Add caching of sharding function

The ddl.bucket_id() function needs to know a sharding function.
It is costly to obtain the function declaration /
definition stored in the _ddl_sharding_func space.

Cache contains:
raw_tuple - raw sharding metadata, used for get_schema()
parsed_func_name - parsed dot notation (like {'foo', 'bar'})
callable - function ready to call, this offloads using of loadstring()
error - string with an error: not nil only if setting callable fails

Cache will be rebuilded if:
* _ddl_sharding_func space changed: cache sets _ddl_sharding_func:on_replace
  trigger
* schema changed: cache checks box.internal.schema_version changes

This patch does not serve hot reload techniques.
This entails an on_replace trigger duplication if hot reload occurs.
Hot reload support will be done in separate task:
#87

Benchmarks - 10000000 bucket_id() calls (file test/bench_cache.lua):
Baseline (no DDL):     3.38s user 0.01s system 99% cpu 3.389 total
After this patch:
DDL with function body: 3.81s user 0.01s system 99% cpu 3.818 total
DDL with function name: 5.49s user 0.00s system 99% cpu 5.495 total
Before patch:
DDL with function body: 55.95s user 0.40s system 99% cpu 56.354 total
DDL with function name: 13.68s user 0.13s system 99% cpu 13.807 total

Closes #82

Author: 0x501D

ddl/cache.lua

@@ -0,0 +1,120 @@
+local cache = nil
+
+local SPACE_NAME_IDX      = 1
+local SHARD_FUNC_NAME_IDX = 2
+local SHARD_FUNC_BODY_IDX = 3
+
+-- Build cache.
+--
+-- Cache structure format:
+--
+-- cache = {
+--   spaces = {
+--     [space_name] = {
+--       -- raw sharding metadata, used for ddl.get()
+--       raw_tuple = <tuple object> (<nil> at error),
+--       -- parsed dot notation (like {'foo', 'bar'})
+--       parsed_func_name = <table> or <nil>
+--       -- a function ready to call
+--       callable = <function> or <nil>,
+--       -- string with an error: not nil only if setting callable fails
+--       error = <string> or <nil>,
+--     }
+--   },
+--   -- current schema version
+--   schema_version = <...>,
+-- }
+
+-- function returns nothing
+local function cache_build()
+    -- clear cache
+    cache.spaces = {}
+
+    if box.space._ddl_sharding_func == nil then
+        return
+    end
+
+    for _, tuple in box.space._ddl_sharding_func:pairs() do
+        local space_name = tuple[SPACE_NAME_IDX]
+        local func_name = tuple[SHARD_FUNC_NAME_IDX]
+        local func_body = tuple[SHARD_FUNC_BODY_IDX]
+
+        cache.spaces[space_name] = {
+            raw_tuple = tuple
+        }
+
+        if func_body ~= nil then
+            local sharding_func, err = loadstring('return ' .. func_body)
+            if sharding_func == nil then
+                cache.spaces[space_name].error =
+                    string.format("Body is incorrect in sharding_func for space (%s): %s",
+                    space_name, err)
+            else
+                cache.spaces[space_name].callable = sharding_func()
+            end
+        elseif func_name ~= nil then
+            -- we cannot save the function itself into the cache,
+            -- because the function can be changed in runtime and
+            -- there is no way to catch this change
+            local chunks = string.split(func_name, '.')
+            cache.spaces[space_name].parsed_func_name = chunks
+        end
+    end
+
+    cache.schema_version = box.internal.schema_version()
+
+end
+
+-- Rebuild cache if _ddl_sharding_func space changed.
+local function cache_set_trigger()
+    if box.space._ddl_sharding_func == nil then
+        return
+    end
+
+    local trigger_found = false
+
+    for _, func in pairs(box.space._ddl_sharding_func:on_replace()) do
+        if func == cache_build then
+            trigger_found = true
+            break
+        end
+    end
+
+    if not trigger_found then
+        box.space._ddl_sharding_func:on_replace(cache_build)
+    end
+end
+
+-- Get data from cache.
+-- Returns all cached data for "space_name" or nil.
+local function cache_get(space_name)
+    if space_name == nil then
+        return nil
+    end
+
+    -- using tarantool internal API.
+    -- this is not reliable, but it is the only way to track
+    -- schema_version changes. Fix it if a public method appears:
+    -- https://github.com/tarantool/tarantool/issues/6544
+    local schema_version = box.internal.schema_version()
+
+    if not cache then
+        cache = {}
+        box.atomic(cache_build)
+        cache_set_trigger()
+    end
+
+    -- rebuild cache if database schema changed
+    if schema_version ~= cache.schema_version then
+        box.atomic(cache_build)
+        cache_set_trigger()
+    end
+
+    return cache.spaces[space_name]
+end
+
+return {
+    internal = {
+        get = cache_get,
+    }
+}

ddl/get.lua

@@ -1,4 +1,5 @@
 local utils = require('ddl.utils')
+local cache = require('ddl.cache')
 local ddl_check = require('ddl.check')
 
 local function _get_index_field_path(space, index_part)
@@ -66,11 +67,18 @@ local function get_metadata(space_name, metadata_name)
 end
 
 local function get_sharding_func(space_name)
-    local record = get_metadata(space_name, "sharding_func")
-    if not record then
+    return cache.internal.get(space_name)
+end
+
+local function get_sharding_func_raw(space_name)
+    local record = cache.internal.get(space_name)
+
+    if not record or not record.raw_tuple then
         return nil
     end
 
+    record = record.raw_tuple
+
     if record.sharding_func_body ~= nil then
         return {body = record.sharding_func_body}
     end
@@ -97,7 +105,7 @@ local function get_space_schema(space_name)
     space_ddl.engine = box_space.engine
     space_ddl.format = box_space:format()
     space_ddl.sharding_key = get_sharding_key(space_name)
-    space_ddl.sharding_func = get_sharding_func(space_name)
+    space_ddl.sharding_func = get_sharding_func_raw(space_name)
     for _, field in ipairs(space_ddl.format) do
         if field.is_nullable == nil then
             field.is_nullable = false
@@ -115,21 +123,20 @@ local function get_space_schema(space_name)
 end
 
 local function prepare_sharding_func_for_call(space_name, sharding_func_def)
-    if type(sharding_func_def) == 'string' then
-        local sharding_func = utils.get_G_function(sharding_func_def)
+    if sharding_func_def.error ~= nil then
+        return nil, sharding_func_def.error
+    end
+
+    if sharding_func_def.parsed_func_name ~= nil then
+        local sharding_func = utils.get_G_function(sharding_func_def.parsed_func_name)
         if sharding_func ~= nil and
            ddl_check.internal.is_callable(sharding_func) == true then
             return sharding_func
         end
     end
 
-    if type(sharding_func_def) == 'table' then
-        local sharding_func, err = loadstring('return ' .. sharding_func_def.body)
-        if sharding_func == nil then
-            return nil, string.format(
-                "Body is incorrect in sharding_func for space (%s): %s", space_name, err)
-        end
-        return sharding_func()
+    if sharding_func_def.callable ~= nil then
+        return sharding_func_def.callable
     end
 
     return nil, string.format(

ddl/utils.lua

@@ -189,9 +189,19 @@ end
 -- split sharding func name in dot notation by dot
 -- foo.bar.baz -> chunks: foo bar baz
 -- foo -> chunks: foo
+--
+-- func_name parameter may be a string in dot notation or table
+-- if func_name type is of type table it is assumed that it is already split
 local function get_G_function(func_name)
-    local chunks = string.split(func_name, '.')
     local sharding_func = _G
+    local chunks
+
+    if type(func_name) == 'string' then
+        chunks = string.split(func_name, '.')
+    else
+        chunks = func_name
+    end
+
     -- check is the each chunk an identifier
     for _, chunk in pairs(chunks) do
         if not check_name_isident(chunk) or sharding_func == nil then

test/bench_cache.lua

@@ -0,0 +1,180 @@
+#!/usr/bin/env tarantool
+
+local db = require('test.db')
+local ddl = require('ddl')
+local helper = require('test.helper')
+
+local ITERS = 10000000
+
+local function mpcrc32(shard_key)
+    local digest = require('digest')
+    if type(shard_key) ~= 'table' then
+        return digest.crc32(tostring(shard_key))
+    else
+        local crc32 = digest.crc32.new()
+        for _, v in ipairs(shard_key) do
+            crc32:update(tostring(v))
+        end
+        return crc32:result()
+    end
+end
+
+local sharding_func_body = [[
+function(shard_key)
+    local digest = require('digest')
+    if type(shard_key) ~= 'table' then
+        return digest.crc32(tostring(shard_key))
+    else
+        local crc32 = digest.crc32.new()
+        for _, v in ipairs(shard_key) do
+            crc32:update(tostring(v))
+        end
+        return crc32:result()
+    end
+end
+]]
+
+local primary_index = {
+    type = 'HASH',
+    unique = true,
+    parts = {
+        {path = 'string_nonnull', is_nullable = false, type = 'string'},
+        {path = 'unsigned_nonnull', is_nullable = false, type = 'unsigned'},
+    },
+    name = 'primary'
+}
+
+local bucket_id_idx = {
+    type = 'TREE',
+    unique = false,
+    parts = {{path = 'bucket_id', type = 'unsigned', is_nullable = false}},
+    name = 'bucket_id'
+}
+
+local function space_init()
+    db.drop_all()
+
+    local space = {
+        engine = 'memtx',
+        is_local = true,
+        temporary = false,
+        format = table.deepcopy(helper.test_space_format())
+    }
+    table.insert(space.format, 1, {
+        name = 'bucket_id', type = 'unsigned', is_nullable = false
+    })
+
+    space.indexes = {
+        table.deepcopy(primary_index),
+        table.deepcopy(bucket_id_idx)
+    }
+    space.sharding_key = {'unsigned_nonnull', 'integer_nonnull'}
+    local schema = {
+        spaces = {
+            space = space,
+        }
+    }
+
+    return schema
+end
+
+local function run_body()
+    db.init()
+    local schema = space_init()
+    schema.spaces.space.sharding_func = {
+        body = sharding_func_body
+    }
+
+    local _, err = ddl.set_schema(schema)
+    if err then
+        print(err)
+        os.exit()
+    end
+
+    for i=1,ITERS do
+        local _, err = ddl.bucket_id('space', i)
+        if err then
+            print(err)
+            os.exit()
+        end
+    end
+
+    print("Done")
+    os.exit()
+end
+
+local function run_baseline()
+    db.init()
+    local schema = space_init()
+    schema.spaces.space.sharding_func = {
+        body = sharding_func_body
+    }
+
+    local _, err = ddl.set_schema(schema)
+    if err then
+        print(err)
+        os.exit()
+    end
+
+    for i=1,ITERS do
+        mpcrc32(i)
+        if err then
+            print(err)
+            os.exit()
+        end
+    end
+
+    print("Done")
+    os.exit()
+end
+
+local function run_name()
+    db.init()
+    local schema = space_init()
+    local sharding_func_name = 'mpcrc32'
+    rawset(_G, sharding_func_name, mpcrc32)
+    schema.spaces.space.sharding_func = sharding_func_name
+
+    local _, err = ddl.set_schema(schema)
+    if err then
+        print(err)
+        os.exit()
+    end
+
+    for i=1,ITERS do
+        local _, err = ddl.bucket_id('space', i)
+        if err then
+            print(err)
+            os.exit()
+        end
+    end
+
+    print("Done")
+    os.exit()
+end
+
+local function main()
+    local getopt = require('posix.unistd').getopt
+
+    if #arg == 0 then
+        print("Usage:", arg[0], "\n  -l - baseline benchmark",
+            "\n  -n - function name caching benchmark",
+            "\n  -b - function body caching benchmark")
+        os.exit()
+    end
+
+    for opt, _, optind in getopt(arg, 'lnb') do
+        if opt == '?' then
+            return print('unrecognized option', arg[optind -1])
+        end
+        if opt == 'l' then
+            run_baseline()
+        elseif opt == 'n' then
+            run_name()
+        elseif opt == 'b' then
+            run_body()
+        end
+    end
+end
+
+main()