blocked SIGPIPE causes uninterruptible dnsjit scripts

[pspacek]

Normally scripts which use stdout can be terminated by signal SIGPIPE, i.e. when another tool in the pipeline cuts the output etc.

Dnsjit blocks SIGPIPE unconditionally and this causes scripts to run until they terminate for other reasons, even if the all the output is being dropped.

This is causing trouble in long testing pipelines which do decompress & process as separate steps in individual processes.

A minimal example:

File /tmp/out.lua

local output = io.stdout
while (true)
do
	output:write("bla\n")
end

LuaJIT behavior:

$ time luajit /tmp/out.lua | head -n1
bla

real	0m0.002s
user	0m0.002s
sys	0m0.000s

Dnsjit behavior:

$ time dnsjit /tmp/out.lua | head -n1
bla
^Ccore fatal: signal 2

real	0m3.654s
user	0m3.622s
sys	0m0.030s

I would submit MR to unblock SIGPIPE, but I'm not sure why generic block for (almost) all signals was put in place.

Current behavior stems from this piece of code:

     48 int main(int argc, char* argv[])
     49 {
...
     60     sigfillset(&set);
     61     if ((err = pthread_sigmask(SIG_BLOCK, &set, 0))) {
     62         glfatal("Unable to set blocked signals with pthread_sigmask()");
     63         return 2;
     64     }
     65 
     66     sigemptyset(&set);
     67     sigaddset(&set, SIGTERM);
     68     sigaddset(&set, SIGQUIT);
     69     sigaddset(&set, SIGINT);
     70 
     71     if ((err = pthread_create(&sighthr, 0, _sighthr, &set))) {
     72         glfatal("Unable to start signal thread with pthread_create()");
     73         return 2;
     74     }

[pspacek]

LOL, I just realized it ignores also SIGSEGV.
killall -11 dnsjit does nothing.

I think that's a bit too much 😂

[jelu]

dnsjit goes wroooooom! no time to stop for signals! 😄 will fix

[jelu]

@pspacek @tkrizek please test this :)

[nicki-krizek]

Signals are now working as expected. We've just had to ignore SIGPIPE in DNS Shotgun to keep its prior behavior, otherwise all looks good. Thanks!