Add Prefix to Redis Cache

[ori-os]

Adding to Thanos' Redis cache config a prefix option.
If configured, all cache keys used by the Thanos component will use this prefix.

This is a pretty small and simple backwards-compatible change that can be added to improve the Redis Cache support.

Using prefixes is a common practice when using Redis and is used to achieve better isolation.

Currently, there is support for choosing a DB, which is a form of isolation. However, prefixes allow for even more customizability in that front with no real downsides.

Below is the suggested new default config redis:

type: REDIS
config:
  addr: ""
  username: ""
  password: ""
  db: 0
  prefix: "" # New
  dial_timeout: 5s
  read_timeout: 3s
  write_timeout: 3s
  max_get_multi_concurrency: 100
  get_multi_batch_size: 100
  max_set_multi_concurrency: 100
  set_multi_batch_size: 100
  tls_enabled: false
  tls_config:
    ca_file: ""
    cert_file: ""
    key_file: ""
    server_name: ""
    insecure_skip_verify: false
  cache_size: 0
  master_name: ""
  max_async_buffer_size: 10000
  max_async_concurrency: 20
  set_async_circuit_breaker_config:
    enabled: false
    half_open_max_requests: 10
    open_duration: 5s
    min_requests: 50
    consecutive_failures: 5
    failure_percent: 0.05
  expiration: 24h0m0s

By setting the default value "", we ensure users who are not interested in this feature won't be affected by it.

We can for example, set the prefix for the query-frontend component to frontend and the prefix for the store-gateway component to store and by that achieve easy and simple isolation between the two, even if they use the same Redis instance and DB.

Needless to say this also helps with integration with other, none-Thanos components which might also use Redis with their own prefix.

I would love to hear more about what everyone has to say on the matter, and I also hope to potentially see this implemented in the near future - as it can be quite handy.

[cnaples79]

I'm working on this.

[GiedriusS]

In what way isolation happens when you have prefixes? It's just easier to type/notice when exploring the values through redis-cli? Or is there some feature somewhere where one can partition the data according to the prefixes?

[ori-os]

In what way isolation happens when you have prefixes? It's just easier to type/notice when exploring the values through redis-cli? Or is there some feature somewhere where one can partition the data according to the prefixes?

It is more so a naming convention. Redis is flat by nature and prefixes act like a namespace divider. Apart from organizing your keys, they also ensure there are no key collisions between different components that use the same Redis (but for different purposes). That's because a collision is impossible when each component has its own unique prefix.