The package uses reflection to decode default types (int, string
etc.) from Redis map results (key-value pair sequences) into
struct fields where the fields are matched to Redis keys by tags.
Similar to how `encoding/json` allows custom decoders using
`UnmarshalJSON()`, the package supports decoding of arbitrary
types into struct fields by defining a `Decode(string) error`
function on types.
The field/type spec of every struct that's passed to Scan() is
cached in the package so that subsequent scans avoid iteration
and reflection of the struct's fields.
As of version 6.0 you can use this 'type' option to ask SCAN to only
return objects that match a given type, allowing you to
iterate through the database looking for keys of a specific type.
You have to set the ReadOnly flag on the ClusterOptions to make it fall back to
non-master nodes on reads. For non-clustered situations where you have multiple
nodes in a single ClusterSlot, it made the non-master node useless because it would
never route requests to it and if you set ReadOnly=true it would issue a READONLY
command to it, which would fail.
See comment in code for more details.
The original text `all sentinels are unreachable` can sometimes be misleading, making some users think that all of their redis sentinels are down.
But it might not be the reality especially when using kubernetes and specify `c.sentinelAddrs` as the Service's domain. In this case, when the sentinel to which the domain is redirected runs into some errors despite other sentinels running healthily, the original text given by redis client will be confusing and delivering false message!
Just got out of a similar dilemma after checking the source code, feel free to correct me if I am wrong :)
