SlideShare a Scribd company logo

Radical NoSQL Scalability with Cassandra - Tim Berglund

JAX London, profile picture
JAX London

The document provides an overview of Cassandra's data model and query language CQL. It describes Cassandra's core abstractions including columns, rows, tables and keyspaces. It also covers writing, reading and updating data as well as secondary indexes, replication strategies and the hash ring-based data distribution model.

Technology
1 of 276
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
Radical NoSQL Scalability with @tlberglund Tuesday, October 16, 12
Data Model for developers Tuesday, October 16, 12
Column Tuesday, October 16, 12
Column full_name: “Tim Berglund” Tuesday, October 16, 12
Column full_name: “Tim Berglund” 20120425T1832 Tuesday, October 16, 12
Column Key/Value pair full_name: “Tim Berglund” 20120425T1832 Tuesday, October 16, 12
Column Key/Value pair full_name: “Tim Berglund” 20120425T1832 Timestamp Tuesday, October 16, 12
Column Tuesday, October 16, 12
Column ‣ Key-value pair Tuesday, October 16, 12
Column ‣ Key-value pair ‣ Optionally typed Tuesday, October 16, 12
Column ‣ Key-value pair ‣ Optionally typed ‣ Timestamped Tuesday, October 16, 12
Column ‣ Key-value pair ‣ Optionally typed ‣ Timestamped ‣ Fundamental unit Tuesday, October 16, 12
Row Tuesday, October 16, 12
Row column Tuesday, October 16, 12
Row column column Tuesday, October 16, 12
Row column column column Tuesday, October 16, 12
Row row key column column column Tuesday, October 16, 12
Row name: bday: role: tlberglund Tim 06-15 teacher Tuesday, October 16, 12
Row Row Key name: bday: role: tlberglund Tim 06-15 teacher Tuesday, October 16, 12
Row Row Key name: bday: role: tlberglund Tim 06-15 teacher Columns Tuesday, October 16, 12
Row name: bday: role: tlberglund Tim 06-15 teacher Tuesday, October 16, 12
Row bday: name: role: tlberglund 06-15 Tim teacher Tuesday, October 16, 12
Row bday: name: role: tlberglund 06-15 Tim teacher Sorted by UTF8Type comparator Tuesday, October 16, 12
Table Tuesday, October 16, 12
Table Tuesday, October 16, 12
Table name: role: status: tim Tim teacher Cool Tuesday, October 16, 12
Table name: role: status: tim Tim teacher Cool name: role: kristen marketing Kristen Tuesday, October 16, 12
Table name: role: status: tim Tim teacher Cool name: role: kristen marketing Kristen billy role: CEO Tuesday, October 16, 12
Table name: role: status: tim Tim teacher Cool name: role: kristen marketing Kristen billy role: CEO name: role: status: matt Matt founder ubercool Tuesday, October 16, 12
Table name: role: status: tim Tim teacher Cool name: role: kristen marketing Kristen billy role: CEO name: role: status: matt Matt founder ubercool Tuesday, October 16, 12
Outer hash key Table name: role: status: tim Tim teacher Cool name: role: kristen marketing Kristen billy role: CEO name: role: status: matt Matt founder ubercool Tuesday, October 16, 12
Outer hash key Table Inner hash key name: role: status: tim Tim teacher Cool name: role: kristen marketing Kristen billy role: CEO name: role: status: matt Matt founder ubercool Tuesday, October 16, 12
Outer hash key Table Inner hash key name: role: status: tim Tim teacher Cool Sparse name: role: kristen marketing Kristen billy role: CEO name: role: status: matt Matt founder ubercool Tuesday, October 16, 12
Database Accounts ClickStream Orders InventoryEvents Tuesday, October 16, 12
Database Accounts ClickStream Tables Orders InventoryEvents Tuesday, October 16, 12
Cluster System Database Application Database Tuesday, October 16, 12
Secondary Indexes Tuesday, October 16, 12
Secondary Indexes ‣ Ubiquitous in relational databases Tuesday, October 16, 12
Secondary Indexes ‣ Ubiquitous in relational databases ‣ Supported in Cassandra, with qualifications Tuesday, October 16, 12
Secondary Indexes name: email: role: tim Tim tb@a.com teacher name: email: kristen k@ds.com Kristen billy role: CEO name: email: status: matt Matt m@ds.com ubercool Tuesday, October 16, 12
Secondary Indexes name: email: role: tim Tim tb@a.com teacher name: email: kristen k@ds.com Kristen billy role: CEO name: email: status: matt Matt m@ds.com ubercool Tuesday, October 16, 12
Secondary Indexes name: email: role: tim Tim tb@a.com teacher name: email: kristen k@ds.com Kristen billy role: CEO name: email: status: matt Matt m@ds.com ubercool Tuesday, October 16, 12
Secondary Indexes Tuesday, October 16, 12
Secondary Indexes ‣ In relational databases: performant for high cardinality Tuesday, October 16, 12
Secondary Indexes ‣ In relational databases: performant for high cardinality ‣ In Cassandra: the reverse Tuesday, October 16, 12
Secondary Indexes ‣ In relational databases: performant for high cardinality ‣ In Cassandra: the reverse ‣ Not suitable for lookup-by-email Tuesday, October 16, 12
Secondary Indexes ‣ In relational databases: performant for high cardinality ‣ In Cassandra: the reverse ‣ Not suitable for lookup-by-email ‣ Suitable for lookup by: region code, gender, state, etc. Tuesday, October 16, 12
Tuesday, October 16, 12
Why ? Tuesday, October 16, 12
Why BigTable? Tuesday, October 16, 12
Why BigTable? Flexibility Tuesday, October 16, 12
Why BigTable? Flexibility Performance Tuesday, October 16, 12
Query Language for developers Tuesday, October 16, 12
Tuesday, October 16, 12
CQL Tuesday, October 16, 12
CQL (Cassandra Query Language) Tuesday, October 16, 12
CREATE Tuesday, October 16, 12
CREATE KEYSPACE Tuesday, October 16, 12
CREATE KEYSPACE CREATE KEYSPACE DemoKeyspace WITH strategy_class='SimpleStrategy' AND strategy_options:replication_factor=1; Tuesday, October 16, 12
CREATE TABLE Tuesday, October 16, 12
CREATE TABLE CREATE TABLE accounts (KEY text PRIMARY KEY) WITH comparator=text AND default_validation=text; Tuesday, October 16, 12
CREATE TABLE Tuesday, October 16, 12
CREATE TABLE CREATE TABLE accounts (KEY text PRIMARY KEY, name text, email text, signed_up_at timestamp) WITH comparator=text; Tuesday, October 16, 12
INSERT Tuesday, October 16, 12
INSERT Tuesday, October 16, 12
INSERT INSERT INTO accounts (KEY, name, email, signed_up_at) VALUES ('tlberglund', 'Tim Berglund', 'tlberglund@gmail.com', '2012-04-25'); Tuesday, October 16, 12
INSERT Tuesday, October 16, 12
INSERT INSERT INTO events (KEY, 0, 1, 2, 3, 4) VALUES ('2012-04-25T11:04:34-0700', 55.4, 56.2, 59.6, 65.3, 79) USING CONSISTENCY QUORUM AND TTL 86400; Tuesday, October 16, 12
SELECT Tuesday, October 16, 12
SELECT Tuesday, October 16, 12
SELECT SELECT * FROM accounts WHERE KEY='tlberglund'; Tuesday, October 16, 12
SELECT Tuesday, October 16, 12
SELECT SELECT 1..3 FROM events WHERE KEY='2012-04-25T11:04:34-0700'; Tuesday, October 16, 12
SELECT Tuesday, October 16, 12
SELECT SELECT * FROM accounts WHERE KEY='tlberglund' USING CONSISTENCY ONE; Tuesday, October 16, 12
UPDATE Tuesday, October 16, 12
UPDATE Tuesday, October 16, 12
UPDATE UPDATE accounts SET last_login='2012-04-25T09:37:35-0700' WHERE KEY='tlberglund'; Tuesday, October 16, 12
Distribution Model Tuesday, October 16, 12
Hash Ring Tuesday, October 16, 12
0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
Writing a Key Tuesday, October 16, 12
0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 name: Tim E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 3D97: Tim E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 C000 4000 3D97: Tim A000 6000 8000 Tuesday, October 16, 12
0000 role: Teacher E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 9C4F: Teacher E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 C000 4000 A000 9C4F: Teacher 6000 8000 Tuesday, October 16, 12
Reading a Key Tuesday, October 16, 12
0000 3D97? E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 3D97? E000 2000 name: Tim C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 9C4F? E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 9C4F? E000 2000 role: Teacher C000 4000 A000 6000 8000 Tuesday, October 16, 12
Replication Tuesday, October 16, 12
Replication Tuesday, October 16, 12
Replication Replication factor (N) Tuesday, October 16, 12
Replication Replication factor (N) Pluggable placement strategies Tuesday, October 16, 12
Replication Strategies Tuesday, October 16, 12
Replication Strategies Simple Tuesday, October 16, 12
Replication Strategies Simple Network Topology Aware Tuesday, October 16, 12
Simple Strategy 0000 E000 2000 C000 4000 N=3 A000 6000 8000 Tuesday, October 16, 12
Simple Strategy 0000 3D97: Tim E000 2000 C000 4000 N=3 A000 6000 8000 Tuesday, October 16, 12
Simple Strategy 0000 E000 2000 C000 4000 3D97: Tim N=3 A000 6000 8000 Tuesday, October 16, 12
Simple Strategy 0000 E000 2000 C000 4000 3D97: Tim N=3 A000 6000 3D97: Tim 8000 Tuesday, October 16, 12
Simple Strategy 0000 E000 2000 C000 4000 3D97: Tim N=3 A000 6000 3D97: Tim 8000 3D97: Tim Tuesday, October 16, 12
Topology Aware DC1 DC2 2000 0000 6000 4000 A000 8000 E000 C000 Tuesday, October 16, 12
Topology Aware DC1 DC2 2000 0000 6000 4000 A000 8000 E000 C000 Tuesday, October 16, 12
Topology Aware 3D97: Tim DC1 DC2 2000 0000 6000 4000 A000 8000 E000 C000 Tuesday, October 16, 12
Topology Aware DC1 DC2 2000 0000 6000 4000 3D97: Tim A000 8000 E000 C000 Tuesday, October 16, 12
Topology Aware DC1 DC2 2000 0000 6000 3D97: Tim 4000 3D97: Tim A000 8000 E000 C000 Tuesday, October 16, 12
Topology Aware DC1 DC2 2000 0000 6000 3D97: Tim 4000 3D97: Tim A000 8000 3D97: Tim E000 C000 Tuesday, October 16, 12
Topology Aware DC1 DC2 2000 0000 6000 3D97: Tim 4000 3D97: Tim A000 3D97: Tim 8000 3D97: Tim E000 C000 Tuesday, October 16, 12
Writing Tuesday, October 16, 12
Client Connections Tuesday, October 16, 12
Client Connections 0000 E000 2000 C000 4000 Client Machine A000 6000 8000 Tuesday, October 16, 12
Client Connections 0000 E000 2000 Client Machine ? C000 4000 A000 6000 8000 Tuesday, October 16, 12
Client Connections 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Client Connections 0000 E000 2000 3D97? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Client Connections 0000 E000 2000 3D97? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Client Connections 0000 E000 2000 3D97? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Client Connections 0000 E000 2000 3D97? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Client Connections 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Client Connections 0000 E000 2000 9C4F? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Client Connections 0000 E000 2000 9C4F? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Client Connections 0000 E000 2000 9C4F? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Client Connections 0000 E000 2000 9C4F? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Load Balancer Tuesday, October 16, 12
Load Balancer Hardware VIP Tuesday, October 16, 12
Load Balancer Hardware VIP HAProxy Tuesday, October 16, 12
Load Balancer Hardware VIP HAProxy Round-robin DNS Tuesday, October 16, 12
Load Balancer Hardware VIP HAProxy Round-robin DNS Client-side Tuesday, October 16, 12
Load Balancer Hardware VIP HAProxy Round-robin DNS Client-side (Hector does this) Tuesday, October 16, 12
Write Consistency Tuesday, October 16, 12
Write Consistency 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Write Consistency 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 14C7 Tuesday, October 16, 12
Write Consistency 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 14C7 Tuesday, October 16, 12
Write Consistency 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 14C7 Tuesday, October 16, 12
Write Consistency 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 14C7 Tuesday, October 16, 12
Write Consistency 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 14C7 Tuesday, October 16, 12
Write Consistency 0000 Coordinator E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 14C7 Tuesday, October 16, 12
Write Consistency 0000 E000 ---- C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Write Consistency 0000 E000 ---- C000 4000 Client Machine 14C7 A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Write Consistency 0000 E000 ---- X C000 4000 Client Machine 14C7 A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Write Consistency 0000 E000 ---- X C000 4000 Client Machine 14C7 A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Write Consistency 0000 E000 ---- X C000 4000 Client Machine 14C7 A000 6000 Load Balancer 8000 Tuesday, October 16, 12
Write Consistency 0000 E000 ---- X C000 4000 Client Machine 14C7 A000 6000 Load Balancer 8000 Coordinator stores a “hint” Tuesday, October 16, 12
Write Consistency Tuesday, October 16, 12
Write Consistency ‣ ANY At least one node (hinted handoffs allowed) Tuesday, October 16, 12
Write Consistency ‣ ANY At least one node (hinted handoffs allowed) ‣ ONE At least one node (no hinted handoffs) Tuesday, October 16, 12
Write Consistency ‣ ANY At least one node (hinted handoffs allowed) ‣ ONE At least one node (no hinted handoffs) ‣ QUORUM (N/2)+1 nodes Tuesday, October 16, 12
Write Consistency Tuesday, October 16, 12
Write Consistency ‣ LOCAL_QUORUM (N/2)+1 nodes in this availability zone Tuesday, October 16, 12
Write Consistency ‣ LOCAL_QUORUM (N/2)+1 nodes in this availability zone ‣ EACH_QUORUM (N/2)+1 nodes in all availability zones Tuesday, October 16, 12
Write Consistency ‣ LOCAL_QUORUM (N/2)+1 nodes in this availability zone ‣ EACH_QUORUM (N/2)+1 nodes in all availability zones ‣ ALL Write successfully to all replicas Tuesday, October 16, 12
Reading Tuesday, October 16, 12
Read Consistency 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
Read Consistency 9C4F? 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
Read Consistency 9C4F? 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
Read Consistency 9C4F? 0000 E000 2000 Tim TODAY C000 4000 A000 6000 8000 Tuesday, October 16, 12
Read Consistency 9C4F? 0000 E000 2000 Tim TODAY Tim TODAY C000 4000 A000 6000 8000 Tuesday, October 16, 12
Read Consistency 9C4F? 0000 E000 2000 Tim TODAY Tim TODAY C000 4000 Tim TODAY A000 6000 8000 Tuesday, October 16, 12
Read Consistency Coordinator 9C4F? 0000 E000 2000 Tim TODAY Tim TODAY C000 4000 Tim TODAY A000 6000 8000 Tuesday, October 16, 12
Read Consistency 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
Read Consistency 9C4F? 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
Read Consistency 9C4F? 0000 E000 2000 Tim TODAY C000 4000 A000 6000 8000 Tuesday, October 16, 12
Read Consistency 9C4F? 0000 E000 2000 Tim TODAY Jim YESTERDAY C000 4000 A000 6000 8000 Tuesday, October 16, 12
Read Consistency 9C4F? 0000 E000 2000 Tim TODAY Jim YESTERDAY C000 4000 Tim TODAY A000 6000 8000 Tuesday, October 16, 12
Read Consistency 9C4F? 0000 E000 2000 Tim TODAY Jim YESTERDAY C000 4000 Tim TODAY A000 6000 8000 Inconsistent Tuesday, October 16, 12
Passive Read Repair Tuesday, October 16, 12
Passive Read Repair ‣ Initiated by coordinator Tuesday, October 16, 12
Passive Read Repair ‣ Initiated by coordinator ‣ Cleans up entropy in a single row Tuesday, October 16, 12
Passive Read Repair ‣ Initiated by coordinator ‣ Cleans up entropy in a single row ‣ Happens regardless of consistency level Tuesday, October 16, 12
Passive Read Repair ‣ Initiated by coordinator ‣ Cleans up entropy in a single row ‣ Happens regardless of consistency level ‣ Just reading the database reduces its entropy Tuesday, October 16, 12
Read Consistency Tuesday, October 16, 12
Read Consistency ‣ ONE Get response from the closest replica Tuesday, October 16, 12
Read Consistency ‣ ONE Get response from the closest replica ‣ QUORUM Get (N/2)+1 nodes, return most recent timestamp Tuesday, October 16, 12
Read Consistency Tuesday, October 16, 12
Read Consistency ‣ LOCAL_QUORUM (N/2)+1 nodes in this availability zone Tuesday, October 16, 12
Read Consistency ‣ LOCAL_QUORUM (N/2)+1 nodes in this availability zone ‣ EACH_QUORUM (N/2)+1 nodes in all availability zones Tuesday, October 16, 12
Read Consistency ‣ LOCAL_QUORUM (N/2)+1 nodes in this availability zone ‣ EACH_QUORUM (N/2)+1 nodes in all availability zones ‣ ALL Wait for all replicas to respond Tuesday, October 16, 12
Tuesday, October 16, 12
But what about Column Families? Tuesday, October 16, 12
Replication For Real Tuesday, October 16, 12
Replication For Real Rows are replicated Tuesday, October 16, 12
Replication For Real Rows are replicated “Key” is row key Tuesday, October 16, 12
Replication For Real Rows are replicated “Key” is row key “Value” is the row data Tuesday, October 16, 12
Replication For Real Rows are replicated “Key” is row key “Value” is the row data Implications for row size Tuesday, October 16, 12
Gossip Tuesday, October 16, 12
Gossip Tuesday, October 16, 12
Gossip ‣ Naive heartbeats don’t scale Tuesday, October 16, 12
Gossip ‣ Naive heartbeats don’t scale ‣ Cluster still needs to know its state Tuesday, October 16, 12
Gossip ‣ Naive heartbeats don’t scale ‣ Cluster still needs to know its state ‣ Computes a real-valued “suspicion” for each node Tuesday, October 16, 12
Gossip ‣ Naive heartbeats don’t scale ‣ Cluster still needs to know its state ‣ Computes a real-valued “suspicion” for each node ‣ Probabilistic Tuesday, October 16, 12
Gossip ‣ Naive heartbeats don’t scale ‣ Cluster still needs to know its state ‣ Computes a real-valued “suspicion” for each node ‣ Probabilistic ‣ Just like the real thing Tuesday, October 16, 12
0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 How are you, E000? C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 I'm cool. C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 C000 4000 How about you, A000? A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 C000 4000 Oh, I'm fine. A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 I trust e000. Not so sure about A000... C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 E000 C000, what 2000 do you know? C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 I'm Great! C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 E000 is doing well. C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 Poor A000 is having trouble lately. C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 Hmmm, so... C000 4000 A000 6000 8000 Tuesday, October 16, 12
0000 E000 2000 ✔ C000 ✔ E000 ✘ A000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
Gossip Config Tuesday, October 16, 12
Gossip Config ‣ A new node needs “seed nodes” Tuesday, October 16, 12
Gossip Config ‣ A new node needs “seed nodes” ‣ Seed nodes configured in $CASSANDRA_HOME/conf/ cassandra.yaml Tuesday, October 16, 12
Storage Tuesday, October 16, 12
Storage Engine Tuesday, October 16, 12
Storage Engine ‣ “Log-structured storage” Tuesday, October 16, 12
Storage Engine ‣ “Log-structured storage” ‣ All writes are sequential Tuesday, October 16, 12
Storage Engine ‣ “Log-structured storage” ‣ All writes are sequential ‣ All writes are immutable Tuesday, October 16, 12
Storage Engine ‣ “Log-structured storage” ‣ All writes are sequential ‣ All writes are immutable ‣ Designed to avoid seeks Tuesday, October 16, 12
Storage Engine ‣ “Log-structured storage” ‣ All writes are sequential ‣ All writes are immutable ‣ Designed to avoid seeks ‣ Writes are faster than reads Tuesday, October 16, 12
Write Sequence Tuesday, October 16, 12
Write Sequence Write from Commit Coordinator Log Tuesday, October 16, 12
Write Sequence Write from Commit Coordinator Log Memtable Tuesday, October 16, 12
Write Sequence Write from Commit Coordinator Log SSTable(s) Memtable (lots of tuning here) Tuesday, October 16, 12
Write Sequence Write from Commit Coordinator Log Compaction (more tuning here) SSTable(s) Memtable (lots of tuning here) Tuesday, October 16, 12
Commit Log Tuesday, October 16, 12
Commit Log ‣ Writes go here first Tuesday, October 16, 12
Commit Log ‣ Writes go here first ‣ Commit succeeds before node reports successful write Tuesday, October 16, 12
Commit Log ‣ Writes go here first ‣ Commit succeeds before node reports successful write ‣ Append-only, sequential writes Tuesday, October 16, 12
Commit Log ‣ Writes go here first ‣ Commit succeeds before node reports successful write ‣ Append-only, sequential writes ‣ One per server Tuesday, October 16, 12
Commit Log ‣ Writes go here first ‣ Commit succeeds before node reports successful write ‣ Append-only, sequential writes ‣ One per server ‣ Good to have a dedicated spindle Tuesday, October 16, 12
MemTable Tuesday, October 16, 12
MemTable ‣ An in-memory structure Tuesday, October 16, 12
MemTable ‣ An in-memory structure ‣ One per column family Tuesday, October 16, 12
MemTable ‣ An in-memory structure ‣ One per column family ‣ Holds most recent row changes Tuesday, October 16, 12
MemTable ‣ An in-memory structure ‣ One per column family ‣ Holds most recent row changes ‣ Tunable memory use Tuesday, October 16, 12
MemTable ‣ An in-memory structure ‣ One per column family ‣ Holds most recent row changes ‣ Tunable memory use ‣ Flushed to disk when “full” Tuesday, October 16, 12
SSTable Tuesday, October 16, 12
SSTable ‣ Memtables flushed to disk here Tuesday, October 16, 12
SSTable ‣ Memtables flushed to disk here ‣ Many SSTable files per column family Tuesday, October 16, 12
SSTable ‣ Memtables flushed to disk here ‣ Many SSTable files per column family ‣ Every SSTable is immutable Tuesday, October 16, 12
SSTable ‣ Memtables flushed to disk here ‣ Many SSTable files per column family ‣ Every SSTable is immutable ‣ SSTables are accessed during reads Tuesday, October 16, 12
SSTable ‣ Memtables flushed to disk here ‣ Many SSTable files per column family ‣ Every SSTable is immutable ‣ SSTables are accessed during reads ‣ Must be compacted Tuesday, October 16, 12
Read Sequence Read from Coordinator Tuesday, October 16, 12
Read Sequence Memtable Tuesday, October 16, 12
Read Sequence Memtable Are all columns here? Tuesday, October 16, 12
Read Sequence Memtable Are all columns here? YES: stop and return result. Tuesday, October 16, 12
Read Sequence Memtable Are all columns here? YES: stop and return result. NO: continue. Tuesday, October 16, 12
Read Sequence Newest SSTable Tuesday, October 16, 12
Read Sequence Newest SSTable Are all columns here? Tuesday, October 16, 12
Read Sequence Newest SSTable Are all columns here? YES: stop and return result. Tuesday, October 16, 12
Read Sequence Newest SSTable Are all columns here? YES: stop and return result. NO: continue. Tuesday, October 16, 12
Read Sequence Next Oldest SSTable Tuesday, October 16, 12
Read Sequence Next Oldest SSTable How about now? Tuesday, October 16, 12
Read Sequence Next Oldest SSTable How about now? YES: great! Tuesday, October 16, 12
Read Sequence Next Oldest SSTable How about now? YES: great! NO: keep looking... Tuesday, October 16, 12
Read Sequence And so on. Tuesday, October 16, 12
Read Sequence And so on. Doesn’t this get old? Tuesday, October 16, 12
Read Sequence Tuesday, October 16, 12
Read Sequence ‣ Check Memtable first Tuesday, October 16, 12
Read Sequence ‣ Check Memtable first ‣ Read SSTables from newest to oldest Tuesday, October 16, 12
Read Sequence ‣ Check Memtable first ‣ Read SSTables from newest to oldest ‣ Bloom filters prevent most reads Tuesday, October 16, 12
Read Sequence ‣ Check Memtable first ‣ Read SSTables from newest to oldest ‣ Bloom filters prevent most reads ‣ Compaction shrinks number of files Tuesday, October 16, 12
Compaction Tuesday, October 16, 12
Compaction Tuesday, October 16, 12
Compaction ‣ Combine many SSTables into one Tuesday, October 16, 12
Compaction ‣ Combine many SSTables into one ‣ Performed in the background Tuesday, October 16, 12
Compaction ‣ Combine many SSTables into one ‣ Performed in the background ‣ Node still operates Tuesday, October 16, 12
Compaction ‣ Combine many SSTables into one ‣ Performed in the background ‣ Node still operates ‣ Requires extra disk space Tuesday, October 16, 12
Compaction ‣ Combine many SSTables into one ‣ Performed in the background ‣ Node still operates ‣ Requires extra disk space ‣ Three tunable varieties Tuesday, October 16, 12
Compaction Tuesday, October 16, 12
Compaction ‣ Major All SSTables are merged into one clean one Tuesday, October 16, 12
Compaction ‣ Major All SSTables are merged into one clean one ‣ Minor Similarly-sized SSTables are merged together after reaching a threshold Tuesday, October 16, 12
Compaction ‣ Major All SSTables are merged into one clean one ‣ Minor Similarly-sized SSTables are merged together after reaching a threshold ‣ Leveled http://www.datastax.com/dev/blog/leveled-compaction-in-apache-cassandra Tuesday, October 16, 12
http://www.datastax.com/ products/community Tuesday, October 16, 12
http://www.datastax.com/ products/community t hi s! n lo ad D ow Tuesday, October 16, 12
Tim Berglund tlberglund@github.com @tlberglund Thank You Tuesday, October 16, 12

More Related Content

PPTX
Cassandra ppt 2
Skillwise Group
 
PDF
Nuove professioni dell'editoria
Flavio Pintarelli
 
PDF
Cassandra Day Chicago 2015: Building Your First Application with Apache Cassa...
DataStax Academy
 
PDF
Cassandra datamodel
lurga
 
PDF
Apache Cassandra
Sperasoft
 
PDF
DataStax: Backup and Restore in Cassandra and OpsCenter
DataStax Academy
 
PDF
Introduction to Cassandra & Data model
Duyhai Doan
 
PDF
Cassandra 0.7, Los Angeles High Scalability Group
jbellis
 
Cassandra ppt 2
Skillwise Group
 
Nuove professioni dell'editoria
Flavio Pintarelli
 
Cassandra Day Chicago 2015: Building Your First Application with Apache Cassa...
DataStax Academy
 
Cassandra datamodel
lurga
 
Apache Cassandra
Sperasoft
 
DataStax: Backup and Restore in Cassandra and OpsCenter
DataStax Academy
 
Introduction to Cassandra & Data model
Duyhai Doan
 
Cassandra 0.7, Los Angeles High Scalability Group
jbellis
 

Viewers also liked (14)

PPTX
Apache Cassandra, part 2 – data model example, machinery
Andrey Lomakin
 
PDF
Cassandra Data Modeling
Ben Knear
 
PPTX
Cassandra Data Modeling - Practical Considerations @ Netflix
nkorla1share
 
PPTX
Cassandra into
DataArt
 
PPT
Cassandra Data Model
ebenhewitt
 
PPTX
Webinar | Introducing DataStax Enterprise 4.6
DataStax
 
PPTX
Data Modeling Basics for the Cloud with DataStax
DataStax
 
PDF
Cassandra Community Webinar | The World's Next Top Data Model
DataStax
 
PDF
Overview of DataStax OpsCenter
DataStax
 
PPTX
Webinar | Target Modernizes Retail with Engaging Digital Experiences
DataStax
 
PDF
Cassandra
Edureka!
 
PDF
Advanced data modeling with apache cassandra
Patrick McFadin
 
PDF
Cassandra at NoSql Matters 2012
jbellis
 
PPTX
Cassandra Performance and Scalability on AWS
Adrian Cockcroft
 
Apache Cassandra, part 2 – data model example, machinery
Andrey Lomakin
 
Cassandra Data Modeling
Ben Knear
 
Cassandra Data Modeling - Practical Considerations @ Netflix
nkorla1share
 
Cassandra into
DataArt
 
Cassandra Data Model
ebenhewitt
 
Webinar | Introducing DataStax Enterprise 4.6
DataStax
 
Data Modeling Basics for the Cloud with DataStax
DataStax
 
Cassandra Community Webinar | The World's Next Top Data Model
DataStax
 
Overview of DataStax OpsCenter
DataStax
 
Webinar | Target Modernizes Retail with Engaging Digital Experiences
DataStax
 
Cassandra
Edureka!
 
Advanced data modeling with apache cassandra
Patrick McFadin
 
Cassandra at NoSql Matters 2012
jbellis
 
Cassandra Performance and Scalability on AWS
Adrian Cockcroft
 
Ad

More from JAX London (20)

PDF
Everything I know about software in spaghetti bolognese: managing complexity
JAX London
 
PDF
Devops with the S for Sharing - Patrick Debois
JAX London
 
PPT
Busy Developer's Guide to Windows 8 HTML/JavaScript Apps
JAX London
 
PDF
It's code but not as we know: Infrastructure as Code - Patrick Debois
JAX London
 
KEY
Locks? We Don't Need No Stinkin' Locks - Michael Barker
JAX London
 
PDF
Worse is better, for better or for worse - Kevlin Henney
JAX London
 
PDF
Java performance: What's the big deal? - Trisha Gee
JAX London
 
PDF
Clojure made-simple - John Stevenson
JAX London
 
PDF
HTML alchemy: the secrets of mixing JavaScript and Java EE - Matthias Wessendorf
JAX London
 
PDF
Play framework 2 : Peter Hilton
JAX London
 
PDF
Complexity theory and software development : Tim Berglund
JAX London
 
PDF
Why FLOSS is a Java developer's best friend: Dave Gruber
JAX London
 
PDF
Akka in Action: Heiko Seeburger
JAX London
 
PDF
NoSQL Smackdown 2012 : Tim Berglund
JAX London
 
PDF
Closures, the next "Big Thing" in Java: Russel Winder
JAX London
 
KEY
Java and the machine - Martijn Verburg and Kirk Pepperdine
JAX London
 
PDF
Mongo DB on the JVM - Brendan McAdams
JAX London
 
PDF
New opportunities for connected data - Ian Robinson
JAX London
 
PDF
HTML5 Websockets and Java - Arun Gupta
JAX London
 
PDF
The Big Data Con: Why Big Data is a Problem, not a Solution - Ian Plosker
JAX London
 
Everything I know about software in spaghetti bolognese: managing complexity
JAX London
 
Devops with the S for Sharing - Patrick Debois
JAX London
 
Busy Developer's Guide to Windows 8 HTML/JavaScript Apps
JAX London
 
It's code but not as we know: Infrastructure as Code - Patrick Debois
JAX London
 
Locks? We Don't Need No Stinkin' Locks - Michael Barker
JAX London
 
Worse is better, for better or for worse - Kevlin Henney
JAX London
 
Java performance: What's the big deal? - Trisha Gee
JAX London
 
Clojure made-simple - John Stevenson
JAX London
 
HTML alchemy: the secrets of mixing JavaScript and Java EE - Matthias Wessendorf
JAX London
 
Play framework 2 : Peter Hilton
JAX London
 
Complexity theory and software development : Tim Berglund
JAX London
 
Why FLOSS is a Java developer's best friend: Dave Gruber
JAX London
 
Akka in Action: Heiko Seeburger
JAX London
 
NoSQL Smackdown 2012 : Tim Berglund
JAX London
 
Closures, the next "Big Thing" in Java: Russel Winder
JAX London
 
Java and the machine - Martijn Verburg and Kirk Pepperdine
JAX London
 
Mongo DB on the JVM - Brendan McAdams
JAX London
 
New opportunities for connected data - Ian Robinson
JAX London
 
HTML5 Websockets and Java - Arun Gupta
JAX London
 
The Big Data Con: Why Big Data is a Problem, not a Solution - Ian Plosker
JAX London
 
Ad

Recently uploaded (20)

PDF
Reach Out and Touch Someone: Haptics and Empathic Computing
Mark Billinghurst
 
PPTX
20250228 LYD VKU AI Blended-Learning.pptx
DatVoNgoc
 
PPTX
Effective Security Operations Center (SOC) A Modern, Strategic, and Threat-In...
ReZa AdineH
 
PPT
“AI and Expert System Decision Support & Business Intelligence Systems”
ZubinRadhakrishnan
 
PDF
Blue Purple Modern Animated Computer Science Presentation.pdf.pdf
Akar16
 
PDF
7 ChatGPT Prompts to Help You Define Your Ideal Customer Profile.pdf
SOFTTECHHUB
 
PPTX
KOM of Painting work and Equipment Insulation REV00 update 25-dec.pptx
muhammadmuneebnaeem7
 
PDF
Shreyas Phanse Resume: Experienced Backend Engineer | Java • Spring Boot • Ka...
SHREYAS PHANSE
 
PDF
CIFDAQ's Market Insight: SEC Turns Pro Crypto
CIFDAQ
 
PPTX
PA Analog/Digital System: The Backbone of Modern Surveillance and Communication
AVTRON Technologies LLC
 
PDF
Empathic Computing: Creating Shared Understanding
Mark Billinghurst
 
PDF
Peak of Data & AI Encore- AI for Metadata and Smarter Workflows
Safe Software
 
PDF
Diabetes mellitus diagnosis method based random forest with bat algorithm
IAESIJAI
 
PDF
NewMind AI Monthly Chronicles - July 2025
NewMind AI
 
PDF
NewMind AI Weekly Chronicles - August'25 Week I
NewMind AI
 
PDF
Build a system with the filesystem maintained by OSTree @ COSCUP 2025
Jian-Hong Pan
 
PDF
Encapsulation_ Review paper, used for researhc scholars
gurumoop
 
PDF
Architecting across the Boundaries of two Complex Domains - Healthcare & Tech...
Peter Tan
 
PDF
Building Integrated photovoltaic BIPV_UPV.pdf
SandeepSingh286037
 
PDF
Electronic commerce courselecture one. Pdf
OmerMohamed64
 
Reach Out and Touch Someone: Haptics and Empathic Computing
Mark Billinghurst
 
20250228 LYD VKU AI Blended-Learning.pptx
DatVoNgoc
 
Effective Security Operations Center (SOC) A Modern, Strategic, and Threat-In...
ReZa AdineH
 
“AI and Expert System Decision Support & Business Intelligence Systems”
ZubinRadhakrishnan
 
Blue Purple Modern Animated Computer Science Presentation.pdf.pdf
Akar16
 
7 ChatGPT Prompts to Help You Define Your Ideal Customer Profile.pdf
SOFTTECHHUB
 
KOM of Painting work and Equipment Insulation REV00 update 25-dec.pptx
muhammadmuneebnaeem7
 
Shreyas Phanse Resume: Experienced Backend Engineer | Java • Spring Boot • Ka...
SHREYAS PHANSE
 
CIFDAQ's Market Insight: SEC Turns Pro Crypto
CIFDAQ
 
PA Analog/Digital System: The Backbone of Modern Surveillance and Communication
AVTRON Technologies LLC
 
Empathic Computing: Creating Shared Understanding
Mark Billinghurst
 
Peak of Data & AI Encore- AI for Metadata and Smarter Workflows
Safe Software
 
Diabetes mellitus diagnosis method based random forest with bat algorithm
IAESIJAI
 
NewMind AI Monthly Chronicles - July 2025
NewMind AI
 
NewMind AI Weekly Chronicles - August'25 Week I
NewMind AI
 
Build a system with the filesystem maintained by OSTree @ COSCUP 2025
Jian-Hong Pan
 
Encapsulation_ Review paper, used for researhc scholars
gurumoop
 
Architecting across the Boundaries of two Complex Domains - Healthcare & Tech...
Peter Tan
 
Building Integrated photovoltaic BIPV_UPV.pdf
SandeepSingh286037
 
Electronic commerce courselecture one. Pdf
OmerMohamed64
 

Radical NoSQL Scalability with Cassandra - Tim Berglund

  • 1. Radical NoSQL Scalability with @tlberglund Tuesday, October 16, 12
  • 2. Data Model for developers Tuesday, October 16, 12
  • 4. Column full_name: “Tim Berglund” Tuesday, October 16, 12
  • 5. Column full_name: “Tim Berglund” 20120425T1832 Tuesday, October 16, 12
  • 6. Column Key/Value pair full_name: “Tim Berglund” 20120425T1832 Tuesday, October 16, 12
  • 7. Column Key/Value pair full_name: “Tim Berglund” 20120425T1832 Timestamp Tuesday, October 16, 12
  • 9. Column ‣ Key-value pair Tuesday, October 16, 12
  • 10. Column ‣ Key-value pair ‣ Optionally typed Tuesday, October 16, 12
  • 11. Column ‣ Key-value pair ‣ Optionally typed ‣ Timestamped Tuesday, October 16, 12
  • 12. Column ‣ Key-value pair ‣ Optionally typed ‣ Timestamped ‣ Fundamental unit Tuesday, October 16, 12
  • 14. Row column Tuesday, October 16, 12
  • 15. Row column column Tuesday, October 16, 12
  • 16. Row column column column Tuesday, October 16, 12
  • 17. Row row key column column column Tuesday, October 16, 12
  • 18. Row name: bday: role: tlberglund Tim 06-15 teacher Tuesday, October 16, 12
  • 19. Row Row Key name: bday: role: tlberglund Tim 06-15 teacher Tuesday, October 16, 12
  • 20. Row Row Key name: bday: role: tlberglund Tim 06-15 teacher Columns Tuesday, October 16, 12
  • 21. Row name: bday: role: tlberglund Tim 06-15 teacher Tuesday, October 16, 12
  • 22. Row bday: name: role: tlberglund 06-15 Tim teacher Tuesday, October 16, 12
  • 23. Row bday: name: role: tlberglund 06-15 Tim teacher Sorted by UTF8Type comparator Tuesday, October 16, 12
  • 26. Table name: role: status: tim Tim teacher Cool Tuesday, October 16, 12
  • 27. Table name: role: status: tim Tim teacher Cool name: role: kristen marketing Kristen Tuesday, October 16, 12
  • 28. Table name: role: status: tim Tim teacher Cool name: role: kristen marketing Kristen billy role: CEO Tuesday, October 16, 12
  • 29. Table name: role: status: tim Tim teacher Cool name: role: kristen marketing Kristen billy role: CEO name: role: status: matt Matt founder ubercool Tuesday, October 16, 12
  • 30. Table name: role: status: tim Tim teacher Cool name: role: kristen marketing Kristen billy role: CEO name: role: status: matt Matt founder ubercool Tuesday, October 16, 12
  • 31. Outer hash key Table name: role: status: tim Tim teacher Cool name: role: kristen marketing Kristen billy role: CEO name: role: status: matt Matt founder ubercool Tuesday, October 16, 12
  • 32. Outer hash key Table Inner hash key name: role: status: tim Tim teacher Cool name: role: kristen marketing Kristen billy role: CEO name: role: status: matt Matt founder ubercool Tuesday, October 16, 12
  • 33. Outer hash key Table Inner hash key name: role: status: tim Tim teacher Cool Sparse name: role: kristen marketing Kristen billy role: CEO name: role: status: matt Matt founder ubercool Tuesday, October 16, 12
  • 34. Database Accounts ClickStream Orders InventoryEvents Tuesday, October 16, 12
  • 35. Database Accounts ClickStream Tables Orders InventoryEvents Tuesday, October 16, 12
  • 36. Cluster System Database Application Database Tuesday, October 16, 12
  • 38. Secondary Indexes ‣ Ubiquitous in relational databases Tuesday, October 16, 12
  • 39. Secondary Indexes ‣ Ubiquitous in relational databases ‣ Supported in Cassandra, with qualifications Tuesday, October 16, 12
  • 40. Secondary Indexes name: email: role: tim Tim tb@a.com teacher name: email: kristen k@ds.com Kristen billy role: CEO name: email: status: matt Matt m@ds.com ubercool Tuesday, October 16, 12
  • 41. Secondary Indexes name: email: role: tim Tim tb@a.com teacher name: email: kristen k@ds.com Kristen billy role: CEO name: email: status: matt Matt m@ds.com ubercool Tuesday, October 16, 12
  • 42. Secondary Indexes name: email: role: tim Tim tb@a.com teacher name: email: kristen k@ds.com Kristen billy role: CEO name: email: status: matt Matt m@ds.com ubercool Tuesday, October 16, 12
  • 44. Secondary Indexes ‣ In relational databases: performant for high cardinality Tuesday, October 16, 12
  • 45. Secondary Indexes ‣ In relational databases: performant for high cardinality ‣ In Cassandra: the reverse Tuesday, October 16, 12
  • 46. Secondary Indexes ‣ In relational databases: performant for high cardinality ‣ In Cassandra: the reverse ‣ Not suitable for lookup-by-email Tuesday, October 16, 12
  • 47. Secondary Indexes ‣ In relational databases: performant for high cardinality ‣ In Cassandra: the reverse ‣ Not suitable for lookup-by-email ‣ Suitable for lookup by: region code, gender, state, etc. Tuesday, October 16, 12
  • 51. Why BigTable? Flexibility Tuesday, October 16, 12
  • 52. Why BigTable? Flexibility Performance Tuesday, October 16, 12
  • 53. Query Language for developers Tuesday, October 16, 12
  • 56. CQL (Cassandra Query Language) Tuesday, October 16, 12
  • 59. CREATE KEYSPACE CREATE KEYSPACE DemoKeyspace WITH strategy_class='SimpleStrategy' AND strategy_options:replication_factor=1; Tuesday, October 16, 12
  • 61. CREATE TABLE CREATE TABLE accounts (KEY text PRIMARY KEY) WITH comparator=text AND default_validation=text; Tuesday, October 16, 12
  • 63. CREATE TABLE CREATE TABLE accounts (KEY text PRIMARY KEY, name text, email text, signed_up_at timestamp) WITH comparator=text; Tuesday, October 16, 12
  • 66. INSERT INSERT INTO accounts (KEY, name, email, signed_up_at) VALUES ('tlberglund', 'Tim Berglund', 'tlberglund@gmail.com', '2012-04-25'); Tuesday, October 16, 12
  • 68. INSERT INSERT INTO events (KEY, 0, 1, 2, 3, 4) VALUES ('2012-04-25T11:04:34-0700', 55.4, 56.2, 59.6, 65.3, 79) USING CONSISTENCY QUORUM AND TTL 86400; Tuesday, October 16, 12
  • 71. SELECT SELECT * FROM accounts WHERE KEY='tlberglund'; Tuesday, October 16, 12
  • 73. SELECT SELECT 1..3 FROM events WHERE KEY='2012-04-25T11:04:34-0700'; Tuesday, October 16, 12
  • 75. SELECT SELECT * FROM accounts WHERE KEY='tlberglund' USING CONSISTENCY ONE; Tuesday, October 16, 12
  • 78. UPDATE UPDATE accounts SET last_login='2012-04-25T09:37:35-0700' WHERE KEY='tlberglund'; Tuesday, October 16, 12
  • 81. 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 82. Writing a Key Tuesday, October 16, 12
  • 83. 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 84. 0000 name: Tim E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 85. 0000 3D97: Tim E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 86. 0000 E000 2000 C000 4000 3D97: Tim A000 6000 8000 Tuesday, October 16, 12
  • 87. 0000 role: Teacher E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 88. 0000 9C4F: Teacher E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 89. 0000 E000 2000 C000 4000 A000 9C4F: Teacher 6000 8000 Tuesday, October 16, 12
  • 90. Reading a Key Tuesday, October 16, 12
  • 91. 0000 3D97? E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 92. 0000 3D97? E000 2000 name: Tim C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 93. 0000 9C4F? E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 94. 0000 9C4F? E000 2000 role: Teacher C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 97. Replication Replication factor (N) Tuesday, October 16, 12
  • 98. Replication Replication factor (N) Pluggable placement strategies Tuesday, October 16, 12
  • 100. Replication Strategies Simple Tuesday, October 16, 12
  • 101. Replication Strategies Simple Network Topology Aware Tuesday, October 16, 12
  • 102. Simple Strategy 0000 E000 2000 C000 4000 N=3 A000 6000 8000 Tuesday, October 16, 12
  • 103. Simple Strategy 0000 3D97: Tim E000 2000 C000 4000 N=3 A000 6000 8000 Tuesday, October 16, 12
  • 104. Simple Strategy 0000 E000 2000 C000 4000 3D97: Tim N=3 A000 6000 8000 Tuesday, October 16, 12
  • 105. Simple Strategy 0000 E000 2000 C000 4000 3D97: Tim N=3 A000 6000 3D97: Tim 8000 Tuesday, October 16, 12
  • 106. Simple Strategy 0000 E000 2000 C000 4000 3D97: Tim N=3 A000 6000 3D97: Tim 8000 3D97: Tim Tuesday, October 16, 12
  • 107. Topology Aware DC1 DC2 2000 0000 6000 4000 A000 8000 E000 C000 Tuesday, October 16, 12
  • 108. Topology Aware DC1 DC2 2000 0000 6000 4000 A000 8000 E000 C000 Tuesday, October 16, 12
  • 109. Topology Aware 3D97: Tim DC1 DC2 2000 0000 6000 4000 A000 8000 E000 C000 Tuesday, October 16, 12
  • 110. Topology Aware DC1 DC2 2000 0000 6000 4000 3D97: Tim A000 8000 E000 C000 Tuesday, October 16, 12
  • 111. Topology Aware DC1 DC2 2000 0000 6000 3D97: Tim 4000 3D97: Tim A000 8000 E000 C000 Tuesday, October 16, 12
  • 112. Topology Aware DC1 DC2 2000 0000 6000 3D97: Tim 4000 3D97: Tim A000 8000 3D97: Tim E000 C000 Tuesday, October 16, 12
  • 113. Topology Aware DC1 DC2 2000 0000 6000 3D97: Tim 4000 3D97: Tim A000 3D97: Tim 8000 3D97: Tim E000 C000 Tuesday, October 16, 12
  • 116. Client Connections 0000 E000 2000 C000 4000 Client Machine A000 6000 8000 Tuesday, October 16, 12
  • 117. Client Connections 0000 E000 2000 Client Machine ? C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 118. Client Connections 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 119. Client Connections 0000 E000 2000 3D97? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 120. Client Connections 0000 E000 2000 3D97? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 121. Client Connections 0000 E000 2000 3D97? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 122. Client Connections 0000 E000 2000 3D97? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 123. Client Connections 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 124. Client Connections 0000 E000 2000 9C4F? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 125. Client Connections 0000 E000 2000 9C4F? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 126. Client Connections 0000 E000 2000 9C4F? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 127. Client Connections 0000 E000 2000 9C4F? C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 129. Load Balancer Hardware VIP Tuesday, October 16, 12
  • 130. Load Balancer Hardware VIP HAProxy Tuesday, October 16, 12
  • 131. Load Balancer Hardware VIP HAProxy Round-robin DNS Tuesday, October 16, 12
  • 132. Load Balancer Hardware VIP HAProxy Round-robin DNS Client-side Tuesday, October 16, 12
  • 133. Load Balancer Hardware VIP HAProxy Round-robin DNS Client-side (Hector does this) Tuesday, October 16, 12
  • 135. Write Consistency 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 136. Write Consistency 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 14C7 Tuesday, October 16, 12
  • 137. Write Consistency 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 14C7 Tuesday, October 16, 12
  • 138. Write Consistency 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 14C7 Tuesday, October 16, 12
  • 139. Write Consistency 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 14C7 Tuesday, October 16, 12
  • 140. Write Consistency 0000 E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 14C7 Tuesday, October 16, 12
  • 141. Write Consistency 0000 Coordinator E000 2000 C000 4000 Client Machine A000 6000 Load Balancer 8000 14C7 Tuesday, October 16, 12
  • 142. Write Consistency 0000 E000 ---- C000 4000 Client Machine A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 143. Write Consistency 0000 E000 ---- C000 4000 Client Machine 14C7 A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 144. Write Consistency 0000 E000 ---- X C000 4000 Client Machine 14C7 A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 145. Write Consistency 0000 E000 ---- X C000 4000 Client Machine 14C7 A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 146. Write Consistency 0000 E000 ---- X C000 4000 Client Machine 14C7 A000 6000 Load Balancer 8000 Tuesday, October 16, 12
  • 147. Write Consistency 0000 E000 ---- X C000 4000 Client Machine 14C7 A000 6000 Load Balancer 8000 Coordinator stores a “hint” Tuesday, October 16, 12
  • 149. Write Consistency ‣ ANY At least one node (hinted handoffs allowed) Tuesday, October 16, 12
  • 150. Write Consistency ‣ ANY At least one node (hinted handoffs allowed) ‣ ONE At least one node (no hinted handoffs) Tuesday, October 16, 12
  • 151. Write Consistency ‣ ANY At least one node (hinted handoffs allowed) ‣ ONE At least one node (no hinted handoffs) ‣ QUORUM (N/2)+1 nodes Tuesday, October 16, 12
  • 153. Write Consistency ‣ LOCAL_QUORUM (N/2)+1 nodes in this availability zone Tuesday, October 16, 12
  • 154. Write Consistency ‣ LOCAL_QUORUM (N/2)+1 nodes in this availability zone ‣ EACH_QUORUM (N/2)+1 nodes in all availability zones Tuesday, October 16, 12
  • 155. Write Consistency ‣ LOCAL_QUORUM (N/2)+1 nodes in this availability zone ‣ EACH_QUORUM (N/2)+1 nodes in all availability zones ‣ ALL Write successfully to all replicas Tuesday, October 16, 12
  • 157. Read Consistency 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 158. Read Consistency 9C4F? 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 159. Read Consistency 9C4F? 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 160. Read Consistency 9C4F? 0000 E000 2000 Tim TODAY C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 161. Read Consistency 9C4F? 0000 E000 2000 Tim TODAY Tim TODAY C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 162. Read Consistency 9C4F? 0000 E000 2000 Tim TODAY Tim TODAY C000 4000 Tim TODAY A000 6000 8000 Tuesday, October 16, 12
  • 163. Read Consistency Coordinator 9C4F? 0000 E000 2000 Tim TODAY Tim TODAY C000 4000 Tim TODAY A000 6000 8000 Tuesday, October 16, 12
  • 164. Read Consistency 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 165. Read Consistency 9C4F? 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 166. Read Consistency 9C4F? 0000 E000 2000 Tim TODAY C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 167. Read Consistency 9C4F? 0000 E000 2000 Tim TODAY Jim YESTERDAY C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 168. Read Consistency 9C4F? 0000 E000 2000 Tim TODAY Jim YESTERDAY C000 4000 Tim TODAY A000 6000 8000 Tuesday, October 16, 12
  • 169. Read Consistency 9C4F? 0000 E000 2000 Tim TODAY Jim YESTERDAY C000 4000 Tim TODAY A000 6000 8000 Inconsistent Tuesday, October 16, 12
  • 170. Passive Read Repair Tuesday, October 16, 12
  • 171. Passive Read Repair ‣ Initiated by coordinator Tuesday, October 16, 12
  • 172. Passive Read Repair ‣ Initiated by coordinator ‣ Cleans up entropy in a single row Tuesday, October 16, 12
  • 173. Passive Read Repair ‣ Initiated by coordinator ‣ Cleans up entropy in a single row ‣ Happens regardless of consistency level Tuesday, October 16, 12
  • 174. Passive Read Repair ‣ Initiated by coordinator ‣ Cleans up entropy in a single row ‣ Happens regardless of consistency level ‣ Just reading the database reduces its entropy Tuesday, October 16, 12
  • 176. Read Consistency ‣ ONE Get response from the closest replica Tuesday, October 16, 12
  • 177. Read Consistency ‣ ONE Get response from the closest replica ‣ QUORUM Get (N/2)+1 nodes, return most recent timestamp Tuesday, October 16, 12
  • 179. Read Consistency ‣ LOCAL_QUORUM (N/2)+1 nodes in this availability zone Tuesday, October 16, 12
  • 180. Read Consistency ‣ LOCAL_QUORUM (N/2)+1 nodes in this availability zone ‣ EACH_QUORUM (N/2)+1 nodes in all availability zones Tuesday, October 16, 12
  • 181. Read Consistency ‣ LOCAL_QUORUM (N/2)+1 nodes in this availability zone ‣ EACH_QUORUM (N/2)+1 nodes in all availability zones ‣ ALL Wait for all replicas to respond Tuesday, October 16, 12
  • 183. But what about Column Families? Tuesday, October 16, 12
  • 184. Replication For Real Tuesday, October 16, 12
  • 185. Replication For Real Rows are replicated Tuesday, October 16, 12
  • 186. Replication For Real Rows are replicated “Key” is row key Tuesday, October 16, 12
  • 187. Replication For Real Rows are replicated “Key” is row key “Value” is the row data Tuesday, October 16, 12
  • 188. Replication For Real Rows are replicated “Key” is row key “Value” is the row data Implications for row size Tuesday, October 16, 12
  • 191. Gossip ‣ Naive heartbeats don’t scale Tuesday, October 16, 12
  • 192. Gossip ‣ Naive heartbeats don’t scale ‣ Cluster still needs to know its state Tuesday, October 16, 12
  • 193. Gossip ‣ Naive heartbeats don’t scale ‣ Cluster still needs to know its state ‣ Computes a real-valued “suspicion” for each node Tuesday, October 16, 12
  • 194. Gossip ‣ Naive heartbeats don’t scale ‣ Cluster still needs to know its state ‣ Computes a real-valued “suspicion” for each node ‣ Probabilistic Tuesday, October 16, 12
  • 195. Gossip ‣ Naive heartbeats don’t scale ‣ Cluster still needs to know its state ‣ Computes a real-valued “suspicion” for each node ‣ Probabilistic ‣ Just like the real thing Tuesday, October 16, 12
  • 196. 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 197. 0000 E000 2000 How are you, E000? C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 198. 0000 E000 2000 I'm cool. C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 199. 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 200. 0000 E000 2000 C000 4000 How about you, A000? A000 6000 8000 Tuesday, October 16, 12
  • 201. 0000 E000 2000 C000 4000 Oh, I'm fine. A000 6000 8000 Tuesday, October 16, 12
  • 202. 0000 E000 2000 I trust e000. Not so sure about A000... C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 203. 0000 E000 2000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 204. 0000 E000 C000, what 2000 do you know? C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 205. 0000 E000 2000 I'm Great! C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 206. 0000 E000 2000 E000 is doing well. C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 207. 0000 E000 2000 Poor A000 is having trouble lately. C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 208. 0000 E000 2000 Hmmm, so... C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 209. 0000 E000 2000 ✔ C000 ✔ E000 ✘ A000 C000 4000 A000 6000 8000 Tuesday, October 16, 12
  • 211. Gossip Config ‣ A new node needs “seed nodes” Tuesday, October 16, 12
  • 212. Gossip Config ‣ A new node needs “seed nodes” ‣ Seed nodes configured in $CASSANDRA_HOME/conf/ cassandra.yaml Tuesday, October 16, 12
  • 215. Storage Engine ‣ “Log-structured storage” Tuesday, October 16, 12
  • 216. Storage Engine ‣ “Log-structured storage” ‣ All writes are sequential Tuesday, October 16, 12
  • 217. Storage Engine ‣ “Log-structured storage” ‣ All writes are sequential ‣ All writes are immutable Tuesday, October 16, 12
  • 218. Storage Engine ‣ “Log-structured storage” ‣ All writes are sequential ‣ All writes are immutable ‣ Designed to avoid seeks Tuesday, October 16, 12
  • 219. Storage Engine ‣ “Log-structured storage” ‣ All writes are sequential ‣ All writes are immutable ‣ Designed to avoid seeks ‣ Writes are faster than reads Tuesday, October 16, 12
  • 221. Write Sequence Write from Commit Coordinator Log Tuesday, October 16, 12
  • 222. Write Sequence Write from Commit Coordinator Log Memtable Tuesday, October 16, 12
  • 223. Write Sequence Write from Commit Coordinator Log SSTable(s) Memtable (lots of tuning here) Tuesday, October 16, 12
  • 224. Write Sequence Write from Commit Coordinator Log Compaction (more tuning here) SSTable(s) Memtable (lots of tuning here) Tuesday, October 16, 12
  • 226. Commit Log ‣ Writes go here first Tuesday, October 16, 12
  • 227. Commit Log ‣ Writes go here first ‣ Commit succeeds before node reports successful write Tuesday, October 16, 12
  • 228. Commit Log ‣ Writes go here first ‣ Commit succeeds before node reports successful write ‣ Append-only, sequential writes Tuesday, October 16, 12
  • 229. Commit Log ‣ Writes go here first ‣ Commit succeeds before node reports successful write ‣ Append-only, sequential writes ‣ One per server Tuesday, October 16, 12
  • 230. Commit Log ‣ Writes go here first ‣ Commit succeeds before node reports successful write ‣ Append-only, sequential writes ‣ One per server ‣ Good to have a dedicated spindle Tuesday, October 16, 12
  • 232. MemTable ‣ An in-memory structure Tuesday, October 16, 12
  • 233. MemTable ‣ An in-memory structure ‣ One per column family Tuesday, October 16, 12
  • 234. MemTable ‣ An in-memory structure ‣ One per column family ‣ Holds most recent row changes Tuesday, October 16, 12
  • 235. MemTable ‣ An in-memory structure ‣ One per column family ‣ Holds most recent row changes ‣ Tunable memory use Tuesday, October 16, 12
  • 236. MemTable ‣ An in-memory structure ‣ One per column family ‣ Holds most recent row changes ‣ Tunable memory use ‣ Flushed to disk when “full” Tuesday, October 16, 12
  • 238. SSTable ‣ Memtables flushed to disk here Tuesday, October 16, 12
  • 239. SSTable ‣ Memtables flushed to disk here ‣ Many SSTable files per column family Tuesday, October 16, 12
  • 240. SSTable ‣ Memtables flushed to disk here ‣ Many SSTable files per column family ‣ Every SSTable is immutable Tuesday, October 16, 12
  • 241. SSTable ‣ Memtables flushed to disk here ‣ Many SSTable files per column family ‣ Every SSTable is immutable ‣ SSTables are accessed during reads Tuesday, October 16, 12
  • 242. SSTable ‣ Memtables flushed to disk here ‣ Many SSTable files per column family ‣ Every SSTable is immutable ‣ SSTables are accessed during reads ‣ Must be compacted Tuesday, October 16, 12
  • 243. Read Sequence Read from Coordinator Tuesday, October 16, 12
  • 244. Read Sequence Memtable Tuesday, October 16, 12
  • 245. Read Sequence Memtable Are all columns here? Tuesday, October 16, 12
  • 246. Read Sequence Memtable Are all columns here? YES: stop and return result. Tuesday, October 16, 12
  • 247. Read Sequence Memtable Are all columns here? YES: stop and return result. NO: continue. Tuesday, October 16, 12
  • 248. Read Sequence Newest SSTable Tuesday, October 16, 12
  • 249. Read Sequence Newest SSTable Are all columns here? Tuesday, October 16, 12
  • 250. Read Sequence Newest SSTable Are all columns here? YES: stop and return result. Tuesday, October 16, 12
  • 251. Read Sequence Newest SSTable Are all columns here? YES: stop and return result. NO: continue. Tuesday, October 16, 12
  • 252. Read Sequence Next Oldest SSTable Tuesday, October 16, 12
  • 253. Read Sequence Next Oldest SSTable How about now? Tuesday, October 16, 12
  • 254. Read Sequence Next Oldest SSTable How about now? YES: great! Tuesday, October 16, 12
  • 255. Read Sequence Next Oldest SSTable How about now? YES: great! NO: keep looking... Tuesday, October 16, 12
  • 256. Read Sequence And so on. Tuesday, October 16, 12
  • 257. Read Sequence And so on. Doesn’t this get old? Tuesday, October 16, 12
  • 259. Read Sequence ‣ Check Memtable first Tuesday, October 16, 12
  • 260. Read Sequence ‣ Check Memtable first ‣ Read SSTables from newest to oldest Tuesday, October 16, 12
  • 261. Read Sequence ‣ Check Memtable first ‣ Read SSTables from newest to oldest ‣ Bloom filters prevent most reads Tuesday, October 16, 12
  • 262. Read Sequence ‣ Check Memtable first ‣ Read SSTables from newest to oldest ‣ Bloom filters prevent most reads ‣ Compaction shrinks number of files Tuesday, October 16, 12
  • 265. Compaction ‣ Combine many SSTables into one Tuesday, October 16, 12
  • 266. Compaction ‣ Combine many SSTables into one ‣ Performed in the background Tuesday, October 16, 12
  • 267. Compaction ‣ Combine many SSTables into one ‣ Performed in the background ‣ Node still operates Tuesday, October 16, 12
  • 268. Compaction ‣ Combine many SSTables into one ‣ Performed in the background ‣ Node still operates ‣ Requires extra disk space Tuesday, October 16, 12
  • 269. Compaction ‣ Combine many SSTables into one ‣ Performed in the background ‣ Node still operates ‣ Requires extra disk space ‣ Three tunable varieties Tuesday, October 16, 12
  • 271. Compaction ‣ Major All SSTables are merged into one clean one Tuesday, October 16, 12
  • 272. Compaction ‣ Major All SSTables are merged into one clean one ‣ Minor Similarly-sized SSTables are merged together after reaching a threshold Tuesday, October 16, 12
  • 273. Compaction ‣ Major All SSTables are merged into one clean one ‣ Minor Similarly-sized SSTables are merged together after reaching a threshold ‣ Leveled http://www.datastax.com/dev/blog/leveled-compaction-in-apache-cassandra Tuesday, October 16, 12
  • 274. http://www.datastax.com/ products/community Tuesday, October 16, 12
  • 275. http://www.datastax.com/ products/community t hi s! n lo ad D ow Tuesday, October 16, 12
  • 276. Tim Berglund tlberglund@github.com @tlberglund Thank You Tuesday, October 16, 12