About Consul ~In Depth~

#hbstudy 60 slides #consul

Transcript

1. ABOUT CONSUL 7/20/2014 @ijin ~In Depth~ hbstudy #60
2. ABOUT ME (@IJIN) • Michael H. Oshita • Japanese American • 10+ Years in Japan • Software/Infra/Cloud Engineer • http://ijin.github.io
3. ABOUT ME (@IJIN) τϥϒϧˑ͠Ύʔͨʔͣ
4. τϥϒϧˑ͠Ύʔͨʔͣ #08 http://bit.ly/torash08
6. CONSUL
7. CONSUL HTTP APIͱDNSͰૢ࡞͕Ͱ͖Δ෼ࢄܕΫϥελ
8. SERVICE DISCOVERY • ؆୯ʹserviceΛొ࿥͠ɺࣗಈͰΫϥελ্Ͱݕग़Մೳ • DNS·ͨ͸HTTPʹΑͬͯࢀর
9. FAILURE DETECTION • ϔϧενΣοΫػೳ • ࢮΜͩϊʔυ΁ͷϦΫΤετᷖճ !
10. MULTI DATACENTER • ෳ਺ͷσʔληϯλʔΛ·͍ͨͰ௨৴ • ଞͷDC্Ͱͷserviceͷࢀর
11. KEY/VALUE STORE • ಈతͳઃఆ৘ใͷஔ͖৔ॴͱͯ͠ • Long PollingͰઃఆมߋͷଈ࣌௨஌
12. TECHNOLOGY • Consensus Protocol (Raft, FSM) In Search of an Understandable Consensus Algorithm - Stanford University • Gossip Protocol (Serf) SWIM: Scalable Weakly-consistent Infection-style Process Group Membership - Cornel University • Sessions (Chuby) The Chubby Lock Service for Loosely-Coupled Distributed Systems - Google • TLS, Jespen Testing !
13. rpc forwarding & gossip protocol server server (leader) server ARCHITECTURE client client client
14. rpc forwarding & gossip protocol server server (leader) server ARCHITECTURE client client client peer set
15. rpc forwarding & gossip protocol server server (leader) server ARCHITECTURE client client client gossip
16. rpc forwarding & gossip protocol server server (leader) server ARCHITECTURE client client client query forward query query forward
17. ARCHITECTURE
18. CONSISTENCY
19. consensus protocol using raft server server (leader) server CONSISTENCY server server
20. consensus protocol using raft server server (leader) server write CONSISTENCY server server
21. consensus protocol using raft server server (leader) server write CONSISTENCY server server write
22. consensus protocol using raft server server (leader) server write CONSISTENCY server server write write
23. consensus protocol using raft server server (leader) server write CONSISTENCY server server write write quorum[(n/2)+1] reached, commited!
24. CONSISTENCY writes: ! "?state" or "?consistent" parameter ɾdefault ɾstrongly consistent ɾstale ! consensus protocol using raft reads: ! Always Strongly Consistent
25. CONSISTENCY default "?state" or "?consistent" parameter ɾMostly Strongly Consistent ɾStale reads possible when leader promotion ɾFast ! server server (old leader) server (new leader) A' A leader lease leader lease
26. CONSISTENCY consistent "?state" or "?consistent" parameter ɾStrongly Consistent ɾLeader validates via Qurum ɾLatency affected by Round-trip ! server server (leader) server A A A leader lease
27. CONSISTENCY stale "?state" or "?consistent" parameter ɾStale Reads, within ~50ms ɾNon-Leader Servers can answer ɾVery Fast ! server server (leader) server A A' A''
28. DEPLOYMENT TABLE Servers Quorum Size Failure Tolerance 1 1 0 2 2 0 3 2 1 4 3 1 5 3 2 6 4 2 7 4 3
29. JEPSEN TESTING Partition tolerance https://github.com/aphyr/jepsen http://www.infoq.com/articles/jepsen http://blog.foundationdb.com/call-me-maybe-foundationdb-vs-jepsen
30. CONSUL AGENT Nodes & Services foo bar baz redis 01.web 02.web node service
31. CONSUL AGENT • DNS Interface • HTTP API • Health Check !
32. CONSUL AGENT DNS Interface <node>.node.<datacenter>.<domain> Node Lookup:
33. CONSUL AGENT DNS Interface
34. CONSUL AGENT DNS Interface <tag>.<service>.service<datacenter>.<domain> Service Lookup:
35. CONSUL AGENT DNS Interface
36. CONSUL AGENT DNS Interface
37. CONSUL AGENT DNS CACHING Stale Reads ɾࢀর͸σϑΥϧτͰLeader NodeͷΈ ɾ༗ޮԽ͢ΔͱଞͷServer Node͔ΒࢀরՄೳ ɾεϧʔϓοτ޲্ ɾmax_staleͰ࣌ؒΛઃఆ ! TTL ɾDNS TTL͸σϑΥϧτͰ0 ɾdnsmasq౳ͰΩϟογϡՄೳʹ { "dns_config": { "service_ttl": { "*": "5s", "web": "30s" } } new! (0.3.0)
38. CONSUL AGENT HTTP API /v1/kv/<key> key/value endpoint: RESTful: GET, PUT, DELETE
39. CONSUL AGENT HTTP API several endpoint categories: • kv - Key/Value store • agent - Agent control • catalog - Manages nodes and services • health - Manages health checks • session - Session manipulation • status - Consul system status • internal - Internal APIs. Purposely
40. CONSUL AGENT HTTP API key/value: base64 raw
41. CONSUL AGENT HTTP API /v1/agent/checks /v1/agent/services /v1/agent/members /v1/agent/self /v1/agent/join/<address> /v1/agent/force-leave/<node> /v1/agent/service/deregister/<serviceID> agent endpoint: /v1/agent/register /v1/agent/deregister/<checkID> /v1/agent/pass/<checkID> /v1/agent/warn/<checkID> /v1/agent/fail/<checkID> /v1/agent/service/register
42. CONSUL AGENT HTTP API /v1/catalog/register /v1/catalog/deregister /v1/catalog/datacenters /v1/catalog/nodes catalog endpoint: /v1/catalog/services /v1/catalog/service/<service> /v1/catalog/node/<node> * supports blocking queries & consistency modes
43. CONSUL AGENT HTTP API /v1/health/node/<node> /v1/health/checks/<service> /v1/health/service/<service> /v1/state/<state> health endpoint: * supports blocking queries & consistency modes
44. CONSUL AGENT HEALTH CHECK { "service": { "check": { "script": "curl localhost > /dev/null 2>&1", "interval": "10s" } } check definition:
45. CONSUL AGENT HEALTH CHECK Exit Code 0 - Check is passing ! Exit Code 1 - Check is warning ! Any Other - Check is failing check script:
46. CONSUL AGENT HEALTH CHECK
47. BLOCKING QUERIES ɾLong PollingΛ࢖ͬͯঢ়ଶมߋΛଈ࣌ݕ஌ ! ɾԠ౴͸λΠϜΞ΢τ͔࣌ҟৗൃੜ࣌ ! ?wait=<interval>&index=<idx>
48. WEB UI
49. USE CASES ɾInternal DNS ! ɾInventory Management ! ɾFailover
50. DEMO
51. BENCHMARK GET 8 CPU Cores, 2Ghz 16GB RAM 160GB SSD disk 1Gbps NIC Requests/sec rps 0 2750 5500 8250 11000 10,469.54 2,115.61 v0.2.0 v0.3.0
52. BENCHMARK GET response time distribution # of responses 0.00 3000.00 6000.00 9000.00 12000.00 response time (s) 0.001 0.004 0.010 0.014 .018 0.023 0.035 0.051 0.067 0.083 v0.2.0 v0.3.0 8 CPU Cores, 2Ghz 16GB RAM 160GB SSD disk 1Gbps NIC
53. BENCHMARK GET 8 CPU Cores, 2Ghz 16GB RAM 160GB SSD disk 1Gbps NIC Requests/sec default consistent stale 0 2750 5500 8250 11000 10,948.2 10,245.562 10,469.54 1,986.771 1,953.555 2,115.61 v0.2.0 v0.3.0
54. BENCHMARK PUT 8 CPU Cores, 2Ghz 16GB RAM 160GB SSD disk 1Gbps NIC Requests/sec rps 0 1250 2500 3750 5000 4,091.67 1,054.031 v0.2.0 v0.3.0
55. BENCHMARK PUT response time distribution # of responses 0.00 2250.00 4500.00 6750.00 9000.00 response time (s) 0.006 0.013 0.017 0.024 0.031 0.038 0.047 0.078 0.109 0.140 0.172 v0.2.0 v0.3.0 8 CPU Cores, 2Ghz 16GB RAM 160GB SSD disk 1Gbps NIC
56. CONSUL-API Library in GO https://github.com/armon/consul-api
57. IN CLOSING consulΠΠΑʂ