Couchbase

[PCBC-139] PHP client needs fastlz compression to be compatible with legacy memcached clients

Dependency — Tue, 11 Dec 2012 19:55:30 -0600

The existing memcached clients for PHP (pecl-memcached and memcache) both default to using fastlz compression for values >= 2000 bytes.

The couchbase client library doesn't have fastlz compression included. It doesn't provide a useful error message when it runs into this problem. Instead, it misinterprets the stored value, tries to allocate a huge block of memory and fails. The error it gives (look, e.g., in the PHP-FPM web-access.log) is:

[01-Nov-2012 20:45:49 UTC] PHP Fatal error: Allowed memory size of 134217728 bytes exhausted (tried to allocate 8460104323416721361 bytes) in /usr/share/nginx/html/cb.php on line 32

To reproduce, just set a value >= 2000 bytes with pecl-memcached client, and try to read it with Couchbase client. Here's a sample:

<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<title>Couchbase Server PHP Test</title>
</head>
<body>

<p>
<?php

$host = "10.4.2.15";
$user = "Administrator";
$password = "password";
$bucketName = "default";
$persistConnection = false;
$cb = new Couchbase($host . ":8091",
        $user, $password, $bucketName, $persistConnection);

$mc = new Memcached('Pool-1');
if (count($mc->getServerList()) == 0) {
    // No servers, this pool needs to be configured
    $mc->addServer($host, 11211);
    //$mc->addServer("other host", 11211);
}

// Change this to 1999 and it will work
$valLength = 2000;
date_default_timezone_set('UTC');
$val = date(DateTime::RFC2822) . ' ';
$val .= str_repeat('X', $valLength - strlen($val));

$mc->set("a", $val);
echo("The value for 'a' is: ");
var_dump($cb->get("a"));

?>
</p>
</body>
</html>

The correct behavior is for the Couchbase client to ship, out of the box, compatible with existing Memcached client implementations. It should include fastlz support directly, or else clearly document how to add such support without requiring any unusual compilations or other hoops. The fastlz code is very small and portable, and hasn't changed since 2007, so it should present a minimal maintenance issue.

In addition, the PHP client documentation should clearly indicate what changes are needed in order for the client to be compatible with the other Couchbase clients (Java, .NET, etc.). Again, without requiring recompilation or extra hoops to jump through.

[PCBC-134] Provide more information on connection failure

Duplicate — Mon, 11 Feb 2013 14:09:04 -0600

When you connect via the new Couchbase(...) is not a direct way to check for whether the connection failed.

It is not possible to check getResultCode()/getResultMessage() since this provides a unintilized couchbase

[PCBC-117] Negative expiry values should throw an error

Wed, 26 Sep 2012 18:13:18 -0500

Negative expiry values should throw an error. It is impossible to determine if the user means to disable expiry, or has accidentally make the expiry value too low (which in C-land will cause a wrap-around).

Currently it just wraps around to a time_t (or doesn't throw an error at least).

[PCBC-115] getStats() returns malformed array

Tue, 18 Sep 2012 11:01:02 -0500

The associative array returned by getStats (OO version) / couchbase_get_stats ("ordinary" function version) has malformed array keys similar to the issue observed in ~~PCBC-66~~. For example, using array_key_exists on "ep_version" returns false, and attempting to reference that array key directly raises and error, even though the key name is visible in the output of a var_dump or similar.

[PCBC-114] getStats() documentation states incorrect parameter options

Thu, 4 Oct 2012 04:37:51 -0500

The documentation for the getStats() function states that parameters may be passed to the function to filter the list of returned statistics (e.g. by statistic name). The functions as implemented however ignore any such parameters, and this is in line with the behavior of the pecl-memcached extension as desired (c.f. http://www.php.net/manual/en/memcached.getstats.php). In other words, the parameters information given in the docs should be removed as the implied filtering capabilities do not exist.

[PCBC-113] getMulti() returns "broken" array

Mon, 10 Sep 2012 18:16:52 -0500

The array's returned by getMulti() are not correctly keyed or are in some other way broken. When one tries to reference the key via it's name, an "undefined index" error is returned. array_key_exists() also returns false.

This paste has some sample code:
https://friendpaste.com/MJS2O8G5b7kRqslp2pMBo

Both the top and bottom cases should end in bool(true).

avsej mentioned in IRC that these two portions of code should be compared:
https://github.com/couchbase/php-ext-couchbase/blob/master/couchbase.c#L940
https://github.com/php-memcached-dev/php-memcached/blob/master/php_memcached.c#L828

Thanks for digging into this issue.

[PCBC-108] version() returns the library version, not server version

Tue, 18 Sep 2012 10:52:13 -0500

Documentation wrongly states that version returns the version of the server; this in fact returns the version of the library

[PCBC-97] tests 006.phpt and 007.phpt fail on multiple nodes owing to how they compare multi results

Wed, 5 Sep 2012 11:05:13 -0500

Currently, in tests 006 and 007, we rely on serialize of multiget responses to make a comparison. Since the results can be in a different order, this will frequently make tests fail.

For example, here's some dump output of v and keys in 006:

v is
string(414) "a:6:{s:23:"couchbase_5025a95b7b7dd";s:29:"couchbase_value_5025a95b7b7e5";s:23:"couchbase_5025a95b7b7bc";s:29:"couchbase_value_5025a95b7b7c8";s:23:"couchbase_5025a95b7b7f2";s:29:"couchbase_value_5025a95b7b7f9";s:23:"couchbase_5025a95b7b801";s:29:"couchbase_value_5025a95b7b807";s:23:"couchbase_5025a95b7b81c";s:29:"couchbase_value_5025a95b7b823";s:23:"couchbase_5025a95b7b80f";s:29:"couchbase_value_5025a95b7b815";}"
keys is
string(414) "a:6:{s:23:"couchbase_5025a95b7b7bc";s:29:"couchbase_value_5025a95b7b7c8";s:23:"couchbase_5025a95b7b7dd";s:29:"couchbase_value_5025a95b7b7e5";s:23:"couchbase_5025a95b7b7f2";s:29:"couchbase_value_5025a95b7b7f9";s:23:"couchbase_5025a95b7b801";s:29:"couchbase_value_5025a95b7b807";s:23:"couchbase_5025a95b7b80f";s:29:"couchbase_value_5025a95b7b815";s:23:"couchbase_5025a95b7b81c";s:29:"couchbase_value_5025a95b7b823";}"

Note that the same stuff is there, it's just not in the same order. In a single node cluster, it comes back in the same order but the results race for position in a multi-node cluster.

[PCBC-75] segfault when using persistent connections

Duplicate — Tue, 18 Sep 2012 10:59:27 -0500

When configuring the PHP client for persistent connections, a segmentation fault is seen upon the second request for a connection within an Apache MPM environment.

The test script is as follows:

<?php

$iterations = 1;
$time_start = microtime(true);
//$cb = new Couchbase("192.168.1.200:8091"); // uses the default bucket
$cb = new Couchbase("192.168.1.200:8091", "newcachebucket", "password", "newcachebucket", true); // uses the default bucket

for ($i = 1; $i <=$iterations; $i++) {
  $cb->set("a" . $i, 1);
  //var_dump($cb->get("a" . $i));
  $cb->get("a" . $i);
}

$time_end = microtime(true);
$time = $time_end - $time_start;

echo "Did $iterations iterations in $time seconds\n";

If making a request of this script with curl, I see the following:
[root@centosb httpd]# for i in {1..16}; do curl http://localhost/;done
Did 1 iterations in 0.009458065032959 seconds

Did 1 iterations in 0.0088930130004883 seconds

Did 1 iterations in 0.0089321136474609 seconds

Did 1 iterations in 0.0089690685272217 seconds

Did 1 iterations in 0.0090639591217041 seconds

Did 1 iterations in 0.0095789432525635 seconds

Did 1 iterations in 0.0090429782867432 seconds

Did 1 iterations in 0.0094709396362305 seconds

curl: (52) Empty reply from server
curl: (52) Empty reply from server
curl: (52) Empty reply from server
curl: (52) Empty reply from server
curl: (52) Empty reply from server
curl: (52) Empty reply from server
curl: (52) Empty reply from server
curl: (52) Empty reply from server

The error_log shows lots of:
[Sun Jun 17 19:22:32 2012] [notice] child pid 2031 exit signal Segmentation fault (11)

[PCBC-74] Constructor segfaults when http:// is used in URL

Mon, 24 Sep 2012 19:06:22 -0500

This works:

$cb = new Couchbase("184.72.2.230:8091"); // uses the default bucket

This segfaults:

$cb = new Couchbase("http://184.72.2.230:8091"); // uses the default bucket

At least some better message to the user would be nice.

Can we also check this for all of our other libraries? The standard way of doing things used to be to put the full URL on there...and still is for .NET at least.

[PCBC-27] getting started screencast is available

Wed, 7 Nov 2012 15:31:07 -0600

getting started screencast is available