greg/machi
1
0
Fork 0
Code Issues 11 Pull requests 1 Projects Releases Packages Wiki Activity Actions

Allow racing trim/fill and read-repair to simply "win".

Browse source 
This exposes a bug in the PULSE model, now that we can have multiple
successful fill/trim for the same LPN.
This commit is contained in:
Scott Lystig Fritchie 2014-02-20 19:19:16 +09:00
parent 5edee3a2cf
commit 431827f65e
1 changed files with 16 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

17
prototype/corfurl/src/corfurl.erl
View file

@ -213,7 +213,22 @@ fill_or_trim_page([], _Epoch, _LPN, _Func) ->
ok; ok;
fill_or_trim_page([H|T], Epoch, LPN, Func) -> fill_or_trim_page([H|T], Epoch, LPN, Func) ->
case corfurl_flu:Func(flu_pid(H), Epoch, LPN) of case corfurl_flu:Func(flu_pid(H), Epoch, LPN) of
ok -> Res when Res == ok; Res == error_trimmed ->
%% Detecting a race here between fills and trims is too crazy,
%% and I don't believe that it *matters*. The ickiest one
%% is a race between Proc A = trim and Proc B = read,
%% chain length of 2 or more:
%% Proc A: trim head -> ok
%% Proc B: read tail -> error_unwritten
%% Proc B: read head -> error_trimmed
%% Proc B: trim tail -> ok
%% Proc A: trim tail -> ??
%%
%% The result that we want that both A & B & any later
%% readers agree that the LPN is trimmed. If the chain is
%% >2, then the procs can win some/all/none of the races
%% to fix up the chain, that's no problem. But don't tell
%% the caller that there was an error during those races.
fill_or_trim_page(T, Epoch, LPN, Func); fill_or_trim_page(T, Epoch, LPN, Func);
Else -> Else ->
%% TODO: worth doing anything here, if we're in the middle of chain? %% TODO: worth doing anything here, if we're in the middle of chain?