1 |
root |
1.1 |
=head1 NAME |
2 |
|
|
|
3 |
|
|
AnyEvent::Watchdog - generic watchdog/program restarter |
4 |
|
|
|
5 |
|
|
=head1 SYNOPSIS |
6 |
|
|
|
7 |
|
|
# MUST be use'd as the very first thing in the main program |
8 |
|
|
use AnyEvent::Watchdog; |
9 |
|
|
|
10 |
|
|
=head1 DESCRIPTION |
11 |
|
|
|
12 |
|
|
This module implements a watchdog that can repeatedly fork the program and |
13 |
|
|
thus effectively restart it - as soon as the module is use'd, it will fork |
14 |
|
|
the program (if possible) and continue to run it normally in the child, |
15 |
|
|
while the parent becomes a supervisor. |
16 |
|
|
|
17 |
|
|
The child can then ask the supervisor to restart itself instead of |
18 |
|
|
exiting, or ask the supervisor to restart it gracefully or forcefully. |
19 |
|
|
|
20 |
|
|
B<NOTE:> This module B<< I<MUST> >> be used as the first thing in the main |
21 |
|
|
program. It will cause weird effects when used from another module, as |
22 |
|
|
perl does not expect to be forked inside C<BEGIN> blocks. |
23 |
|
|
|
24 |
|
|
=head1 RECIPES |
25 |
|
|
|
26 |
|
|
Use AnyEvent::Watchdog solely as a convinient on-demand-restarter: |
27 |
|
|
|
28 |
|
|
use AnyEvent::Watchdog; |
29 |
|
|
|
30 |
|
|
# and whenever you wnat to restart (e.g. to upgrade code): |
31 |
|
|
AnyEvent::Watchdog::restart; |
32 |
|
|
|
33 |
|
|
Use AnyEvent::Watchdog to kill the program and exit when the event loop |
34 |
|
|
fails to run for more than two minutes: |
35 |
|
|
|
36 |
|
|
use AnyEvent::Watchdog qw(autorestart heartbeat=120); |
37 |
|
|
|
38 |
|
|
Use AnyEvent::Watchdog to automatically restart the program |
39 |
|
|
when it fails to handle events for longer than 5 minutes: |
40 |
|
|
|
41 |
|
|
use AnyEvent::Watchdog qw(autorestart heartbeat=300); |
42 |
|
|
|
43 |
|
|
=head1 FUNCTIONS |
44 |
|
|
|
45 |
|
|
The module supports the following functions: |
46 |
|
|
|
47 |
|
|
=over 4 |
48 |
|
|
|
49 |
|
|
=cut |
50 |
|
|
|
51 |
|
|
package AnyEvent::Watchdog; |
52 |
|
|
|
53 |
|
|
# load modules we will use later anyways |
54 |
|
|
use common::sense; |
55 |
|
|
|
56 |
|
|
use Carp (); |
57 |
|
|
|
58 |
|
|
our $VERSION = '0.1'; |
59 |
|
|
|
60 |
|
|
our $PID; # child pid |
61 |
|
|
our $ENABLED = 1; |
62 |
|
|
our $AUTORESTART; # actually exit |
63 |
|
|
our $HEARTBEAT; |
64 |
|
|
our ($P, $C); |
65 |
|
|
|
66 |
|
|
sub poll($) { |
67 |
|
|
(vec my $v, fileno $P, 1) = 1; |
68 |
|
|
CORE::select $v, undef, undef, $_[0] |
69 |
|
|
} |
70 |
|
|
|
71 |
|
|
sub server { |
72 |
|
|
my $expected;# do we expect a program exit? |
73 |
|
|
my $heartbeat; |
74 |
|
|
|
75 |
|
|
$AUTORESTART = 0; |
76 |
|
|
|
77 |
|
|
local $SIG{HUP} = 'IGNORE'; |
78 |
|
|
local $SIG{INT} = 'IGNORE'; |
79 |
|
|
local $SIG{TERM} = 'IGNORE'; |
80 |
|
|
|
81 |
|
|
while () { |
82 |
|
|
if ($heartbeat) { |
83 |
|
|
unless (poll $heartbeat) { |
84 |
|
|
$expected = 1; |
85 |
|
|
warn "AnyEvent::Watchdog: heartbeat failed. killing.\n"; |
86 |
|
|
kill 9, $PID; |
87 |
|
|
last; |
88 |
|
|
} |
89 |
|
|
} |
90 |
|
|
|
91 |
|
|
sysread $P, my $cmd, 1 |
92 |
|
|
or last; |
93 |
|
|
|
94 |
|
|
if ($cmd eq chr 0) { |
95 |
|
|
$AUTORESTART = 0; |
96 |
|
|
|
97 |
|
|
} elsif ($cmd eq chr 1) { |
98 |
|
|
$AUTORESTART = 1; |
99 |
|
|
|
100 |
|
|
} elsif ($cmd eq chr 2) { |
101 |
|
|
sysread $P, my $timeout, 1 |
102 |
|
|
or last; |
103 |
|
|
|
104 |
|
|
$timeout = ord $timeout; |
105 |
|
|
|
106 |
|
|
unless (poll $timeout) { |
107 |
|
|
warn "AnyEvent::Watchdog: program attempted restart, but failed to do so within $timeout seconds. killing.\n"; |
108 |
|
|
kill 9, $PID; |
109 |
|
|
} |
110 |
|
|
|
111 |
|
|
if (sysread $P, my $dummy, 1) { |
112 |
|
|
warn "AnyEvent::Watchdog: unexpected program output. killing.\n"; |
113 |
|
|
kill 9, $PID; |
114 |
|
|
} |
115 |
|
|
|
116 |
|
|
$expected = 1; |
117 |
|
|
last; |
118 |
|
|
|
119 |
|
|
} elsif ($cmd eq chr 3) { |
120 |
|
|
sysread $P, my $interval, 1 |
121 |
|
|
or last; |
122 |
|
|
|
123 |
|
|
$heartbeat = ord $interval |
124 |
|
|
unless defined $heartbeat; |
125 |
|
|
|
126 |
|
|
} elsif ($cmd eq chr 4) { |
127 |
|
|
# heartbeat |
128 |
|
|
# TODO: should only reset heartbeat timeout with \005 |
129 |
|
|
|
130 |
|
|
} else { |
131 |
|
|
warn "AnyEvent::Watchdog: unexpected program output. killing.\n"; |
132 |
|
|
kill 9, $PID; |
133 |
|
|
last; |
134 |
|
|
} |
135 |
|
|
} |
136 |
|
|
|
137 |
|
|
waitpid $PID, 0; |
138 |
|
|
|
139 |
|
|
require POSIX; |
140 |
|
|
|
141 |
|
|
my $termsig = POSIX::WIFSIGNALED ($?) && POSIX::WTERMSIG ($?); |
142 |
|
|
|
143 |
|
|
if ($termsig == POSIX::SIGINT () || $termsig == POSIX::SIGTERM ()) { |
144 |
|
|
$AUTORESTART = 0; |
145 |
|
|
$expected = 1; |
146 |
|
|
} |
147 |
|
|
|
148 |
|
|
unless ($expected) { |
149 |
|
|
warn "AnyEvent::Watchdog: program exited unexpectedly with status $?.\n" |
150 |
|
|
if $? >> 8; |
151 |
|
|
} |
152 |
|
|
|
153 |
|
|
if ($AUTORESTART) { |
154 |
|
|
warn "AnyEvent::Watchdog: attempting automatic restart.\n"; |
155 |
|
|
} else { |
156 |
|
|
if ($termsig) { |
157 |
|
|
$SIG{$_} = 'DEFAULT' for keys %SIG; |
158 |
|
|
kill $termsig, $$; |
159 |
|
|
POSIX::_exit (127); |
160 |
|
|
} else { |
161 |
|
|
POSIX::_exit ($? >> 8); |
162 |
|
|
} |
163 |
|
|
} |
164 |
|
|
} |
165 |
|
|
|
166 |
|
|
our %SEEKPOS; |
167 |
|
|
# due to bugs in perl, try to remember file offsets for all fds, and restore them later |
168 |
|
|
# (the parser otherwise exhausts the input files) |
169 |
|
|
|
170 |
|
|
# this causes perlio to flush it's handles internally, so |
171 |
|
|
# seek offsets become correct. |
172 |
|
|
exec "."; # toi toi toi |
173 |
|
|
#{ |
174 |
|
|
# local $SIG{CHLD} = 'DEFAULT'; |
175 |
|
|
# my $pid = fork; |
176 |
|
|
# |
177 |
|
|
# if ($pid) { |
178 |
|
|
# waitpid $pid, 0; |
179 |
|
|
# } else { |
180 |
|
|
# kill 9, $$; |
181 |
|
|
# } |
182 |
|
|
#} |
183 |
|
|
|
184 |
|
|
# now records all fd positions |
185 |
|
|
for (0 .. 1023) { |
186 |
|
|
open my $fh, "<&$_" or next; |
187 |
|
|
$SEEKPOS{$_} = (sysseek $fh, 0, 1 or next); |
188 |
|
|
} |
189 |
|
|
|
190 |
|
|
while () { |
191 |
|
|
if ($^O =~ /mswin32/i) { |
192 |
|
|
require AnyEvent::Util; |
193 |
|
|
($P, $C) = AnyEvent::Util::portable_socketpair () |
194 |
|
|
or Carp::croak "AnyEvent::Watchdog: unable to create restarter pipe: $!\n"; |
195 |
|
|
} else { |
196 |
|
|
require Socket; |
197 |
|
|
socketpair $P, $C, Socket::AF_UNIX (), Socket::SOCK_STREAM (), 0 |
198 |
|
|
or Carp::croak "AnyEvent::Watchdog: unable to create restarter pipe: $!\n"; |
199 |
|
|
} |
200 |
|
|
|
201 |
|
|
local $SIG{CHLD} = 'DEFAULT'; |
202 |
|
|
|
203 |
|
|
$PID = fork; |
204 |
|
|
|
205 |
|
|
unless (defined $PID) { |
206 |
|
|
warn "AnyEvent::Watchdog: '$!', retrying in one second...\n"; |
207 |
|
|
sleep 1; |
208 |
|
|
} elsif ($PID) { |
209 |
|
|
close $C; |
210 |
|
|
server; |
211 |
|
|
} else { |
212 |
|
|
# restore seek offsets |
213 |
|
|
while (my ($k, $v) = each %SEEKPOS) { |
214 |
|
|
open my $fh, "<&$k" or next; |
215 |
|
|
sysseek $fh, $v, 0; |
216 |
|
|
} |
217 |
|
|
|
218 |
|
|
# continue the program normally |
219 |
|
|
close $P; |
220 |
|
|
last; |
221 |
|
|
} |
222 |
|
|
} |
223 |
|
|
|
224 |
|
|
=item AnyEvent::Watchdog::restart [$timeout] |
225 |
|
|
|
226 |
|
|
Tells the supervisor to restart the process when it exits, or forcefully |
227 |
|
|
after C<$timeout> seconds (minimum 1, maximum 255, default 60). |
228 |
|
|
|
229 |
|
|
Calls C<exit 0> to exit the process cleanly. |
230 |
|
|
|
231 |
|
|
=cut |
232 |
|
|
|
233 |
|
|
sub restart(;$) { |
234 |
|
|
my ($timeout) = @_; |
235 |
|
|
|
236 |
|
|
$timeout = 60 unless defined $timeout; |
237 |
|
|
$timeout = 1 if $timeout < 1; |
238 |
|
|
$timeout = 255 if $timeout > 255; |
239 |
|
|
|
240 |
|
|
syswrite $C, "\x01\x02" . chr $timeout; |
241 |
|
|
exit 0; |
242 |
|
|
} |
243 |
|
|
|
244 |
|
|
=item AnyEvent::Watchdog::autorestart [$boolean] |
245 |
|
|
|
246 |
|
|
=item use AnyEvent::Watchdog qw(autorestart[=$boolean]) |
247 |
|
|
|
248 |
|
|
Enables or disables autorestart (initially disabled, default for |
249 |
|
|
C<$boolean> is to enable): By default, the supervisor will exit if the |
250 |
|
|
program exits or dies in any way. When enabling autorestart behaviour, |
251 |
|
|
then the supervisor will try to restart the program after it dies. |
252 |
|
|
|
253 |
|
|
Note that the supervisor will never autorestart when the child died with |
254 |
|
|
SIGINT or SIGTERM. |
255 |
|
|
|
256 |
|
|
=cut |
257 |
|
|
|
258 |
|
|
sub autorestart(;$) { |
259 |
|
|
syswrite $C, !@_ || $_[0] ? "\x01" : "\x00"; |
260 |
|
|
} |
261 |
|
|
|
262 |
|
|
=item AnyEvent::Watchdog::heartbeat [$interval] |
263 |
|
|
|
264 |
|
|
=item use AnyEvent::Watchdog qw(heartbeat[=$interval]) |
265 |
|
|
|
266 |
|
|
Tells the supervisor to automatically kill the program if it doesn't |
267 |
|
|
react for C<$interval> seconds (minium 1, maximum 255, default 60) , then |
268 |
|
|
installs an AnyEvent timer the sends a regular heartbeat to the supervisor |
269 |
|
|
twice as often. |
270 |
|
|
|
271 |
|
|
Exit behaviour isn't changed, so if you want a restart instead of an exit, |
272 |
|
|
you have to call C<autorestart>. |
273 |
|
|
|
274 |
|
|
Once enabled, the heartbeat cannot be switched off. |
275 |
|
|
|
276 |
|
|
=cut |
277 |
|
|
|
278 |
|
|
sub heartbeat(;$) { |
279 |
|
|
my ($interval) = @_; |
280 |
|
|
|
281 |
|
|
$interval = 60 unless defined $interval; |
282 |
|
|
$interval = 1 if $interval < 1; |
283 |
|
|
$interval = 255 if $interval > 255; |
284 |
|
|
|
285 |
|
|
syswrite $C, "\x03" . chr $interval; |
286 |
|
|
|
287 |
|
|
require AE; |
288 |
|
|
$HEARTBEAT = AE::timer (0, $interval * 0.5, sub { |
289 |
|
|
syswrite $C, "\x04"; |
290 |
|
|
}); |
291 |
|
|
} |
292 |
|
|
|
293 |
|
|
sub import { |
294 |
|
|
shift; |
295 |
|
|
|
296 |
|
|
for (@_) { |
297 |
|
|
if (/^autorestart(?:=(.*))?$/) { |
298 |
|
|
autorestart defined $1 ? $1 : 1; |
299 |
|
|
} elsif (/^heartbeat(?:=(.*))?$/) { |
300 |
|
|
heartbeat $1; |
301 |
|
|
} else { |
302 |
|
|
Carp::croak "AnyEvent::Watchdog: '$_' is not a valid import argument"; |
303 |
|
|
} |
304 |
|
|
} |
305 |
|
|
} |
306 |
|
|
|
307 |
|
|
=head1 SEE ALSO |
308 |
|
|
|
309 |
|
|
L<AnyEvent>. |
310 |
|
|
|
311 |
|
|
=head1 AUTHOR |
312 |
|
|
|
313 |
|
|
Marc Lehmann <schmorp@schmorp.de> |
314 |
|
|
http://home.schmorp.de/ |
315 |
|
|
|
316 |
|
|
=cut |
317 |
|
|
|
318 |
|
|
1 |
319 |
|
|
|