net: shutdown gracefully #1164

indutny · 2015-03-16T20:56:34Z

Wait for the shutdown request completion before emitting the finish
event and destroying the socket. While this might not be that relevant
in case of plain TCP sockets, the TLS implementation sends close-notify
packet during shutdown request. Destroying socket whilst this write is
in progress tends to cause ECONNRESET errors in our tests.

indutny · 2015-03-16T20:57:07Z

cc @piscisaureus should fix that nasty issue on windows. Going to test it now.

calvinmetcalf · 2015-03-16T21:13:03Z

you beat me to opening this, through I was going call it _end for symmetry with the end method

indutny · 2015-03-16T21:36:16Z

Do not land this yet, it doesn't fix the issue on windows. Needs some investigation.

indutny · 2015-03-16T22:08:45Z

Going to push here a couple of commits just for my own testing, sorry about this!

indutny · 2015-03-17T01:21:36Z

Ok, now it seems to be ready. @piscisaureus : this fixes the issue on windows once and for free ;)

Fishrock123 · 2015-03-17T01:49:50Z

CI: https://jenkins-iojs.nodesource.com/view/iojs/job/iojs+any-pr+multi/309/

piscisaureus · 2015-03-17T02:01:15Z

@indutny

First of all, great work for digging into this and finding a fix.

I have one question though. Back in the day there we deliberately chose not to gracefully close TCP connections when serving http, for performance reasons. Instead we'd just wait until all writes were completed, and then close the connection. The kernel then takes care of doing a graceful close (note that it's still a graceful close!).

This is also the source of the subtle differences between the various finalization methods for TCP connections:

end() flushes all the writes, then calls shutdown(3), then close(3).
destroySoon() flushes all the writes and calls close(3) to do a graceful shutdown.
destroy() does a hard close, potentially causing a RST packet to be sent (so that the other side gets an ECONNRESET error).

I think it would be best if the TLS wrap also supported this kind of behavior.

I am not sure I completely understand how this patch affects that behavior.

@bnoordhuis I would also like to hear some deep thoughts from you.

indutny · 2015-03-17T02:05:09Z

@piscisaureus I guess we could use it for TLS only then, putting this __flush method to TLSSocket.

Fishrock123 · 2015-03-17T02:08:13Z

Looks like this cropped up on both windows machines:

=== release test-tls-server-verify ===
Path: parallel/test-tls-server-verify
connecting with agent1
Running 'Do not request certs. Everyone is unauthorized.'
- unauthed connection: null
  * unauthed
events.js:141
      throw er; // Unhandled 'error' event
            ^
Error: read ECONNRESET
    at exports._errnoException (util.js:734:11)
    at TLSWrap.onread (net.js:514:26)
Command: c:\workspace\iojs+any-pr+multi\nodes\iojs-win2008r2\Release\iojs.exe c:\workspace\iojs+any-pr+multi\nodes\iojs-win2008r2\test\parallel\test-tls-server-verify.js

indutny · 2015-03-17T02:58:22Z

@Fishrock123 yeah, I see it on my windows box too. Looking.

indutny · 2015-03-18T00:02:15Z

Ok, let's hope this one is a green now: https://jenkins-iojs.nodesource.com/job/iojs+any-pr+multi/316/

bnoordhuis · 2015-03-18T00:09:58Z

What's the reason for always calling shutdown() on TCP sockets now? I have a gut feeling that's going to impact performance for no good reason.

Fishrock123 · 2015-03-18T00:21:33Z

@indutny

=== release test-child-process-stdout-flush-exit ===
Path: parallel/test-child-process-stdout-flush-exit
assert.js:87
  throw new assert.AssertionError({
        ^
AssertionError: false == true
    at ChildProcess.<anonymous> (c:\workspace\iojs+any-pr+multi\nodes\iojs-win2012r2\test\parallel\test-child-process-stdout-flush-exit.js:43:5)
    at emitTwo (events.js:87:13)
    at ChildProcess.emit (events.js:169:7)
    at maybeClose (child_process.js:984:16)
    at Process.ChildProcess._handle.onexit (child_process.js:1057:5)
Command: c:\workspace\iojs+any-pr+multi\nodes\iojs-win2012r2\Release\iojs.exe c:\workspace\iojs+any-pr+multi\nodes\iojs-win2012r2\test\parallel\test-child-process-stdout-flush-exit.js

~~Edit: this doesn't seem related? Weird. Haven't seen this before/recently..~~

indutny · 2015-03-18T00:27:28Z

@Fishrock123 is it related or not?

@bnoordhuis I was trying to figure out shutdown in the middle thing, but it might be not worth it, indeed.

Fishrock123 · 2015-03-18T00:29:14Z

@indutny I think you'd be able to tell better than me, but I'd wager to say it might be, given it tests functionality related to flush and this changes some stream flush code.

indutny · 2015-03-18T17:51:25Z

@Fishrock123 I was wrong to close it yesterday :) Now I see that it should be still useful.

indutny · 2015-03-18T21:06:54Z

CI again: https://jenkins-iojs.nodesource.com/job/iojs+any-pr+multi/324/

I don't see any failures except fs-access on my box.

indutny · 2015-03-18T21:53:41Z

It appears to me that parallel/test-net-reconnect-error is just very slow on windows.

indutny · 2015-03-18T21:53:47Z

Looking into other tests too.

indutny · 2015-03-18T21:57:39Z

parallel/test-http-curl-chunk-problem works just fine without timing out.

indutny · 2015-03-18T22:00:05Z

Same about parallel/test-process-kill-null, all of the rest works too.

indutny · 2015-03-18T22:11:04Z

@piscisaureus may I ask you to take a look at these? I'm not sure how I could reproduce...

Fishrock123 · 2015-03-19T00:58:10Z

@indutny Odd, that test failure did not manifest itself again.

The timeouts seem standard and almost certainly are not related to the PR. (That should be for a different issue.)

indutny · 2015-03-19T01:56:40Z

New CI: https://jenkins-iojs.nodesource.com/job/iojs+any-pr+multi/334/

jbergstroem · 2015-03-19T02:06:38Z

Looks like the os x box fails sequential/test-child-process-fork-getconnections.js. Related?

Fishrock123 · 2015-03-19T02:06:42Z

An OS X failure has popped up, I reported this one in #1100 as I had previously seen it locally. This is the first time I've seen it on the CI. (should be unrelated)

It's not clear what additional tests are wanted. The current malformed URL test seems adequate. PR-URL: nodejs#1875 Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Evan Lucas <[email protected]>

`printDeprecationMessage` is used to deprecate modules and execution branches. PR-URL: nodejs#1822 Reviewed-By: Ben Noordhuis <[email protected]>

It makes no sense to allow people use constants from `smalloc`, since it will be removed completely eventually. PR-URL: nodejs#1822 Reviewed-By: Ben Noordhuis <[email protected]>

When loading session, OCSP response, SNI, always check that the `self._handle` is present. If it is not - the socket was closed - and we should emit the error instead of throwing an uncaught exception. Fix: nodejs/node-v0.x-archive#8780 Fix: nodejs#1696 PR-URL: nodejs#1702 Reviewed-By: Trevor Norris <[email protected]>

* Destroy `SSL*` and friends on a next tick to make sure that we are not doing it in one of the OpenSSL callbacks * Add more checks to the C++ methods that might be invoked during destructor's pending queue cleanup Fix: nodejs/node-v0.x-archive#8780 Fix: nodejs#1696 PR-URL: nodejs#1702 Reviewed-By: Trevor Norris <[email protected]>

PR-URL: nodejs#1702 Reviewed-By: Trevor Norris <[email protected]>

Add string encoding option for fs.createReadStream and fs.createWriteStream. and check argument type more strictly PR-URL: nodejs#1845 Reviewed-By: Colin Ihrig <[email protected]>

When using `iojs debug -p <pid>` with an invalid pid, the debugger printed an internal error message because it wasn't smart enough to figure out that the target process didn't exist. Now it is. PR-URL: nodejs#1863 Reviewed-By: Ben Noordhuis <[email protected]> Reviewed-By: Fedor Indutny <[email protected]> Reviewed-By: Jeremiah Senkpiel <[email protected]> Reviewed-By: Yosuke Furukawa <[email protected]>

This allows `graceful-fs` to evaluate `fs` source without access to internals. This is a temporary workaround that makes npm work. See: isaacs/node-graceful-fs#41 Fixes: nodejs#1898 PR-URL: nodejs#1903 Reviewed-By: Ben Noordhuis <[email protected]>

bnoordhuis · 2015-06-05T13:34:05Z

@indutny Is this PR still relevant? I remember when I last reviewed it I didn't really get the point. Perhaps some TLS regression tests would help?

indutny · 2015-06-05T14:01:44Z

Yes, it is relevant. I will work on regression test for this.

PR-URL: nodejs#1905 Refs: nodejs#1791 Reviewed-By: Ben Noordhuis <[email protected]> Reviewed-By: Jeremiah Senkpiel <[email protected]>

On case insensitive platforms, the rule was catching the debug module under npm and eslint. See: nodejs#1899 (comment) PR-URL: nodejs#1908 Reviewed-By: Jeremiah Senkpiel <[email protected]>

os.homedir() calls libuv's uv_os_homedir() to retrieve the current user's home directory. PR-URL: nodejs#1791 Reviewed-By: Ben Noordhuis <[email protected]> Reviewed-By: Rod Vagg <[email protected]>

Wait for the `shutdown` request completion before emitting the `finish` event and destroying the socket. While this might not be that relevant in case of plain TCP sockets, the TLS implementation sends close-notify packet during shutdown request. Destroying socket whilst this write is in progress tends to cause ECONNRESET errors in our tests.

indutny · 2015-06-06T17:51:04Z

Thinking about it, it looks like it is no longer needed. Thanks for raising the question, @bnoordhuis .

indutny mentioned this pull request Mar 16, 2015

Proposal: add _flush to Writable streams nodejs/readable-stream#112

Closed

indutny closed this Mar 18, 2015

indutny reopened this Mar 18, 2015

indutny force-pushed the fix/graceful-shutdown-in-tls branch from c57e567 to aff0cb3 Compare March 19, 2015 01:56

Trott and others added 9 commits June 4, 2015 11:38

test: remove TODO

6537fd4

It's not clear what additional tests are wanted. The current malformed URL test seems adequate. PR-URL: nodejs#1875 Reviewed-By: Colin Ihrig <[email protected]> Reviewed-By: Evan Lucas <[email protected]>

util: introduce printDeprecationMessage function

628845b

`printDeprecationMessage` is used to deprecate modules and execution branches. PR-URL: nodejs#1822 Reviewed-By: Ben Noordhuis <[email protected]>

smalloc: deprecate whole module

6d95f4f

It makes no sense to allow people use constants from `smalloc`, since it will be removed completely eventually. PR-URL: nodejs#1822 Reviewed-By: Ben Noordhuis <[email protected]>

tls_wrap: invoke queued callbacks in DestroySSL

59d9734

PR-URL: nodejs#1702 Reviewed-By: Trevor Norris <[email protected]>

fs: Add string encoding option for Stream method

353e26e

Add string encoding option for fs.createReadStream and fs.createWriteStream. and check argument type more strictly PR-URL: nodejs#1845 Reviewed-By: Colin Ihrig <[email protected]>

fs: use kMaxLength from binding

2dcef83

This allows `graceful-fs` to evaluate `fs` source without access to internals. This is a temporary workaround that makes npm work. See: isaacs/node-graceful-fs#41 Fixes: nodejs#1898 PR-URL: nodejs#1903 Reviewed-By: Ben Noordhuis <[email protected]>

saghul and others added 4 commits June 5, 2015 22:12

deps: update libuv to version 1.6.1

a5bd466

PR-URL: nodejs#1905 Refs: nodejs#1791 Reviewed-By: Ben Noordhuis <[email protected]> Reviewed-By: Jeremiah Senkpiel <[email protected]>

gitignore: don't ignore the debug npm module

02c3450

On case insensitive platforms, the rule was catching the debug module under npm and eslint. See: nodejs#1899 (comment) PR-URL: nodejs#1908 Reviewed-By: Jeremiah Senkpiel <[email protected]>

os: add homedir()

6e78e5f

os.homedir() calls libuv's uv_os_homedir() to retrieve the current user's home directory. PR-URL: nodejs#1791 Reviewed-By: Ben Noordhuis <[email protected]> Reviewed-By: Rod Vagg <[email protected]>

indutny force-pushed the fix/graceful-shutdown-in-tls branch from bd4daca to f3503ba Compare June 6, 2015 16:26

indutny closed this Jun 6, 2015

indutny deleted the fix/graceful-shutdown-in-tls branch June 6, 2015 17:51

indutny restored the fix/graceful-shutdown-in-tls branch June 6, 2015 17:51

indutny deleted the fix/graceful-shutdown-in-tls branch August 8, 2015 22:43

snyk-bot mentioned this pull request Jul 17, 2020

[Snyk] Security upgrade standard from 11.0.1 to 12.0.0 adamlaska/node#37

Open

snyk-bot mentioned this pull request Aug 11, 2020

[Snyk] Fix for 3 vulnerabilities turkdevops/node#11

Merged

This was referenced Apr 24, 2023

[Snyk] Fix for 3 vulnerabilities aliscco/alisco-node#105

Open

[Snyk] Fix for 3 vulnerabilities aliscco/alisco-node#113

Open

This was referenced Apr 25, 2023

[Snyk] Fix for 10 vulnerabilities aliscco/alisco-node#134

Open

[Snyk] Security upgrade standard from 11.0.1 to 12.0.0 aliscco/alisco-node#136

Open

aliscco mentioned this pull request Apr 25, 2023

[Snyk] Fix for 3 vulnerabilities aliscco/alisco-node#166

Open

snyk-bot mentioned this pull request Apr 26, 2023

[Snyk] Fix for 9 vulnerabilities aliscco/alisco-node#210

Open

aliscco mentioned this pull request Apr 28, 2023

[Snyk] Fix for 9 vulnerabilities aliscco/alisco-node#311

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

net: shutdown gracefully #1164

net: shutdown gracefully #1164

indutny commented Mar 16, 2015

indutny commented Mar 16, 2015

calvinmetcalf commented Mar 16, 2015

indutny commented Mar 16, 2015

indutny commented Mar 16, 2015

indutny commented Mar 17, 2015

Fishrock123 commented Mar 17, 2015

piscisaureus commented Mar 17, 2015

indutny commented Mar 17, 2015

Fishrock123 commented Mar 17, 2015

indutny commented Mar 17, 2015

indutny commented Mar 18, 2015

bnoordhuis commented Mar 18, 2015

Fishrock123 commented Mar 18, 2015

indutny commented Mar 18, 2015

Fishrock123 commented Mar 18, 2015

indutny commented Mar 18, 2015

indutny commented Mar 18, 2015

indutny commented Mar 18, 2015

indutny commented Mar 18, 2015

indutny commented Mar 18, 2015

indutny commented Mar 18, 2015

indutny commented Mar 18, 2015

Fishrock123 commented Mar 19, 2015

indutny commented Mar 19, 2015

jbergstroem commented Mar 19, 2015

Fishrock123 commented Mar 19, 2015

bnoordhuis commented Jun 5, 2015

indutny commented Jun 5, 2015

indutny commented Jun 6, 2015

net: shutdown gracefully #1164

net: shutdown gracefully #1164

Conversation

indutny commented Mar 16, 2015

indutny commented Mar 16, 2015

calvinmetcalf commented Mar 16, 2015

indutny commented Mar 16, 2015

indutny commented Mar 16, 2015

indutny commented Mar 17, 2015

Fishrock123 commented Mar 17, 2015

piscisaureus commented Mar 17, 2015

indutny commented Mar 17, 2015

Fishrock123 commented Mar 17, 2015

indutny commented Mar 17, 2015

indutny commented Mar 18, 2015

bnoordhuis commented Mar 18, 2015

Fishrock123 commented Mar 18, 2015

indutny commented Mar 18, 2015

Fishrock123 commented Mar 18, 2015

indutny commented Mar 18, 2015

indutny commented Mar 18, 2015

indutny commented Mar 18, 2015

indutny commented Mar 18, 2015

indutny commented Mar 18, 2015

indutny commented Mar 18, 2015

indutny commented Mar 18, 2015

Fishrock123 commented Mar 19, 2015

indutny commented Mar 19, 2015

jbergstroem commented Mar 19, 2015

Fishrock123 commented Mar 19, 2015

bnoordhuis commented Jun 5, 2015

indutny commented Jun 5, 2015

indutny commented Jun 6, 2015