Make page cache level configurable

priv/leveled.schema

@@ -72,7 +72,7 @@
   {datatype, integer}
 ]}.
 
-%% @doc The approximate size (in bytes) when a Journal file should be rolled.
+%% @doc The approximate count of objects when a Journal file should be rolled.
 %% This time measured in object count, a file will be rolled if either the
 %% object count or the journal size limit is reached.  Default 200K.
 %% Note that on startup an actual maximum size will be chosen which varies by
@@ -83,6 +83,14 @@
   {datatype, integer}
 ]}.
 
+%% @doc The level of the ledger to be pre-loaded into the page cache
+%% Depending on how much memory is available for the page cache, and how much
+%% disk I/O activity can be tolerated at startup - then the level at which the
+%% ledger is forced into the page cache can be controlled by configuration.
+{mapping, "leveled.ledger_pagecachelevel", "leveled.ledger_pagecachelevel", [
+  {default, 4},
+  {datatype, integer}
+]}.
 
 %% @doc The number of journal compactions per vnode per day
 %% The higher the value, the more compaction runs, and the sooner space is 

priv/leveled_multi.schema

@@ -66,7 +66,7 @@
   {datatype, integer}
 ]}.
 
-%% @doc The approximate size (in bytes) when a Journal file should be rolled.
+%% @doc The approximate count of objects when a Journal file should be rolled.
 %% This time measured in object count, a file will be rolled if either the
 %% object count or the journal size limit is reached.  Default 200K.
 %% Note that on startup an actual maximum size will be chosen which varies by
@@ -77,6 +77,16 @@
   {datatype, integer}
 ]}.
 
+%% @doc The level of the ledger to be pre-loaded into the page cache
+%% Depending on how much memory is available for the page cache, and how much
+%% disk I/O activity can be tolerated at startup - then the level at which the
+%% ledger is forced into the page cache can be controlled by configuration.
+{mapping, "multi_backend.$name.leveled.ledger_pagecachelevel", "riak_kv.multi_backend", [
+  {default, 4},
+  {datatype, integer}
+]}.
+
+
 %% @doc The number of journal compactions per vnode per day
 %% The higher the value, the more compaction runs, and the sooner space is 
 %% recovered.  But each run has a cost

src/leveled_bookie.erl

@@ -144,6 +144,7 @@
                 {maxrunlength_compactionpercentage, 70.0},
                 {reload_strategy, []},
                 {max_pencillercachesize, ?MAX_PCL_CACHE_SIZE},
+                {ledger_preloadpagecache_level, ?SST_PAGECACHELEVEL_LOOKUP},
                 {compression_method, ?COMPRESSION_METHOD},
                 {compression_point, ?COMPRESSION_POINT},
                 {log_level, ?LOG_LEVEL},
@@ -320,6 +321,10 @@
             % The minimum size 400 - attempt to set this vlaue lower will be 
             % ignored.  As a rule the value should be at least 4 x the Bookie's
             % cache size
+        {ledger_preloadpagecache_level, pos_integer()} |
+            % To which level of the ledger should the ledger contents be
+            % pre-loaded into the pagecache (using fadvise on creation and
+            % startup)
         {compression_method, native|lz4} |
             % Compression method and point allow Leveled to be switched from
             % using bif based compression (zlib) to using nif based compression
@@ -1184,12 +1189,14 @@ init([Opts]) ->
                     ok
             end,
 
+            PageCacheLevel = proplists:get_value(ledger_preloadpagecache_level, Opts),    
+
             {HeadOnly, HeadLookup, SSTPageCacheLevel} = 
                 case proplists:get_value(head_only, Opts) of 
                     false ->
-                        {false, true, ?SST_PAGECACHELEVEL_LOOKUP};
+                        {false, true, PageCacheLevel};
                     with_lookup ->
-                        {true, true, ?SST_PAGECACHELEVEL_LOOKUP};
+                        {true, true, PageCacheLevel};
                     no_lookup ->
                         {true, false, ?SST_PAGECACHELEVEL_NOLOOKUP}
                 end,