We have a request offering to sponsor work on callbacks for cases where AMQP_MANDATORY AND AMQP_IMMEDIATE are not satisfiable by the broker.

Also there is this: https://bugs.php.net/bug.php?id=62786

Any volunteers wanting to get in contact with the sponsor mail @pdezwart, @lstrojny or me.

According to AMQP Specification, section 4.8 - "Error Handling" on some kind or exception as they declared in section 1.2. - "AMQP­ defined Constants" we have to close channel or even connection (with all of it channels) and mark those object as disconnected to prevent their further usage in case of try-catch application survive.

In case of re-declaring already existent exchange with different params we doesn't properly close channel, which may lead in unexpected behavior when same channel being used.

See test case for this issue.

For example, when we utilize such channel for message publishing we'll receive successful result for AMQPExchange::publish, but no message is published in fact.

P.S.: there are may be some other places in sources where we do not handle properly AMQP exceptions.

I was searching for memory leaks in the evening (bug #63350) and found that we use 255 chars for string fields including terminator character '\0'. In fact broker use pure 255 characters for identifier. I didn't check that on envelope fields, but i have tested queue and exchange name and implement consumer tag with this fix (see my pull request #40).

This is a bit trivial bug and I will write a fix for it tomorrow evening, but maybe there is some hidden reason why it is so (255 and not 256)?

According to AMQP 0-9-1 spec (section 2.3.2 Data Types, and 4.2.5.3 Strings) short strings is 255 byts of UTF-8(?) data without binary zero, and we in fact wipe the last one with terminator when we receive string that is 255 bytes. It's an edge case, but I vote up to support specs.

See http://pastebin.com/XPkZatck for queues and http://pastebin.com/179GGfpQ for exchanges naming issue demonstration.

I tried to connect to RabbitMQ with wrong login/password.
I catched the exception "Library error: Die Ressource ist zur Zeit nicht verf\xc3\xbcgbar - Potential login failure.".
But on return out of my funktion there is a "child pid 17084 exit signal Segmentation fault (11)".

PHP 5.3.15

amqp
Version => 1.2.0
Revision => $Revision: 327551 $
Compiled => Aug 28 2013 @ 11:39:32
AMQP protocol version => 0-9-1
librabbitmq version => 0.3.0

1.0.9

I confirm on Mac OS X 10.8.2. php version PHP 5.4.10 (tarball from 20 Dec 2012 with md5 cb716b657a30570b9b468b9e7bc551a1) from official php.net site, php-amqp from v1.0.9 branch from pdezwart repo. Strange that Instruments doesn't show any leeks.

I slightly modify initial script to reproduce the bug, just declare queue, because without any binded to exchange queue broker will drop message and we also cannot get message from nonexistent queue.

• bug_63350_demo.php
• /Users/pinepain/bin/php bug_63350_demo.php
• /Users/pinepain/bin/php -v
• /Users/pinepain/bin/php --ri amqp
• sw_vers
• cc --version

P.S.: looking on that with valgrind doesn’t looks optimistic.
valgrind --tool=memcheck --num-callers=30 --log-file=php.log /Users/pinepain/bin/php bug_63350_demo.php

• terminal output
• cat php.log

Strange, that Instruments (tool from Xcode package) doesn't show any leaks or other problems.

According to specs AMQPQueue::delete is idempotent, so in php-amqp implementation it can trow exception or return some value, which is true for now, but in fact we should return deleted messages count as to AMQP specs.

But should this be done in current release or moved to next major one?

Logically checking whether AMQPQueue::delete return true, especially doing strict check makes no sense, so adding this fix in current release should not break significantly users codebase.

P.S.: here is a gist with implementation

Hi @pdezwart

First of all, I really thank you that you've shared this great extension with community, I'm working in a project in which we are using it inside a Daemon (php client mode).

After been using it for a while we believe that the following improvement would be nice , as it would allow users to use consume method without blocking the execution. I will detail below how could be the flow using consume in non blocking way.

1. Execute AMQPQueue::consume ( null, $flags )
2. If consume method was not previously sent, send it, and store that consume has been enabled ( probably in an property called "consumer" that then could be returned calling a new AMQPQueue::isConsumer method)
3. Call read_message_from_channel to retrieve a message (if any) from queue
4. Continue exactly as AMQPQueue::get method is (return a new AMQPEnvelope if message was retrieved or false otherwise)

So to pass it clean two modifications to extension interface need to be done

1. Change consume method to allow it to be called without callback:
   public void AMQPQueue::consume ( callable $callback = null [, int $flags = AMQP_NOPARAM ] )
2. Add a new method
   public bool AMQPQueue::isConsumer ( )
   to retrieve if consume mechanism is enabled on it

We certainly believe this would be a great improvement to the extension, without involving a lot of development, moreover we can give you a hand with it.

What do you think?

Be well!

When server is down, client can connect to the machine that hosts the server due to its security settings. But, connection hangs for a couple of minutes, and then it throws AMQPConnectionException with "Socket error: could not connect to host." message. Setting timeout value is not working.

Introduced in bb523aa

Calling setHost() repeatedly results in a memory leak.

https://bugs.php.net/bug.php?id=66848

for ($ii = 0; $ii < 200; $ii++) {
    $amqp_connection = new \AMQPConnection();

    if ($ii % 50 == 0) echo $ii.' '.number_format(memory_get_usage()/1024).'KiB'.PHP_EOL;
}
echo $ii.' '.number_format(memory_get_usage()/1024).'KiB'.PHP_EOL.PHP_EOL;

for ($ii = 0; $ii < 200; $ii++) {
    $amqp_connection = new \AMQPConnection();
    $amqp_connection->setHost('127.0.0.1');           // only difference to above loop

    if ($ii % 50 == 0) echo $ii.' '.number_format(memory_get_usage()/1024).'KiB'.PHP_EOL;
}
echo $ii.' '.number_format(memory_get_usage()/1024).'KiB'.PHP_EOL.PHP_EOL;

Typical output showing how memory usage continues to grow in the second loop

0 224KiB
50 224KiB
100 224KiB
150 224KiB
200 224KiB

0 224KiB
50 226KiB
100 227KiB
150 229KiB
200 230KiB

Is 1.3.0 a stable release?

Are you able to shed some light on why we should be receiving the following log entries when using version 1.4.0-1? Attempts to connect fail with {handshake_timeout,frame_header}.

These are intermittent and do not appear to relate to high load on the server which has a load average of 0.0x most of the time. The publish rate to RabbitMQ is less than 0.5/sec.

=INFO REPORT==== 27-Apr-2014::05:05:02 ===
accepting AMQP connection <0.22398.229> (127.0.0.1:36025 -> 127.0.0.1:5672)

=INFO REPORT==== 27-Apr-2014::05:05:02 ===
accepting AMQP connection <0.22401.229> (127.0.0.1:36026 -> 127.0.0.1:5672)

=ERROR REPORT==== 27-Apr-2014::05:05:12 ===
closing AMQP connection <0.22398.229> (127.0.0.1:36025 -> 127.0.0.1:5672):
{handshake_timeout,frame_header}

=ERROR REPORT==== 27-Apr-2014::05:05:12 ===
closing AMQP connection <0.22401.229> (127.0.0.1:36026 -> 127.0.0.1:5672):
{handshake_timeout,frame_header}

# php -v
PHP 5.5.10 (cli) (built: Mar  6 2014 10:44:40)
Copyright (c) 1997-2014 The PHP Group
Zend Engine v2.5.0, Copyright (c) 1998-2014 Zend Technologies

# php --ri amqp

amqp

Version => 1.4.0
Revision => $Revision: 327551 $
Compiled => Apr 23 2014 @ 09:03:55
AMQP protocol version => 0-9-1
librabbitmq version => 0.5.0

Directive => Local Value => Master Value
amqp.host => localhost => localhost
amqp.vhost => / => /
amqp.port => 5672 => 5672
amqp.timeout => no value => no value
amqp.read_timeout => 0 => 0
amqp.write_timeout => 0 => 0
amqp.connect_timeout => 0 => 0
amqp.login => guest => guest
amqp.password => guest => guest
amqp.auto_ack => 0 => 0
amqp.prefetch_count => 3 => 3

# yum list installed | grep rabbit
librabbitmq.x86_64               0.5.0-1.el6.remi          @remi
librabbitmq-devel.x86_64         0.5.0-1.el6.remi          @remi
librabbitmq-tools.x86_64         0.5.0-1.el6.remi          @remi
rabbitmq-server.noarch           3.3.0-1                   installed

# yum list installed | grep php
php55u.x86_64                    5.5.10-1.ius.el6          @rackspace-rhel-x86_64-server-6-ius
php55u-cli.x86_64                5.5.10-1.ius.el6          @rackspace-rhel-x86_64-server-6-ius
php55u-common.x86_64             5.5.10-1.ius.el6          @rackspace-rhel-x86_64-server-6-ius
php55u-devel.x86_64              5.5.10-1.ius.el6          @rackspace-rhel-x86_64-server-6-ius
php55u-gd.x86_64                 5.5.10-1.ius.el6          @rackspace-rhel-x86_64-server-6-ius
php55u-gmp.x86_64                5.5.10-1.ius.el6          @rackspace-rhel-x86_64-server-6-ius
php55u-imap.x86_64               5.5.10-1.ius.el6          @rackspace-rhel-x86_64-server-6-ius
php55u-mbstring.x86_64           5.5.10-1.ius.el6          @rackspace-rhel-x86_64-server-6-ius
php55u-mysqlnd.x86_64            5.5.10-1.ius.el6          @rackspace-rhel-x86_64-server-6-ius
php55u-pdo.x86_64                5.5.10-1.ius.el6          @rackspace-rhel-x86_64-server-6-ius
php55u-pear.noarch               1:1.9.4-20.ius.el6        @rackspace-rhel-x86_64-server-6-ius
php55u-pecl-jsonc.x86_64         1.3.5-1.ius.el6           @rackspace-rhel-x86_64-server-6-ius
php55u-process.x86_64            5.5.10-1.ius.el6          @rackspace-rhel-x86_64-server-6-ius
php55u-xml.x86_64                5.5.10-1.ius.el6          @rackspace-rhel-x86_64-server-6-ius

# cat /etc/redhat-release
Red Hat Enterprise Linux Server release 6.5 (Santiago)

I just did some digging around https://bugs.php.net/bug.php?id=63309 and http://stackoverflow.com/questions/12936726/php-amqp-bug-with-channel-creation .

Our bookkeeping of channels in the connection_resource slots can be out of sync with the server reality when an exception renders a channel unusable and a new channel is created right afterwards (before the channel destructor hit).

I have tried to use AMQPQueue::consume with "return false" on two different queues on the same channel. However messages published to queue2 are received by the consumer of queue1. I don't believe this is expected behavior but I'm not totally sure if I'm using the methods as they are supposed to be used.

I have written a test proving the issue here:
https://github.com/empi89/php-amqp/blob/master/tests/amqpqueue_consume_basic_multiple_queues.phpt

You could find the travis-ci build here:
https://travis-ci.org/empi89/php-amqp/jobs/7545321

Since the documentation is practically non-existent, it would be nice if somebody would implement tutorials from the RabbitMQ site.

Maybe this would be less time consuming then trying to update the whole documentation and it would certainly be of much help to the newcomers.

Repo with tutorials for other languages and implementations is located at rabbitmq/rabbitmq-tutorials

We have an offer to sponsor work on publisher acknowledgements as described here

We'll gladly supply the contact to any volunteers. Just ping @pdezwart, @lstrojny or me for the details.

RabbitMQ allows for header exchanges. Binding a queue to an exchange should be able send additional arguments as a list of matching headers.

It looks like the arguments value that is passed over to the c client sets arguments to null.

I'd expect the following to work in a near future release:

$queue->bind('exchange name', 'routing key', ['x-match' => 'all', 'type' =>'some value']);

Also see https://bugs.php.net/bug.php?id=62099

I'm pretty sure this already worked. But looking at connection->connection_resource->fd from multiple forked processes that created new connections, they all are the same socket.

A test for this probably required exposing the fd through debug_info.

Hey guys, is the heartbeat problem fixed?

👍

In stubs and official docs method AMQPConnection::connect shown as one that may return boolean FALSE, but in fact it will never do so according to php_amqp_connect C function which return positive result (integer 1) only if connection established, otherwise exception thrown.

I propose to make AMQPConnection::connect method void to prevent users to make useless checks, like that one from official docs.

For backward compatibility we can't do that immediately, so for now it would nice do change just docs and stubs to reflect that the FALSE result is impossible.

$exchange = new \AMQPExchange($this->amqp_channel);
// or just don't bother calling setname
$exchange->setName('');
$exchange->setType(AMQP_EX_TYPE_DIRECT);
$exchange->setFlags(AMQP_PASSIVE);

$exchange->publish('msg', 'routingkey');

Results in:

Fatal error: Uncaught exception 'AMQPExchangeException' with message 'Invalid exchange name given, must be between 1 and 255 characters long.'

This exception should not be thrown as a empty exchange name is a valid exchange. Publishing to an empty name exchange results in the broker routing the message to a queue with a name that is equal to the routing key.

Potential sponsor reports memory leaks with pconnect() when used in a loop.
Possibly related: https://bugs.php.net/bug.php?id=63350

Not sure if this is the correct place to report this, but I get a Segmentation fault when installing amqp-1.4.0beta2

[root@localhost ~]# uname -a
Linux localhost.localdomain 2.6.32-358.el6.x86_64 #1 SMP Fri Feb 22 00:31:26 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux

Last few lines of PECL install

See any operating system documentation about shared libraries for
more information, such as the ld(1) and ld.so(8) manual pages.
----------------------------------------------------------------------

Build complete.
Don't forget to run 'make test'.

running: make INSTALL_ROOT="/var/tmp/pear-build-rootye2Jud/install-amqp-1.4.0beta2" install
Installing shared extensions:     /var/tmp/pear-build-rootye2Jud/install-amqp-1.4.0beta2/usr/lib64/php/modules/
running: find "/var/tmp/pear-build-rootye2Jud/install-amqp-1.4.0beta2" | xargs ls -dils
663766   4 drwxr-xr-x. 3 root root   4096 Mar 17 09:49 /var/tmp/pear-build-rootye2Jud/install-amqp-1.4.0beta2
663800   4 drwxr-xr-x. 3 root root   4096 Mar 17 09:49 /var/tmp/pear-build-rootye2Jud/install-amqp-1.4.0beta2/usr
788331   4 drwxr-xr-x. 3 root root   4096 Mar 17 09:49 /var/tmp/pear-build-rootye2Jud/install-amqp-1.4.0beta2/usr/lib64
788332   4 drwxr-xr-x. 3 root root   4096 Mar 17 09:49 /var/tmp/pear-build-rootye2Jud/install-amqp-1.4.0beta2/usr/lib64/php
788333   4 drwxr-xr-x. 2 root root   4096 Mar 17 09:49 /var/tmp/pear-build-rootye2Jud/install-amqp-1.4.0beta2/usr/lib64/php/modules
788334 364 -rwxr-xr-x. 1 root root 371736 Mar 17 09:49 /var/tmp/pear-build-rootye2Jud/install-amqp-1.4.0beta2/usr/lib64/php/modules/amqp.so

Build process completed successfully
Installing '/usr/lib64/php/modules/amqp.so'
install ok: channel://pecl.php.net/amqp-1.4.0beta2
configuration option "php_ini" is not set to php.ini location
You should add "extension=amqp.so" to php.ini
Segmentation fault
[root@localhost ~]#

Hello. I think this is a bug when using consume with AMQP_AUTOACK parameter. If there is a exception in the function that processes the messages, not only one message disappears from queue but multiple messages. I have some cases in which 5 messages disappear, some situations in which more than 5 disappear.

Switching to manual acknowledgements behaves as expected, only the message currently being processed disappears from the queue.

The version i am currently using is: 1.4.0beta2

Hi,

Maybe it's not an issue but is there a way to break a consume when there are no messages in a queue ? I now use pcntl_alarm but this causes Interrupted system call. is there a better way to do that ?

@pdezwart could you activate Travis CI integration?

Can somebody give explanation for TODO file items:

• Fix php_st_error_handling deprecation
• Fix parameter parsing (see PECL email)
• Add non-busy wait consumption

What do you think?

https://github.com/pdezwart/php-amqp/blob/master/amqp_exchange.c#L596

In this line, if the buffer is not ok, amqp_maybe_release_buffers (https://github.com/alanxz/rabbitmq-c/blob/b852f846c46a3bb9451f99844edfbea41f3289f1/librabbitmq/amqp_connection.c#L402) will not release the buffer.

The behaviour is ok if you don't catch the exception.. but once you catch the exception, the php process still stuck due to the buffer...

Should call amqp_release_buffers directly and release the buffer instead of amqp_maybe_release_buffers ?

it's nice if it can support composer . thank you

Version 1.4.0beta2 reports as 1.4.0beta1.

1/ missing files

• amqp_object_store.c
• amqp_object_store.h

2/ bad version

  $ grep Version amqp-1.0.8/amqp.c
      php_info_print_table_header(2, "Version",                 "1.0.6");

3/ bad link from PECL site to sources

http://svn.php.net/repository/pecl/amqp instead of https://github.com/pdezwart/php-amqp

I am getting the following error upon attempting to invoke pecl install amqp:

/tmp/pear/temp/amqp/amqp_queue.c: In function 'parse_amqp_table': /tmp/pear/temp/amqp/amqp_queue.c:188:9: error: 'AMQP_FIELD_KIND_U64' undeclared (first use in this function) /tmp/pear/temp/amqp/amqp_queue.c:188:9: note: each undeclared identifier is reported only once for each function it appears in /tmp/pear/temp/amqp/amqp_queue.c: In function 'zim_amqp_queue_class_nack': /tmp/pear/temp/amqp/amqp_queue.c:1092:2: error: unknown type name 'amqp_basic_nack_t' /tmp/pear/temp/amqp/amqp_queue.c:1112:3: error: request for member 'delivery_tag' in something not a structure or union /tmp/pear/temp/amqp/amqp_queue.c:1113:3: error: request for member 'multiple' in something not a structure or union /tmp/pear/temp/amqp/amqp_queue.c:1114:3: error: request for member 'requeue' in something not a structure or union /tmp/pear/temp/amqp/amqp_queue.c:1119:3: error: 'AMQP_BASIC_NACK_METHOD' undeclared (first use in this function) make: *** [amqp_queue.lo] Error 1 ERROR: make' failed`

There's a ridonkulous amount of open amqp bugs on bugs.php.net.
I replied to most of them, and the majority seems to be either due to outdated documentation, or bugs fixed with verision 1.0.9.

Still there are some that deserve investigating:

• https://bugs.php.net/bug.php?id=60352
• https://bugs.php.net/bug.php?id=62786
• https://bugs.php.net/bug.php?id=63346
• https://bugs.php.net/bug.php?id=63350
• https://bugs.php.net/bug.php?id=63309

Is there a way to change the "bugs" link in the pecl page to point to github issues? I find the php-bugtracker awkward to use, to put it mildly.

with AMQP_HEARTBEAT == 0 in some cases the rabbitmq server doesn't close dead connections, that causes socket descriptors overflow

Hi,

It looks like we have tagged 1.2.0 which includes the rabbitmq-server 3.1 fix, however, I am not seeing this on pecl yet. Is there a different distribution mechanism I should be aware of?

This test:
https://github.com/pdezwart/php-amqp/blob/master/tests/amqpqueue_bind_basic_headers_arguments.phpt

Prints "null" instead of expected "bool(true)". It's means that binding not realy happend

On version 1.3.0 on mac

RabbitMQ version: 3.2.3

Our rabbitmq server was not responding for a while, and we found that the connect_timeout setting in AMQPConnection was not working. Instead, connect() waits for 60 seconds before failing (Socket error: could not connect to host.).

        $conn = new \AMQPConnection(array(
            'host' => $config["host"],
            'port' => 5672,
            'login' => $config["login"],
            'password' => $config["password"],
            'connect_timeout' => 1,
            'read_timeout' => 1,
            'write_timeout' => 1
        ));

A quick way to reproduce this is to set the host to an unroutable IP, like 10.255.255.1.

It would be great if library will support "Consumer Cancel Notification" - http://www.rabbitmq.com/consumer-cancel.html

With the latest fixes and improvements, should we have another release?

I just updated my RabbitMQ Server to Version 3.1.0-1. After that, all (except the first) of my AMQPQueue::get-Calls ran into the "connection closed unexpectedly" AMQPConnectionException. I don't know if its related to php-amqp, the rabbitmq-c library or the server itself. However, I could kind of workaround the problem by setting setPrefetchCount to 0 on the AMQPChannel.

I have shared the serverlog file here: https://gist.github.com/empi89/5508316

It would be really nice if someone who knows the deeper details of the involved software could clarify where (or if) to report the problem upstream. Thanks!

When reading a message from a channel, we currently only have limited support for arrays.
For example, when a message is moved to an error queue, the header structure of AMQPEnvelope contains a key 'x-death', which should contain an array with the reasons why the message ended up there.

With our current implementation there is no way to retrieve this information; header 'x-death' always gives an empty array.

The reason lies in the way we parse the values in amqp_queue.c
https://github.com/pdezwart/php-amqp/blob/master/amqp_queue.c#L354-L370

I don't think there is a limit on the nesting depth, so parsing amqp arrays into zval arrays should probably be refactored into a separate function, so we can apply it recursively.

This is more of a note about the state of things, patches highly welcome.
Anybody want to give it a shot? Instant fame and eternal gratitude guaranteed.

Hi all!.

The qos settings should be setted ONLY when a queue::consume is called. For example, my web app only publish messages, whence qos settings no sense.
For resolve this issue, the channel must have a flag indicating if qos was setted or not and in the consume method of queue, it must be ensured that qos options were setted (in case that not, defaults qos settings must be setted).

https://github.com/pdezwart/php-amqp/blob/master/amqp_channel.c#L195

Regards!.

I did manage to install the library now. However, I am receiving this error:

PHP Warning: PHP Startup: Unable to load dynamic library '/usr/lib/php5/20100525/amqp.so' - /usr/lib/php5/20100525/amqp.so: undefined symbol: amqp_basic_qos in Unknown on line 0

Then I get a fatal error saying that the Class 'AMQPConnection' is not found.

1. When you published a message in an unknown exchange, then a message in a known exchange, the second one is not published (except with a new Channel).
2. If a new exchange is used, a Fatal Error occured when trying to get the message (even with a new Channel).

You can find a full example in this gist.

if connection was closed - isConnected returns true until any api method called

It would be great if library will support "Confirms" - http://www.rabbitmq.com/confirms.html

The documentation on php.net is out of date and misleading at times. How does the process of updating it work? Anything we can do to help?