predis/README.md

# Predis #

[![Software license][ico-license]](LICENSE)
[![Latest stable][ico-version-stable]][link-releases]
[![Latest development][ico-version-dev]][link-releases]
[![Monthly installs][ico-downloads-monthly]][link-downloads]
[![Build status][ico-build]][link-actions]
[![Coverage Status][ico-coverage]][link-coverage]

A flexible and feature-complete [Redis](http://redis.io) / [Valkey](https://github.com/valkey-io/valkey) client for PHP 7.2 and newer.

More details about this project can be found on the [frequently asked questions](FAQ.md).


## Main features ##

- Support for Redis from __3.0__ to __8.0__.
- Support for clustering using client-side sharding and pluggable keyspace distributors.
- Support for [redis-cluster](http://redis.io/topics/cluster-tutorial) (Redis >= 3.0).
- Support for master-slave replication setups and [redis-sentinel](http://redis.io/topics/sentinel).
- Transparent key prefixing of keys using a customizable prefix strategy.
- Command pipelining on both single nodes and clusters (client-side sharding only).
- Abstraction for Redis transactions (Redis >= 2.0) and CAS operations (Redis >= 2.2).
- Abstraction for Lua scripting (Redis >= 2.6) and automatic switching between `EVALSHA` or `EVAL`.
- Abstraction for Hinted Hash Templates (`HIMPORT`, Redis >= 8.10) with automatic per-connection fieldset replay.
- Abstraction for `SCAN`, `SSCAN`, `ZSCAN` and `HSCAN` (Redis >= 2.8) based on PHP iterators.
- Connections are established lazily by the client upon the first command and can be persisted.
- Connections can be established via TCP/IP (also TLS/SSL-encrypted) or UNIX domain sockets.
- Support for custom connection classes for providing different network or protocol backends.
- Flexible system for defining custom commands and override the default ones.


## How to _install_ and use Predis ##

This library can be found on [Packagist](http://packagist.org/packages/predis/predis) for an easier
management of projects dependencies using [Composer](http://packagist.org/about-composer).
Compressed archives of each release are [available on GitHub](https://github.com/predis/predis/releases).

```shell
composer require predis/predis
```


### Loading the library ###

Predis relies on the autoloading features of PHP to load its files when needed and complies with the
[PSR-4 standard](https://github.com/php-fig/fig-standards/blob/master/accepted/PSR-4-autoloader.md).
Autoloading is handled automatically when dependencies are managed through Composer, but it is also
possible to leverage its own autoloader in projects or scripts lacking any autoload facility:

```php
// Prepend a base path if Predis is not available in your "include_path".
require 'Predis/Autoloader.php';

Predis\Autoloader::register();
```


### Connecting to Redis ###

When creating a client instance without passing any connection parameter, Predis assumes `127.0.0.1`
and `6379` as default host and port. The default timeout for the `connect()` operation is 5 seconds:

```php
$client = new Predis\Client();
$client->set('foo', 'bar');
$value = $client->get('foo');
```

Connection parameters can be supplied either in the form of URI strings or named arrays. The latter
is the preferred way to supply parameters, but URI strings can be useful when parameters are read
from non-structured or partially-structured sources:

```php
// Parameters passed using a named array:
$client = new Predis\Client([
    'scheme' => 'tcp',
    'host'   => '10.0.0.1',
    'port'   => 6379,
]);

// Same set of parameters, passed using an URI string:
$client = new Predis\Client('tcp://10.0.0.1:6379');
```

Password protected servers can be accessed by adding `password` to the parameters set. When ACLs are
enabled on Redis >= 6.0, both `username` and `password` are required for user authentication.

It is also possible to connect to local instances of Redis using UNIX domain sockets, in this case
the parameters must use the `unix` scheme and specify a path for the socket file:

```php
$client = new Predis\Client(['scheme' => 'unix', 'path' => '/path/to/redis.sock']);
$client = new Predis\Client('unix:/path/to/redis.sock');
```

The client can leverage TLS/SSL encryption to connect to secured remote Redis instances without the
need to configure an SSL proxy like stunnel. This can be useful when connecting to nodes running on
various cloud hosting providers. Encryption can be enabled with using the `tls` scheme and an array
of suitable [options](http://php.net/manual/context.ssl.php) passed via the `ssl` parameter:

```php
// Named array of connection parameters:
$client = new Predis\Client([
  'scheme' => 'tls',
  'ssl'    => ['cafile' => 'private.pem', 'verify_peer' => true],
]);

// Same set of parameters, but using an URI string:
$client = new Predis\Client('tls://127.0.0.1?ssl[cafile]=private.pem&ssl[verify_peer]=1');
```

The connection schemes [`redis`](http://www.iana.org/assignments/uri-schemes/prov/redis) (alias of
`tcp`) and [`rediss`](http://www.iana.org/assignments/uri-schemes/prov/rediss) (alias of `tls`) are
also supported, with the difference that URI strings containing these schemes are parsed following
the rules described on their respective IANA provisional registration documents.

Since Redis 8.6, you can authenticate a client using the Subject CN from its TLS client certificate (mTLS).
When this is enabled on the server, the client is authenticated during the TLS handshake, so you don’t need
to send an AUTH command.

To use this, configure:

- a CA certificate used to verify the server certificate (cafile),
- a client certificate (local_cert) signed by a CA trusted by the Redis server for client authentication,
- the corresponding private key (local_pk).

Make sure:

- the Redis server certificate is signed by a CA trusted by the client, and
- the client certificate is signed by a CA trusted by the Redis server (mTLS).

```php
// Named array of connection parameters:
$client = new Predis\Client([
    'scheme' => 'tls',
    'ssl' => [
        'cafile'      => 'ca.pem',          // CA used to verify the server certificate
        'local_cert'  => 'client.crt',      // client certificate (Subject CN maps to ACL user)
        'local_pk'    => 'client.key',      // client private key
        'verify_peer' => true,
    ],
]);

// ACL user must exist and match the certificate Subject CN (example: CN=CN_NAME).
// Enable the user and grant permissions as needed:
$client->acl->setUser('CN_NAME', 'on', '>clientpass', 'allcommands', 'allkeys')

echo $client->acl->whoami() // CN_NAME
```

The actual list of supported connection parameters can vary depending on each connection backend so
it is recommended to refer to their specific documentation or implementation for details.

Predis can aggregate multiple connections when providing an array of connection parameters and the
appropriate option to instruct the client about how to aggregate them (clustering, replication or a
custom aggregation logic). Named arrays and URI strings can be mixed when providing configurations
for each node:

```php
$client = new Predis\Client([
    'tcp://10.0.0.1?alias=first-node', ['host' => '10.0.0.2', 'alias' => 'second-node'],
], [
    'cluster' => 'predis',
]);
```

See the [aggregate connections](#aggregate-connections) section of this document for more details.

Connections to Redis are lazy meaning that the client connects to a server only if and when needed.
While it is recommended to let the client do its own stuff under the hood, there may be times when
it is still desired to have control of when the connection is opened or closed: this can easily be
achieved by invoking `$client->connect()` and `$client->disconnect()`. Please note that the effect
of these methods on aggregate connections may differ depending on each specific implementation.

#### Persistent connections ####

To increase a performance of your application you may set up a client to use persistent TCP connection, this way
client saves a time on socket creation and connection handshake. By default, connection is created on first-command
execution and will be automatically closed by GC before the process is being killed.
However, if your application is backed by PHP-FPM the processes are idle, and you may set up it to be persistent and
reusable across multiple script execution within the same process.

To enable the persistent connection mode you should provide following configuration:

```php
// Standalone
$client = new Predis\Client(['persistent' => true]);

// Cluster
$client = new Predis\Client(
    ['tcp://host:port', 'tcp://host:port', 'tcp://host:port'],
    ['cluster' => 'redis', 'parameters' => ['persistent' => true]]
);
```

**Important**

If you operate on multiple clients within the same application, and they communicate with the same resource, by default
they will share the same socket (that's the default behaviour of persistent sockets). So in this case you would need
to additionally provide a `conn_uid` identifier for each client, this way each client will create its own socket so
the connection context won't be shared across clients. This socket behaviour explained
[here](https://www.php.net/manual/en/function.stream-socket-client.php#105393)

```php
// Standalone
$client1 = new Predis\Client(['persistent' => true, 'conn_uid' => 'id_1']);
$client2 = new Predis\Client(['persistent' => true, 'conn_uid' => 'id_2']);

// Cluster
$client1 = new Predis\Client(
    ['tcp://host:port', 'tcp://host:port', 'tcp://host:port'],
    ['cluster' => 'redis', 'parameters' => ['persistent' => true, 'conn_uid' => 'id_1']]
);
$client2 = new Predis\Client(
    ['tcp://host:port', 'tcp://host:port', 'tcp://host:port'],
    ['cluster' => 'redis', 'parameters' => ['persistent' => true, 'conn_uid' => 'id_2']]
);
```

### Client configuration ###

Many aspects and behaviors of the client can be configured by passing specific client options to the
second argument of `Predis\Client::__construct()`:

```php
$client = new Predis\Client($parameters, ['prefix' => 'sample:']);
```

Options are managed using a mini DI-alike container and their values can be lazily initialized only
when needed. The client options supported by default in Predis are:

  - `prefix`: prefix string applied to every key found in commands.
  - `exceptions`: whether the client should throw or return responses upon Redis errors.
  - `connections`: list of connection backends or a connection factory instance.
  - `cluster`: specifies a cluster backend (`predis`, `redis` or callable).
  - `replication`: specifies a replication backend (`predis`, `sentinel` or callable).
  - `aggregate`: configures the client with a custom aggregate connection (callable).
  - `parameters`: list of default connection parameters for aggregate connections.
  - `commands`: specifies a command factory instance to use through the library.
  - `readTimeout`: (cluster only) Timeout between read operations while loop over connections.

Users can also provide custom options with values or callable objects (for lazy initialization) that
are stored in the options container for later use through the library.


### Aggregate connections ###

Aggregate connections are the foundation upon which Predis implements clustering and replication and
they are used to group multiple connections to single Redis nodes and hide the specific logic needed
to handle them properly depending on the context. Aggregate connections usually require an array of
connection parameters along with the appropriate client option when creating a new client instance.

#### Cluster ####

Predis can be configured to work in clustering mode with a traditional client-side sharding approach
to create a cluster of independent nodes and distribute the keyspace among them. This approach needs
some sort of external health monitoring of nodes and requires the keyspace to be rebalanced manually
when nodes are added or removed:

```php
$parameters = ['tcp://10.0.0.1', 'tcp://10.0.0.2', 'tcp://10.0.0.3'];
$options    = ['cluster' => 'predis'];

$client = new Predis\Client($parameters);
```

Along with Redis 3.0, a new supervised and coordinated type of clustering was introduced in the form
of [redis-cluster](http://redis.io/topics/cluster-tutorial). This kind of approach uses a different
algorithm to distribute the keyspaces, with Redis nodes coordinating themselves by communicating via
a gossip protocol to handle health status, rebalancing, nodes discovery and request redirection. In
order to connect to a cluster managed by redis-cluster, the client requires a list of its nodes (not
necessarily complete since it will automatically discover new nodes if necessary) and the `cluster`
client options set to `redis`:

```php
$parameters = ['tcp://10.0.0.1', 'tcp://10.0.0.2', 'tcp://10.0.0.3'];
$options    = ['cluster' => 'redis'];

$client = new Predis\Client($parameters, $options);
```

#### Redis Gears with cluster ####

Since Redis v7.2, Redis Gears module is a part of Redis Stack bundle. Client supports a variety of
Redis Gears commands that can be used with OSS cluster API. Currently, before using any Redis
Gears commands against OSS cluster Redis server needs to be aware of cluster topology.

`REDISGEARS_2.REFRESHCLUSTER` command should be called against **each master node** (read replicas
should be ignored) **on cluster creation and each time cluster topology changes**.

In most cases this actions should be performed from the CLI interface by the administrator, DevOPS
or even Kubernetes, depends on your infrastructure managing process. However, client provides an API
to do this programmatically.

```php
/** @var \Predis\Connection\Cluster\ClusterInterface $connection */
$connection->executeCommandOnEachNode(
    new \Predis\Command\RawCommand('REDISGEARS_2.REFRESHCLUSTER')
);
```

#### Replication ####

The client can be configured to operate in a single master / multiple slaves setup to provide better
service availability. When using replication, Predis recognizes read-only commands and sends them to
a random slave in order to provide some sort of load-balancing and switches to the master as soon as
it detects a command that performs any kind of operation that would end up modifying the keyspace or
the value of a key. Instead of raising a connection error when a slave fails, the client attempts to
fall back to a different slave among the ones provided in the configuration.

The basic configuration needed to use the client in replication mode requires one Redis server to be
identified as the master (this can be done via connection parameters by setting the `role` parameter
to `master`) and one or more slaves (in this case setting `role` to `slave` for slaves is optional):

```php
$parameters = ['tcp://10.0.0.1?role=master', 'tcp://10.0.0.2', 'tcp://10.0.0.3'];
$options    = ['replication' => 'predis'];

$client = new Predis\Client($parameters, $options);
```

The above configuration has a static list of servers and relies entirely on the client's logic, but
it is possible to rely on [`redis-sentinel`](http://redis.io/topics/sentinel) for a more robust HA
environment with sentinel servers acting as a source of authority for clients for service discovery.
The minimum configuration required by the client to work with redis-sentinel is a list of connection
parameters pointing to a bunch of sentinel instances, the `replication` option set to `sentinel` and
the `service` option set to the name of the service:

```php
$sentinels = ['tcp://10.0.0.1', 'tcp://10.0.0.2', 'tcp://10.0.0.3'];
$options   = ['replication' => 'sentinel', 'service' => 'mymaster'];

$client = new Predis\Client($sentinels, $options);
```

If the master and slave nodes are configured to require an authentication from clients, a password
must be provided via the global `parameters` client option. This option can also be used to specify
a different database index. The client options array would then look like this:

```php
$options = [
    'replication' => 'sentinel',
    'service' => 'mymaster',
    'parameters' => [
        'password' => $secretpassword,
        'database' => 10,
    ],
];
```

While Predis is able to distinguish commands performing write and read-only operations, `EVAL` and
`EVALSHA` represent a corner case in which the client switches to the master node because it cannot
tell when a Lua script is safe to be executed on slaves. While this is indeed the default behavior,
when certain Lua scripts do not perform write operations it is possible to provide an hint to tell
the client to stick with slaves for their execution:

```php
$parameters = ['tcp://10.0.0.1?role=master', 'tcp://10.0.0.2', 'tcp://10.0.0.3'];
$options    = ['replication' => function () {
    // Set scripts that won't trigger a switch from a slave to the master node.
    $strategy = new Predis\Replication\ReplicationStrategy();
    $strategy->setScriptReadOnly($LUA_SCRIPT);

    return new Predis\Connection\Replication\MasterSlaveReplication($strategy);
}];

$client = new Predis\Client($parameters, $options);
$client->eval($LUA_SCRIPT, 0);             // Sticks to slave using `eval`...
$client->evalsha(sha1($LUA_SCRIPT), 0);    // ... and `evalsha`, too.
```

The [`examples`](examples/) directory contains a few scripts that demonstrate how the client can be
configured and used to leverage replication in both basic and complex scenarios.


### Command pipelines ###

Pipelining can help with performances when many commands need to be sent to a server by reducing the
latency introduced by network round-trip timings. Pipelining also works with aggregate connections.
The client can execute the pipeline inside a callable block or return a pipeline instance with the
ability to chain commands thanks to its fluent interface:

```php
// Executes a pipeline inside the given callable block:
$responses = $client->pipeline(function ($pipe) {
    for ($i = 0; $i < 1000; $i++) {
        $pipe->set("key:$i", str_pad($i, 4, '0', 0));
        $pipe->get("key:$i");
    }
});

// Returns a pipeline that can be chained thanks to its fluent interface:
$responses = $client->pipeline()->set('foo', 'bar')->get('foo')->execute();
```


### Transactions ###

The client provides an abstraction for Redis transactions based on `MULTI` and `EXEC` with a similar
interface to command pipelines:

```php
// Executes a transaction inside the given callable block:
$responses = $client->transaction(function ($tx) {
    $tx->set('foo', 'bar');
    $tx->get('foo');
});

// Returns a transaction that can be chained thanks to its fluent interface:
$responses = $client->transaction()->set('foo', 'bar')->get('foo')->execute();
```

This abstraction can perform check-and-set operations thanks to `WATCH` and `UNWATCH` and provides
automatic retries of transactions aborted by Redis when `WATCH`ed keys are touched. For an example
of a transaction using CAS you can see [the following example](examples/transaction_using_cas.php).

#### Support for clustered connections ####

Since Predis v3.0 transactions could be used with clustered connections. However, it has some limitations due to the
fact that Redis doesn't support distributed transactions. All keys in the transaction context should operate on the same
hash slot, due to this limitation it's recommended to use `{}` syntax to make sure that all keys will be mapped to the same hash
slot. Apart from it no additional configuration needed on a client side.

```php
$redis = $this->getClient();

$response = $redis->transaction(function (MultiExec $tx) {
    $tx->set('{foo}foo', 'value');
    $tx->set('{foo}bar', 'value');
    $tx->set('{foo}baz', 'value');
});

// ['OK', 'OK', 'OK']
```


### Hinted Hash Templates (HIMPORT) ###

> **Experimental:** this feature is experimental and its API (the `himport` container and option) may
> change in a future release.

`HIMPORT` (Redis >= 8.10) speeds up loading many hashes that share the same field names. The field
names are sent once with `HIMPORT PREPARE`, registering them under a fieldset name, and hashes are
then created with `HIMPORT SET` by sending only the values. The server stores such hashes in a
memory-efficient encoding where the field names are kept only once; the resulting keys are ordinary
hashes that work with every regular hash command.

Predis exposes the command family through the `himport` container:

```php
$client->himport->prepare('users', ['name', 'email', 'age']);

$client->himport->set('user:1', 'users', ['alice', 'alice@example.com', '25']);
$client->himport->set('user:2', 'users', ['bob', 'bob@example.com', '30']);

$client->himport->discard('users');   // 1
$client->himport->discardAll();        // number of fieldsets removed
```

Values are paired by position with the fields supplied to `prepare()`, in the caller's order — Predis
never reorders them. Hash enumeration order (e.g. `HGETALL`) is not guaranteed to match the `PREPARE`
order, only the value-to-field pairing is.

A fieldset is **server-side session state that lives on a single physical connection** and is lost when
that connection is dropped (reconnect, `RESET`, cluster failover). To keep this transparent, Predis
tracks the fieldsets prepared through the container and:

- replays each `PREPARE` automatically when a connection is re-established (at most once per physical
  connection); this always happens and does not depend on retries;
- if a `HIMPORT SET` still reports `no such fieldset` (for example on a connection created after a
  cluster redirection), re-prepares the fieldset on the executing connection and retries the write.

The re-prepare-and-retry step **reuses the client's configured retry policy** — it is not a separate
mechanism. It therefore only happens when retries are enabled (via the `retry` connection parameter),
and never more than the configured number of attempts; with retries disabled the `no such fieldset`
error propagates unchanged. It can additionally be turned off, even when retries are enabled, with the
`himport` option:

```php
$client = new Predis\Client(
    $parameters + ['retry' => new Predis\Retry\Retry(new Predis\Retry\Strategy\ExponentialBackoff(), 3)],
    ['himport' => ['auto_prepare' => false]] // opt out of HIMPORT re-prepare specifically
);
```

Fieldsets can also be declared up front through the `himport` option. Fieldsets declared this way are
prepared on demand the first time a `HIMPORT SET` references them on a connection, so the application
never has to call `prepare()` for them (this uses the re-prepare-and-retry path above, so it requires
retries to be enabled):

```php
$client = new Predis\Client($parameters + ['retry' => new Predis\Retry\Retry(new Predis\Retry\Strategy\ExponentialBackoff(), 3)], [
    'himport' => [
        'fieldsets' => [
            'users' => ['name', 'email', 'age'],
        ],
    ],
]);

// No prepare() call needed — "users" is known from configuration:
$client->himport->set('user:1', 'users', ['alice', 'alice@example.com', '25']);
```

On a cluster, `prepare()`, `discard()` and `discardAll()` fan out to every master shard, while `set()`
is routed by the hash slot of its key like any other write. This ensures a `HIMPORT SET` succeeds on
whichever shard owns its key.

The raw command form (`$client->himport('PREPARE', 'users', 'name', 'email')`) is also available and is
the one to use inside pipelines and transactions; it performs no client-side tracking or recovery, so
`PREPARE` and the dependent `SET` commands must run on the same connection (which pipelines and
transactions guarantee).


### Adding new commands ###

While we try to update Predis to stay up to date with all the commands available in Redis, you might
prefer to stick with an old version of the library or provide a different way to filter arguments or
parse responses for specific commands. To achieve that, Predis provides the ability to implement new
command classes to define or override commands in the default command factory used by the client:

```php
// Define a new command by extending Predis\Command\Command:
class BrandNewRedisCommand extends Predis\Command\Command
{
    public function getId()
    {
        return 'NEWCMD';
    }
}

// Inject your command in the current command factory:
$client = new Predis\Client($parameters, [
    'commands' => [
        'newcmd' => 'BrandNewRedisCommand',
    ],
]);

$response = $client->newcmd();
```

There is also a method to send raw commands without filtering their arguments or parsing responses.
Users must provide the list of arguments for the command as an array, following the signatures as
defined by the [Redis documentation for commands](http://redis.io/commands):

```php
$response = $client->executeRaw(['SET', 'foo', 'bar']);
```


### Script commands ###

While it is possible to leverage [Lua scripting](http://redis.io/commands/eval) on Redis 2.6+ using
directly [`EVAL`](http://redis.io/commands/eval) and [`EVALSHA`](http://redis.io/commands/evalsha),
Predis offers script commands as an higher level abstraction built upon them to make things simple.
Script commands can be registered in the command factory used by the client and are accessible as if
they were plain Redis commands, but they define Lua scripts that get transmitted to the server for
remote execution. Internally they use [`EVALSHA`](http://redis.io/commands/evalsha) by default and
identify a script by its SHA1 hash to save bandwidth, but [`EVAL`](http://redis.io/commands/eval)
is used as a fall back when needed:

```php
// Define a new script command by extending Predis\Command\ScriptCommand:
class ListPushRandomValue extends Predis\Command\ScriptCommand
{
    public function getKeysCount()
    {
        return 1;
    }

    public function getScript()
    {
        return <<<LUA
math.randomseed(ARGV[1])
local rnd = tostring(math.random())
redis.call('lpush', KEYS[1], rnd)
return rnd
LUA;
    }
}

// Inject the script command in the current command factory:
$client = new Predis\Client($parameters, [
    'commands' => [
        'lpushrand' => 'ListPushRandomValue',
    ],
]);

$response = $client->lpushrand('random_values', $seed = mt_rand());
```


### Customizable connection backends ###

Predis can use different connection backends to connect to Redis. The builtin Relay integration
leverages the [Relay](https://github.com/cachewerk/relay) extension for PHP for major performance
gains, by caching a partial replica of the Redis dataset in PHP shared runtime memory.

```php
$client = new Predis\Client('tcp://127.0.0.1', [
    'connections' => 'relay',
]);
```

Developers can create their own connection classes to support whole new network backends, extend
existing classes or provide completely different implementations. Connection classes must implement
`Predis\Connection\NodeConnectionInterface` or extend `Predis\Connection\AbstractConnection`:

```php
class MyConnectionClass implements Predis\Connection\NodeConnectionInterface
{
    // Implementation goes here...
}

// Use MyConnectionClass to handle connections for the `tcp` scheme:
$client = new Predis\Client('tcp://127.0.0.1', [
    'connections' => ['tcp' => 'MyConnectionClass'],
]);
```

For a more in-depth insight on how to create new connection backends you can refer to the actual
implementation of the standard connection classes available in the `Predis\Connection` namespace.

### Retry exceptions

You can enable automatic retry that is disabled by default, to be able to reduce the amount of
false-positives in case of network issues. By default, we're retrying on any connection,
timeout or socket initialization exception, but you can update the list of retry
exceptions. For now `EqualBackoff` and `ExponentialBackoff` strategies are available,
but you may provide your custom one. Retry may be configured with any type of communication
(standalone node, cluster, pipeline, transaction, replication). Here's an example of
configuration:

```php
// Standalone client
$client = new Predis\Client([
    'retry' => new \Predis\Retry\Retry(
        new \Predis\Retry\Strategy\ExponentialBackoff(1000, 10000), // Base and cap configuration in microseconds
        3                                                           // Number of retries
    ),
]);

// Cluster configuration
$options = [
    'parameters' => [
        'retry' => new \Predis\Retry\Retry(new \Predis\Retry\Strategy\ExponentialBackoff(1000, 10000), 3),
    ],
];

$client = new Predis\Client(['tcp://host:port', 'tcp://host:port', 'tcp://host:port'], $options);

$retry = new \Predis\Retry\Retry(
    new \Predis\Retry\Strategy\ExponentialBackoff(1000, 10000),
    3
);

// Update a list of exceptions to catch
$retry->updateCatchableExceptions([Exception::class]);
```

## RESP3 ##

### Connection ###
To establish the connection using the [RESP3](https://github.com/redis/redis-specifications/blob/master/protocol/RESP3.md) protocol, you need to set parameter `protocol => 3`. The default protocol is RESP2.

You can pass parameter as configuration option in array or as a query parameter in `redis_url`

```php
  // Configuration option
  $client = new \Predis\Client(['protocol' => 3]);

  // Redis URL
  $client = new \Predis\Client('redis://localhost:6379?protocol=3');

  // ["proto" => "3"]
  $client->executeRaw(['HELLO']);
```

### Command responses ###
RESP3 protocol introduce a variety of new [response types](https://github.com/redis/redis-specifications/blob/master/protocol/RESP3.md#resp3-types),
so on the client-side we have more explicit understanding on data types we retrieve from server. Here's some examples to show the difference
between RESP2 and RESP3 responses.

#### Float responses ####
``` php
// RESP2 connection
$client = new \Predis\Client();

$client->geoadd('my_geo', 11.111, 22.222, 'member1');

// [[0 => string(20) "11.11099988222122192", 1 => string(20) "22.22200052541037252"]]
// RESP2 returns float values as simple strings.
var_dump($client->geopos('my_geo', ['member1']));

// RESP3 connection
$client = new \Predis\Client(['protocol' => 3]);

// [[0 => float(11.110999882221222), 1 => float(22.222000525410373)]]
// RESP3 introduces new double type, that corresponds to PHP float.
var_dump($client->geopos('my_geo', ['member1']));
```

#### Aggregate types ####
In RESP3 new aggregate type [Map](https://github.com/redis/redis-specifications/blob/master/protocol/RESP3.md#map-type)
was introduced, that represents the sequence of field-value pairs. So it simplifies parsing, since we don't need to specify
parsing strategy per command (RESP2) and instead relies on the type defined by protocol (RESP3).

In most cases RESP2 responses shouldn't differ from RESP3, since we added additional parsing for those
command that return field-value pairs. However, since RESP2 requires additional parsing, it could be that some commands
had lack of it and return unhandled responses. In this case there would be difference like this:

```php
$client = new \Predis\Client();

// RESP2: ['field', 'value]
$client->commandThatReturnsFieldValuePair('key');

$client = new \Predis\Client(['protocol' => 3]);

// RESP3: ['field' => 'value]
$client->commandThatReturnsFieldValuePair('key');
```

Feel free to open PR or GitHub issue if you face those protocol mismatching.

### Push notifications ###
RESP3 introduce a concept of [push connection](https://github.com/redis/redis-specifications/blob/master/protocol/RESP3.md#push-type),
is the one where server could send asynchronous data to client which was not explicitly requested. Predis 3.0 provides
an API to establish this kind of connection as separate blocking process (worker) and invoke callbacks depends on push
notification message type.

#### Consumer ####
First of all, you need to set up a consumer connection and provide an optional callback that will be executed before
event loop will be started. It allows you to subscribe on channels, enable keys invalidations tracking or enable monitor
connection, any Redis command to let server know that you want to receive push notification within this connection.

```php
// Make sure that RESP3 protocol enabled and read_write_timeout set 0,
// so connection won't be killed by timeout.
$client = new Predis\Client(['read_write_timeout' => 0, 'protocol' => 3]);

// Create push notifications consumer.
// Provides callback where current consumer subscribes to few channels before
// enter the loop.
$push = $client->push(function (ClientInterface $client) {
    $response = $client->subscribe('channel', 'control');
    $status = ($response[2] === 1) ? 'OK' : 'FAILED';
    echo "Channel subscription status: {$status}\n";
});
```

#### Dispatcher loop ####
Dispatcher object allows you to attach a callback to given push notification type and run the actual worker process that
listen for incoming push notifications. To be able to stop blocking process in runtime you can specify a condition and
call `$dispatcher->stop()` method from given callback. In this example we're waiting for specific message `terminate`
within `control` channel that we subscribed to before entering the loop.

```php
// Storage for incoming notifications.
$messages = [];

// Create dispatcher for push notifications.
$dispatcher = new Predis\Consumer\Push\DispatcherLoop($push);

$dispatcher->attachCallback(
    PushResponseInterface::MESSAGE_DATA_TYPE,
    static function (array $payload, DispatcherLoopInterface $dispatcher) {
        global $messages;
        [$channel, $message] = $payload;

        if ($channel === 'control' && $message === 'terminate') {
            echo "Terminating notification consumer.\n";
            $dispatcher->stop();

            return;
        }

        $messages[] = $message;
        echo "Received message: {$message}\n";
    }
);

// Run consumer loop with attached callbacks.
$dispatcher->run();

// Count all messages that were received during consumer loop.
$messagesCount = count($messages);
echo "We received: {$messagesCount} messages\n";
```

This example shows a simple script to count all incoming messages from push notifications that we receive from
subscribed channels until stop condition will be met. Examples available in `examples/` folder.

### Sharded pub/sub ###
From Redis 7.0, sharded Pub/Sub is introduced in which shard channels are assigned to slots by the same algorithm used
to assign keys to slots.

Predis 3.0 provides an API that allows to use pub/sub for Cluster connections using sharded pub/sub from Redis.
You don't need to specify any additional configuration to enable sharded pub/sub, it will be automatically enabled if
Cluster connection is using.

Implementation looks pretty much the same as Push notification, so you need to set up consumer
and run it over Dispatcher loop object. All examples available in `examples/` folder.
## Development ##


### Reporting bugs and contributing code ###

Contributions to Predis are highly appreciated either in the form of pull requests for new features,
bug fixes, or just bug reports. We only ask you to adhere to issue and pull request templates.


### Test suite ###

__ATTENTION__: Do not ever run the test suite shipped with Predis against instances of Redis running
in production environments or containing data you are interested in!

Predis has a comprehensive test suite covering every aspect of the library and that can optionally
perform integration tests against a running instance of Redis (required >= 2.4.0 in order to verify
the correct behavior of the implementation of each command. Integration tests for unsupported Redis
commands are automatically skipped. If you do not have Redis up and running, integration tests can
be disabled. See [the tests README](tests/README.md) for more details about testing this library.

Predis uses GitHub Actions for continuous integration and the history for past and current builds can be
found [on its actions page](https://github.com/predis/predis/actions).

### License ###

The code for Predis is distributed under the terms of the MIT license (see [LICENSE](LICENSE)).

[ico-license]: https://img.shields.io/github/license/predis/predis.svg?style=flat-square
[ico-version-stable]: https://img.shields.io/github/v/tag/predis/predis?label=stable&style=flat-square
[ico-version-dev]: https://img.shields.io/github/v/tag/predis/predis?include_prereleases&label=pre-release&style=flat-square
[ico-downloads-monthly]: https://img.shields.io/packagist/dm/predis/predis.svg?style=flat-square
[ico-build]: https://img.shields.io/github/actions/workflow/status/predis/predis/tests.yml?branch=main&style=flat-square
[ico-coverage]: https://img.shields.io/coverallsCoverage/github/predis/predis?style=flat-square

[link-releases]: https://github.com/predis/predis/releases
[link-actions]: https://github.com/predis/predis/actions
[link-downloads]: https://packagist.org/packages/predis/predis/stats
[link-coverage]: https://coveralls.io/github/predis/predis