line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
package HPC::Runner::Command::watch_db; |
2
|
|
|
|
|
|
|
|
3
|
1
|
|
|
1
|
|
24885
|
use MooseX::App::Command; |
|
1
|
|
|
|
|
8
|
|
|
1
|
|
|
|
|
24
|
|
4
|
1
|
|
|
1
|
|
14201
|
use Data::Dumper; |
|
1
|
|
|
|
|
5
|
|
|
1
|
|
|
|
|
84
|
|
5
|
1
|
|
|
1
|
|
7
|
use Log::Log4perl qw(:easy); |
|
1
|
|
|
|
|
4
|
|
|
1
|
|
|
|
|
15
|
|
6
|
|
|
|
|
|
|
|
7
|
|
|
|
|
|
|
extends 'HPC::Runner::Command'; |
8
|
|
|
|
|
|
|
with 'HPC::Runner::Command::Plugin::Logger::Sqlite'; |
9
|
|
|
|
|
|
|
|
10
|
|
|
|
|
|
|
command_short_description 'Watch the sqlitedb and exit when job submissions are complete.'; |
11
|
|
|
|
|
|
|
command_long_description 'Watch the sqlitedb for one or more submission ids. This is only really useful for testing. In a real world application it is probably best to just have the scheduler email you on completion, unless you are submitting more jobs than you want emails.'; |
12
|
|
|
|
|
|
|
|
13
|
|
|
|
|
|
|
has 'total_processes' => ( |
14
|
|
|
|
|
|
|
traits => ['Number'], |
15
|
|
|
|
|
|
|
is => 'rw', |
16
|
|
|
|
|
|
|
isa => 'Num', |
17
|
|
|
|
|
|
|
default => 0, |
18
|
|
|
|
|
|
|
handles => { |
19
|
|
|
|
|
|
|
set_total_processes => 'set', |
20
|
|
|
|
|
|
|
add_total_processes => 'add', |
21
|
|
|
|
|
|
|
}, |
22
|
|
|
|
|
|
|
); |
23
|
|
|
|
|
|
|
|
24
|
|
|
|
|
|
|
option 'exit_on_fail' => ( |
25
|
|
|
|
|
|
|
traits => ['Bool'], |
26
|
|
|
|
|
|
|
is => 'rw', |
27
|
|
|
|
|
|
|
isa => 'Bool', |
28
|
|
|
|
|
|
|
default => 0, |
29
|
|
|
|
|
|
|
documentation => 'Fail if any jobs have an exit code besides 0 - whether all tasks have completed or not', |
30
|
|
|
|
|
|
|
); |
31
|
|
|
|
|
|
|
|
32
|
|
|
|
|
|
|
option 'sleep_interval' => ( |
33
|
|
|
|
|
|
|
is => 'rw', |
34
|
|
|
|
|
|
|
isa => 'Int', |
35
|
|
|
|
|
|
|
default => 10, |
36
|
|
|
|
|
|
|
documentation => 'Sleep interval in seconds to query sqlite db. For software testing you should leave as is. For longer running analyses you probably want to increase this.', |
37
|
|
|
|
|
|
|
); |
38
|
|
|
|
|
|
|
|
39
|
|
|
|
|
|
|
option 'verbose' => ( |
40
|
|
|
|
|
|
|
traits => ['Bool'], |
41
|
|
|
|
|
|
|
is => 'rw', |
42
|
|
|
|
|
|
|
isa => 'Bool', |
43
|
|
|
|
|
|
|
default => 0, |
44
|
|
|
|
|
|
|
documentation => 'Enable verbose logging', |
45
|
|
|
|
|
|
|
); |
46
|
|
|
|
|
|
|
|
47
|
|
|
|
|
|
|
sub execute { |
48
|
0
|
|
|
0
|
|
|
my $self = shift; |
49
|
|
|
|
|
|
|
|
50
|
0
|
0
|
|
|
|
|
if($self->submission_id){ |
51
|
0
|
|
|
|
|
|
$self->app_log->info("Watching Submission Id : " . $self->submission_id); |
52
|
|
|
|
|
|
|
} |
53
|
|
|
|
|
|
|
else{ |
54
|
0
|
|
|
|
|
|
$self->app_log->info("No submission id specified. We will watch the whole database"); |
55
|
|
|
|
|
|
|
} |
56
|
|
|
|
|
|
|
|
57
|
0
|
|
|
|
|
|
my $results = $self->query_submissions; |
58
|
0
|
|
|
|
|
|
$self->total_tasks($results); |
59
|
|
|
|
|
|
|
|
60
|
0
|
|
|
|
|
|
while (1){ |
61
|
|
|
|
|
|
|
|
62
|
0
|
0
|
|
|
|
|
if($self->verbose){ |
63
|
0
|
|
|
|
|
|
$self->app_log->debug("Watching again..."); |
64
|
|
|
|
|
|
|
} |
65
|
|
|
|
|
|
|
|
66
|
0
|
|
|
|
|
|
my $results = $self->query_submissions; |
67
|
|
|
|
|
|
|
|
68
|
0
|
|
|
|
|
|
my $jobs = $results->search_related('jobs'); |
69
|
0
|
|
|
|
|
|
my $tasks = $jobs->search_related('tasks'); |
70
|
|
|
|
|
|
|
|
71
|
0
|
|
|
|
|
|
$self->query_task($tasks); |
72
|
|
|
|
|
|
|
|
73
|
0
|
|
|
|
|
|
sleep ($self->sleep_interval); |
74
|
|
|
|
|
|
|
} |
75
|
|
|
|
|
|
|
} |
76
|
|
|
|
|
|
|
|
77
|
|
|
|
|
|
|
sub query_task { |
78
|
0
|
|
|
0
|
|
|
my $self = shift; |
79
|
0
|
|
|
|
|
|
my $task_rs = shift; |
80
|
|
|
|
|
|
|
|
81
|
0
|
0
|
|
|
|
|
if($self->verbose){ |
82
|
0
|
|
|
|
|
|
$self->app_log->debug("Tasks in DB are ".$task_rs->count); |
83
|
|
|
|
|
|
|
} |
84
|
|
|
|
|
|
|
|
85
|
|
|
|
|
|
|
#If exit on fail we don't care if we have completed the number of processes - just fail |
86
|
0
|
0
|
|
|
|
|
if ($self->exit_on_fail){ |
87
|
0
|
|
|
|
|
|
$self->check_exit_code($task_rs); |
88
|
|
|
|
|
|
|
} |
89
|
|
|
|
|
|
|
|
90
|
0
|
0
|
|
|
|
|
if ($task_rs->count != $self->total_processes){ |
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
91
|
|
|
|
|
|
|
#We have |
92
|
0
|
|
|
|
|
|
return; |
93
|
|
|
|
|
|
|
} |
94
|
|
|
|
|
|
|
elsif($task_rs->count == $self->total_processes){ |
95
|
0
|
|
|
|
|
|
$self->app_log->info("We have completed ".$self->total_processes." tasks. Exiting successfully"); |
96
|
0
|
|
|
|
|
|
exit 0; |
97
|
|
|
|
|
|
|
} |
98
|
|
|
|
|
|
|
elsif($task_rs->count >= $self->total_processes){ |
99
|
0
|
|
|
|
|
|
$self->app_log->info("More tasks were completed than were in the databases ".$self->total_processes." tasks."); |
100
|
0
|
|
|
|
|
|
$self->app_log->info("Were jobs restarted manually?"); |
101
|
0
|
|
|
|
|
|
$self->app_log->info("Exiting successfully"); |
102
|
0
|
|
|
|
|
|
exit 0; |
103
|
|
|
|
|
|
|
} |
104
|
|
|
|
|
|
|
else{ |
105
|
0
|
|
|
|
|
|
$self->app_log->debug("Not sure how we got here..."); |
106
|
|
|
|
|
|
|
} |
107
|
|
|
|
|
|
|
|
108
|
|
|
|
|
|
|
} |
109
|
|
|
|
|
|
|
|
110
|
|
|
|
|
|
|
sub check_exit_code { |
111
|
0
|
|
|
0
|
|
|
my $self = shift; |
112
|
0
|
|
|
|
|
|
my $task_rs = shift; |
113
|
|
|
|
|
|
|
|
114
|
0
|
|
|
|
|
|
my $exit_codes = $task_rs->get_column('exit_code'); |
115
|
|
|
|
|
|
|
|
116
|
0
|
|
|
|
|
|
while ( my $res = $task_rs->next ) { |
117
|
0
|
0
|
|
|
|
|
if ($res->exit_code != 0){ |
118
|
0
|
|
|
|
|
|
$self->app_log->error("A task has failed! ".$res->task_pi); |
119
|
0
|
|
|
|
|
|
exit 1; |
120
|
|
|
|
|
|
|
} |
121
|
|
|
|
|
|
|
} |
122
|
|
|
|
|
|
|
} |
123
|
|
|
|
|
|
|
|
124
|
|
|
|
|
|
|
sub total_tasks { |
125
|
0
|
|
|
0
|
|
|
my $self = shift; |
126
|
0
|
|
|
|
|
|
my $results = shift; |
127
|
|
|
|
|
|
|
|
128
|
0
|
|
|
|
|
|
while ( my $res = $results->next ) { |
129
|
0
|
|
|
|
|
|
$self->add_total_processes( $res->total_processes ); |
130
|
|
|
|
|
|
|
} |
131
|
|
|
|
|
|
|
} |
132
|
|
|
|
|
|
|
|
133
|
|
|
|
|
|
|
sub query_submissions { |
134
|
0
|
|
|
0
|
|
|
my $self = shift; |
135
|
|
|
|
|
|
|
|
136
|
0
|
|
|
|
|
|
my $results; |
137
|
|
|
|
|
|
|
|
138
|
0
|
0
|
|
|
|
|
if($self->project){ |
|
|
0
|
|
|
|
|
|
139
|
0
|
|
|
|
|
|
$results = $self->schema->resultset('Submission') |
140
|
|
|
|
|
|
|
->search( { 'project' => $self->project } ); |
141
|
|
|
|
|
|
|
} |
142
|
|
|
|
|
|
|
elsif ($self->submission_id){ |
143
|
0
|
|
|
|
|
|
$results = $self->schema->resultset('Submission') |
144
|
|
|
|
|
|
|
->search( { 'submission_pi' => $self->submission_id } ); |
145
|
|
|
|
|
|
|
} |
146
|
|
|
|
|
|
|
else{ |
147
|
0
|
|
|
|
|
|
$results = $self->schema->resultset('Submission') |
148
|
|
|
|
|
|
|
->search(); |
149
|
|
|
|
|
|
|
} |
150
|
|
|
|
|
|
|
|
151
|
0
|
|
|
|
|
|
return $results; |
152
|
|
|
|
|
|
|
|
153
|
|
|
|
|
|
|
|
154
|
|
|
|
|
|
|
} |
155
|
|
|
|
|
|
|
|
156
|
|
|
|
|
|
|
#TODO To keep or not to keep? |
157
|
|
|
|
|
|
|
|
158
|
|
|
|
|
|
|
sub query_job { |
159
|
0
|
|
|
0
|
|
|
my $self = shift; |
160
|
0
|
|
|
|
|
|
my $job_rs = shift; |
161
|
|
|
|
|
|
|
|
162
|
|
|
|
|
|
|
#$job_rs->result_class('DBIx::Class::ResultClass::HashRefInflator'); |
163
|
|
|
|
|
|
|
#while ( my $res = $job_rs->next ) { |
164
|
|
|
|
|
|
|
#print Dumper($res); |
165
|
|
|
|
|
|
|
#} |
166
|
|
|
|
|
|
|
} |
167
|
|
|
|
|
|
|
|
168
|
|
|
|
|
|
|
# sub query_related { |
169
|
|
|
|
|
|
|
# my $self = shift; |
170
|
|
|
|
|
|
|
# |
171
|
|
|
|
|
|
|
# #$ENV{DBIC_TRACE} = 1; |
172
|
|
|
|
|
|
|
# |
173
|
|
|
|
|
|
|
# print "In query related\n"; |
174
|
|
|
|
|
|
|
# $self->schema->storage->debug(1); |
175
|
|
|
|
|
|
|
# |
176
|
|
|
|
|
|
|
# my $results = $self->schema->resultset('Submission') |
177
|
|
|
|
|
|
|
# ->search( {}, { 'prefetch' => { jobs => 'tasks' } } ); |
178
|
|
|
|
|
|
|
# |
179
|
|
|
|
|
|
|
# $results->result_class('DBIx::Class::ResultClass::HashRefInflator'); |
180
|
|
|
|
|
|
|
# |
181
|
|
|
|
|
|
|
# while ( my $res = $results->next ) { |
182
|
|
|
|
|
|
|
# print "Here is a result!\n"; |
183
|
|
|
|
|
|
|
# print Dumper($res); |
184
|
|
|
|
|
|
|
# } |
185
|
|
|
|
|
|
|
# |
186
|
|
|
|
|
|
|
# } |
187
|
|
|
|
|
|
|
|
188
|
|
|
|
|
|
|
1; |