Redis sorted set by string values - language-agnostic

Question

Suppose you have a hash 'users' whose entries map numeric IDs to JSON-encoded arrays, so, for instance, the integer 1 maps to the string {name: 'John', surname: 'Doe', occupation: 'plumber'}.
The numeric IDs of items in the hash are stored in various lists. Thus, if 'foobar' is one of these lists, to retrieve the actual data from it I would run a simple Lua script (implementing a server-side join operation). Or, as I've just learned, I could use something like
SORT foobar BY inexistent_key GET user:*
but that implies storing each user's data into a separate key which seems expensive (in my case I have many small collections so I want to take advantage of Redis compression of hashes).
The question is finally this: I need to keep these lists ordered alphabetically by, say, each user's surname, then by name. What is the best way to achieve this, without changing too much the underlying data model (if possible)?
So far the best I could think of was using the SORT command together with the BY and STORE clauses, that is, running
SORT foobar BY surname:* STORE foobar
whenever the list is changed, but that way I'd need many keys. If I could use a hash in the BY clause that would be the ideal solution, it seems to me.
If the fields I want to sort by were somehow limited (as in, just a couple hundred surnames) I could think of mapping strings to integers and use a Redis sorted set, but this doesn't seem to be the case.

Answer 1

You can sort by hash keys, without the complexity of Lua scripts, but you will have to duplicate the keys in your Json structure to Redis' hash keys.
The below example has the following structure:
users is a set with the user id's in query.
user:X is the Redis hash which contains the duplicated name / surname data.
userdata:X is your original users Json hash.
redis> hmset user:1 name First surname User
OK
redis> set userdata:1 "{occupation: 'Tester'}"
OK
redis> hmset user:2 name Last surname Violet
OK
redis> set userdata:2 "{occupation: 'Bookseller'}"
OK
redis> hmset user:3 name Middle surname Veredict
OK
redis> set userdata:3 "{occupation: 'Judge'}"
OK
redis> hmset user:4 name Ultimate surname Veredict
OK
redis> set userdata:4 "{occupation: 'Ultimate Judge'}"
OK
redis> sadd users 1
(integer) 1
redis> sadd users 2
(integer) 1
redis> sadd users 3
(integer) 1
redis> sadd users 4
(integer) 1
redis> sort users by user:*->surname get user:*->name get user:*->surname get userdata:* alpha
1) "First"
2) "User"
3) "{occupation: 'Tester'}"
4) "Middle"
5) "Veredict"
6) "{occupation: 'Judge'}"
7) "Ultimate"
8) "Veredict"
9) "{occupation: 'Ultimate Judge'}"
10) "Last"
11) "Violet"
12) "{occupation: 'Bookseller'}"
Edit
I have overlooked that multiple By only consider the last clause. So you can't sort by more than one key in one command.
Also the SORT command used for lexicographic ordering now requires alpha modifier.

Answer 2

In the end it seems to me that a way to approach my problem is to use the table.sort() function available in Lua. In addition to the small hash 'users' and the small list of IDs 'foobar' I introduced another small hash, say 'users:sort-strings', where I store the strings by which I'd like to sort the IDs in 'foobar' (in the fictitious example a combination of surname and name). To sort 'foobar' I would then run the following Lua snippet in Redis, passing as keys 'foobar', 'users:sort-strings' and 'foobar:tmp' (a temporary key).
local lst = redis.call('LRANGE', KEYS[1], 0, -1)
local sort_function = function (id0, id1)
local s0 = redis.call('HGET', KEYS[2], id0)
local s1 = redis.call('HGET', KEYS[2], id1)
return (s0 < s1)
end
table.sort(lst, sort_function)
for key, value in ipairs(lst) do
redis.call('RPUSH', KEYS[3], value)
end
redis.call('DEL', KEYS[1])
redis.call('RENAME', KEYS[3], KEYS[1])