Investigate flaky parallel/test-debug-port-numbers

[Trott]

This appears to be the (or one) cause of the recent firestorm of EADDRINUSE failures on CI. (@jbergstroem reported that the test was running and not exiting on CI hosts exhibiting the EADDRINUSE issue.)

Ran a stress test to confirm, and sure enough...

https://ci.nodejs.org/job/node-stress-single-test/nodes=freebsd102-64/748/console

+ OK=5
+ echo '5   OK: 5   NOT OK: 0   TOTAL: 999'
5   OK: 5   NOT OK: 0   TOTAL: 999
+ for i in '`seq $RUN_TIMES`'
+ python tools/test.py -p tap --mode=release parallel/test-debug-port-numbers
1..1
ok 1 parallel/test-debug-port-numbers
  ---
  duration_ms: 1.376
  ...
+ OK=6
+ echo '6   OK: 6   NOT OK: 0   TOTAL: 999'
6   OK: 6   NOT OK: 0   TOTAL: 999
+ for i in '`seq $RUN_TIMES`'
+ python tools/test.py -p tap --mode=release parallel/test-debug-port-numbers
1..1
not ok 1 parallel/test-debug-port-numbers
# TIMEOUT
# debug> debug> debug> debug> �< Debugger listening on port 12346�< Debugger listening on port 12347
# debug> �connecting to 127.0.0.1:12347 ...
# ��debug> < Debugger listening on port 12348< Error: listen EADDRINUSE :::12349�connecting to 127.0.0.1:12346 ...
# debug> �connecting to 127.0.0.1:12348 ..
# debug> �connecting to 127.0.0.1:12349 .... ok
# debug>  ok
# debug>  ok
# debug> �<     at Object.exports._errnoException (util.js:1007:11)
# <     at exports._exceptionWithHostPort (util.js:1030:20)
# <     at Agent.Server._listen2 (net.js:1253:14)
# <     at listen (net.js:1289:10)
# <     at Agent.Server.listen (net.js:1385:5)
# <     at Object.start (_debug_agent.js:21:9)
# <     at startup (node.js:74:44)
# <     at node.js:444:3
# debug>  ok
# debug> �break in test/parallel/test-debug-port-numbers.js:1
# �> 1 'use strict';
# �  2 
# �  3 const common = require('../common');
# debug> �break in test/parallel/test-debug-port-numbers.js:1
# �> 1 'use strict';
# �  2 
# �  3 const common = require('../common');
# debug> �break in test/parallel/test-debug-port-numbers.js:1
# �> 1 'use strict';
# �  2 
# �  3 const common = require('../common');
# debug> 
  ---
  duration_ms: 60.65
  ...
+ NOK=1
+ echo '7   OK: 6   NOT OK: 1   TOTAL: 999'
7   OK: 6   NOT OK: 1   TOTAL: 999
+ for i in '`seq $RUN_TIMES`'
+ python tools/test.py -p tap --mode=release parallel/test-debug-port-numbers
1..1
not ok 1 parallel/test-debug-port-numbers
# TIMEOUT
# debug> debug> debug> debug> ��< Error: listen EADDRINUSE :::12349
# < Debugger listening on port 12347debug> �connecting to 127.0.0.1:12349 ..
# .�< Debugger listening on port 12346
# debug> �connecting to 127.0.0.1:12347 ...��< Debugger listening on port 12348<     at Object.exports._errnoException (util.js:1007:11)
# <     at exports._exceptionWithHostPort (util.js:1030:20)
# <     at Agent.Server._listen2 (net.js:1253:14)
# <     at listen (net.js:1289:10)
# <     at Agent.Server.listen (net.js:1385:5)
# <     at Object.start (_debug_agent.js:21:9)
# <     at startup (node.js:74:44)
# <     at node.js:444:3
# 
# debug> debug> �connecting to 127.0.0.1:12348 ..debug>  ok
# . ok
# debug> debug> �connecting to 127.0.0.1:12346 ... ok
# debug>  ok
# debug> �break in test/parallel/test-debug-port-numbers.js:1
# �> 1 'use strict';
# �  2 
# �  3 const common = require('../common');
# debug> �break in test/parallel/test-debug-port-numbers.js:1
# �> 1 'use strict';
# �  2 
# �  3 const common = require('../common');
# debug> �break in test/parallel/test-debug-port-numbers.js:1
# �> 1 'use strict';
# �  2 
# �  3 const common = require('../common');
# debug> 

The commit that landed this test is dated October, but it only landed two days ago.

/cc @bnoordhuis

[bnoordhuis]

I can reproduce and I figured it out, see #7037 for a workaround.

The problem is that the debugger process spawns the debuggee (a.k.a. the inferior) but doesn't always terminate it on exit. If you modify the test to send a SIGKILL to the debugger, the test always fails on the second run, because the inferior stays around, hogging the port.