Cassandra Summit Tokyo 2017 Keynote

by Nate McCall at Cassandra Summit Tokyo 2017

Transcript

1. APACHE CASSANDRA WHERE WE ARE NOW AND THE ROAD AHEAD
2. Nate McCall @zznate CTO, The Last Pickle Apache Cassandra Committer PMC Chair
3. AGENDA * Community * Apache Cassandra project status * 4.0 * The road ahead
4. BUT FIRST...
5. October 31st, 2010
8. LET'S TALK ABOUT COMMUNITY
9. PROJECTS WITH COMPANIES THAT PROVIDE DISTRIBUTIONS • Cloudera|Hortonworks|MapR: Hadoop • DataBricks: Spark • Confluent: Kafka • Data Artisans: Flink • IBM: *.* 9
10. EXAMPLE ASF VENDOR COMPATIBILITY MATRIX
11. UNFORTUNATELY, WE ARE STARTING TO SEE THIS TOO!
12. NO COMMUNITY SUPPORT IS A LONELY ROAD TO TRAVEL.
13. WHAT DOES THIS MEAN FOR OUR COMMUNITY? • Most committers and community participants are running clusters • Users on the mailing lists tend to have a lot of experience • No corporate message in project communication • No external pressures on developing features for marketing timelines • Events will be by the community for the community (like this one!) 13
14. https://twitter.com/jericevans/status/912765106739712001
15. COMMUNITY STRENGTHS • Very active mail lists • Progressively better JIRA grooming • Documentation improvements • Fast turn around on serious bugs 15
16. OTHER COMMUNITY HIGHLIGHTS Donation of build resources
17. https://builds.apache.org/computer/
18. OTHER COMMUNITY HIGHLIGHTS Contribution of cassandra-dtest project
19. Donated to the ASF
20. The takeaway: You don't need a vendor to run Apache Cassandra.
21. APACHE CASSANDRA PROJECT STATUS
22. (YEAH, WE ALL GOT CONFUSED) "TICK TOCK"
23. TICK TOC LEGACY: GOOD • Much larger emphasis on integration testing • Longer term support of existing releases (2.1.19 just released!) • Commitment to "green test board" before release
24. TICK TOC LEGACY: BAD 24 git am -3 mypatch-3.0.patch git merge cassandra-3.0 -s ours git apply -3 mypatch-3.11.patch git commit -amend git merge cassandra-3.11 -s ours git apply -3 mypatch-trunk.patch git commit -amend git push origin cassandra-3.0 cassandra-3.3 trunk -atomic Committing to 3.0, 3.11, and trunk:
25. TICK TOC LEGACY: BAD 25 git am -3 mypatch-3.0.patch git merge cassandra-3.0 -s ours git apply -3 mypatch-3.11.patch git commit -amend git merge cassandra-3.11 -s ours git apply -3 mypatch-trunk.patch git commit -amend git push origin cassandra-3.0 cassandra-3.3 trunk -atomic Committing to 3.0, 3.11, and trunk:
26. TICK TOC LEGACY: BAD 26 git am -3 mypatch-3.0.patch git merge cassandra-3.0 -s ours git apply -3 mypatch-3.11.patch git commit -amend git merge cassandra-3.11 -s ours git apply -3 mypatch-trunk.patch git commit -amend git push origin cassandra-3.0 cassandra-3.3 trunk -atomic Committing to 3.0, 3.11, and trunk:
27. GREAT. WHAT VERSION SHOULD I USE? 2.2 3.0 3.11
28. IT'S TIME FOR REAL TALK.
29. VERSION BREAKDOWN • 2.2: low risk, works well • 3.0: "stable" release of tic/toc • 3.11: the new stable (as of 3.11.1) 29
30. VERSION BREAKDOWN • 2.2: low risk, works well • 3.0: "stable" release of tic/toc • 3.11: the new stable (as of 3.11.1) 30
31. BUT WHAT ABOUT MATERIALIZED VIEWS? • *JUST* fixed https://issues.apache.org/jira/browse/CASSANDRA-11500 • Partition deletion causes entire view partition to be held in memory (CASSANDRA-12783) • Can only filter on primary key columns (CASSANDRA-13826) 31
32. MATERIALIZED VIEWS: THE HAPPY PATH IS VERY NARROW
33. NEW IN 4.0
34. NO MORE THRIFT :(
35. FASTER, BETTER, CHEAPER REPAIR
36. REPAIR BUG FIXES AND IMPROVEMENTS • Over 20 bugs fixed! • New metrics provide table and keyspace level visibility: CASSANDRA-13531, CASSANDRA-13598 • More repair information on nodetool tablestats: CASSANDRA-13774 • Incremental repair works more consistently • Substantial repair logging improvements: CASSANDRA-13468 36
37. REPAIR STREAMING PREVIEW 37 Total estimated streaming: 41384 ranges, 10 sstables, 159.007MiB bytes sandwich_long.test_data - 41384 ranges, 10 sstables, 159.007MiB bytes /127.0.0.2 -> /127.0.0.1: 10346 ranges, 3 sstables, 62.888MiB bytes /127.0.0.1 -> /127.0.0.2: 10346 ranges, 2 sstables, 16.616MiB bytes /127.0.0.1 -> /127.0.0.3: 10346 ranges, 2 sstables, 16.616MiB bytes /127.0.0.3 -> /127.0.0.1: 10346 ranges, 3 sstables, 62.887MiB bytes nodetool repair --preview
38. REPAIR STREAMING PREVIEW 38 Total estimated streaming: 41384 ranges, 10 sstables, 159.007MiB bytes sandwich_long.test_data - 41384 ranges, 10 sstables, 159.007MiB bytes /127.0.0.2 -> /127.0.0.1: 10346 ranges, 3 sstables, 62.888MiB bytes /127.0.0.1 -> /127.0.0.2: 10346 ranges, 2 sstables, 16.616MiB bytes /127.0.0.1 -> /127.0.0.3: 10346 ranges, 2 sstables, 16.616MiB bytes /127.0.0.3 -> /127.0.0.1: 10346 ranges, 3 sstables, 62.887MiB bytes nodetool repair --preview who what
39. PULL REPAIR 39 Total estimated streaming: 3550 ranges, 3 sstables, 10.525MiB bytes sandwich_long.test_data - 3550 ranges, 3 sstables, 10.525MiB bytes /127.0.0.2 -> /127.0.0.1: 1775 ranges, 3 sstables, 10.525MiB bytes /127.0.0.1 -> /127.0.0.2: 1775 ranges, 0 sstables, 0.000KiB bytes nodetool repair --pull -hosts 127.0.0.1,127.0.0.2 -st -3074457345618258603 -et 0 --preview
40. PULL REPAIR 40 Total estimated streaming: 3550 ranges, 3 sstables, 10.525MiB bytes sandwich_long.test_data - 3550 ranges, 3 sstables, 10.525MiB bytes /127.0.0.2 -> /127.0.0.1: 1775 ranges, 3 sstables, 10.525MiB bytes /127.0.0.1 -> /127.0.0.2: 1775 ranges, 0 sstables, 0.000KiB bytes nodetool repair --pull -hosts 127.0.0.1,127.0.0.2 -st -3074457345618258603 -et 0 --preview
41. PULL REPAIR 41 Total estimated streaming: 3550 ranges, 3 sstables, 10.525MiB bytes sandwich_long.test_data - 3550 ranges, 3 sstables, 10.525MiB bytes /127.0.0.2 -> /127.0.0.1: 1775 ranges, 3 sstables, 10.525MiB bytes /127.0.0.1 -> /127.0.0.2: 1775 ranges, 0 sstables, 0.000KiB bytes nodetool repair --pull -hosts 127.0.0.1,127.0.0.2 -st -3074457345618258603 -et 0 --preview
42. PULL REPAIR 42 Total estimated streaming: 3550 ranges, 3 sstables, 10.525MiB bytes sandwich_long.test_data - 3550 ranges, 3 sstables, 10.525MiB bytes /127.0.0.2 -> /127.0.0.1: 1775 ranges, 3 sstables, 10.525MiB bytes /127.0.0.1 -> /127.0.0.2: 1775 ranges, 0 sstables, 0.000KiB bytes nodetool repair --pull -hosts 127.0.0.1,127.0.0.2 -st -3074457345618258603 -et 0 --preview
43. PULL REPAIR 43 Total estimated streaming: 3550 ranges, 3 sstables, 10.525MiB bytes sandwich_long.test_data - 3550 ranges, 3 sstables, 10.525MiB bytes /127.0.0.2 -> /127.0.0.1: 1775 ranges, 3 sstables, 10.525MiB bytes /127.0.0.1 -> /127.0.0.2: 1775 ranges, 0 sstables, 0.000KiB bytes nodetool repair --pull -hosts 127.0.0.1,127.0.0.2 -st -3074457345618258603 -et 0 --preview who what
44. REPAIR STREAMING PREVIEW 44 Total estimated streaming: 41384 ranges, 10 sstables, 159.007MiB bytes sandwich_long.test_data - 41384 ranges, 10 sstables, 159.007MiB bytes /127.0.0.2 -> /127.0.0.1: 10346 ranges, 3 sstables, 62.888MiB bytes /127.0.0.1 -> /127.0.0.2: 10346 ranges, 2 sstables, 16.616MiB bytes /127.0.0.1 -> /127.0.0.3: 10346 ranges, 2 sstables, 16.616MiB bytes /127.0.0.3 -> /127.0.0.1: 10346 ranges, 3 sstables, 62.887MiB bytes nodetool repair --preview who what
45. NEW CONVENIENCE FUNCTIONS CQL ADDITIONS
46. TIMESTAMP FUNCTIONS 46 CREATE TABLE times ( pk int, time date, val int, PRIMARY KEY (pk, time));
47. TIMESTAMP FUNCTIONS 47 CREATE TABLE times ( pk int, time date, val int, PRIMARY KEY (pk, time));
48. TIMESTAMP FUNCTIONS 48 SELECT * FROM times WHERE pk = 1; pk | time | val ----+------------+----- 1 | 2016-09-28 | 2 1 | 2016-09-30 | 2 SELECT * FROM times WHERE pk = 1 AND time < '2017-10-01' - 1y2d; pk | time | val ----+------------+----- 1 | 2016-09-28 | 2
49. TIMESTAMP FUNCTIONS 49 SELECT * FROM times WHERE pk = 1; pk | time | val ----+------------+----- 1 | 2016-09-28 | 2 1 | 2016-09-30 | 2 SELECT * FROM times WHERE pk = 1 AND time < '2017-10-01' - 1y2d; pk | time | val ----+------------+----- 1 | 2016-09-28 | 2
50. TIMESTAMP FUNCTIONS 50 SELECT * FROM times WHERE pk = 1; pk | time | val ----+------------+----- 1 | 2016-09-28 | 2 1 | 2016-09-30 | 2 SELECT * FROM times WHERE pk = 1 AND time < '2017-10-01' - 1y2d; pk | time | val ----+------------+----- 1 | 2016-09-28 | 2
51. TIMESTAMP FUNCTIONS 51 SELECT * FROM times WHERE pk = 1; pk | time | val ----+------------+----- 1 | 2016-09-28 | 2 1 | 2016-09-30 | 2 SELECT * FROM times WHERE pk = 1 AND time < '2017-10-01' - 1y2d; pk | time | val ----+------------+----- 1 | 2016-09-28 | 2
52. ARITHMETIC OPERATORS 52 CREATE TABLE arithmetic ( key text, ts bigint, val text, PRIMARY KEY (key, ts));
53. ARITHMETIC OPERATORS 53 CREATE TABLE arithmetic ( key text, ts bigint, val text, PRIMARY KEY (key, ts));
54. ARITHMETIC OPERATORS 54 SELECT * from arithmetic WHERE key = 'mykey'; key | ts | val -------+----+------------ mykey | 10 | some value
55. ARITHMETIC OPERATORS 55 SELECT ts * 10 FROM arithmetic WHERE key = 'mykey'; ts * 10 --------- 100 SELECT ts % 10 FROM arithmetic WHERE key = 'mykey'; ts % 10 --------- 0
56. ARITHMETIC OPERATORS 56 https://github.com/apache/cassandra/blob/trunk/doc/source/cql/operators.rst
57. INTERNODE MESSAGING RE-WRITE 57
58. INTERNODE MESSAGING RE-WRITE 58 @15:45 Achievement Unblocked: Switching Apache Cassandra to Netty and non-blocking I/O - A followup
59. EASIER TO IMPLEMENT CLUSTER SECURITY
60. EASIER CLUSTER SECURITY 60 Because: It is really easy to attack an un-protected cluster
61. EASIER CLUSTER SECURITY 61 HA! A bin-packed message format with no source verification!* Ease of scalability comes with a price * <currently reading o.a.c.net.MessageIn#read>
62. EASIER CLUSTER SECURITY 62 nmap -Pn -p7000 \ -oG logs/cass.gnmap 54.88.0.0/14 HA! A bin-packed message format with no source verification!*
63. IMPORTANT BECAUSE... 63 Fun fact 1: It takes a single Message to insert an admin account into the system table
64. IMPORTANT BECAUSE... 64 Fun fact 2: It takes a single Message to truncate a table
65. IMPORTANT BECAUSE... 65 -Dcassandra.write_survey=true How to steal writes in real time:
66. node to node encryption and SSL client certificate authentication to cluster traffic THE EASY FIX: 66 http://thelastpickle.com/blog/2015/09/30/hardening-cassandra-step-by-step-part-1-server-to-server.html
67. node to node encryption and SSL client certificate authentication to cluster traffic THE EASY FIX: 67 Bonus: can be done with NO downtime!!! http://thelastpickle.com/blog/2015/09/30/hardening-cassandra-step-by-step-part-1-server-to-server.html
68. node to node encryption and SSL client certificate authentication to cluster traffic THE EASY FIX: 68 Bonus: can be done with NO downtime!!! http://thelastpickle.com/blog/2015/09/30/hardening-cassandra-step-by-step-part-1-server-to-server.html CASSANDRA-10404
69. THE EASY FIX: 69 When you are done it should look like:
70. THE EASY FIX: 70 Don't forget this part!!! When you are done it should look like:
71. WHAT COMES NEXT?
72. VIRTUAL TABLES
73. 73 cqlsh> select * from virtual.tables5 where keyspace_name = 'tlp_ks' and metric > 'memtable' and metric < 'memtableZ' ALLOW FILTERING; keyspace_name | table_name | metric | value ---------------+--------------------------+----------------------+---------------- tlp_ks | monitoring_example | memtableOnHeapSize | {"value":95201} tlp_ks | monitoring_example | memtableOffHeapSize | {"value":44811} tlp_ks | monitoring_example | memtableLiveDataSize | {"value":42128} tlp_ks | monitoring_example | memtableColumnsCount | {"value":248} tlp_ks | monitoring_example | memtableSwitchCount | {"count":4} ...
74. 74 cqlsh> select * from virtual.tables5 where keyspace_name = 'tlp_ks' and metric > 'memtable' and metric < 'memtableZ' ALLOW FILTERING; keyspace_name | table_name | metric | value ---------------+--------------------------+----------------------+---------------- tlp_ks | monitoring_example | memtableOnHeapSize | {"value":95201} tlp_ks | monitoring_example | memtableOffHeapSize | {"value":44811} tlp_ks | monitoring_example | memtableLiveDataSize | {"value":42128} tlp_ks | monitoring_example | memtableColumnsCount | {"value":248} tlp_ks | monitoring_example | memtableSwitchCount | {"count":4} ...
76. CASSANDRA-7622, CASSANDRA-9233
77. PLUGGABLE STORAGE ENGINE
78. https://www.slideshare.net/sunsuk7tp/mycassandra
79. https://www.slideshare.net/sunsuk7tp/mycassandra CASSANDRA-2995
80. ROCKS DB INTEGRATION
81. ROCKS DB INTEGRATION
82. ROCKS DB INTEGRATION Thanks to Dikang Gu and the Instagram team!
87. NOT A PANACEA • Some challenges with streaming • Big surface area of changes 87
89. PLUGGABLE STORAGE AND ROCKSDB - FOLLOW THESE ISSUES: • CASSANDRA-13474 • CASSANDRA-13475 • CASSANDRA-13476 • CASSANDRA-13553 89
90. EXPERIMENTAL FEATURES OFF BY DEFAULT 90 "-Dcassandra.mv.allow_filtering_nonkey_columns_unsafe=true" MV keys on non primary key columns (from CASSANDRA-13826):
91. FIXING BUGS! (DISCOVERED BY REAL WORKLOADS) 91 Our developers are running clusters. Sometimes *very* big clusters.
92. Future Cassandra Operators
93. THANKS! @ZZNATE