Authentication for sentinels was implemented in v1.1.5 (commit 2e76410)
but ended up being bugged (see ISSUE #658). This is now postponed as it
requires a more thorough investigation.
Password-based authentication for sentinels has been added in Redis 5.
Predis was actively ignoring any "password" parameter for sentinels when
creating connections to them to avoid issues when this parameter is set
in the default "parameters" array passed via client options, as they are
applied to **every** connection created by Predis (see #346).
We need to find a better way to specify a common password for sentinels
to be handled in a different way than the ones for Redis nodes. For now
each sentinel node protected by password must have an explicit password
set in its parameters list even if this password, by design, is the same
for all sentinels. Since we cannot use default "parameters" as explained
above but we still need to pass a common value for all sentinels an idea
could be using a dedicated client option like we did with "service", but
we will see later.
In this commit we also explicitly reset any "username" parameter as it
would trigger an `AUTH $username $password` but sentinels do not support
ACL authentication.
Fixes#594.
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.
This prevents an early failure of the command execution on the client
when one slave gets back online but is still loading the dataset from
disk (when this happens, Redis returns the -LOADING error response).
This commit fixes#280.
When the connection to a node in the cluster fails in the attempt to
execute a command, Predis now removes the failed connection from the
cluster pool and contacts a random node to ask for a fresh slots map
and tries to execute the command once again.
When the cluster is configured to have each master replicated to one
or more slaves, one the slaves is automatically promoted to the role
of master by redis-cluster with this change being reflected in the
output of CLUSTER SLOTS, so the next execution should run just fine.
Our current approach is relatively naive as CLUSTER SLOTS is executed
against a random master node, meaning that the client must open a new
connection and execute one more roundtrip only to fetch the new slots
map. For now it is enough, it is still better than having the client
fail when you actually have somes slaves in your redis-cluster setup,
but one improvement could consist in caching the list of slaves for
each master returned in the response of CLUSTER SLOTS so that when a
connection fails the client can try to guess which connection should
use for the next attempt.
This commit closes#173, closes#215, and closes#314.
When using redis-cluster it is now also possible to pass one slot or
non-contiguous ranges of slots via connection parameters in order to
improve the ability to pre-configure the slots map on the client.
Here is an example:
$parameters = [
'tcp://10.0.0.1:6379?slots=0-5460,5500-5600,11000',
'tcp://10.0.0.2:6379?slots=5461-5499,5600-10921',
'tcp://10.0.0.3:6379?slots=10922-10999,11001-16383',
];
This commit fixes#312 (props to @kenotr0n for the original PR).
Now the client can discover the whole replication configuration by
asking to one of the servers (master has the precedence) using the
INFO REPLICATION command. This is obviously a best-effort fallback
and there is no strong guarantee about reliability and efficiency.
By enabling auto-discovery, the client automates this process when
the execution of a command fails because one of the target servers
is unreachable. The replication connection requires an instance of
connection factory associated to it in order to be able to create
new connections on the fly.
It is possible to enable the auto-discovery procedure easily via
client options:
$client = new Predis\Client($servers, [
'replication' => true,
'autodiscovery' => true,
]);
Internally the replication class uses this order to pick which server
it should connect to: current connection, one of the slaves, master.
If there is at least 1 slave, connect() will not fail even if master
is undefined. If there are no slaves, connect() will pick master. If
there are no connections registered for replication, connect() will
fail immediatly.
Using IPv6 with Predis was basically impossible due to various inconsistencies
and bugs through the library, now it is supported by all the connection classes.
Following the standard for IPv6 literal addresses in URI strings, the IP literal
must be enclosed within square brackets when passing the parameters as a string:
$parameters = 'tcp://[2001:db8:0:f101::1]:6379';
See https://tools.ietf.org/html/rfc3986#section-3.2.2 for further details.
This commit also fixes#239 making redis-cluster usable with nodes using IPv6.
Serializing an instance of a redis-cluster connection with its nodes
can take more than 1 seconds on extremely slow hardware.
See issue #220 for reference.
This optimization makes it possible to fetch the slots map directly
from the server indicated by the -MOVED response eliminating the need
to use a random node in the pool, which in turn could require Predis
to open a new and useless connection.
This is a more sane default as it allows users to indicate only a few
servers of the whole cluster composition, while it previously required
a more complex configuration of the client using client options.
This feature can be disabled using the "enableAutoSlotsMap()" method.
This change is possible because, after a few changes in redis-cluster,
our default cluster strategy used for client side sharding and the one
used for redis-cluster turned out to be exactly the same, except for
the hashing function used to calculate distribution.
Actually some checks used to enforce correctness are redundant in the
context of redis-cluster (e.g. the one used to make sure that keys in
requests performing cross-keys operations will hash to the same slot,
which is performed by the server) so we could also add a more dumb and
permissive cluster strategy that relies on checks performed by Redis.
Differently to v0.8, the strategy for client-side sharding now uses
the same rules for extracting hash tags from keys especially when
empty tags are found in the string.
CLUSTER SLOTS returns a structured response easier to handle compared
to the bulk response of CLUSTER NODES which must be parsed making the
whole thing more fragile.
CLUSTER SLOTS has been added in Redis 3.0.0b7.
This is more consistent with the actual purpose of this method and
more in-line with a possible future change in the underlying command
used to retrieve the slots map if redis-cluster will implement the
CLUSTER SLOTS command.
* Renamed SingleConnectionInterface to NodeConnectionInterface since
this name is better and makes even more sense in the context of
cluster and replication scenarios.
* Moved specialized aggregate connections (the ones implementing both
predis and redis cluster and master/slave replication) in a newly
created Predis\Connection\Aggregate sub-namespace.
* Removed the "Connection" part from names of aggregate connection
interfaces in the Predis\Connection\Aggregate sub-namespace.
* Changed "Composable" to "Composite" in the name of interfaces and
classes that can use pluggable protocol processors.
Daniele Alessandri
Mark Fettig
Pascal Hofmann