Fixed bug #49664: --round-robin does not complete.

This commit is contained in:
Ole Tange 2016-12-14 15:04:57 +01:00
parent 2dcc9db3ca
commit 987cc8c30b
5 changed files with 225 additions and 199 deletions

View file

@ -453,7 +453,9 @@ sub spreadstdin {
}
}
}
if(not $anything_written and not eof($in)) {
if(not $anything_written
and not eof($in)
and not $Global::no_autoexpand_block) {
# Nothing was written - maybe the block size < record size?
# Increase blocksize exponentially up to 2GB-1 (2GB causes problems)
if($blocksize < $two_gb) {
@ -555,6 +557,7 @@ sub nindex {
{
my @robin_queue;
my $sleep = 1;
sub round_robin_write {
# Input:
@ -568,31 +571,41 @@ sub nindex {
# Returns:
# $something_written = amount of bytes written
my ($header_ref,$buffer_ref,$recstart,$recend,$endpos) = @_;
my $something_written = 0;
my $written = 0;
my $block_passed = 0;
my $sleep = 1;
while(not $block_passed) {
# Continue flushing existing buffers
# until one is empty and a new block is passed
if(@robin_queue) {
# Rotate queue once so new blocks get a fair chance
# to be given to another block
push @robin_queue, shift @robin_queue;
} else {
# Make a queue to spread the blocks evenly
if(not @robin_queue) {
push @robin_queue, (sort { $a->seq() <=> $b->seq() }
values %Global::running);
}
while(my $job = shift @robin_queue) {
do {
$written = 0;
for my $job (@robin_queue) {
if($job->block_length() > 0) {
$something_written += $job->non_blocking_write();
$written += $job->non_blocking_write();
} else {
$job->set_block($header_ref,$buffer_ref,$endpos,$recstart,$recend);
$job->set_block($header_ref,$buffer_ref,
$endpos,$recstart,$recend);
$block_passed = 1;
$job->set_virgin(0);
$something_written += $job->non_blocking_write();
$written += $job->non_blocking_write();
last;
}
}
if($written) {
$sleep = $sleep/1.5+0.001;
}
} while($written and not $block_passed);
$sleep = ::reap_usleep($sleep);
}
return $something_written;
return $written;
}
}
@ -6753,7 +6766,9 @@ sub non_blocking_write {
substr($self->{'block'},$self->{'block_pos'}));
if (!defined($rv) && $! == EAGAIN) {
# would block - but would have written
$something_written = 1;
$something_written = 0;
# avoid triggering auto expanding block
$Global::no_autoexpand_block ||= 1;
} elsif ($self->{'block_pos'}+$rv != $self->{'block_length'}) {
# incomplete write
# Remove the written part

View file

@ -132,5 +132,21 @@ par_compress_fail() {
seq 12 | parallel --compress -k seq {} 1000000 | md5sum
}
par_distribute_input_by_ability() {
echo "### Distribute input to jobs that are ready"
echo "Job-slot n is 50% slower than n+1, so the order should be 1..7"
seq 20000000 |
parallel --tagstring {#} -j7 --block 300k --round-robin --pipe \
'pv -qL{=$_=$job->seq()**3+9=}0000 |wc -c' |
sort -nk2 | field 1
}
par_round_robin_blocks() {
echo "bug #49664: --round-robin does not complete"
seq 20000000 | parallel --block 10M --round-robin --pipe wc -c | wc -l
}
export -f $(compgen -A function | grep par_)
compgen -A function | grep par_ | sort | parallel -j6 --tag -k '{} 2>&1'

View file

@ -3,73 +3,82 @@
# Simple jobs that never fails
# Each should be taking 3-10s and be possible to run in parallel
# I.e.: No race conditions, no logins
cat <<'EOF' | sed -e 's/;$/; /;s/$SERVER1/'$SERVER1'/;s/$SERVER2/'$SERVER2'/' | stdout parallel -vj0 -k --joblog /tmp/jl-`basename $0` -L1
par_results_arg_256() {
echo '### bug #42089: --results with arg > 256 chars (should be 1 char shorter)'
parallel --results parallel_test_dir echo ::: 1234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456;
ls parallel_test_dir/1/
rm -rf parallel_test_dir
}
echo '**'
echo '### Test slow arguments generation - https://savannah.gnu.org/bugs/?32834';
par_slow_args_generation() {
echo '### Test slow arguments generation - https://savannah.gnu.org/bugs/?32834'
seq 1 3 | parallel -j1 "sleep 2; echo {}" | parallel -kj2 echo
}
echo '**'
par_kill_term_twice() {
echo '### Are children killed if GNU Parallel receives TERM twice? There should be no sleep at the end'
parallel -q bash -c 'sleep 120 & pid=$!; wait $pid' ::: 1 &
T=$!;
sleep 5;
pstree $$;
kill -TERM $T;
sleep 1;
pstree $$;
kill -TERM $T;
sleep 1;
pstree $$;
echo '**'
T=$!
sleep 5
pstree $$
kill -TERM $T
sleep 1
pstree $$
kill -TERM $T
sleep 1
pstree $$
}
par_kill_int_twice() {
echo '### Are children killed if GNU Parallel receives INT twice? There should be no sleep at the end'
parallel -q bash -c 'sleep 120 & pid=$!; wait $pid' ::: 1 &
T=$!;
sleep 5;
pstree $$;
kill -INT $T;
sleep 1;
pstree $$;
echo '**'
T=$!
sleep 5
pstree $$
kill -INT $T
sleep 1
pstree $$
}
par_children_receive_sig() {
echo '### Do children receive --termseq signals'
show_signals() {
perl -e 'for(keys %SIG) { $SIG{$_} = eval "sub { print STDERR \"Got $_\\n\"; }";} while(1){sleep 1}';
};
export -f show_signals;
echo | stdout parallel --termseq TERM,200,TERM,100,TERM,50,KILL,25 -u --timeout 1 show_signals;
echo | stdout parallel --termseq INT,200,TERM,100,KILL,25 -u --timeout 1 show_signals;
sleep 3;
}
export -f show_signals
echo | stdout parallel --termseq TERM,200,TERM,100,TERM,50,KILL,25 -u \
--timeout 1 show_signals
echo '**'
echo | stdout parallel --termseq INT,200,TERM,100,KILL,25 -u \
--timeout 1 show_signals
sleep 3
}
par_wrong_slot_rpl_resume() {
echo '### bug #47644: Wrong slot number replacement when resuming'
seq 0 20 | parallel -kj 4 --delay 0.2 --joblog /tmp/parallel-bug-47558 'sleep 1; echo {%} {=$_==10 and exit =}';
seq 0 20 | parallel -kj 4 --resume --delay 0.2 --joblog /tmp/parallel-bug-47558 'sleep 1; echo {%} {=$_==110 and exit =}'
echo '**'
seq 0 20 |
parallel -kj 4 --delay 0.2 --joblog /tmp/parallel-bug-47558 \
'sleep 1; echo {%} {=$_==10 and exit =}'
seq 0 20 |
parallel -kj 4 --resume --delay 0.2 --joblog /tmp/parallel-bug-47558 \
'sleep 1; echo {%} {=$_==110 and exit =}'
}
par_pipepart_block() {
echo '### --pipepart --block -# (# < 0)'
seq 1000 > /run/shm/parallel$$;
parallel -j2 -k --pipepart echo {#} :::: /run/shm/parallel$$;
parallel -j2 -k --block -1 --pipepart echo {#}-2 :::: /run/shm/parallel$$;
parallel -j2 -k --block -2 --pipepart echo {#}-4 :::: /run/shm/parallel$$;
parallel -j2 -k --block -10 --pipepart echo {#}-20 :::: /run/shm/parallel$$;
seq 1000 > /run/shm/parallel$$
parallel -j2 -k --pipepart echo {#} :::: /run/shm/parallel$$
parallel -j2 -k --block -1 --pipepart echo {#}-2 :::: /run/shm/parallel$$
parallel -j2 -k --block -2 --pipepart echo {#}-4 :::: /run/shm/parallel$$
parallel -j2 -k --block -10 --pipepart echo {#}-20 :::: /run/shm/parallel$$
rm /run/shm/parallel$$
}
echo '**'
EOF
export -f $(compgen -A function | grep par_)
compgen -A function | grep par_ | sort | parallel -j6 --tag -k '{} 2>&1'

View file

@ -1,6 +1,15 @@
par_compress_fail ### bug #41609: --compress fails
par_compress_fail 24812dd0f24a26d08a780f988b9d5ad2 -
par_compress_fail 24812dd0f24a26d08a780f988b9d5ad2 -
par_distribute_input_by_ability ### Distribute input to jobs that are ready
par_distribute_input_by_ability Job-slot n is 50% slower than n+1, so the order should be 1..7
par_distribute_input_by_ability 1
par_distribute_input_by_ability 2
par_distribute_input_by_ability 3
par_distribute_input_by_ability 4
par_distribute_input_by_ability 5
par_distribute_input_by_ability 6
par_distribute_input_by_ability 7
par_first_print_halt_on_error_1 ### Test first dying print --halt-on-error 1
par_first_print_halt_on_error_1 0
par_first_print_halt_on_error_1 parallel: This job failed:
@ -529,6 +538,8 @@ par_print_before_halt_on_error 2 exit code 1
par_print_before_halt_on_error 2 0 0
par_print_before_halt_on_error 2 parallel: This job failed:
par_print_before_halt_on_error 2 perl -e sleep\ \$ARGV\[0\]\;print\ STDERR\ @ARGV,\"\\n\"\;\ 2\ \>\ 0\ \?\ exit\ shift\ :\ exit\ not\ shift\; 1
par_round_robin_blocks bug #49664: --round-robin does not complete
par_round_robin_blocks 2
par_testhalt ### testhalt --halt now,fail=0
par_testhalt parallel: This job failed:
par_testhalt sleep 1.3; exit 1

View file

@ -1,120 +1,95 @@
echo '### bug #42089: --results with arg > 256 chars (should be 1 char shorter)'
### bug #42089: --results with arg > 256 chars (should be 1 char shorter)
parallel --results parallel_test_dir echo ::: 1234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456; ls parallel_test_dir/1/
1234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456
123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345
rm -rf parallel_test_dir
echo '**'
**
echo '### Test slow arguments generation - https://savannah.gnu.org/bugs/?32834'; seq 1 3 | parallel -j1 "sleep 2; echo {}" | parallel -kj2 echo
### Test slow arguments generation - https://savannah.gnu.org/bugs/?32834
1
2
3
echo '**'
**
echo '### Are children killed if GNU Parallel receives TERM twice? There should be no sleep at the end'
### Are children killed if GNU Parallel receives TERM twice? There should be no sleep at the end
parallel -q bash -c 'sleep 120 & pid=$!; wait $pid' ::: 1 & T=$!; sleep 5; pstree $$; kill -TERM $T; sleep 1; pstree $$; kill -TERM $T; sleep 1; pstree $$; echo '**'
bash-+-perl---bash---sleep
`-pstree
bash-+-perl---bash---sleep
`-pstree
bash---pstree
**
parallel: SIGTERM received. No new jobs will be started.
parallel: Waiting for these 1 jobs to finish. Send SIGTERM again to stop now.
parallel: bash -c sleep\ 120\ \&\ pid\=\$\!\;\ wait\ \$pid 1
echo '### Are children killed if GNU Parallel receives INT twice? There should be no sleep at the end'
### Are children killed if GNU Parallel receives INT twice? There should be no sleep at the end
parallel -q bash -c 'sleep 120 & pid=$!; wait $pid' ::: 1 & T=$!; sleep 5; pstree $$; kill -INT $T; sleep 1; pstree $$; echo '**'
bash-+-perl---bash---sleep
`-pstree
bash---pstree
**
echo '### Do children receive --termseq signals'
### Do children receive --termseq signals
show_signals() { perl -e 'for(keys %SIG) { $SIG{$_} = eval "sub { print STDERR \"Got $_\\n\"; }";} while(1){sleep 1}'; }; export -f show_signals; echo | stdout parallel --termseq TERM,200,TERM,100,TERM,50,KILL,25 -u --timeout 1 show_signals; echo | stdout parallel --termseq INT,200,TERM,100,KILL,25 -u --timeout 1 show_signals; sleep 3; echo '**'
Got TERM
Got TERM
Got TERM
Got INT
Got TERM
**
echo '### bug #47644: Wrong slot number replacement when resuming'
### bug #47644: Wrong slot number replacement when resuming
seq 0 20 | parallel -kj 4 --delay 0.2 --joblog /tmp/parallel-bug-47558 'sleep 1; echo {%} {=$_==10 and exit =}'; seq 0 20 | parallel -kj 4 --resume --delay 0.2 --joblog /tmp/parallel-bug-47558 'sleep 1; echo {%} {=$_==110 and exit =}'
1 0
2 1
3 2
4 3
1 4
2 5
3 6
4 7
1 8
2 9
3 10
4 11
1 12
2 13
3 14
4 15
1 16
2 17
3 18
4 19
1 20
echo '**'
**
echo '### --pipepart --block -# (# < 0)'
### --pipepart --block -# (# < 0)
seq 1000 > /run/shm/parallel$$; parallel -j2 -k --pipepart echo {#} :::: /run/shm/parallel$$; parallel -j2 -k --block -1 --pipepart echo {#}-2 :::: /run/shm/parallel$$; parallel -j2 -k --block -2 --pipepart echo {#}-4 :::: /run/shm/parallel$$; parallel -j2 -k --block -10 --pipepart echo {#}-20 :::: /run/shm/parallel$$; rm /run/shm/parallel$$
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
1-2
2-2
1-4
2-4
3-4
4-4
1-20
2-20
3-20
4-20
5-20
6-20
7-20
8-20
9-20
10-20
11-20
12-20
13-20
14-20
15-20
16-20
17-20
18-20
19-20
20-20
echo '**'
**
par_children_receive_sig ### Do children receive --termseq signals
par_children_receive_sig Got TERM
par_children_receive_sig Got TERM
par_children_receive_sig Got TERM
par_children_receive_sig Got INT
par_children_receive_sig Got TERM
par_kill_int_twice ### Are children killed if GNU Parallel receives INT twice? There should be no sleep at the end
par_kill_int_twice bash-+-perl---bash---sleep
par_kill_int_twice `-pstree
par_kill_int_twice bash---pstree
par_kill_term_twice ### Are children killed if GNU Parallel receives TERM twice? There should be no sleep at the end
par_kill_term_twice bash-+-perl---bash---sleep
par_kill_term_twice `-pstree
par_kill_term_twice parallel: SIGTERM received. No new jobs will be started.
par_kill_term_twice parallel: Waiting for these 1 jobs to finish. Send SIGTERM again to stop now.
par_kill_term_twice parallel: bash -c sleep\ 120\ \&\ pid\=\$\!\;\ wait\ \$pid 1
par_kill_term_twice bash-+-perl---bash---sleep
par_kill_term_twice `-pstree
par_kill_term_twice bash---pstree
par_pipepart_block ### --pipepart --block -# (# < 0)
par_pipepart_block 1
par_pipepart_block 2
par_pipepart_block 3
par_pipepart_block 4
par_pipepart_block 5
par_pipepart_block 6
par_pipepart_block 7
par_pipepart_block 8
par_pipepart_block 9
par_pipepart_block 10
par_pipepart_block 11
par_pipepart_block 12
par_pipepart_block 13
par_pipepart_block 14
par_pipepart_block 15
par_pipepart_block 16
par_pipepart_block 17
par_pipepart_block 18
par_pipepart_block 19
par_pipepart_block 20
par_pipepart_block 1-2
par_pipepart_block 2-2
par_pipepart_block 1-4
par_pipepart_block 2-4
par_pipepart_block 3-4
par_pipepart_block 4-4
par_pipepart_block 1-20
par_pipepart_block 2-20
par_pipepart_block 3-20
par_pipepart_block 4-20
par_pipepart_block 5-20
par_pipepart_block 6-20
par_pipepart_block 7-20
par_pipepart_block 8-20
par_pipepart_block 9-20
par_pipepart_block 10-20
par_pipepart_block 11-20
par_pipepart_block 12-20
par_pipepart_block 13-20
par_pipepart_block 14-20
par_pipepart_block 15-20
par_pipepart_block 16-20
par_pipepart_block 17-20
par_pipepart_block 18-20
par_pipepart_block 19-20
par_pipepart_block 20-20
par_results_arg_256 ### bug #42089: --results with arg > 256 chars (should be 1 char shorter)
par_results_arg_256 1234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456
par_results_arg_256 123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234567890123456789012345
par_slow_args_generation ### Test slow arguments generation - https://savannah.gnu.org/bugs/?32834
par_slow_args_generation 1
par_slow_args_generation 2
par_slow_args_generation 3
par_wrong_slot_rpl_resume ### bug #47644: Wrong slot number replacement when resuming
par_wrong_slot_rpl_resume 1 0
par_wrong_slot_rpl_resume 2 1
par_wrong_slot_rpl_resume 3 2
par_wrong_slot_rpl_resume 4 3
par_wrong_slot_rpl_resume 1 4
par_wrong_slot_rpl_resume 2 5
par_wrong_slot_rpl_resume 3 6
par_wrong_slot_rpl_resume 4 7
par_wrong_slot_rpl_resume 1 8
par_wrong_slot_rpl_resume 2 9
par_wrong_slot_rpl_resume 3 10
par_wrong_slot_rpl_resume 4 11
par_wrong_slot_rpl_resume 1 12
par_wrong_slot_rpl_resume 2 13
par_wrong_slot_rpl_resume 3 14
par_wrong_slot_rpl_resume 4 15
par_wrong_slot_rpl_resume 1 16
par_wrong_slot_rpl_resume 2 17
par_wrong_slot_rpl_resume 3 18
par_wrong_slot_rpl_resume 4 19
par_wrong_slot_rpl_resume 1 20