File 0830-Rewrite-RR-cache-access-time-update-to-pre... of Package erlang

Overview Repositories Revisions Requests Users Attributes Meta

File 0830-Rewrite-RR-cache-access-time-update-to-preserve-tabl.patch of Package erlang (Revision 9f7ba17abd41d405759f6f11f4d7ded6)

Currently displaying revision 9f7ba17abd41d405759f6f11f4d7ded6 , Show latest

From 49e73257bb19634518cc7606eb02df7f710d2e28 Mon Sep 17 00:00:00 2001
From: Raimo Niskanen <raimo@erlang.org>
Date: Mon, 21 Aug 2023 17:23:02 +0200
Subject: [PATCH] Rewrite RR cache access time update to preserve table insert
 order

The cache ETS table is a bag table that in itself preserves insert order,
and when access times in the table are update, it is possible to
do the insertion of RRs with updated access time in the same order,
which overall preserves insert order.

To preserve the RR order; rewrite the deduplication code to not store
the deduplicated result in a map, but instead use a map for deduplication
lookup and store the result in a list.

All this to not get in the way of DNS servers that e.g. implement load balancing
by returning RRS in some specific order.
---
 lib/kernel/src/inet_db.erl | 111 +++++++++++++++++++++++--------------
 1 file changed, 68 insertions(+), 43 deletions(-)

diff --git a/lib/kernel/src/inet_db.erl b/lib/kernel/src/inet_db.erl
index b7d5afcc64..6b4a0d4450 100644
--- a/lib/kernel/src/inet_db.erl
+++ b/lib/kernel/src/inet_db.erl
@@ -1022,7 +1022,7 @@ handle_call(Request, From, #state{db=Db}=State) ->
 
 	{add_rrs, RRs} ->
 	    ?dbg("add_rrs: ~p~n", [RRs]),
-	    {reply, do_add_rrs(RRs, Db, State), State};
+	    {reply, do_add_rrs(RRs, Db, State#state.cache), State};
 
 	{del_rr, RR} when is_record(RR, dns_rr) ->
 	    Cache = State#state.cache,
@@ -1667,14 +1667,11 @@ is_reqname(_) -> false.
 %% #dns_rr.cnt is used to store the access time
 %% instead of number of accesses.
 %%
-do_add_rrs(RRs, Db, State) ->
-    CacheDb = State#state.cache,
-    do_add_rrs(RRs, Db, State, CacheDb).
-
-do_add_rrs([], _Db, _State, _CacheDb) ->
+do_add_rrs([], _Db, _CacheDb) ->
     ok;
-do_add_rrs([RR | RRs], Db, State, CacheDb) ->
-    case alloc_entry(Db, CacheDb, #dns_rr.tm) of
+do_add_rrs([RR | RRs], Db, CacheDb) ->
+    Size = ets:lookup_element(Db, cache_size, 2),
+    case alloc_entry(CacheDb, #dns_rr.tm, Size) of
 	true ->
             %% Add to cache
             %%
@@ -1698,7 +1695,7 @@ do_add_rrs([RR | RRs], Db, State, CacheDb) ->
                             DelRR <- DeleteRRs],
                     ok
             end,
-            do_add_rrs(RRs, Db, State, CacheDb);
+            do_add_rrs(RRs, Db, CacheDb);
 	false ->
 	    ok
     end.
@@ -1745,47 +1742,76 @@ lookup_cache_data(LcDomain, Type) ->
 %% in the table, i.e identical domain, class, type and data.
 %% We embrace that and eliminate duplicates here.
 %%
-%% Look up all matching objects.  The still valid ones
-%% should be returned, and updated with a new cnt time.
-%% All expired ones should be deleted.
+%% Look up all matching objects.
+%% The still valid ones should be returned and updated
+%% in the ETS table with a new access time (#dns_rr.cnt).
+%% All expired ones should be deleted from the ETS table.
 %%
 match_rr(MatchRR) ->
     CacheDb = inet_cache,
     RRs = ets:match_object(CacheDb, MatchRR),
-    match_rr(CacheDb, RRs, times(), #{}, #{}, []).
+    match_rr(CacheDb, RRs, times(), [], []).
 %%
-match_rr(CacheDb, [], _Time, ResultRRs, InsertRRs, DeleteRRs) ->
-    %% We insert first so an RR always is present,
+match_rr(CacheDb, [], Time, KeepRRs, DeleteRRs) ->
+    %%
+    %% Keep the first duplicate RR in KeepRRs (reversed)
+    %% that is; the last in RRs
+    ResultRRs = match_rr_dedup(KeepRRs),
+    %%
+    %% We insert before delete so an RR always is present,
     %% which may create duplicates
-    _ = [ets:insert(CacheDb, RR) || RR <- maps:values(InsertRRs)],
+    _ = [ets:insert(CacheDb, RR#dns_rr{cnt = Time})
+         || RR <- ResultRRs,
+            %%
+            %% Insert only if access time changes
+            RR#dns_rr.cnt < Time],
     _ = [ets:delete_object(CacheDb, RR) || RR <- DeleteRRs],
-    maps:values(ResultRRs);
-match_rr(CacheDb, [RR | RRs], Time, ResultRRs, InsertRRs, DeleteRRs) ->
+    ResultRRs;
+%%
+%% Updating the access time (#dns_rr.cnt) is done by first inserting
+%% an updated RR and then deleting the old, both done above.
+%%
+%% This does not work if the access time for the inserted record
+%% is the same as for the deleted record because then both records
+%% are identical and we end up with the record being deleted
+%% instead of updated.
+%%
+%% When the access time is unchanged, within the time granularity,
+%% the RR should not be updated so it is not put on the delete list
+%% (below) and not re-inserted (above).  Both parts of this
+%% split operation has to use the same condition; RR#dns_rr.cnt < Time,
+%% for this to work.
+%%
+match_rr(CacheDb, [RR | RRs], Time, KeepRRs, DeleteRRs) ->
     %%
-    #dns_rr{ttl = TTL, tm = TM, cnt = Cnt} = RR,
+    #dns_rr{ttl = TTL, tm = TM} = RR,
     if
         TM + TTL < Time ->
-            %% Expired, delete
-            match_rr(
-              CacheDb, RRs, Time,
-              ResultRRs, InsertRRs, [RR | DeleteRRs]);
-        Time =< Cnt ->
-            %% Valid and just updated, return and do not update
-            Key = match_rr_key(RR),
-            match_rr(
-              CacheDb, RRs, Time,
-              ResultRRs#{Key => RR}, InsertRRs, DeleteRRs);
+            %% Expired
+            match_rr(CacheDb, RRs, Time, KeepRRs, [RR | DeleteRRs]);
+        RR#dns_rr.cnt < Time -> % Delete only if access time changes
+            %% Not expired
+            match_rr(CacheDb, RRs, Time, [RR | KeepRRs], [RR | DeleteRRs]);
+        true -> % Cnt == Time since Time is monotonically increasing
+            %% Not expired
+            match_rr(CacheDb, RRs, Time, [RR | KeepRRs], DeleteRRs)
+    end.
+
+%% Remove all duplicate RRs (according to match_rr_key/1)
+%% - keep the first, return reversed list
+%%
+match_rr_dedup(RRs) ->
+    match_rr_dedup(RRs, #{}, []).
+%%
+match_rr_dedup([], _Seen, Acc) ->
+    Acc;
+match_rr_dedup([RR | RRs], Seen, Acc) ->
+    Key = match_rr_key(RR),
+    case erlang:is_map_key(Key, Seen) of
         true ->
-            %% Valid; return and re-insert with updated cnt time.
-            %% The clause above ensures that the cnt field is changed
-            %% which is essential to not accidentally delete
-            %% a record we also insert.
-            Key = match_rr_key(RR),
-            match_rr(
-              CacheDb, RRs, Time,
-              ResultRRs#{Key => RR},
-              InsertRRs#{Key => RR#dns_rr{cnt = Time}},
-              [RR | DeleteRRs])
+            match_rr_dedup(RRs, Seen, Acc);
+        false ->
+            match_rr_dedup(RRs, Seen#{Key => []}, [RR | Acc])
     end.
 
 -compile({inline, [match_rr_key/1]}).
@@ -1927,8 +1953,7 @@ delete_expired(CacheDb, TM) ->
 %% Returns: true if space for a new entry otherwise false
 %% (true if we have a cache since we always make room for new).
 %% -------------------------------------------------------------------
-alloc_entry(Db, CacheDb, TM) ->
-    Size = ets:lookup_element(Db, cache_size, 2),
+alloc_entry(CacheDb, TM, Size) ->
     if
 	Size =< 0 ->
 	    false;
@@ -1947,11 +1972,11 @@ alloc_entry(Db, CacheDb, TM) ->
 %% This deletion should always give some room since
 %% it removes a percentage of the oldest entries.
 %%
-%% Fetch all cnt times, sort them, calculate a limit
+%% Fetch all access times (#dns_rr.cnt), sort them, calculate a limit
 %% as the earliest of the time 1/3 from the oldest to now,
 %% and the 1/10 oldest entry,.
 %%
-%% Delete all entries with a cnt time older than that,
+%% Delete all entries with an access time (#dns_rr.cnt) older than that,
 %% and all expired (tm + ttl < now).
 %%
 delete_oldest(CacheDb, TM, N) ->
-- 
2.35.3

Places

File 0830-Rewrite-RR-cache-access-time-update-to-preserve-tabl.patch of Package erlang (Revision 9f7ba17abd41d405759f6f11f4d7ded6)

Places