Sending SIGUSR2 signal to a stopped process bricks the whole cluster

What's going wrong?

If one sends a SIGUSR2 to a stopped process the other processes stop responding.

How could we reproduce this issue?

Checkout pm2 repo.

# Setup cluster
cd examples/cluster-http
echo 'process.on("SIGUSR2", () => { console.log('SIGUSR2'); });' >> http-no-exit.js
pm2 start ecosystem.config.js
curl 'http://localhost:8002' && echo "ok" || echo "fail" 
# ^^ It works!!!


pm2 sendSignal SIGUSR2 0 &
curl 'http://localhost:8002' && echo "ok" || echo "fail" 
# ^^ It works!!!

pm2 stop 0 &
curl 'http://localhost:8002' && echo "ok" || echo "fail" 
# ^^ It works!!!

pm2 sendSignal SIGUSR2 0 &
curl 'http://localhost:8002' && echo "ok" || echo "fail" 
# ^^ It does not work

Supporting information

$ pm2 report
--- PM2 report ----------------------------------------------------------------
Date                 : Fri Dec 16 2022 14:26:20 GMT-0500 (Eastern Standard Time)
===============================================================================
--- Daemon -------------------------------------------------
pm2d version         : 5.2.2
node version         : 16.11.1
node path            : /Users/<USER>/.nvm/versions/node/v16.11.1/bin/pm2
argv                 : /Users/<USER>/.nvm/versions/node/v16.11.1/bin/node,/Users/<USER>/.nvm/versions/node/v16.11.1/lib/node_modules/pm2/lib/Daemon.js
argv0                : node
user                 : oleg
uid                  : 501
gid                  : 20
uptime               : 3min
===============================================================================
--- CLI ----------------------------------------------------
local pm2            : 5.2.2
node version         : 16.11.1
node path            : /Users/<USER>/.nvm/versions/node/v16.11.1/bin/pm2
argv                 : /Users/<USER>/.nvm/versions/node/v16.11.1/bin/node,/Users/<USER>/.nvm/versions/node/v16.11.1/bin/pm2,report
argv0                : node
user                 : oleg
uid                  : 501
gid                  : 20
===============================================================================
--- System info --------------------------------------------
arch                 : arm64
platform             : darwin
type                 : Darwin
cpus                 : Apple M1 Max
cpus nb              : 10
freemem              : 11419484160
totalmem             : 68719476736
home                 : /Users/<USER>
===============================================================================

List of processes or logs does not contain anything interesting.

Reproduce the failure from examples/cluster-http using ecosystem.config.js and the appended http-no-exit.js handler, comparing curl responses before and after stopping process 0 and sending SIGUSR2. Trace the pm2 sendSignal command for the stopped process; done means the remaining cluster processes continue responding to requests.