Redis plugin for the Salt caching subsystem.
New in version 2017.7.0.
Changed in version 3005.
To enable this cache plugin, the master will need the python client for redis installed. This can be easily installed with pip:
salt \* pip.install redis
As Redis provides a simple mechanism for very fast key-value store, in order to provide the necessary features for the Salt caching subsystem, the following conventions are used:
A Redis key consists of the bank name and the cache key separated by /
, e.g.:
$KEY_minions/alpha/stuff
where minions/alpha
is the bank name
and stuff
is the key name.
As the caching subsystem is organised as a tree, we need to store the caching
path and identify the bank and its offspring. At the same time, Redis is
linear and we need to avoid doing keys <pattern>
which is very
inefficient as it goes through all the keys on the remote Redis server.
Instead, each bank hierarchy has a Redis SET associated which stores the list
of sub-banks. By default, these keys begin with $BANK_
.
In addition, each key name is stored in a separate SET of all the keys within
a bank. By default, these SETs begin with $BANKEYS_
.
For example, to store the key my-key
under the bank root-bank/sub-bank/leaf-bank
,
the following hierarchy will be built:
127.0.0.1:6379> SMEMBERS $BANK_root-bank
1) "sub-bank"
127.0.0.1:6379> SMEMBERS $BANK_root-bank/sub-bank
1) "leaf-bank"
127.0.0.1:6379> SMEMBERS $BANKEYS_root-bank/sub-bank/leaf-bank
1) "my-key"
127.0.0.1:6379> GET $KEY_root-bank/sub-bank/leaf-bank/my-key
"my-value"
There are four types of keys stored:
$BANK_*
is a Redis SET containing the list of banks under the current bank.
$BANKEYS_*
is a Redis SET containing the list of keys under the current bank.
$KEY_*
keeps the value of the key.
$TSTAMP_*
stores the last updated timestamp of the key.
These prefixes and the separator can be adjusted using the configuration options:
$BANK
The prefix used for the name of the Redis key storing the list of sub-banks.
$BANKEYS
The prefix used for the name of the Redis key storing the list of keys under a certain bank.
$KEY
The prefix of the Redis keys having the value of the keys to be cached under a certain bank.
$TSTAMP
The prefix for the last modified timestamp for keys.
New in version 3005.
_
The separator between the prefix and the key body.
The connection details can be specified using:
localhost
The hostname of the Redis server.
6379
The Redis server port.
False
Whether cluster_mode is enabled or not
A list of host, port dictionaries pointing to cluster members. At least one is required but multiple nodes are better
cache.redis.cluster.startup_nodes
- host: redis-member-1
port: 6379
- host: redis-member-2
port: 6379
False
Some cluster providers restrict certain redis commands such as CONFIG for enhanced security. Set this option to true to skip checks that required advanced privileges.
Note
Most cloud hosted redis clusters will require this to be set to True
'0'
The database index.
Note
The database index must be specified as string not as integer value!
Redis connection password.
unix_socket_path:
New in version 2018.3.1.
Path to a UNIX socket for access. Overrides host / port.
Configuration Example:
cache.redis.host: localhost
cache.redis.port: 6379
cache.redis.db: '0'
cache.redis.password: my pass
cache.redis.bank_prefix: #BANK
cache.redis.bank_keys_prefix: #BANKEYS
cache.redis.key_prefix: #KEY
cache.redis.timestamp_prefix: #TICKS
cache.redis.separator: '@'
Cluster Configuration Example:
cache.redis.cluster_mode: true
cache.redis.cluster.skip_full_coverage_check: true
cache.redis.cluster.startup_nodes:
- host: redis-member-1
port: 6379
- host: redis-member-2
port: 6379
cache.redis.db: '0'
cache.redis.password: my pass
cache.redis.bank_prefix: #BANK
cache.redis.bank_keys_prefix: #BANKEYS
cache.redis.key_prefix: #KEY
cache.redis.separator: '@'
Checks if the specified bank contains the specified key.
Fetch data from the Redis cache.
Remove the key from the cache bank with all the key content. If no key is specified, remove the entire bank with all keys and sub-banks inside. This function is using the Redis pipelining for best performance. However, when removing a whole bank, in order to re-create the tree, there are a couple of requests made. In total:
one for node in the hierarchy sub-tree, starting from the bank node
one pipelined request to get the keys under all banks in the sub-tree
one pipeline request to remove the corresponding keys
This is not quite optimal, as if we need to flush a bank having a very long list of sub-banks, the number of requests to build the sub-tree may grow quite big.
An improvement for this would be loading a custom Lua script in the Redis instance of the user
(using the register_script
feature) and call it whenever we flush.
This script would only need to build this sub-tree causing problems. It can be added later and the behaviour
should not change as the user needs to explicitly allow Salt inject scripts in their Redis instance.
Effectively a noop. Return an empty dictionary.
Lists entries stored in the specified bank.
Store the data in a Redis key.
Return the Unix Epoch timestamp of when the key was last updated. Return None if key is not found.