From b560527ee3d5605310f3793dbcbf86a9b4844c91 Mon Sep 17 00:00:00 2001 From: Christian Grothoff Date: Sun, 20 Mar 2022 16:52:38 +0100 Subject: [PATCH] ideas for mboss --- src/exchangedb/exchange-0001.sql | 34 ++++++++++++++++++++++++++++++++ 1 file changed, 34 insertions(+) diff --git a/src/exchangedb/exchange-0001.sql b/src/exchangedb/exchange-0001.sql index 08e6fa5f0..80ade238c 100644 --- a/src/exchangedb/exchange-0001.sql +++ b/src/exchangedb/exchange-0001.sql @@ -629,6 +629,40 @@ CREATE TABLE IF NOT EXISTS deposits ,UNIQUE (shard, coin_pub, merchant_pub, h_contract_terms) ) PARTITION BY HASH (shard); -- FIXME: why not BY RANGE? RANGE would seem better for 'deposits_get_ready'! +-- FIXME: +-- new idea: partition deposits by coin_pub (remove deposits_by_coin) +-- define 'ready' == ! (tiny || done || blocked) +-- add new deposits_by_ready (on shard + wire_deadline), select by shard, then ready + deadline +-- -- use triggers to ONLY include 'ready' deposits (delete on update)! +-- -- use multi-level partitions: Hash(shard) + Range(wire_deadline/sec) +-- add new deposits_by_match (on shard + refund_deadline) +-- -- use triggers to ONLY include 'ready' deposits (delete on update)! +-- -- use multi-level partitions: Hash(shard) + Range(refund_deadline/sec) +-- => first we select per-merchant shard, basically stay on the same system as other ops for the same merchant +-- => second we select by deadline, use enough values so that _usually_ the aggregator +-- and the 'insert' process _can_ work on different shards! +-- => the latter could be achieved by dynamically (!) creating/deleting partitions: +-- create new partitions 'as needed', drop old ones once the aggregator has made +-- them empty; as 'new' deposits will always have deadlines in the future, this +-- would basically guarantee no conflict between aggregator and exchange service! +-- SEE also: https://www.cybertec-postgresql.com/en/automatic-partition-creation-in-postgresql/ +-- (article is slightly wrong, as this works:) +--CREATE TABLE tab ( +-- id bigint GENERATED ALWAYS AS IDENTITY, +-- ts timestamp NOT NULL, +-- data text +-- PARTITION BY LIST ((ts::date)); +-- CREATE TABLE tab_def PARTITION OF tab DEFAULT; +-- BEGIN +-- CREATE TABLE tab_part2 (LIKE tab); +-- insert into tab_part2 (id,ts, data) values (5,'2022-03-21', 'foo'); +-- alter table tab attach partition tab_part2 for values in ('2022-03-21'); +-- commit; +-- Naturally, to ensure this is actually 100% conflict-free, we'd +-- need to create tables at the granularity of the wire/refund deadlines; +-- that is right now seconds (!). But I see no problem with changing the +-- aggregator to basically always run 1 minute behind and use minutes instead! + COMMENT ON TABLE deposits IS 'Deposits we have received and for which we need to make (aggregate) wire transfers (and manage refunds).';