forgot to add PHPUnit version --- 3.5.14

The selenium project does not support the PHP bindings. Please can you try with either
python, ruby, .NET or Java to see if you can replicate.

If the issue is not there please can you raise this with the owner of the PHP bindings

I believe the problem is in "-role hub"

If I understand correctly, HUB should act the same way as RC from 'outside'. Thus running
tests on HUB and RC should return same results in the same way.

but this is not the case
java -jar selenium-server-standalone-2.3.0.jar
runs OK

java -jar selenium-server-standalone-2.3.0.jar -role hub (with one RC connected)
error - Could not connect to the Selenium RC server

same machine, same tests

I have the same problem, as soon as I run -role hub I get the "Session not available"
error. No problem without the role

Attached you can find my fix for PHPUnit support.
I don't know if it's a good fix, but it solved my problems :)

Can someone vet this patch please?

What's the patch aiming to do?  The reply to the client should originate on one of the
nodes, not the hub.  The hub just operates as a proxy, unless there's an issue, in
which case it can return error messages.

@joc...@netlog.com,
can you send the exact command you use to launch the hub, the node, and the stacktrace
you have on both when your problem appear.

thanks,
François

To launch the hub:

sudo java -jar build/java/server/src/org/openqa/grid/selenium/selenium-standalone.jar
-host 10.0.1.4 -role hub

Launch the rc:
java -jar selenium-standalone.jar -debug -role rc -hub http://10.0.1.4:4444/grid/register

Log entry from hub:
Sep 5, 2011 8:45:17 PM org.openqa.grid.web.servlet.handler.RequestHandler getSession
WARNING: Cannot find session null in the registry.

No long entry from RC.

Got the latest version from svn and compiled with ./go selenium-server-standalone

If I use PHPUnit to connect directly to an RC then it works fine, only an issue when
I connect to the hub from PHPUnit.

Let me know if you need more info.

Having the same issue here using PHPUnit 3.5.15 and Selenium standalone 2.7.0. Setting
the testsuite to use port 4444 (the hub) results in the following error:

Sep 26, 2011 4:48:53 PM org.openqa.grid.web.servlet.handler.RequestHandler getSession
WARNING: Cannot find session null in the registry.

Connecting to the RC on port 5555 works fine, except I lose all the advantages of Grid.

Any luck with this? I'm developing nodejs bindings and run into the exact same problem.

Also running into this exact issue.
I'm running PHPunit 3.6 and selenium-standalone-2.12.0.jar

I was running the command direct against the hub without using the PHPUnit-Framework.
As a result I received the same bug as you guys. Hence the bug doesn't seem to be effected
by PHPUnit.

Command:
http://localhost:4444/selenium-server/driver/?cmd=getNewBrowserSession&1=*firefox&2=http://www.google.de

Configuration:
Hub runs on localhost:4444
Remote Control runs on localhost:5555

Result of command:
HTTP ERROR: 500
org.openqa.grid.common.exception.GridException: Session [null] not available - []
RequestURI=/selenium-server/driver/

Greetz Michael

i was trying selenium-server-standalone-2.13.0.jar and selenium-server-standalone-2.14.0.jar
w/ and w/out phpunit 3.5.15

Configuration:
Hub runs on somehost1:4444
Remote Control runs on somehost2:5556

Command (does NOT work):
http://somehost1:4444/selenium-server/driver/?cmd=getNewBrowserSession&1=*firefox&2=http://www.google.com

Result of command:
HTTP ERROR: 500
java.lang.NullPointerException
RequestURI=/selenium-server/driver/
Powered by Jetty://

Command (works):
http://somehost2:5556/selenium-server/driver/?cmd=getNewBrowserSession&1=*firefox&2=http://www.google.com

Result of command:
OK,55986f5f68bf40ffbedfbec1f1e89440

The previous fix works until r14920 (when TestSession.java was rewritten).
Please find attached a patch that does the same, but from r14920.

Patch in #18 worked after recompiling.

Tested with PHPUnit 3.6.7 and Grid2

psemuu, why is the patch necessary?  What's different about the PHP bindings?

I can confirm patch #18 solves the issue.

With the current release build (2.17.0) PHPUnit can launch selenium tests directly
to a node, but not through the hub. Hub-node communication does work, since the same
setup launches tests nicely through the hub when using one of the newer PHP WebDriver
implementations.

Unfortunately I'm not well enough versed in the Selenium RC protocol and the Selenium
architecture to be able to judge if the problem is due to how the PHPUnit Selenium
implementation sends requests, or if this is some kind of edge case bug in the hub/node
communication implementation. Comment #14 does make it clear this is not really a PHP/PHPUnit
issue, but rather something else.

Is there a reason why this patch cannot be applied to the offical source tree? I applied
the patch (manually) against r15573, works fine from what I can tell so far.

successfully applyed patch, all working, except local russian language. native language
(such as $this->click("link=Зарегистрироваться"); ) directly to RC works fine, but
via HUB converts to something unreadable, and hangs with FAILED link.

The patches attached previously append the proper query part to the url of the proxy
request (created in TestSession.java in method prepareProxyRequest). Without the query
part the command request does not reach the selenium server (running as a grid node)
and a java.lang.NullPointerException is raised.
The bug can be reproduced easily as shown in comments 14 and 15.

The second bug mentioned in comment 23 is connected with the fact that the same proxyRequest
is executed by Apache's DefaultHttpClient which has ISO-8859-1 as the default charset.
A request that goes directly to the selenium server (in case of PHPUnit at least) has
no Content-Type and thus the url params are decoded correctly using utf-8.
But a request that is proxied through the grid has a Content-Type with a ISO-8859-1
charset and the url params are incorrectly decoded resulting in errors.
The obvious fix is to add a header with a proper charset (utf-8), whenever the initial
request is missing one. Attached you can find a patch fixing the problem.

References:
http://hc.apache.org/httpcomponents-client-ga/httpclient/apidocs/index.html?org/apache/http/impl/client/DefaultHttpClient.html
http://hc.apache.org/httpcomponents-core-ga/httpcore/apidocs/index.html?constant-values.html

romantschuk,

The reason I want to better understand before applying this patch is I really don't
want to break the bindings in every other language.  So far all I've seen is "this
makes it work," which is valuable, but I really want to know why.

I'll try to match the patch up with comment #14 to see what might be going on.  I know
that this hasn't been an issue with Ruby or Java bindings (and I believe the Python
ones as well).

psemu,

Thanks for the clarification.  I'll take a look.

I just compared the python client with the PHPUnit extension and all became clear.
PHPUnit uses a GET request to send a command to the selenium server, e.g.:
http://host:port/selenium-server/driver/?cmd=getNewBrowserSession&1=*firefox&2=http://www.google.com
The python client does this by sending a request to the server by POST, e.g.:
URL: http://host:port/selenium-server/driver/
headers: Content-Type: application/x-www-form-urlencoded; charset=utf-8
body: cmd=getNewBrowserSession&1=*firefox&2=http://www.google.com

That's why the patches where necessary - PHPUnit does it the wrong way. And so did
we by testing the grid with a web browser.

I'm going to post an issue on https://github.com/sebastianbergmann/phpunit tomorrow
and propose a fix there. It will address both of the issues discussed here (passing
the query string and incorrect encoding). Hopefully it will be merged into PHPUnit's
driver implementation. I will also add a link to it here for continuity.

Sorry for the commotion.

Nice find, was there a documented change for dropping GET in either Selenium server
or PHPUnit_Selenium?

the webdriver JSON protocol is very well documented/specified.
The selenium 1 legacy protocol isn't. Each implementation does things slightly differently.One
more reason to move to webdriver :)


thanks,
François 

Thank you very much! 

patch works perfect!

Nice to know that the core reason was determined. :)

Please link the PHPUnit issue once it is created. At the time of writing it seems the
issue isn't there yet:
https://github.com/sebastianbergmann/phpunit/issues

Our organization does plan to move to WebDriver, but with four implementations for
PHP (that I've found so far) and no "official" upgrade path with PHPUnit we're waiting
and seeing for a little while longer.

FYI on the WebDriver migration for PHP, I worked on creating an interface for PHPUnit
to run WebDriver instead, but haven't completed it yet. I'll post up when I get chance
to complete.

Sadly in all the existing implementations I've seen, none care to focus on migration
and backward compatibility support from PHPUnit-Selenium (RC). Rather the ones that
have PHPUnit support implement their own extension of PHPUnit separate from PHPUnit-Selenium.

My approach is to create a wrapper interface between PHPUnit and WebDriver that provides
same API as PHPUnit-Selenium (essentially a WebDriverBackedSelenium in PHP). And as
such, existing PHP Selenium RC code could in most cases run with this on WebDriver
mostly unmodified. I know it won't be perfect but beats rewriting/porting from scratch.
This wrapper interface can also be ported among the existing PHP WebDriver implementations,
so one can choose their preferred implementation to use with PHPUnit. Those who want
pure WebDriver API can then unwrap my WebDriverBackedSelenium API to expose WebDriver
and not use the Selenium RC API.

Giorgio is already working on WebDriver in PHPUnit_Selenium.

See PHPUnit_Selenium 1.2.1 (PHPUnit/Extensions/Selenium2TestCase)

Can someone with access add the bug/feature request for the GRID problem on the phpunit-selenium
issues list?  Might get more attention then on the larger phpunit list.  I haven't
tried the patch on the server side.  Is it pretty simple to
make the curl requests use POST instead of GET in the Driver?

forgot the link: https://github.com/sebastianbergmann/phpunit-selenium

Good to know the project PHPUnit-Selenium is finally getting Selenium 2 support. Last
I heard, Sebastian wasn't going to add support for WebDriver, guess the new maintainer
Giorgio decided otherwise.

As I have no time to add the issue right now, you can find attached the patch that fixes
the PHPUnit_Selenium extension (in version 1.2.0) to work with Grid2. Apply it in your
PHPUnit/Extensions/SeleniumTestCase/ directory.
The patched driver sends the request just like clients in other languages (checked
python and ruby clients) - the command and it's arguments are sent in the body by POST
with proper headers.

I would also like to mention that the patches that I posted earlier do 2 things:
- adjust Grid to process requests just like the RC server (don't know about WebDriver)
- it accepts both POST and GET requests
- add a sane default for the proxy request's charset if the initial request is missing
one (it can also be done elsewhere)
They may actually turn out to be quite useful.

At the same time I must emphasize the difficulty of migrating a really large number
of tests to WebDriver. So please take it into cosideration when you (the project members)
set a low priority for issues connected with RC.

Here is the PHPUnit_Selenium issue: https://github.com/sebastianbergmann/phpunit-selenium/issues/85

Thanks!  I did the same thing last night, and except for a few characters different
my patch looks exactly like yours.  I quickly tested it against RC, and will try Grid
today.

Thanks for the patch to Driver.php, psemuu! 

In terms of the aforementioned low priority for RC issues and the suggestion to migrate
to WebDriver, I have yet to find any definitive documentation on how to migrate tests
and/or which WebDriver wrapper to use. 

Can anyone point me in the right direction for documentation on how to migrate to WebDriver
while still using phpunit? The only wrapper I've seen that has some promise is https://github.com/chibimagic/WebDriver-PHP,
but there are certain things that aren't ideal such as the apparent lack of support
for specifying browser via XML configuration file.

Any tips?

Regarding migration path to WebDriver, there are 2 options for any supported language:

* do extreme migration of porting RC test code to WebDriver (different APIs)
* migrate using WebDriverBackedSelenium

WebDriverBackedSelenium is currently only available to Java (and I think .NET, though
not widely or well documented). There is plan to support it as well within JSONWireProtocol
as part of server feature, but no ETA, search for the issue on the Selenium project.
But in the meantime, one could also write their own wrapper that maps RC API to WebDriver,
it's not that hard, just time consuming.

As for migration for PHP/PHPUnit, I see 2 current options:

* use PHPUnit's Selenium2TestCase support (try while in development, or wait 'til it's
complete)
* use one of the other PHP WebDriver implementations

I think going forward, it would be nice to abstract out the PHPUnit extension class
for Selenium2/WebDriver. It would provide an interface for one to integrate/implement
with PHP WebDriver implementation of choice. Because PHPUnit-Selenium extension is
the defacto standard for Selenium testing with PHP compared to chibimagic's WebDriver
PHPUnit interface, etc. I'm sure some folks might prefer a particular PHP WebDriver
implementation but want to use the full featured PHPUnit capability as well. Maybe
do this by forking phpunit-selenium project.

My suggestion for migrating PHPUnit code is to find/build a WebDriverBackedSelenium(TestCase)
and thus can use existing code as is and migrate over to WebDriver API more gradually.
I created issue for that for PHPUnit extension:

https://github.com/sebastianbergmann/phpunit-selenium/issues/86

The issue over at PHPUnit_Selenium has been fixed & closed: https://github.com/sebastianbergmann/phpunit-selenium/issues/85

Can anybody confirm that the current PHPUnit can work with Grid2, after https://github.com/sebastianbergmann/phpunit-selenium/issues/85
was fixed?

Closing as it appeared to be an issue in PHPUnit and that issue is now closed