I’ve added search and trim to my expiring records module in Erlang. This started out as an in-memory key/value store, that I then migrated over to using Mnesia and eventually to a replicated Mnesia table.

The fetch/1 function is already doing a simple query, with match_object.

Result = mnesia:match_object(expiring_records, #record{key = Key, value = '_', expires_at = '_'}, read)

The three parameters there are the name of the table - expiring_records, the matching pattern and the lock type (read lock).

The fetch/1 function looks up the key as it was added to the table with store/3. If the key is a tuple, we can also do a partial match:

Result = mnesia:match_object(expiring_records, #record{key = {'_', "bongo"}, value = '_', expires_at = '_'}, read)

I’ve added a search/1 function the module that takes in a matching pattern and returns a list of items where the key matches the pattern. Here’s the test for the search/1 function:

search_partial_key(_Config) ->
    ok = expiring_records:store({"bingo", "parlor"}, "bongo", erlang:system_time(second) + 3600),
    ok = expiring_records:store("bingo", "bongo", erlang:system_time(second) + 3600),
    ok = expiring_records:store({"smoking", "parlor"}, "bongo", erlang:system_time(second) + 3600),
    ok = expiring_records:store("smoking", "bongo", erlang:system_time(second) + 3600),
    ok = expiring_records:store("parlor", "bongo", erlang:system_time(second) + 3600),
    ok = expiring_records:store({"reading", "parlor"}, "bongo", erlang:system_time(second) - 1),
    [A, B] = expiring_records:search({'_', "parlor"}),
    {"bingo", "parlor"} = A#record.key,
    {"smoking", "parlor"} = B#record.key.

For more complex queries we can use select. This function takes in a match specification that goes beyond the pattern matching done by match_object. The trim/0 function finds records where the expiration time has passed:

handle_call(trim, _From, State) ->
    Trans = fun() ->
        Now = erlang:system_time(second),
        MatchHead = #record{key='$1', expires_at = '$2', _='_'},
        Guard = {'>', Now, '$2'},
        Result = '$1',
        ExpiredKeys = mnesia:select(expiring_records, [{MatchHead, [Guard], [Result]}]),
        delete_records(ExpiredKeys)
    end,
    {atomic, ok} = mnesia:transaction(Trans),
    {reply, ok, State};

The MatchHead specifies the things we care about in the record and gives them labels that we can refer in the other parts of the match specification. The key is labelled $1 and is referred to in the Result. The expires_at field is labelled $2 and is referred to in the Guard. This guard expression is quite simple - Now should be larger than the expiration time of the record. This select call returns a list of keys for records that have expired, that will in turn get deleted.

I need to experiment with the performance of these sort of queries. Mnesia tables can have secondary indices that ought to help, but I’m sure queries end up being a sequential scan of all entries, applying the pattern matching or guard expression to each entry in turn.