This abstraction makes it possible to issue a WAIT command using a
client instance connected to a single Redis node or configured with
either a replication or cluster backend:
$client = new Predis\Client($parameters, $options);
$ack = new Predis\Replication\WaitContext($client);
$ack->set('key:1', 'value:1');
$ack->set('key:2', 'value:2');
if ($ack->wait(2, 500)) {
// Writes acknowledged by at least 2 slaves.
} else {
// Writes were not acknowledged by at least 2 slaves in 500ms.
}
When the client is operating in replication mode, WAIT is executed
against the connection currently in use by the underlying backend. On
the other hand when it is operating in cluster mode WAIT is executed
against only one connection as cross-slot operations are not allowed.
This is just a first draft in response to #298 that serves as a base
for further improvements and changes. Tests are still missing.
The following code, despite not being something you would do in real
world scenarios, eventually ended up generating an exception for "Too
many open files".
for ($i = 0; $i < 10000; $i++) {
$client = new Predis\Client();
$client->connect();
}
The reason was that the destructor for the connection was not invoked
by PHP as expected when the client instance went out of scope, so the
underlying stream resource were not being properly released.
Even without an actual "connect()" operation, the memory usage of PHP
kept growing until exhausting the configured value of max memory.
The source of the issue is related to the use of closures as handlers
for the phpiredis reader resource, to be more precise it seems to be
some kind of circular reference memory leak. Apparently PHP does not
like the fact that a closure, automatically bound to "$this" starting
with PHP 5.4, is stored in the reader resource which in turn is kept
referenced by "$this". This ends up the refcount not going down thus
the GC does not collect those connection objects going out of scope.
This is confirmed by the fact that this issue is not triggered when
using PHP 5.3 where the ZE does not automatically bind closures to
"$this", unless you capture "$this" with the "use()" directive (and
the usual "$that = $this" trick).
Using a static assignement instead of simply returning new closures
in "getStatusHandler()" and "getErrorHandler()" is kind of an hack
that seems to be working fine, the added value of this solution is
that we do not have to break the internal API of the three connection
backends based on ext-phpiredis.
This commit fixes#345.
This should not break existing code but allows users to retrieve more
easily the current dispatcher loop instance without resorting to some
tricks (like relying on the "use()" directive with closures).
This new method acts exactly like "getClientFor()" by returning a new
client instance for the specified node unless a callback is passed as
the second argument, in this case the callback is invoked and the new
client instance is passed to it. The value returned by the callback
is used as the return value of the "on()" method.
Promoted the "switchToMaster()" and "switchToSlave()" methods to be
part of the replication connection interface and demoted the method
"switchTo($connection)".
All option classes have been moved in the Predis\Configuration\Option
namespace and some have been optimized to have less impact on client
initialization timings.
Furthermore the accepted values for some options have been changed,
this is the complete list of accepted values:
- _aggregate_: callable returning an aggregate connection.
- cluster: string value ("predis", "redis"), callable returning an
aggregate connection.
- replication: string value ("predis", "sentinel"), callable returning
an aggregate connection.
- commands: command factory, named array mapping command IDs to PHP
classes, callable returning a command factory or a named array.
- connections: connection factory, callable returning a connection
factory, named array mapping connection schemes to PHP classes.
- _prefix_: string value, command processor, callable.
- _exceptions_: boolean value.
Note that the cluster and replication options now return a closure
acting as initializer instead of an aggregate connection.
By default Predis now uses a convention-over-configuration approach
by looking for a command class in the Predis\Command\Redis namespace
if it is not already defined in the commands class map.
This change allow us to decrease the time needed to load Predis on
each request since we removed 99% of the mappings in the commands
class map. Classes defined in the internal class map still take the
precedence over this mechanism, so users can still define their own
command classes to handle each command.
This change reduces some unnecessary complexity in the library, Redis
commands do not change much after all. Developers can still implement
their own commands factory, inject new commands or override existing
ones. The "profile" client options has been renamed to "commands" and
it accepts instances of Predis\Command\FactoryInterface.
The test suite checks at runtime the version of the running instance
of Redis for integration tests to adapt itself automatically.
We check if the string value is different than the casted int value,
if so it means that the integer is beyond PHP_INT_MAX or PHP_INT_MIN
and we simply return the string value. This is also useful on Windows
builds of PHP since the maximum integer size (prior to PHP 7.0) is 32
bits even for 64 bit builds.
Iterating over Predis\Connection\Aggregate\RedisCluster returns all
the connections currently mapped in the slots map instead of just the
ones initialized in the pool.
When the slots map is retrieved from Redis (which by default is done
automatically) this allows to iterate over all of the current master
nodes of the cluster. When the underlying use of "CLUSTER SLOTS" is
disabled the iteration returns only connections with a slots range
associated in their parameters or initialized by `-MOVED` responses
to make the behaviour of the iteration consistent between the two
modes of operation.
When various nodes in the configuration are unreachable while trying
to send a command, we should attempt to contact a reachable node to
fetch an updated slots map up to $retryLimit times or until there are
no more servers in the pool before giving up.
It is possible that the slots map fetched from Redis contains stale
data and points to a dead server, this happens when the nodes still
have to agree that a master server is down before promoting a slave
to the role of master. In this case no further attempts to execute
the command are performed and an exception is thrown.
This still needs some more testing and will delay v1.0.4 a few days
past its scheduled release.
Daniele Alessandri
Pascal Hofmann
orvice