test: fix net-server-max-connections-close-makes-more-available

[santigimeno]

• Handle connection 'error' event so the test ends.

[santigimeno]

Before this change I was getting this error:

creating connection 0
sending message: 0
received message: 0
connection 0 received response
creating connection 1
sending message: 1
events.js:141
      throw er; // Unhandled 'error' event
            ^
Error: read ECONNRESET
    at exports._errnoException (util.js:846:11)
    at TCP.onread (net.js:540:26)

Not sure if this change is correct though, as I'm not familiar with Promises

[targos]

I am not getting this error :

➜  io.js git:(master) ✗ ./iojs test/parallel/test-net-server-max-connections-close-makes-more-available.js
creating connection 0
sending message: 0
received message: 0
connection 0 received response
creating connection 1
sending message: 1
ending 1
closing connection 0
ending 0
creating connection 2
sending message: 2
received message: 2
connection 2 received response
creating connection 3
sending message: 3
ending 3
closing connection 2
ending 2

Is this error event supposed to be emitted ?

[santigimeno]

I'm getting this error consistently in a FreeBSD 10.1 box

[bnoordhuis]

Refs #1855 and /cc @brendanashworth @Trott

[Trott]

Perhaps add a check to make sure the err.code is ECONNRESET just to make sure some other problem doesn't mask an issue here?

[Trott]

This isn't happening for me (just built from master on OS X). If some setups send ECONNRESET and others just silently drop the connection, that's probably a bigger bug than just tweaking this test, yeah? (Although I suppose there may be OS-specific limitations on what types of errors can be retrieved. That doesn't seem likely to be an issue with this, though.)

[silverwind]

This test failed on a bunch of platforms before: https://jenkins-iojs.nodesource.com/job/iojs+any-pr+multi/758/

[brendanashworth]

I should have run the test on the CI, sorry for that. But there does seem to be an issue if an ECONNRESET is thrown on some platforms and not others.

[Trott]

This version in this pull request is failing for me on OS X:

assert.js:88
  throw new assert.AssertionError({
        ^
AssertionError: [] deepEqual [ 1, 3 ]
    at process.<anonymous> (/Users/trott/io.js/test/parallel/test-net-server-max-connections-close-makes-more-available.js:87:10)
    at emitOne (events.js:82:20)
    at process.emit (events.js:169:7)

This would seem to suggest that whether or not an error event is triggered when a connection is rejected may be platform-dependent. It would also seem to me that the FreeBSD behavior is correct (and the test should more resemble what's in this pull request) and the OS X behavior is a bug.

[silverwind]

CI with this PR: https://jenkins-iojs.nodesource.com/job/iojs+any-pr+multi/759/

[santigimeno]

@Trott I have applied your suggestions
Also I have removed the failed check so it passes in other platforms, but I agree that it looks like a bug

[brendanashworth]

CI is happy, I'm happy.

[Trott]

I created an issue for the not-triggering-ECONNRESET thing.

[brendanashworth]

We should get this in. LGTM.

[Trott]

LGTM

(Not sure if my LGTM counts for anything, but if it doesn't count for nothing, then there it is.)

[brendanashworth]

Thanks everyone, landed in a804026. (I had to fix the commit log a little bit)