net: persist net.Socket options before connect #880

evanlucas · 2015-02-18T18:17:00Z

Remember net.Socket options called before connect and retroactively
applies them after the handle has been created.

This change makes the following function calls more user-friendly:

setKeepAlive()
setNoDelay()
ref()
unref()

I'm not thrilled with how test/parallel/test-net-persistent-nodelay.js is implemented, but I have been having trouble reliably testing the functionality of setting setNoDelay() prior to handle creation.

Related: nodejs/node-v0.x-archive#7077 and nodejs/node-v0.x-archive#8572

R= @cjihrig @bnoordhuis

vkurchatkin · 2015-02-18T18:22:27Z

lib/net.js

@@ -935,12 +958,16 @@ Socket.prototype.connect = function(options, cb) {
 Socket.prototype.ref = function() {
  if (this._handle)
    this._handle.ref();
+  else
+    this._handleOptions.ref = true;


I think that ref should actually be an int and we should ++ and -- it accordingly.

scratch that. I thought that's what libuv does, but it's just a flag

cjihrig · 2015-02-18T18:26:49Z

I have a similar solution for ref()/unref() open in nodejs/node-v0.x-archive#9115. I kind of like the idea of keeping the flags separate, instead of a _handleOptions object.

evanlucas · 2015-02-18T19:50:02Z

@cjihrig I'd be happy to refactor it to work similar to your PR. Perhaps we should open nodejs/node-v0.x-archive#9115 here as well to make it complete for both client and server?

cjihrig · 2015-02-18T22:02:59Z

@evanlucas sure, I'll open it over here.

mscdex · 2015-02-18T23:42:29Z

+1 I've run into this issue before when setting up a socket before calling connect()...

evanlucas · 2015-02-19T00:51:24Z

I went ahead and changed over to keeping the flags separate.

yorkie · 2015-02-19T12:10:24Z

lib/net.js

  if (this._handle)
    this._handle.ref();
 };


 Socket.prototype.unref = function() {
+  this._unref = new Date();


how about using Date.now(), not big deal just own opinion :)

This should probably just be true

yup true would be okay, i guess @evanlucas just wanna record the time of last calling unref()

I had originally used a date after looking at nodejs/node-v0.x-archive#9115, but I went ahead and changed it to just using a boolean

evanlucas · 2015-02-20T17:49:51Z

I added the fix for self._setNoDelay

cjihrig · 2015-02-20T18:24:00Z

lib/net.js

 };


 Socket.prototype.setKeepAlive = function(setting, msecs) {
  if (this._handle && this._handle.setKeepAlive)
    this._handle.setKeepAlive(setting, ~~(msecs / 1000));
+  else


I thought I commented on this before, but maybe not. By using an if-else, don't you take the chance of the _setKeepAlive flag having a different value from what is actually being used by the handle (same for _setNoDelay)? For example, if you call setKeepAlive() after the handle is created, then _setKeepAlive will always be null. It's not a huge deal since it's internal, but it could be confusing during debugging.

What if you did:

this._setKeepAlive = [setting, ~~(msecs / 1000)]; if (this._handle && this._handle.setKeepAlive) this._handle.setKeepAlive(setting, ~~(msecs / 1000));

evanlucas · 2015-02-20T21:57:03Z

Requested changes made :]

cjihrig · 2015-02-22T03:32:42Z

lib/net.js

@@ -121,6 +121,9 @@ function Socket(options) {
  this._hadError = false;
  this._handle = null;
  this._host = null;
+  this._unref = false;
+  this._setNoDelay = null;


Should this default to false?

cjihrig · 2015-02-22T03:44:13Z

This also needs a test for ref()/unref().

bnoordhuis · 2015-02-22T06:28:40Z

lib/net.js

  if (this._handle && this._handle.setNoDelay)
-    this._handle.setNoDelay(enable === undefined ? true : !!enable);
+    this._handle.setNoDelay(this._setNoDelay);
 };


I don't think you need new flags for this:

Socket.prototype.setNoDelay = function(enable) { if (this._connected === false) { this.once('connect', this.setNoDelay.bind(this, enable)); return; } // ... };

The use of Function#bind() is intentional, it avoids the unconditional closure allocation that would happen if you wrote the uncommon path as this.once('connect', function() { this.setNoDelay(enable); }).

evanlucas · 2015-02-23T04:10:48Z

Ok, I went ahead and went the route @bnoordhuis suggested. It seems cleaner to me. Thoughts?

Also added test for ref/unref

evanlucas · 2015-03-06T18:43:58Z

any objections to the changed implementation of this?

bnoordhuis · 2015-03-06T20:55:04Z

lib/net.js

-  if (this._handle)
-    this._handle.ref();
+  if (!this._handle) {
+    this.once('connect', this.ref.bind(this));


The .bind(this) is superfluous, just this.once('connect', this.ref) will do.

bnoordhuis · 2015-03-06T21:04:33Z

Left some comments. I don't know about the tests, they really only verify that the method is called, not that the action associated with the method is actually carried out.

evanlucas · 2015-03-19T16:11:28Z

Will get the comments addressed. As far as the tests go, I'm having trouble reliably testing the functionality. Will keep working on it now though

silverwind · 2015-04-03T16:52:27Z

Which functionality are you having issues testing?

evanlucas · 2015-04-03T23:07:37Z

tbh, the reliably testing setNoDelay (and confirming that it actually disables the Nagle Algorithm) is the one that I am having issues with as of now

This commit: - fixes development branch (v1.x -> master) - updates stability index wording - use iojs binary instead of node PR-URL: nodejs#1466 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

This reverts commit 06cfff9. Reverted because it introduced a regression where (because options were modified in the later functionality) options.host and options.port would be overridden with values provided in other, supported ways. PR-URL: nodejs#1467 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

This commit adds a test to ensure all options are NOT modified after passing them to http.request. Specifically options.host and options.port are the most prominent that would previously error, but add the other options that have default values. options.host and options.port were overridden for the one-argument net.createConnection(options) call. PR-URL: nodejs#1467 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

This brings in the '%PYTHON%' revert, and restores the correct NODE_MODULE_VERSION. PR-URL: nodejs#1482 Reviewed-By: Jeremiah Senkpiel <fishrock123@rocketmail.com>

Update AUTHORS list using tools/update-authors.sh PR-URL: nodejs#1476 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

PR-URL: nodejs#1503 Reviewed-By: Colin Ihrig <cjihrig@gmail.com> Reviewed-By: Evan Lucas <evanlucas@me.com>

as per nodejs#1502 PR-URL: nodejs#1507 Reviewed-By: Chris Dickinson <christopher.s.dickinson@gmail.com> Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

tls.connect(options) with no options.host should accept a certificate with CN: 'localhost'. Fix Error: Hostname/IP doesn't match certificate's altnames: "Host: undefined. is not cert's CN: localhost" 'localhost' is not added directly to defaults because that is not always desired (for example, when using options.socket) PR-URL: nodejs#1493 Fixes: nodejs#1489 Reviewed-By: Brendan Ashworth <brendan.ashworth@me.com> Reviewed-By: Roman Reiss <me@silverwind.io>

If `$NODE_PATH` contains trailing separators, `Module.globalPaths` will contains empty strings. When `Module` try to resolve a module's path, `path.resolve('', 'index.js')` will boil down to `$PWD/index.js`, which makes sub modules can access global modules and get unexpected result. PR-URL: nodejs#1488 Reviewed-By: Roman Reiss <me@silverwind.io>

Update the remaining markdown files to refer to the master branch. PR-URL: nodejs#1511 Reviewed-By: Jeremiah Senkpiel <fishrock123@rocketmail.com>

When buffer list less than 2, no need to calculate the length. The change's benchmark result is here: https://gist.github.com/JacksonTian/2c9e2bdec00018e010e6 It improve 15% ~ 25% speed when list only have one buffer, to other cases no effect. PR-URL: nodejs#1437 Reviewed-By: Trevor Norris <trev.norris@gmail.com> Reviewed-By: Brendan Ashworth <brendan.ashworth@me.com>

brendanashworth · 2015-04-24T01:37:11Z

lib/net.js

+              enable ? this.setNoDelay : this.setNoDelay.bind(this, enable));
+    return;
+  }
+
  // backwards compatibility: assume true when `enable` is omitted
  if (this._handle && this._handle.setNoDelay)


this._handle would always be truthy at this point, right? Is the check still necessary?

Separates out the lookup logic for net.Socket. In the event the `host` property is an IP address, the lookup is skipped. PR-URL: nodejs#1505 Reviewed-By: Chris Dickinson <christopher.s.dickinson@gmail.com> Reviewed-By: Yosuke Furukawa <yosuke.furukawa@gmail.com>

Allows customization of the lookup function used when Socket.prototype.connect is called using a hostname. PR-URL: nodejs#1505 Reviewed-By: Chris Dickinson <christopher.s.dickinson@gmail.com> Reviewed-By: Yosuke Furukawa <yosuke.furukawa@gmail.com>

Remembers net.Socket options called before connect and retroactively applies them after the handle has been created. This change makes the following function calls more user-friendly: - setKeepAlive() - setNoDelay() - ref() - unref() Related: nodejs/node-v0.x-archive#7077 and nodejs/node-v0.x-archive#8572

evanlucas · 2015-04-24T14:35:22Z

I opened a new PR pointed at master. #1518

vkurchatkin reviewed Feb 18, 2015
View reviewed changes

evanlucas force-pushed the net-persistent-opts branch 2 times, most recently from db308b3 to 9fa5e6a Compare February 19, 2015 00:49

yorkie reviewed Feb 19, 2015
View reviewed changes

evanlucas force-pushed the net-persistent-opts branch from 9fa5e6a to 7eccc69 Compare February 20, 2015 17:48

cjihrig reviewed Feb 20, 2015
View reviewed changes

evanlucas force-pushed the net-persistent-opts branch from 7eccc69 to ac3fbc1 Compare February 20, 2015 21:55

cjihrig reviewed Feb 22, 2015
View reviewed changes

bnoordhuis reviewed Feb 22, 2015
View reviewed changes

evanlucas force-pushed the net-persistent-opts branch from ac3fbc1 to d0df008 Compare February 23, 2015 04:04

silverwind mentioned this pull request Mar 6, 2015

net: enable TCP_NODELAY by default on all platforms #906

Closed

bnoordhuis reviewed Mar 6, 2015
View reviewed changes

evanlucas force-pushed the net-persistent-opts branch 2 times, most recently from b96e2a0 to 87212d4 Compare March 19, 2015 16:31

mscdex added the net Issues and PRs related to the net subsystem. label Mar 22, 2015

Working on v2.0.0

e61ee49

chrisdickinson added the land-on-master label Apr 17, 2015

brendanashworth and others added 11 commits April 18, 2015 13:13

doc: update CONTRIBUTING.md

6870764

This commit: - fixes development branch (v1.x -> master) - updates stability index wording - use iojs binary instead of node PR-URL: nodejs#1466 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

Merge v1.8.1.

59a5c98

This brings in the '%PYTHON%' revert, and restores the correct NODE_MODULE_VERSION. PR-URL: nodejs#1482 Reviewed-By: Jeremiah Senkpiel <fishrock123@rocketmail.com>

doc: update AUTHORS list

2632775

Update AUTHORS list using tools/update-authors.sh PR-URL: nodejs#1476 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

doc: add spaces to child.kill example

b16a328

PR-URL: nodejs#1503 Reviewed-By: Colin Ihrig <cjihrig@gmail.com> Reviewed-By: Evan Lucas <evanlucas@me.com>

doc: add Fishrock123 to the TC

22aafa5

as per nodejs#1502 PR-URL: nodejs#1507 Reviewed-By: Chris Dickinson <christopher.s.dickinson@gmail.com> Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

doc: update branch to master

bb254b5

Update the remaining markdown files to refer to the master branch. PR-URL: nodejs#1511 Reviewed-By: Jeremiah Senkpiel <fishrock123@rocketmail.com>

brendanashworth reviewed Apr 24, 2015
View reviewed changes

evanlucas added 3 commits April 24, 2015 07:51

evanlucas force-pushed the net-persistent-opts branch from 87212d4 to 102a993 Compare April 24, 2015 14:32

evanlucas mentioned this pull request Apr 24, 2015

net: persist net.Socket options before connect #1518

Closed

evanlucas closed this Apr 24, 2015

snyk-bot mentioned this pull request Jan 31, 2022

[Snyk] Security upgrade shelljs from 0.7.8 to 0.8.5 enterstudio/node#26

Open

aliscco mentioned this pull request Apr 27, 2023

[Snyk] Fix for 30 vulnerabilities aliscco/alisco-node#279

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

net: persist net.Socket options before connect #880

net: persist net.Socket options before connect #880

evanlucas commented Feb 18, 2015

vkurchatkin Feb 18, 2015

vkurchatkin Feb 18, 2015

cjihrig commented Feb 18, 2015

evanlucas commented Feb 18, 2015

cjihrig commented Feb 18, 2015

mscdex commented Feb 18, 2015

evanlucas commented Feb 19, 2015

yorkie Feb 19, 2015

cjihrig Feb 19, 2015

yorkie Feb 19, 2015

evanlucas Feb 20, 2015

evanlucas commented Feb 20, 2015

cjihrig Feb 20, 2015

evanlucas commented Feb 20, 2015

cjihrig Feb 22, 2015

cjihrig commented Feb 22, 2015

bnoordhuis Feb 22, 2015

evanlucas commented Feb 23, 2015

evanlucas commented Mar 6, 2015

bnoordhuis Mar 6, 2015

bnoordhuis commented Mar 6, 2015

evanlucas commented Mar 19, 2015

silverwind commented Apr 3, 2015

evanlucas commented Apr 3, 2015

brendanashworth Apr 24, 2015

evanlucas Apr 24, 2015

evanlucas commented Apr 24, 2015

net: persist net.Socket options before connect #880

net: persist net.Socket options before connect #880

Conversation

evanlucas commented Feb 18, 2015

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cjihrig commented Feb 18, 2015

evanlucas commented Feb 18, 2015

cjihrig commented Feb 18, 2015

mscdex commented Feb 18, 2015

evanlucas commented Feb 19, 2015

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

evanlucas commented Feb 20, 2015

Choose a reason for hiding this comment

evanlucas commented Feb 20, 2015

Choose a reason for hiding this comment

cjihrig commented Feb 22, 2015

Choose a reason for hiding this comment

evanlucas commented Feb 23, 2015

evanlucas commented Mar 6, 2015

Choose a reason for hiding this comment

bnoordhuis commented Mar 6, 2015

evanlucas commented Mar 19, 2015

silverwind commented Apr 3, 2015

evanlucas commented Apr 3, 2015

Choose a reason for hiding this comment

Choose a reason for hiding this comment

evanlucas commented Apr 24, 2015