ways in which you
   6 can run them.
   7
   8 When fixing the tools or adding enhancements, you are strongly
   9 encouraged to add tests in this directory to cover what you are
  10 trying to fix or enhance.  The later part of this short document
  11 describes how your test scripts should be organized.
  12
  13
  14 Running Tests
  15 -------------
  16
  17 The easiest way to run tests is to say "make".  This runs all
  18 the tests on the current git repository.
  19
  20     === Running 2 tests in this tree ===
  21     [...]
  22     Test                                     this tree
  23     ---------------------------------------------------------
  24     0001.1: rev-list --all                   0.54(0.51+0.02)
  25     0001.2: rev-list --all --objects         6.14(5.99+0.11)
  26     7810.1: grep worktree, cheap regex       0.16(0.16+0.35)
  27     7810.2: grep worktree, expensive regex   7.90(29.75+0.37)
  28     7810.3: grep --cached, cheap regex       3.07(3.02+0.25)
  29     7810.4: grep --cached, expensive regex   9.39(30.57+0.24)
  30
  31 Output format is in seconds "Elapsed(User + System)"
  32
  33 You can compare multiple repositories and even git revisions with the
  34 'run' script:
  35
  36     $ ./run . origin/next /path/to/git-tree p0001-rev-list.sh
  37
  38 where . stands for the current git tree.  The full invocation is
  39
  40     ./run [<revision|directory>...] [--] [<test-script>...]
  41
  42 A '.' argument is implied if you do not pass any other
  43 revisions/directories.
  44
  45 You can also manually test this or another git build tree, and then
  46 call the aggregation script to summarize the results:
  47
  48     $ ./p0001-rev-list.sh
  49     [...]
  50     $ ./run /path/to/other/git -- ./p0001-rev-list.sh
  51     [...]
  52     $ ./aggregate.perl . /path/to/other/git ./p0001-rev-list.sh
  53
  54 aggregate.perl has the same invocation as 'run', it just does not run
  55 anything beforehand.
  56
  57 You can set the following variables (also in your config.mak):
  58
  59     GIT_PERF_REPEAT_COUNT
  60         Number of times a test should be repeated for best-of-N
  61         measurements.  Defaults to 3.
  62
  63     GIT_PERF_MAKE_OPTS
  64         Options to use when automatically building a git tree for
  65         performance testing. E.g., -j6 would be useful. Passed
  66         directly to make as "make $GIT_PERF_MAKE_OPTS".
  67
  68     GIT_PERF_MAKE_COMMAND
  69         An arbitrary command that'll be run in place of the make
  70         command, if set the GIT_PERF_MAKE_OPTS variable is
  71         ignored. Useful in cases where source tree changes might
  72         require issuing a different make command to different
  73         revisions.
  74
  75         This can be (ab)used to monkeypatch or otherwise change the
  76         tree about to be built. Note that the build directory can be
  77         re-used for subsequent runs so the make command might get
  78         executed multiple times on the same tree, but don't count on
  79         any of that, that's an implementation detail that might change
  80         in the future.
  81
  82     GIT_PERF_REPO
  83     GIT_PERF_LARGE_REPO
  84         Repositories to copy for the performance tests.  The normal
  85         repo should be at least git.git size.  The large repo should
  86         probably be about linux.git size for optimal results.
  87         Both default to the git.git you are running from.
  88
  89     GIT_PERF_EXTRA
  90         Boolean to enable additional tests. Most test scripts are
  91         written to detect regressions between two versions of Git, and
  92         the output will compare timings for individual tests between
  93         those versions. Some scripts have additional tests which are not
  94         run by default, that show patterns within a single version of
  95         Git (e.g., performance of index-pack as the number of threads
  96         changes). These can be enabled with GIT_PERF_EXTRA.
  97
  98 You can also pass the options taken by ordinary git tests; the most
  99 useful one is:
 100
 101 --root=<directory>::
 102         Create "trash" directories used to store all temporary data during
 103         testing under <directory>, instead of the t/ directory.
 104         Using this option with a RAM-based filesystem (such as tmpfs)
 105         can massively speed up the test suite.
 106
 107
 108 Naming Tests
 109 ------------
 110
 111 The performance test files are named as:
 112
 113         pNNNN-commandname-details.sh
 114
 115 where N is a decimal digit.  The same conventions for choosing NNNN as
 116 for normal tests apply.
 117
 118
 119 Writing Tests
 120 -------------
 121
 122 The perf script starts much like a normal test script, except it
 123 sources perf-lib.sh:
 124
 125         #!/bin/sh
 126         #
 127         # Copyright (c) 2005 Junio C Hamano
 128         #
 129
 130         test_description='xxx performance test'
 131         . ./perf-lib.sh
 132
 133 After that you will want to use some of the following:
 134
 135         test_perf_fresh_repo    # sets up an empty repository
 136         test_perf_default_repo  # sets up a "normal" repository
 137         test_perf_large_repo    # sets up a "large" repository
 138
 139         test_perf_default_repo sub  # ditto, in a subdir "sub"
 140
 141         test_checkout_worktree  # if you need the worktree too
 142
 143 At least one of the first two is required!
 144
 145 You can use test_expect_success as usual. In both test_expect_success
 146 and in test_perf, running "git" points to the version that is being
 147 perf-tested. The $MODERN_GIT variable points to the git wrapper for the
 148 currently checked-out version (i.e., the one that matches the t/perf
 149 scripts you are running).  This is useful if your setup uses commands
 150 that only work with newer versions of git than what you might want to
 151 test (but obviously your new commands must still create a state that can
 152 be used by the older version of git you are testing).
 153
 154 For actual performance tests, use
 155
 156         test_perf 'descriptive string' '
 157                 command1 &&
 158                 command2
 159         '
 160
 161 test_perf spawns a subshell, for lack of better options.  This means
 162 that
 163
 164 * you _must_ export all variables that you need in the subshell
 165
 166 * you _must_ flag all variables that you want to persist from the
 167   subshell with 'test_export':
 168
 169         test_perf 'descriptive string' '
 170                 foo=$(git rev-parse HEAD) &&
 171                 test_export foo
 172         '
 173
 174   The so-exported variables are automatically marked for export in the
 175   shell executing the perf test.  For your convenience, test_export is
 176   the same as export in the main shell.
 177
 178   This feature relies on a bit of magic using 'set' and 'source'.
 179   While we have tried to make sure that it can cope with embedded
 180   whitespace and other special characters, it will not work with
 181   multi-line data.
 182
 183 Rather than tracking the performance by run-time as `test_perf` does, you
 184 may also track output size by using `test_size`. The stdout of the
 185 function should be a single numeric value, which will be captured and
 186 shown in the aggregated output. For example:
 187
 188         test_perf 'time foo' '
 189                 ./foo >foo.out
 190         '
 191
 192         test_size 'output size'
 193                 wc -c <foo.out
 194         '
 195
 196 might produce output like:
 197
 198         Test                origin           HEAD
 199         -------------------------------------------------------------
 200         1234.1 time foo     0.37(0.79+0.02)  0.26(0.51+0.02) -29.7%
 201         1234.2 output size             4.3M             3.6M -14.7%
 202
 203 The item being measured (and its units) is up to the test; the context
 204 and the test title should make it clear to the user whether bigger or
 205 smaller numbers are better. Unlike test_perf, the test code will only be
 206 run once, since output sizes tend to be more deterministic than timings.