Support Unix domain socket (UDS) server #120

andig · 2017-10-07T14:00:41Z


Resolves / closes #25

clue

Awesome, would love to get this in! 🎉

Can you add the missing tests and documentation in the README file? Also, this should probably be integrated into the main Server class.

Keep it up! 👍

andig · 2017-10-07T16:21:22Z

Can you add the missing tests and documentation in the README file? Also, this should probably be integrated into the main Server class.

Done :)

jsor · 2017-10-07T18:11:27Z

src/UnixServer.php

+        if (strpos($path, '://') === false) {
+            $path = 'unix://' . $path;
+        } elseif (substr($path, 0, 7) !== 'unix://') {
+            throw new \InvalidArgumentException('Given URI "' . $path . '" is InvalidArgumentException');


...is InvalidArgumentException -> ...is invalid

clue · 2017-10-07T21:03:41Z

Thank you for the quick update!

I don't see any major issues after eyeballing this and would love to get this in I'll try to get back to this tomorrow and address some minor outstanding formatting issues myself 👍

WyriHaximus · 2017-10-07T21:27:38Z

README.md

+To listen on a Unix domain socket, the uri MUST be prefixed with the `unix://` scheme:
+
+```php
+$server = new Server('unix:///tmp/.sock', $loop);


WyriHaximus · 2017-10-07T21:45:32Z

tests/UnixServerTest.php

+    }
+
+    /**
+     * @covers React\EventLoop\StreamSelectLoop::tick


Why does this test covers React\EventLoop\StreamSelectLoop::tick?

WyriHaximus · 2017-10-07T21:45:48Z

tests/UnixServerTest.php

+    }
+
+    /**
+     * @covers React\EventLoop\StreamSelectLoop::tick


Why does this test covers React\EventLoop\StreamSelectLoop::tick?

I must admit that the tests are not original art but all copied from TcpServer- so that's a copy/paste artefact. Remove the covers hint here?

Yeah noticed that, but to be honest that is alright. There are few distinct differences that I'm not bothered by it 😄 . The hints are fine as long as they make sense, covering a loop tick in a server test doesn't =D

Should I change the TcpServerTest, too?

Nah out of scope for this PR

WyriHaximus · 2017-10-07T21:48:32Z

tests/UnixServerTest.php

+
+    private function getRandomSocketUri()
+    {
+        return "unix://" . sys_get_temp_dir() . DIRECTORY_SEPARATOR . uniqid(rand(), true) . '.sock';


Just nitpicking here but why not go with uniqid('reactphp-socket-tests-unix-server-', true) to make it explicit when debugging that it's a socket for the UnixServer test?

Turned out to not be a good idea, see 9993763. I've added a check for the client as well.

Interesting, what happened?

Warning is shown (in the tests without @) that the path is truncated. My thinking was that we should rather be explicit with failure instead of the user potentially reusing an existing socket due to same name.

kelunik · 2017-10-08T11:08:13Z

src/UnixConnector.php

@@ -31,8 +31,9 @@ public function connect($path)
            return Promise\reject(new InvalidArgumentException('Given URI "' . $path . '" is invalid'));
        }

-        if (strlen($path) > 104) {
-            return Promise\reject(new InvalidArgumentException('Given URI "' . $path . '" exceeds maximum allowed length of 104 bytes'));
+        $maxLen = PHP_OS === 'Darwin' ? 104 : 108;


Is the max-length including or excluding the trailing NUL byte? A link would probably be good for future reference.

kelunik · 2017-10-08T11:12:17Z

tests/UnixServerTest.php

-        // string > 104 characters
-        $path = "unix://" . sys_get_temp_dir() . DIRECTORY_SEPARATOR . str_repeat('_', 100) . '.sock';
+        // string > 104/108 characters
+        $path = "unix://" . sys_get_temp_dir() . DIRECTORY_SEPARATOR . str_repeat('_', 104) . '.sock';


It should probably subtract the length of the sys_get_temp_dir() for the str_repeat.

It just needs to be long enough to trigger the exception? I did not want to go into complex scenarios here.

Yes, but it should trigger on exactly X byte and succeed otherwise, no? Probably should have a tests that it works with one less character.

Thanks for the discussion so far! I wonder if we really need this check in the first place? It looks like stream_socket_server() rejects an invalid path anyway, so why not rely on this?

https://3v4l.org/Gff8D

This would allow us to remove any platform specific constants / assumptions. What do you think about this?

Never mind, it looks like it actually still succeeds despite this warning and instead creates a truncated socket path. Sorry for the confusion, go ahead :-)

right, your test is invalid, see https://3v4l.org/4OPt0

kelunik · 2017-10-08T12:42:05Z

tests/UnixConnectorTest.php

@@ -65,12 +65,28 @@ public function testValid()

    public function testConnectWithOverlyLongAddress()
    {
+        https://serverfault.com/questions/641347/check-if-a-path-exceeds-maximum-for-unix-domain-socket


Missing //.

the-eater · 2017-10-11T08:34:27Z

What are currently the blocking issues with this PR? I need it in my project and would like to get rid of the dev-unix as v0.8.4, also wouldn't it be sane to stop supporting PHP5.3-5.5 as they are all EOL?

andig · 2017-10-11T09:26:20Z

I will remove the loop->tick from the tests, then it should imho be ready for final review

Ready for squashed commit?

WyriHaximus · 2017-10-11T20:01:02Z

@andig is there a twitter handle or email address we can reach? If you don't want to post it publicly feel free to send an email to the email address on my profile.

the-eater · 2017-10-11T20:14:49Z

src/Server.php

-            $server = new SecureServer($server, $loop, $context['tls']);
+            if ($scheme === 'tls') {
+                $server = new SecureServer($server, $loop, $context['tls']);
+            }


Maybe export the Server constructing logic to a separate method like this: the-eater@2defb1b?diff=split

Functionally LGTM, however I don't currently see this covered in the test suite (plus minor CS issue above). Can you update this? 👍

@clue what's your expectation here? Testing the type of the create private server member? Would need a hint how to do this due to the private nature.

This class is a facade that mostly contains some glue code to make the underlying APIs more easily accessible. Just check the other high-level functional tests and the code coverage. IMHO it's sufficient to create an Unix domain socket server with this facade and check a client connection can be established.

Gotcha. Tests added.

andig · 2017-10-19T16:41:33Z

Ping @WyriHaximus what is missing to get this aging pr merged? Final comment above?

WyriHaximus · 2017-10-19T17:58:03Z

@andig I will have a look this weekend but last time I ran into some issues. Will ping you with some questions then 👍

clue · 2017-10-20T13:53:03Z

Thank you @andig! We're currently trying to focus our efforts on stabilizing things so we can get a stable v1.0.0 release out as soon as possible. This features was originally planned for v1.1.0 via #25, but given the high quality of this PR I'm open to getting this in before the v1.0.0 release.

I would like to get the outstanding v0.8.5 bug fix release out in the next days first before introducing this new feature. Does it make sense for you to postpone this to the next v0.8.6 release so we can look into this immediately after the bug fix release is out?

andig · 2017-10-20T14:26:20Z

@clue I was hoping to get this into 0.8.5. It is a new feature but does not break compatibility. Only exception is the handling of maximum UDS path lengths that would otherwise potentially have caused unexpected behaviour. So- the sooner the better as ppm depends on it.

clue · 2017-10-22T16:52:18Z

src/UnixConnector.php

+
+        $maxLen = PHP_OS === 'Darwin' ? 104 : 108;
+        if (strlen($path) > $maxLen) {
+            return Promise\reject(new InvalidArgumentException('Given URI "' . $path . '" exceeds maximum allowed length'));


I think you're addressing a relevant issue here, thank you! However, I would both get a better understanding of this problem before getting this in and also believe that this is unrelated to this new feature. Does it make sense to you to remove this for now and file this a separate PR? 👍

Some thoughts (short brain dump only):

The path limit seems to apply to the path as given, a prior chdir() may be used to work around this for use use cases

The unix:// scheme is not included in this limit

The limit is system dependent and 108 for Linux and 104 for BSD (Mac) and lower for other systems

Linux man pages suggest 108 is a valid length, but PHP already raises a notice here

The E_NOTICE was apparently introduced via https://bugs.php.net/bug.php?id=60106 in PHP 5.4.1+

All relevant versions (including < PHP 5.4.1) already (silently) truncated the path

My experiments: https://3v4l.org/hSRCU

Thanks for the experiments :)

The path limit seems to apply to the path as given, a prior chdir() may be used to work around this for use use cases

Would need to double-check.

The unix:// scheme is not included in this limit

Why do you think so? The previous unit tests seemed to indicate that it does?

The limit is system dependent and 108 for Linux and 104 for BSD (Mac) and lower for other systems

Unless we could identify from some setting which those are we could only live with assumptions. Custom kernels might have yet other limits?

Linux man pages suggest 108 is a valid length, but PHP already raises a notice here

I can only imagine PHP takes the trailing 0 into account for that check. Might be a bug?

All relevant versions (including < PHP 5.4.1) already (silently) truncated the path

Meaning your suggestion would be to accept this situation?

I'd be happy to not re-raise this PR but thought it would be good as a precaution as I've only accidentally hit this limit and it might have gone unnoticed.

Long story short: do you think it is worth pursuing this or rather entirely skip it?

WyriHaximus · 2017-10-22T16:59:43Z

@andig could you share your nginx config? And how do you deal with existing .sock files when you haven't started the process yet?

andig · 2017-10-22T17:49:23Z

@clue thank you for the review. I have removed all path length stuff and will raise WIP PR once this is in and I can rebase on top of it.

could you share your nginx config?

@WyriHaximus I can't- no nginx in use, sorry.

And how do you deal with existing .sock files when you haven't started the process yet?

Not sure. If the socket is not in use the server will allow this. What is the expected behaviour?

clue

Awesome, thank you for keeping up with this @andig, much appreciated! 👍 I've squashed the 16 commits into 2 commits and made some minor adjustments to documentation, otherwise LGTM, now let's get this in! 🎉

andig · 2017-11-13T06:17:18Z

Will check the test failures, but those are unrelated.

andig · 2017-11-13T08:17:54Z

Should the OSX test failures re-appear please check travis-ci/travis-ci#8552 (comment)

clue · 2017-11-13T16:12:11Z

Unrelated Mac OS X test failure has been resolved by restarting build, see also #134 for follow-up PR

WyriHaximus

🚢 🇮🇹

andig · 2017-11-17T09:57:29Z

@clue is this something that could get packaged into 0.8.6 soonish?

clue · 2017-11-17T09:58:24Z

@andig Very soon-ish, stay tuned

cboden · 2017-11-17T15:26:57Z

Thanks @andig! Fantastic work!!! 🎉

andig mentioned this pull request Oct 7, 2017

Cannot use HttpServer with UDS reactphp/http#231

Closed

clue added help wanted new feature labels Oct 7, 2017

WyriHaximus requested review from jsor, WyriHaximus and clue October 7, 2017 14:10

clue reviewed Oct 7, 2017

View reviewed changes

andig mentioned this pull request Oct 7, 2017

Upgrade React/Http to v0.8 php-pm/php-pm-httpkernel#63

Merged

4 tasks

andig force-pushed the unix branch from be35aef to 85f646b Compare October 7, 2017 16:20

jsor reviewed Oct 7, 2017

View reviewed changes

WyriHaximus reviewed Oct 7, 2017

View reviewed changes

kelunik reviewed Oct 8, 2017

View reviewed changes

clue mentioned this pull request Oct 10, 2017

Add support for Unix domain sockets (UDS) #17

Closed

andig force-pushed the unix branch from a4c4576 to 45f121d Compare October 11, 2017 10:40

the-eater reviewed Oct 11, 2017

View reviewed changes

andig mentioned this pull request Oct 22, 2017

Upgrade React/Http to v0.8 php-pm/php-pm#271

Merged

5 tasks

clue reviewed Oct 22, 2017

View reviewed changes

clue added this to the v0.8.6 milestone Oct 23, 2017

andig added 2 commits November 13, 2017 00:56

Add advanced UnixServer

7f4814f

Add Unix domain socket (UDS) support to Server with unix:// URI scheme

9257632

clue force-pushed the unix branch from ea2b0c7 to 9257632 Compare November 12, 2017 23:58

clue approved these changes Nov 13, 2017

View reviewed changes

clue changed the title ~~Add unix server~~ Support Unix domain socket (UDS) server Nov 13, 2017

jsor approved these changes Nov 16, 2017

View reviewed changes

WyriHaximus approved these changes Nov 16, 2017

View reviewed changes

WyriHaximus merged commit 626bf5b into reactphp:master Nov 16, 2017

andig deleted the unix branch November 17, 2017 09:56

Support Unix domain socket (UDS) server #120

Support Unix domain socket (UDS) server #120

Conversation

andig commented Oct 7, 2017 • edited by clue Loading

clue left a comment

Choose a reason for hiding this comment

andig commented Oct 7, 2017

Choose a reason for hiding this comment

clue commented Oct 7, 2017

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

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

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

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

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

the-eater commented Oct 11, 2017 • edited Loading

andig commented Oct 11, 2017 • edited Loading

WyriHaximus commented Oct 11, 2017

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

Choose a reason for hiding this comment

andig commented Oct 19, 2017

WyriHaximus commented Oct 19, 2017

clue commented Oct 20, 2017

andig commented Oct 20, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

WyriHaximus commented Oct 22, 2017

andig commented Oct 22, 2017

clue left a comment

Choose a reason for hiding this comment

andig commented Nov 13, 2017

andig commented Nov 13, 2017

clue commented Nov 13, 2017

WyriHaximus left a comment

Choose a reason for hiding this comment

andig commented Nov 17, 2017

clue commented Nov 17, 2017

cboden commented Nov 17, 2017

andig commented Oct 7, 2017 •

edited by clue

Loading

the-eater commented Oct 11, 2017 •

edited

Loading

andig commented Oct 11, 2017 •

edited

Loading