2 # Copyright (C) 2019-2021 all contributors <meta@public-inbox.org>
3 # License: AGPL-3.0+ <https://www.gnu.org/licenses/agpl-3.0.txt>
5 # Parallel test runner which preloads code and reuses worker processes
6 # to give a nice speedup over prove(1). It also generates per-test
7 # .log files (similar to automake tests).
9 # *.t files run by this should not rely on global state.
11 # Usage: $PERL -I lib -w t/run.perl -j4
12 # Or via prove(1): prove -lvw t/run.perl :: -j4
15 use IO::Handle; # ->autoflush
16 use PublicInbox::TestCommon;
17 use PublicInbox::Spawn;
18 use Getopt::Long qw(:config gnu_getopt no_ignore_case auto_abbrev);
21 use POSIX qw(_POSIX_PIPE_BUF WNOHANG);
26 our $log_suffix = '.log';
27 my ($shuffle, %pids, @err);
28 GetOptions('j|jobs=i' => \$jobs,
29 'repeat=i' => \$repeat,
30 'log=s' => \$log_suffix,
31 's|shuffle' => \$shuffle,
32 ) or die "Usage: $0 [-j JOBS] [--log=SUFFIX] [--repeat RUNS]";
33 if (($ENV{TEST_RUN_MODE} // 2) == 0) {
34 die "$0 is not compatible with TEST_RUN_MODE=0\n";
36 my @tests = scalar(@ARGV) ? @ARGV : glob('t/*.t');
37 open my $cwd_fh, '<', '.' or die "open .: $!";
38 open my $OLDOUT, '>&STDOUT' or die "dup STDOUT: $!";
39 open my $OLDERR, '>&STDERR' or die "dup STDERR: $!";
40 $OLDOUT->autoflush(1);
41 $OLDERR->autoflush(1);
43 my ($run_log, $tmp_rl);
44 my $rl = $ENV{TEST_RUN_LOG};
46 $tmp_rl = File::Temp->new(CLEANUP => 1);
47 $rl = $tmp_rl->filename;
49 open $run_log, '+>>', $rl or die "open $rl: $!";
50 $run_log->autoflush(1); # one reader, many writers
52 key2sub($_) for @tests; # precache
54 my ($for_destroy, $lei_env, $lei_daemon_pid, $owner_pid);
55 if (!$ENV{TEST_LEI_DAEMON_PERSIST_DIR} &&
56 (PublicInbox::Spawn->can('recv_cmd4') ||
57 eval { require Socket::MsgHdr })) {
59 ($lei_env->{XDG_RUNTIME_DIR}, $for_destroy) = tmpdir;
60 $ENV{TEST_LEI_DAEMON_PERSIST_DIR} = $lei_env->{XDG_RUNTIME_DIR};
61 run_script([qw(lei daemon-pid)], $lei_env, { 1 => \$lei_daemon_pid });
62 chomp $lei_daemon_pid;
63 $lei_daemon_pid =~ /\A[0-9]+\z/ or die "no daemon pid: $lei_daemon_pid";
64 kill(0, $lei_daemon_pid) or die "kill $lei_daemon_pid: $!";
65 if (my $t = $ENV{GNU_TAIL}) {
66 system("$t --pid=$lei_daemon_pid -F " .
67 "$lei_env->{XDG_RUNTIME_DIR}/lei/errors.log >&2 &");
69 if (my $strace_cmd = $ENV{STRACE_CMD}) {
70 system("$strace_cmd -p $lei_daemon_pid &");
77 } elsif (open(my $prove_state, '<', '.prove') && eval { require YAML::XS }) {
78 # reuse "prove --state=save" data to start slowest tests, first
79 my $state = YAML::XS::Load(do { local $/; <$prove_state> });
80 my $t = $state->{tests};
82 ($t->{$b}->{elapsed} // 0) <=> ($t->{$a}->{elapsed} // 0)
86 our $tb = Test::More->builder;
93 our ($worker, $worker_test);
96 $? = 255 if $? == 0 && !$tb->is_passing;
97 my $status = $? ? 'not ok' : 'ok';
98 chdir($cwd_fh) or DIE "fchdir: $!";
99 if ($log_suffix ne '') {
100 my $log = $worker_test;
101 $log =~ s/\.t\z/$log_suffix/;
103 if (open my $fh, '<', $log) {
104 my @not_ok = grep(!/^(?:ok |[ \t]*#)/ms, <$fh>);
105 my $last = $not_ok[-1] // '';
106 pop @not_ok if $last =~ /^[0-9]+\.\.[0-9]+$/;
107 my $pfx = "# $log: ";
108 print $OLDERR map { $pfx.$_ } @not_ok;
109 seek($fh, 0, SEEK_SET) or die "seek: $!";
111 # show unique skip texts and the number of times
112 # each text was skipped
114 my @sk = (<$fh> =~ m/^ok [0-9]+ (# skip [^\n]+)/mgs);
117 my @err = grep { !$nr{$_}++ } @sk;
118 print $OLDERR "$pfx$_ ($nr{$_})\n" for @err;
119 $skip = ' # total skipped: '.scalar(@sk);
122 print $OLDERR "could not open: $log: $!\n";
124 print $OLDOUT "$status $worker_test$skip\n";
128 # Test::Builder or Test2::Hub may call exit() from plan(skip_all => ...)
129 END { test_status() if (defined($worker_test) && $worker == $$) }
133 syswrite($run_log, "$$ $test\n");
135 if ($log_suffix ne '') {
137 $log =~ s/\.[^\.]+\z/$log_suffix/ or DIE "can't log for $test";
138 open $log_fh, '>', $log or DIE "open $log: $!";
139 $log_fh->autoflush(1);
140 $tb->output($log_fh);
141 $tb->failure_output($log_fh);
142 $tb->todo_output($log_fh);
143 open STDOUT, '>&', $log_fh or DIE "1>$log: $!";
144 open STDERR, '>&', $log_fh or DIE "2>$log: $!";
146 $worker_test = $test;
149 $worker_test = undef;
150 push @err, "$test ($?)" if $?;
153 sub UINT_SIZE () { 4 }
155 # worker processes will SIGUSR1 the producer process when it
156 # sees EOF on the pipe. On FreeBSD 11.2 and Perl 5.30.0,
157 # sys/ioctl.ph gives the wrong value for FIONREAD().
159 my $eof; # we stop respawning if true
161 my $start_worker = sub {
162 my ($j, $rd, $wr, $todo) = @_;
163 my $pid = fork // DIE "fork: $!";
168 my $r = sysread($rd, my $buf, UINT_SIZE);
174 DIE "short read $r" if $r != UINT_SIZE;
175 my $t = unpack('I', $buf);
176 run_test($todo->[$t]);
179 kill 'USR1', $producer if !$eof; # sets $eof in $producer
180 DIE join('', map { "E: $_\n" } @err) if @err;
187 # negative $repeat means loop forever:
188 for (my $i = $repeat; $i != 0; $i--) {
189 my @todo = $shuffle ? List::Util::shuffle(@tests) : @tests;
191 # single-producer, multi-consumer queue relying on POSIX pipe semantics
192 # POSIX.1-2008 stipulates a regular file should work, but Linux <3.14
193 # had broken read(2) semantics according to the read(2) manpage
194 pipe(my ($rd, $wr)) or DIE "pipe: $!";
196 # fill the queue before forking so children can start earlier
197 my $n = (_POSIX_PIPE_BUF / UINT_SIZE);
199 print $wr join('', map { pack('I', $_) } (0..$#todo)) or DIE;
201 } else { # write what we can...
203 print $wr join('', map { pack('I', $_) } (0..$n)) or DIE;
204 $n += 1; # and send more ($n..$#todo), later
207 local $SIG{USR1} = sub { $eof = 1 };
210 my $flags = $sig ? WNOHANG : 0;
212 my $pid = waitpid(-1, $flags) or return;
214 my $j = delete $pids{$pid};
216 push @err, "reaped unknown $pid ($?)";
220 seek($run_log, 0, SEEK_SET);
221 chomp(my @t = grep(/^$pid /, <$run_log>));
222 $t[0] //= "$pid unknown";
223 push @err, "job[$j] ($?) PID=$t[-1]";
225 # skip_all can exit(0), respawn if needed:
227 print $OLDERR "# respawning job[$j]\n";
228 $start_worker->($j, $rd, $wr, \@todo);
233 # start the workers to consume the queue
234 for (my $j = 0; $j < $jobs; $j++) {
235 $start_worker->($j, $rd, $wr, \@todo);
238 local $SIG{CHLD} = $sigchld;
239 # too many tests to fit in the pipe before starting workers,
240 # send the rest now the workers are running
241 print $wr join('', map { pack('I', $_) } ($n..$#todo)) or DIE;
245 $sigchld->(0) while scalar(keys(%pids));
246 DIE join('', map { "E: $_\n" } @err) if @err;
249 print $OLDOUT "1..".($repeat * scalar(@tests))."\n" if $repeat >= 0;
250 if ($lei_env && $$ == $owner_pid) {
251 my $opt = { 1 => $OLDOUT, 2 => $OLDERR };
252 run_script([qw(lei daemon-kill)], $lei_env, $opt);