Tarantool как платформа для микросервисов / Антон Резников, Владимир Перепелица (Mail.Ru)
3 likes3,281 views
The document discusses Tarantool as a microservices platform, highlighting its performance capabilities, including handling over 100k requests per second per CPU core. It outlines various features such as SQL support, smart caching, complex queues, and its ability to handle OAuth tokens and user requests efficiently. Additionally, it emphasizes the importance of accessing data quickly and computing close to where the data is stored for optimal performance.
![8 lines
Lua: join by hand
function custom_join( key )
local result = {}
for tuple in space:iterator('EQ', key) do
local joined = other:select(0,tuple[1])
table.insert(result,{ tuple, joined })
end
return result
end](https://image.slidesharecdn.com/9-160115111534/85/Tarantool-Mail-Ru-29-320.jpg)
![20 lines!
Aggregation: example
function select(key)
local tuple = cache:select {key}
if not tuple then
if locks[key] then
local ch = fiber.channel(1)
locks[key][ch] = ch
return ch:get()
else
locks[key] = {}
local ret = sql:execute('select * from ...')
for ch in pairs(locks[key]) do
ch:put(ret)
end
locks[key] = nil
return ret
end
else
return tuple
end
end](https://image.slidesharecdn.com/9-160115111534/85/Tarantool-Mail-Ru-42-320.jpg)
Tarantool как платформа для микросервисов / Антон Резников, Владимир Перепелица (Mail.Ru)
- 1. Tarantool as a microservices platform Anton Reznikov Vladimir Perepelitsa
- 2. 100k+ RPS per cpu core
- 3. $1 000 000+ comparing to big cluster
- 4. What else…
- 5. SQL?
- 6. Memcached?
- 7. Queues?
- 8. $$$
- 9. Roadmap
- 11. Roadmap KV with expiration Smart caching proxy
- 12. Roadmap KV with expiration Smart caching proxy Complex queues
- 13. Roadmap KV with expiration Smart caching proxy Complex queues Application server
- 14. Roadmap KV with expiration Smart caching proxy Complex queues Application server Custom cluster solutions
- 15. Get your data in RAM Compute close to data Enjoy the performance
- 16. Average performance Operation Req / s / CPU Select: 150k Insert: 75k Update: 75k Delete: 90k LuaCall: 75k
- 17. Indexes: HASH
- 18. Indexes: HASH TREE: SG, AVL (1.5) | B+* (1.6)
- 19. Indexes: HASH TREE: SG, AVL (1.5) | B+* (1.6) RTREE (1.6)
- 20. Indexes: HASH TREE: SG, AVL (1.5) | B+* (1.6) RTREE (1.6) BITMAP
- 21. Indexes: HASH TREE: SG, AVL (1.5) | B+* (1.6) RTREE (1.6) BITMAP Composite
- 22. Indexes: HASH TREE: SG, AVL (1.5) | B+* (1.6) RTREE (1.6) BITMAP Composite Multiple
- 23. Lua
- 24. Luajit!
- 25. Luajit! fibers (green threads) Cooperative multitasking
- 26. Luajit! fibers (green threads) Cooperative multitasking channels (fiber communicaion) Synchronization primitives
- 27. Luajit! fibers (green threads) Cooperative multitasking channels (fiber communicaion) Synchronization primitives sockets (client & server) Pure raw, like in C and with helpers
- 28. Luajit! fibers (green threads) Cooperative multitasking channels (fiber communicaion) Synchronization primitives sockets (client & server) Pure raw, like in C and with helpers FFI (foreign function interface) Ability to call C/C++ (dlopen)
- 29. 8 lines Lua: join by hand function custom_join( key ) local result = {} for tuple in space:iterator('EQ', key) do local joined = other:select(0,tuple[1]) table.insert(result,{ tuple, joined }) end return result end
- 30. Access tokens Provides access by URL to specific feature without additional authorization Durable Temporary Consistent
- 31. Token Payload
- 32. Token Payload
- 35. Classic SQL App Web application with 20k requests for select and 2k requests for update Have memcached Have SQL replica
- 41. chanscache query 1 query SQL query 2 reply 2 chan SQL responsereply 1 query 3 reply 3
- 42. 20 lines! Aggregation: example function select(key) local tuple = cache:select {key} if not tuple then if locks[key] then local ch = fiber.channel(1) locks[key][ch] = ch return ch:get() else locks[key] = {} local ret = sql:execute('select * from ...') for ch in pairs(locks[key]) do ch:put(ret) end locks[key] = nil return ret end else return tuple end end
- 43. OAuth tokens Storage for 10M OAuth access+refresh tokens with deadline refreshing Keep minimal expired tokens Balance load over time
- 47. Queue priority to refresh no need to refreshexpired 60s first order old age second order 5 min third order
- 49. Subscription verification Mobile application provide subscription id, that must be verified in store Don't duplicate requests Reply as soon as possible Provide service if client gone
- 50. App Worker wait create task wait have task x3 done task id Apple
- 51. Database application Replicated database with a lot of business logic inside Multiple APIs Failsafe deploy Monitoring Informative logs
- 52. App
- 53. API v1 App Lua
- 56. Sharding is unavoidable Data does not fit in RAM of a single instance or many CPU cores required Database as a service Consistency checking
- 57. #1 #2 #3 App Sh prx f(k) App App Sh prx f(k) Sh prx f(k)
- 58. #1 #2 #3 App Sh prx f1(k) App App Sh prx f(k) Sh prx f(k)
- 59. #1 #2 ?? #3 App Sh prx f(k) App App Sh prx f(k) Sh prx f(k)
- 61. Access tokens (v2) Provides access by URL to specific feature without additional authorization High availability Allow loss of majority
- 65. Get your data in RAM Compute close to data Enjoy the performance
- 66. tarantool.orgDocumentation: try.tarantool.orgTry it online: github.com/tarantoolExplore it: github.com/mons/tnt-luasLuas: Questions?