Riak PG: Distributed Process Groups on Dynamo-style Distributed Storage

Erlang Workshop '13

Transcript

1. Riak PG Distributed Process Groups on Dynamo-style Distributed Storage Christopher Meiklejohn Basho Technologies, Inc. Erlang Workshop ’13 Wednesday, October 2, 13
2. The Goal Wednesday, October 2, 13
3. Build a highly-available, fault-tolerant registry. The Goal Wednesday, October 2, 13
4. Understand the tradeo!s. The Goal Wednesday, October 2, 13
5. The Problem Wednesday, October 2, 13
6. Highly-available distributed process groups. The Problem Wednesday, October 2, 13
7. Examples: pg2, gproc The Problem Wednesday, October 2, 13
8. Reappearing groups; synchronous global writes. The pg2 Problem Wednesday, October 2, 13
9. Election deadlocks; con"icts; dynamic clusters. The gproc Problem Wednesday, October 2, 13
10. Wednesday, October 2, 13
11. A more serious problem with using gen_leader is that it requires advance knowledge of all candidate nodes. “Extended Process Registry for Erlang”, Ulf T. Wiger, Erlang Workshop ’07 Wednesday, October 2, 13
12. Perhaps even more serious is gen_leader’s lack of support for dynamically recon#gured networks, and for de-con"icting the states of two leaders (which is presumably the most di$cult part of adding nodes on the "y). “Extended Process Registry for Erlang”, Ulf T. Wiger, Erlang Workshop ’07 Wednesday, October 2, 13
13. The Challenges Wednesday, October 2, 13
14. Dynamic addition and removal of nodes. The 3 Challenges Wednesday, October 2, 13
15. Coordination of state mutation. The 3 Challenges Wednesday, October 2, 13
16. Resolution of con"icting values. The 3 Challenges Wednesday, October 2, 13
17. Riak PG Wednesday, October 2, 13
18. Dynamic membership through virtual nodes. Riak PG Wednesday, October 2, 13
19. Replicated state; quorum reads and writes. Riak PG Wednesday, October 2, 13
20. Con"ict-free resolution with CRDTs. Riak PG Wednesday, October 2, 13
21. Eventually consistent; harvest vs. yield tradeo!. Riak PG Wednesday, October 2, 13
22. Eventual consistency is a consistency model used in distributed computing that informally guarantees that, if no new updates are made to a given data item, eventually all accesses to that item will return the last updated value. “Eventual Consistency”, Wikipedia Wednesday, October 2, 13
23. Our approaches tolerate partial failures by emphasizing simple composition mechanisms that promote fault containment, and by translating possible partial failure modes into engineering mechanisms that provide smoothly degrading functionality rather than lack of availability of the service as a whole. “Harvest, Yield, and Scalable Tolerant Systems”, Fox and Brewer Wednesday, October 2, 13
24. The Requirements Wednesday, October 2, 13
25. Structured names. The Requirements Wednesday, October 2, 13
26. Multiple non-unique names per process. The Requirements Wednesday, October 2, 13
27. Dynamic cluster membership. The Requirements Wednesday, October 2, 13
28. Partition tolerance and con"ict resolution. The Requirements Wednesday, October 2, 13
29. The Applications Wednesday, October 2, 13
30. Service lookup pattern; publish and subscribe. The Applications Wednesday, October 2, 13
31. Trade consistency for availability. The Applications Wednesday, October 2, 13
32. Riak Core; CRDTs The Background Wednesday, October 2, 13
33. Riak Core Wednesday, October 2, 13
34. Erlang implementation of Dynamo. Riak Core Wednesday, October 2, 13
35. Consistent hashing. Riak Core Wednesday, October 2, 13
36. Hash-space partitioning. Riak Core Wednesday, October 2, 13
37. Dynamic membership. Riak Core Wednesday, October 2, 13
38. Replication factor. Riak Core Wednesday, October 2, 13
39. Observed-Removed Set Wednesday, October 2, 13
40. CvRDT; bounded join-semilattice. Observed-Removed Set Wednesday, October 2, 13
41. Set; with merge function computing a LUB. Observed-Removed Set Wednesday, October 2, 13
42. Two G-Sets; preserves monotonicity. Observed-Removed Set Wednesday, October 2, 13
43. [ [{1, a}], [] ] [ [{1, a}], [] ] Wednesday, October 2, 13
44. [ [{1, a}], [] ] [ [{1, a}], [] ] [ [{1, a}, {2, b}], [] ] Wednesday, October 2, 13
45. [ [{1, a}], [] ] [ [{1, a}], [] ] [ [{1, a}, {2, b}], [] ] [ [{1, a}], [{1, a}] ] Wednesday, October 2, 13
46. [ [{1, a}], [] ] [ [{1, a}], [] ] [ [{1, a}, {2, b}], [] ] [ [{1, a}], [{1, a}] ] [ [{1, a}, {2, b}], [{1, a}] ] Wednesday, October 2, 13
47. [ [{1, a}], [] ] [ [{1, a}], [] ] Wednesday, October 2, 13
48. [ [{1, a}], [] ] [ [{1, a}], [] ] [ [{1, a}], [{1, a}] ] [ [{1, a}], [{1, a}] ] Wednesday, October 2, 13
49. [ [{1, a}], [] ] [ [{1, a}], [] ] [ [{1, a}], [{1, a}] ] [ [{1, a}], [{1, a}] ] [ [{1, a}, {2, a}], [{1, a}] ] Wednesday, October 2, 13
50. [ [{1, a}], [] ] [ [{1, a}], [] ] [ [{1, a}], [{1, a}] ] [ [{1, a}], [{1, a}] ] [ [{1, a}, {2, a}], [{1, a}] ] [ [{1, a}, {2, a}], [{1, a}] ] Wednesday, October 2, 13
51. The Implementation Wednesday, October 2, 13
52. Same as pg2; create, join, leave, and members. The Implementation Wednesday, October 2, 13
53. Extended with connected members. The Implementation Wednesday, October 2, 13
54. Membership vnode stores registrations. The Implementation Wednesday, October 2, 13
55. Con"ict-free resolution with OR-set. The Implementation Wednesday, October 2, 13
56. Process pruning; lack of monitors. The Implementation Wednesday, October 2, 13
57. Code examples. The Implementation Wednesday, October 2, 13
58. The Virtual Node Wednesday, October 2, 13
59. %% @doc Respond to a join request. handle_command({join, {ReqId, _}, Group, Pid}, _Sender, #state{groups=Groups0, partition=Partition}=State) -> %% Find existing list of Pids, and add object to it. Pids0 = pids(Groups0, Group, riak_dt_vvorset:new()), Pids = riak_dt_vvorset:update({add, Pid}, Partition, Pids0), %% Store back into the dict. Groups = dict:store(Group, Pids, Groups0), %% Return updated groups. {reply, {ok, ReqId}, State#state{groups=Groups}}; %% @doc Return pids from the dict. -spec pids(dict(), atom(), term()) -> term(). pids(Groups, Group, Default) -> case dict:find(Group, Groups) of {ok, Object} -> Object; _ -> Default end. riak_pg/src/riak_pg_memberships_vnode.erl Wednesday, October 2, 13
60. %% @doc Respond to a leave request. handle_command({leave, {ReqId, _}, Group, Pid}, _Sender, #state{groups=Groups0, partition=Partition}=State) -> %% Find existing list of Pids, and remove object from it. Pids0 = pids(Groups0, Group, riak_dt_vvorset:new()), Pids = riak_dt_vvorset:update({remove, Pid}, Partition, Pids0), %% Store back into the dict. Groups = dict:store(Group, Pids, Groups0), %% Return updated groups. {reply, {ok, ReqId}, State#state{groups=Groups}}; %% @doc Return pids from the dict. -spec pids(dict(), atom(), term()) -> term(). pids(Groups, Group, Default) -> case dict:find(Group, Groups) of {ok, Object} -> Object; _ -> Default end. riak_pg/src/riak_pg_memberships_vnode.erl Wednesday, October 2, 13
61. The Write Coordinator Wednesday, October 2, 13
62. %% @doc Execute the request. execute(timeout, #state{preflist=Preflist, req_id=ReqId, coordinator=Coordinator, group=Group, pid=Pid}=State) -> riak_pg_memberships_vnode:join(Preflist, {ReqId, Coordinator}, Group, Pid), {next_state, waiting, State}. %% @doc Attempt to write to every single node responsible for this %% group. waiting({ok, ReqId}, #state{responses=Responses0, from=From}=State0) -> Responses = Responses0 + 1, State = State0#state{responses=Responses}, case Responses =:= ?W of true -> From ! {ReqId, ok}, {stop, normal, State}; false -> {next_state, waiting, State} end. riak_pg/src/riak_pg_memberships_vnode.erl Wednesday, October 2, 13
63. The Read Coordinator Wednesday, October 2, 13
64. %% @doc Pull a unique list of memberships from replicas, and %% relay the message to it. waiting({ok, _ReqId, IndexNode, Reply}, #state{from=From, req_id=ReqId, num_responses=NumResponses0, replies=Replies0}=State0) -> NumResponses = NumResponses0 + 1, Replies = [{IndexNode, Reply}|Replies0], State = State0#state{num_responses=NumResponses, replies=Replies}, case NumResponses =:= ?R of true -> Pids = riak_dt_vvorset:value(merge(Replies)), From ! {ReqId, ok, Pids}, case NumResponses =:= ?N of true -> {next_state, finalize, State, 0}; false -> {next_state, waiting_n, State} end; false -> {next_state, waiting, State} end. riak_pg/src/riak_pg_members_fsm.erl Wednesday, October 2, 13
65. %% @doc Perform merge of replicas. merge(Replies) -> lists:foldl(fun({_, Pids}, Acc) -> riak_dt_vvorset:merge(Pids, Acc) end, riak_dt_vvorset:new(), Replies). riak_pg/src/riak_pg_members_fsm.erl Wednesday, October 2, 13
66. %% @doc Wait for the remainder of responses from replicas. waiting_n({ok, _ReqId, IndexNode, Reply}, #state{num_responses=NumResponses0, replies=Replies0}=State0) -> NumResponses = NumResponses0 + 1, Replies = [{IndexNode, Reply}|Replies0], State = State0#state{num_responses=NumResponses, replies=Replies}, case NumResponses =:= ?N of true -> {next_state, finalize, State, 0}; false -> {next_state, waiting_n, State} end. riak_pg/src/riak_pg_members_fsm.erl Wednesday, October 2, 13
67. %% @doc Perform read repair. finalize(timeout, #state{replies=Replies}=State) -> Merged = merge(Replies), Pruned = prune(Merged), ok = repair(Replies, State#state{pids=Pruned}), {stop, normal, State}. riak_pg/src/riak_pg_members_fsm.erl Wednesday, October 2, 13
68. %% @doc If the node is connected, and the process is not alive, prune %% it. prune_pid(Pid) when is_pid(Pid) -> lists:member(node(Pid), nodes()) andalso (is_process_alive(node(Pid), Pid) =:= false). %% @doc Remote call to determine if process is alive or not; assume if %% the node fails communication it is, since we have no proof it %% is not. is_process_alive(Node, Pid) -> case rpc:call(Node, erlang, is_process_alive, [Pid]) of {badrpc, _} -> true; Value -> Value end. %% @doc Based on connected nodes, prune out processes that no longer %% exist. prune(Set) -> Pids0 = riak_dt_vvorset:value(Set), lists:foldl(fun(Pid, Pids) -> case prune_pid(Pid) of true -> riak_dt_vvorset:update({remove, Pid}, none, Pids); false -> Pids end end, Set, Pids0). riak_pg/src/riak_pg_members_fsm.erl Wednesday, October 2, 13
69. The Evaluation Wednesday, October 2, 13
70. Partitions heal without con"icts. The Evaluation Wednesday, October 2, 13
71. Howl; CloudI Process Groups; Riak Pipe The Related Work Wednesday, October 2, 13
72. The Future Work Wednesday, October 2, 13
73. CRDT garbage collection. The Future Work Wednesday, October 2, 13
74. Active anti-entropy mechanism. The Future Work Wednesday, October 2, 13
75. The Conclusion Wednesday, October 2, 13
76. http://github.com/cmeiklejohn/riak_pg Thanks! Questions? Wednesday, October 2, 13