Github user jaydoane commented on a diff in the pull request:
https://github.com/apache/couchdb-couch/pull/185#discussion_r71279401
--- Diff: src/couch_file.erl ---
@@ -524,6 +525,99 @@ handle_info({'EXIT', _, Reason}, Fd) ->
{stop, Reason, Fd}.
+%% first try to read header at end, falling back to chunked reads on
failure
+find_last_header(Fd, FileSize) ->
+ ok = file:advise(Fd, 0, FileSize, dont_need), % minimize disk cache
pollution
+ Pos = (FileSize div ?SIZE_BLOCK) * ?SIZE_BLOCK,
+ case file:pread(Fd, Pos, FileSize - Pos) of
+ eof ->
+ no_valid_header;
+ {ok, Data} ->
+ try match_header(Fd, Pos, Data, 0) of
+ {ok, HeaderBin} ->
+ {ok, HeaderBin}
+ catch
+ error:{badmatch, _} ->
+ find_header_in_chunks(Fd, Pos, 1)
+ end
+ end.
+
+-define(DEFAULT_CHUNK_MAX_SIZE, ?SIZE_BLOCK).
+-define(DEFAULT_CHUNK_EXPONENT_BASE, 1).
+
+chunk_max_size() ->
+ config:get_integer("couchdb", "chunk_max_size",
?DEFAULT_CHUNK_MAX_SIZE).
+
+%% chunk size exponentially increases by iteration, up to some maximum,
+%% and should never exceed the current position in the file
+chunk_size(Pos, Multiplier) ->
+ Size = min(Multiplier * ?SIZE_BLOCK, chunk_max_size()),
+ min(Pos, Size).
+
+multiplier(Size, Multiplier) ->
+ case Size < chunk_max_size() of
+ true ->
+ Multiplier * config:get_integer(
+ "couchdb", "chunk_exponent_base",
?DEFAULT_CHUNK_EXPONENT_BASE);
+ false ->
+ Multiplier
+ end.
+
+find_header_in_chunks(_Fd, Pos, _Multiplier) when Pos < 0 ->
+ no_valid_header;
+find_header_in_chunks(Fd, Pos, Multiplier) ->
+ Size = chunk_size(Pos, Multiplier),
+ case Size > 0 of
+ false ->
+ no_valid_header;
+ true ->
+ {ok, Chunk} = file:pread(Fd, Pos - Size, Size),
+ case find_header_in_chunk(Fd, Pos, Chunk, Size) of
+ {ok, HeaderBin, _Offset} ->
+ %% io:format("found header at ~p multiplier ~p
chunksize ~p~n",
+ %% [Pos - Size + _Offset, Multiplier, Size]),
+ {ok, HeaderBin};
+ not_found ->
+ NewMultiplier = multiplier(Size, Multiplier),
+ find_header_in_chunks(Fd, Pos - Size, NewMultiplier)
+ end
+ end.
+
+find_header_in_chunk(_Fd, _Pos, _Chunk, Offset) when Offset < 0 ->
+ not_found;
+find_header_in_chunk(Fd, Pos, Chunk, Offset) ->
+ try match_header(Fd, Pos, Chunk, Offset) of
+ {ok, HeaderBin} ->
+ {ok, HeaderBin, Offset}
+ catch
+ error:{badmatch, _} ->
+ find_header_in_chunk(Fd, Pos, Chunk, Offset - ?SIZE_BLOCK)
+ end.
+
+read_size(Data, Offset) ->
+ min(size(Data), Offset + ?SIZE_BLOCK) - Offset.
+
+match_header(Fd, Pos, Data, Offset) ->
+ <<1, HeaderSize:32/integer-unsigned, RestBlock/binary>> =
+ binary:part(Data, Offset, read_size(Data, Offset)),
+ PrefixSize = 5, % 1 (version) + 4 (size) bytes
+ TotalSize = total_read_size(PrefixSize, HeaderSize),
+ HeaderPos = Pos - byte_size(Data) + Offset,
+ RawBin = raw_bin(Fd, HeaderPos, TotalSize, RestBlock),
+ <<Hash:16/binary, HeaderBin/binary>> =
+ iolist_to_binary(remove_block_prefixes(PrefixSize, RawBin)),
+ Hash = crypto:hash(md5, HeaderBin),
--- End diff --
`match_header` can also raise a badmatch on line 601, especially since only
header blocks start with a <<1>>. While I was tempted to avoid try ... catch
and just use case statements, it seemed to make the code significantly more
difficult to understand.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
