In banking system demands are very tight. Database should be at least tripled, stand-by nodes should pick up master reads from failover node, writes should be accepted on a reasonble quorum, failover must be followed by recovery, database should be able to scale even with the RAM/DISC limitations.
No data should be treated as written otherwise that commited to all replicas. All this circumstances leads us to chain replication protocol as a simple and natural feedback to this challenge.
Different replication techniques exists to satisfy replication demands. Master-slave replication is most widely known type of replication used before in such products like GFS, HDFS, mongodb, etc. Quorum Intersection is another technique used in databases like Cassandra or Amazon Dynamo. They mostly provide a consistent distributed repository for event tables or for file storage. In banking industry we synchronize account balances and need simple and managable protocol for storage consistency issuing high demand on system integrity.
There are several classes of error usually implied when dealing with failure detection. The most weak class is fail-stop events, when the outage is normal or predictable. The second class is crash-failures, the ubnormal terminations and outages. The most strong type of failures are byzantine failures resistant to bit-flips, hacked parties or any types of compromising the transaction objects. For banking applications the byzantine fault tolerance is desired, despite it affects the latency.
- Highly-available CP database :-)
- 2N+1 nodes tolerates N failures
- Consistent hashing DHT
- RAFT for managing server configurations timeline
- HMAC signing for Byzantine capabilities
- Various database backends: mnesia, riak, redis, fs, sql
- High-performance non-blocking TCP acceptor
- Separate endpoints for HEART, CLIENT and SERVER protocols
- Pure, clean and understandable codebase
- Article about CR implementation details: https://synrc.com/apps/cr/doc/cr.htm
- Business Processing Erlang book: https://synrc.com/apps/bpe/doc/book.pdf
make console NAME=cr
make console NAME=cr2
make console NAME=cr3
You could start all nodes in separate console sesions or you
can make start NAME=cr2
nodes and later attach to them with make attach NAME=cr2
.
Also the start is compatible within single folders, which cause no single problem.
> timer:tc(cr,test,[500]).
=INFO REPORT==== 7-Apr-2015::00:56:34 ===
cr:Already in Database: 14020
New record will be applied: 500
{214369,{transactions,11510}}
Fore generating sample data, let say 500 transactions you may run with cr:test(500)
.
By measuring accepring performance it's like 2000 Req/s
.
> cr:dump().
vnode i n top latency
121791803110908576516973736059690251637994378581 1 1 391 2/198/64
243583606221817153033947472119380503275988757162 2 1 400 2/183/72
365375409332725729550921208179070754913983135743 3 1 388 3/195/64
487167212443634306067894944238761006551977514324 4 1 357 2/183/53
608959015554542882584868680298451258189971892905 5 2 12994 2/198/67
730750818665451459101842416358141509827966271486 6 2 13017 3/184/66
852542621776360035618816152417831761465960650067 7 2 13019 2/201/75
974334424887268612135789888477522013103955028648 8 2 13020 3/178/62
1096126227998177188652763624537212264741949407229 9 3 13021 2/190/68
1217918031109085765169737360596902516379943785810 10 3 13028 3/206/65
1339709834219994341686711096656592768017938164391 11 3 13030 2/208/55
1461501637330902918203684832716283019655932542972 12 3 13031 2/185/58
ok
The latency in last column ~70 ms
means the moment data is stored on all mnesia
replicas.
The latency in a given example is for storing async_dirty using KVS
chain linking (from 1 to 3
msg per write operation, from 1 to 2
msg for lookups)
clustered in 3 nodes
with same replicas number.
Let's say we want to see all the operations log of a given replica 391
.
> cr:dump(391).
operation id prev i size
transaction:389:feed::false: 391 387 1 480
transaction:399:feed::false: 387 382 1 500
transaction:375:feed::false: 382 379 1 446
transaction:373:feed::false: 379 378 1 446
transaction:383:feed::false: 378 376 1 473
transaction:392:feed::false: 376 374 1 500
transaction:360:feed::false: 374 371 1 446
transaction:366:feed::false: 371 370 1 473
transaction:370:feed::false: 370 369 1 446
transaction:371:feed::false: 369 368 1 446
ok
You may check this from the other side. First retrieve the operation and then retrieve the transaction created during operation.
> kvs:get(operation,391).
{ok,#operation{id = 391,version = undefined,container = log,
feed_id = {121791803110908576516973736059690251637994378581,1}, % VNODE
prev = 387,next = undefined,feeds = [],guard = false,
etc = undefined,
body = {prepare,{<0.41.0>,{1428,358105,840469}},
[{121791803110908576516973736059690251637994378581,1}, % SIGNATURES
{608959015554542882584868680298451258189971892905,2}],
#transaction{id = 389,version = undefined,container = feed,
feed_id = undefined,prev = undefined,next = undefined,
feeds = [],guard = false,etc = undefined,
timestamp = undefined,beneficiary = undefined,...}},
name = prepare,status = pending}}
The transaction. For linking transaction to the link you should use full XA protocol with two-stage confirmation (1) the PUT operation followed with (2) LINK operation to some feed, such as user account or customer admin list.
> kvs:get(transaction,389).
{ok,#transaction{id = 389,version = undefined,
container = feed, feed_id = undefined, prev = undefined,
next = undefined, feeds = [], guard = false, etc = undefined,
timestamp = [], beneficiary = [],
subsidiary = [], amount = [],tax = [],
ballance = [], currency = [],
description = [], info = [],
prevdate = [], rate = [], item = []}}
The actiual Erlang business logic, banking transaction from db
schema
application is stored under 389 id. So you can easlity grab it unlinked
as it was stored as atomic PUT.
- consensus protols 1) raft and 2) paxos are distributed under the terms of Apache 2.0 http://www.apache.org/licenses/LICENSE-2.0.html
- cr itself is distributed under the DHARMA license: http://5ht.github.io/license.htm
Copyright (c) 2015 Synrc Research Center s.r.o.
- Maxim Sokhatsky
- Vladimir Kirillov
- Sergey Klimenko
- Valery Maleshkin
- Victor Sovietov
OM A HUM