0. はじめに
・本記載は、Cassandra-stressについて、ディストリビューション、バージョン = DSE 4.8.16をベースに記載
・記載が古いバージョン対応だし、サブコマンドやオプションはもっといろいろあるので、詳しくはapacheやdatastaxのガイドを参照
1. userモードの概要
・userモードを使用することで、任意のテーブルとそのテーブルに対する任意のクエリのストレステストを実行可能
2. yamlファイル定義
yamlファイルの定義説明
① keyspace: keyspace_definition:
・キースペース指定
・CREATE KEYSPACEでのキースペース定義
※既に作成済みのキースペースを利用する場合、キースペース定義は省略可能
< 定義の記載例 >
keyspace: stest01
keyspace_definition:
CREATE KEYSPACE stest01 WITH replication = {'class': 'NetworkTopologyStrategy', 'DC1': 3};
② table: table_definition:
・テーブル名の定義
・CREATE TABLEでのテーブル定義
※既に作成済みのテーブルを利用する場合、テーブル定義は省略可能
< 定義の記載例 >
table: test01
table_definition: |
CREATE TABLE test01 (
name text,
date timestamp,
address inet,
ino int,
iname text,
uid timeuuid,
PRIMARY KEY(name, date)
) WITH compaction = { 'class':'SizeTieredCompactionStrategy' }
③ columnspec:
・テーブルの列フィールドの長さや値の範囲、partitionごとのclustering行数のカスタマイズを指定
< 定義の記載例 >
columnspec:
- name: name
size: uniform(1..20)
- name: date
cluster: fixed(1)
- name: address
population: exp(1..4)
- name: ino
population: exp(1..1000)
- name: iname
size: fixed(8)
population: uniform(1..5)
< データパターン指定の項目 >
| パラメタ | 意味 | デフォルト |
|---|---|---|
| size | text,blob,set,listのcolumnのフィールドの長さを定義、例えば、fixed(5)指定すると、固定で長さ5になるし、uniform(1..10)とすると長さ1～10の分布となる。 | UNIFORM(4..8) |
| population | 列の値が変動する範囲を定義、テキスト項目だと母集団の数の指定になるし、数値的な項目だと指定した範囲の値となる。例えば、text型でuniform(1..10)とすると値の母集団10とおりとなるし、int型でfixed(10)だと値がすべて10になる。 | UNIFORM(1..100B) |
| cluster | 各partitionごとのclustering keyの数を制御。例えば、fixed(4)指定すると、partitionごとにclustering key 4レコードずつ生成 | FIXED(1) |
< 分布の指定 >
| パラメタ | 意味 | |
|---|---|---|
| EXP(min..max) | 最小～最大範囲の指数分布(ランダム) | |
| EXTREME(min..max,shape) | 最小～最大範囲の極値分布 | |
| GAUSSIAN(min..max,stdvrng) | 指定された標準偏差による正規分布 | |
| GAUSSIAN(min..max,mean,stdvrng) | 指定された標準偏差による正規分布 | |
| UNIFORM(min..max) | 最小～最大範囲のの均一分布 | |
| FIXED(val) | 常に同じ値を返す |
④ insert:
・ストレステスト実行時のデータの挿入方法を指定
例えば、fixed(10)とすると、固定数の行のinsertを実行
< 定義の記載例 >
insert:
partitions: fixed(10)
batchtype: UNLOGGED
< 指定項目 >
| 指定 | 意味 | デフォルト |
|---|---|---|
| Partitions | insertするパーティションの数 | FIXED(1) |
| Batch type | 使用するCQLバッチのタイプ UNLOGGED指定でパフォーマンスを向上可能 | LOGGED |
⑤ queries:
・テーブルのCQLクエリを指定
・fieldsは、バインド変数を同じ行から選択するか、パーティション内のすべての行から選択するかを指定
< 定義の記載例 >
queries:
select1:
cql: select address from test01 where name = ? and date = ?
fields: samerow
3. 実行方法
・YAMLファイル指定なので、userオプションを指定し、profile=でyamlファイルを指定
・opsで実行する操作とそれぞれの操作数を指定 ※"("と")"のところでエスケープバックスラッシュ要
・duration=やn=で実行時間や操作回数を指定
< サブオプション,追加パラメタの一部 >
| パラメタ | パラメタ種別 | 意味 | デフォルト | 備考 |
|---|---|---|---|---|
| -rate | サブオプション | threads=?とかthreads<=?とかで実行スレッド数を指定 | 4スレッドから開始し、スループットが飽和するまでスレッドを増加して実行していく(最大1000まで) | ・「4.実行例」を参照 |
| profile=? | 追加パラメタ | userオプション指定時、YAMLファイルを指定 | - | |
| truncate=? | 追加パラメタ | 作成したテーブルをtruncateする | never(once,alwaysの選択可) | |
| duration=? | 追加パラメタ | 1回あたりの実行時間を指定 秒/分/時間で指定 | - | ・n=と同時指定はできない |
| n=? | 追加パラメタ | 実行する操作回数 | - | ・durationと同時指定はできない。例えばn=100でinsert=3,simpe1=1なら、insert:75回、simple1操作:25回実行 |
| n>? | 追加パラメタ | 不確定収束を受け入れる前に実行する最小実行回数を指定 | - | ・durationと同時指定はできない |
| n<? | 追加パラメタ | 不確定収束を受け入れる前に実行する最大実行回数を指定 | - | ・durationと同時指定はできない |
4.実行例
① insert 100回実行を1スレッド固定で実行
# cassandra-stress user profile=stest02.yaml ops\(insert=1\) n=100 -rate threads=1
Connected to cluster: Test Cluster
Datatacenter: DC1; Host: localhost/127.0.0.1; Rack: RAC1
Created schema. Sleeping 1s for propagation.
Sleeping 2s...
Running [insert] with 1 threads for 100 iteration
type, total ops, op/s, pk/s, row/s, mean, med, .95, .99, .999, max, time, stderr, errors, gc: #, max ms, sum ms, sdv ms, mb
Generating batches with [5..5] partitions and [20..20] rows (of [20..20] total rows in the partitions)
total, 20, 90, 452, 1809, 10.5, 8.1, 21.0, 21.0, 21.0, 21.0, 0.2, 0.00000, 0, 0, 0, 0, 0, 0
Results:
op rate : 90 [insert:90]
partition rate : 452 [insert:452]
row rate : 1809 [insert:1809]
latency mean : 10.5 [insert:10.5]
latency median : 8.1 [insert:8.1]
latency 95th percentile : 21.0 [insert:21.0]
latency 99th percentile : 21.0 [insert:21.0]
latency 99.9th percentile : 21.0 [insert:21.0]
latency max : 21.0 [insert:21.0]
Total partitions : 100 [insert:100]
Total errors : 0 [insert:0]
total gc count : 0
total gc mb : 0
total gc time (s) : 0
avg gc time(ms) : NaN
stdev gc time(ms) : 0
Total operation time : 00:00:00
END
#
② insert 100回実行をスレッド条件増加しながら16スレッドまで
# cassandra-stress user profile=stest02.yaml ops\(insert=1\) n=100 -rate threads\<=16
Connected to cluster: Test Cluster
Datatacenter: DC1; Host: localhost/127.0.0.1; Rack: RAC1
Created schema. Sleeping 1s for propagation.
Sleeping 2s...
Running with 4 threadCount
Running [insert] with 4 threads for 100 iteration
type, total ops, op/s, pk/s, row/s, mean, med, .95, .99, .999, max, time, stderr, errors, gc: #, max ms, sum ms, sdv ms, mb
Generating batches with [5..5] partitions and [20..20] rows (of [20..20] total rows in the partitions)
total, 20, 95, 477, 1906, 39.4, 31.7, 125.7, 125.7, 125.7, 125.7, 0.2, 0.00000, 0, 0, 0, 0, 0, 0
Results:
op rate : 95 [insert:95]
partition rate : 477 [insert:477]
row rate : 1906 [insert:1906]
latency mean : 39.4 [insert:39.4]
latency median : 31.7 [insert:31.7]
latency 95th percentile : 125.7 [insert:125.7]
latency 99th percentile : 125.7 [insert:125.7]
latency 99.9th percentile : 125.7 [insert:125.7]
latency max : 125.7 [insert:125.7]
Total partitions : 100 [insert:100]
Total errors : 0 [insert:0]
total gc count : 0
total gc mb : 0
total gc time (s) : 0
avg gc time(ms) : NaN
stdev gc time(ms) : 0
Total operation time : 00:00:00
Sleeping for 15s
Running with 8 threadCount
Running [insert] with 8 threads for 100 iteration
type, total ops, op/s, pk/s, row/s, mean, med, .95, .99, .999, max, time, stderr, errors, gc: #, max ms, sum ms, sdv ms, mb
total, 20, 247, 1234, 4937, 51.8, 53.0, 78.6, 78.6, 78.6, 78.6, 0.1, 0.00000, 0, 0, 0, 0, 0, 0
Results:
op rate : 247 [insert:247]
partition rate : 1234 [insert:1234]
row rate : 4937 [insert:4937]
latency mean : 51.8 [insert:51.8]
latency median : 53.0 [insert:53.0]
latency 95th percentile : 78.6 [insert:78.6]
latency 99th percentile : 78.6 [insert:78.6]
latency 99.9th percentile : 78.6 [insert:78.6]
latency max : 78.6 [insert:78.6]
Total partitions : 100 [insert:100]
Total errors : 0 [insert:0]
total gc count : 0
total gc mb : 0
total gc time (s) : 0
avg gc time(ms) : NaN
stdev gc time(ms) : 0
Total operation time : 00:00:00
Improvement over 4 threadCount: 159%
Sleeping for 15s
Running with 16 threadCount
Running [insert] with 16 threads for 100 iteration
type, total ops, op/s, pk/s, row/s, mean, med, .95, .99, .999, max, time, stderr, errors, gc: #, max ms, sum ms, sdv ms, mb
total, 20, 208, 1042, 4166, 109.1, 118.0, 148.4, 148.4, 148.4, 148.4, 0.1, 0.00000, 0, 0, 0, 0, 0, 0
Results:
op rate : 208 [insert:208]
partition rate : 1042 [insert:1042]
row rate : 4166 [insert:4166]
latency mean : 109.1 [insert:109.1]
latency median : 118.0 [insert:118.0]
latency 95th percentile : 148.4 [insert:148.4]
latency 99th percentile : 148.4 [insert:148.4]
latency 99.9th percentile : 148.4 [insert:148.4]
latency max : 148.4 [insert:148.4]
Total partitions : 100 [insert:100]
Total errors : 0 [insert:0]
total gc count : 0
total gc mb : 0
total gc time (s) : 0
avg gc time(ms) : NaN
stdev gc time(ms) : 0
Total operation time : 00:00:00
Improvement over 8 threadCount: -15%
id, type, total ops, op/s, pk/s, row/s, mean, med, .95, .99, .999, max, time, stderr, errors, gc: #, max ms, sum ms, sdv ms, mb
4 threadCount, insert, 20, 95, 477, 1906, 39.4, 31.7, 125.7, 125.7, 125.7, 125.7, 0.2, 0.00000, 0, 0, 0, 0, 0, 0
4 threadCount, total, 20, 95, 477, 1906, 39.4, 31.7, 125.7, 125.7, 125.7, 125.7, 0.2, 0.00000, 0, 0, 0, 0, 0, 0
8 threadCount, insert, 20, 247, 1234, 4937, 51.8, 53.0, 78.6, 78.6, 78.6, 78.6, 0.1, 0.00000, 0, 0, 0, 0, 0, 0
8 threadCount, total, 20, 247, 1234, 4937, 51.8, 53.0, 78.6, 78.6, 78.6, 78.6, 0.1, 0.00000, 0, 0, 0, 0, 0, 0
16 threadCount, insert, 20, 208, 1042, 4166, 109.1, 118.0, 148.4, 148.4, 148.4, 148.4, 0.1, 0.00000, 0, 0, 0, 0, 0, 0
16 threadCount, total, 20, 208, 1042, 4166, 109.1, 118.0, 148.4, 148.4, 148.4, 148.4, 0.1, 0.00000, 0, 0, 0, 0, 0, 0
END
#
② insert 100回実行をスレッド条件増加しながら実行 ( 飽和するまで )
※-rate auto指定と同じ (たぶん)
# cassandra-stress user profile=stest02.yaml ops\(insert=1\) n=100
Connected to cluster: Test Cluster
Datatacenter: DC1; Host: localhost/127.0.0.1; Rack: RAC1
Created schema. Sleeping 1s for propagation.
Sleeping 2s...
Running with 4 threadCount
Running [insert] with 4 threads for 100 iteration
type, total ops, op/s, pk/s, row/s, mean, med, .95, .99, .999, max, time, stderr, errors, gc: #, max ms, sum ms, sdv ms, mb
Generating batches with [5..5] partitions and [20..20] rows (of [20..20] total rows in the partitions)
total, 20, 97, 486, 1943, 38.0, 24.9, 150.4, 150.4, 150.4, 150.4, 0.2, 0.00000, 0, 0, 0, 0, 0, 0
Results:
op rate : 97 [insert:97]
partition rate : 486 [insert:486]
row rate : 1943 [insert:1943]
latency mean : 38.0 [insert:38.0]
latency median : 24.9 [insert:24.9]
latency 95th percentile : 150.4 [insert:150.4]
latency 99th percentile : 150.4 [insert:150.4]
latency 99.9th percentile : 150.4 [insert:150.4]
latency max : 150.4 [insert:150.4]
Total partitions : 100 [insert:100]
Total errors : 0 [insert:0]
total gc count : 0
total gc mb : 0
total gc time (s) : 0
avg gc time(ms) : NaN
stdev gc time(ms) : 0
Total operation time : 00:00:00
Sleeping for 15s
Running with 8 threadCount
Running [insert] with 8 threads for 100 iteration
type, total ops, op/s, pk/s, row/s, mean, med, .95, .99, .999, max, time, stderr, errors, gc: #, max ms, sum ms, sdv ms, mb
total, 20, 263, 1317, 5270, 44.3, 47.8, 75.5, 75.5, 75.5, 75.5, 0.1, 0.00000, 0, 0, 0, 0, 0, 0
Results:
op rate : 263 [insert:263]
partition rate : 1317 [insert:1317]
row rate : 5270 [insert:5270]
latency mean : 44.3 [insert:44.3]
latency median : 47.8 [insert:47.8]
latency 95th percentile : 75.5 [insert:75.5]
latency 99th percentile : 75.5 [insert:75.5]
latency 99.9th percentile : 75.5 [insert:75.5]
latency max : 75.5 [insert:75.5]
Total partitions : 100 [insert:100]
Total errors : 0 [insert:0]
total gc count : 0
total gc mb : 0
total gc time (s) : 0
avg gc time(ms) : NaN
stdev gc time(ms) : 0
Total operation time : 00:00:00
Improvement over 4 threadCount: 171%
Sleeping for 15s
Running with 16 threadCount
Running [insert] with 16 threads for 100 iteration
type, total ops, op/s, pk/s, row/s, mean, med, .95, .99, .999, max, time, stderr, errors, gc: #, max ms, sum ms, sdv ms, mb
total, 20, 259, 1296, 5185, 83.1, 85.4, 122.9, 122.9, 122.9, 122.9, 0.1, 0.00000, 0, 0, 0, 0, 0, 0
Results:
op rate : 259 [insert:259]
partition rate : 1296 [insert:1296]
row rate : 5185 [insert:5185]
latency mean : 83.1 [insert:83.1]
latency median : 85.4 [insert:85.4]
latency 95th percentile : 122.9 [insert:122.9]
latency 99th percentile : 122.9 [insert:122.9]
latency 99.9th percentile : 122.9 [insert:122.9]
latency max : 122.9 [insert:122.9]
Total partitions : 100 [insert:100]
Total errors : 0 [insert:0]
total gc count : 0
total gc mb : 0
total gc time (s) : 0
avg gc time(ms) : NaN
stdev gc time(ms) : 0
Total operation time : 00:00:00
Improvement over 8 threadCount: -2%
Sleeping for 15s
Running with 24 threadCount
Running [insert] with 24 threads for 100 iteration
type, total ops, op/s, pk/s, row/s, mean, med, .95, .99, .999, max, time, stderr, errors, gc: #, max ms, sum ms, sdv ms, mb
total, 20, 4137, 20685, 82738, 134.0, 132.2, 159.2, 159.2, 159.2, 159.2, 0.0, -0.00000, 0, 0, 0, 0, 0, 0
Results:
op rate : 4137 [insert:4137]
partition rate : 20685 [insert:20685]
row rate : 82738 [insert:82738]
latency mean : 134.0 [insert:134.0]
latency median : 132.2 [insert:132.2]
latency 95th percentile : 159.2 [insert:159.2]
latency 99th percentile : 159.2 [insert:159.2]
latency 99.9th percentile : 159.2 [insert:159.2]
latency max : 159.2 [insert:159.2]
Total partitions : 100 [insert:100]
Total errors : 0 [insert:0]
total gc count : 0
total gc mb : 0
total gc time (s) : 0
avg gc time(ms) : NaN
stdev gc time(ms) : 0
Total operation time : 00:00:00
Improvement over 16 threadCount: 1497%
Sleeping for 15s
Running with 36 threadCount
Running [insert] with 36 threads for 100 iteration
type, total ops, op/s, pk/s, row/s, mean, med, .95, .99, .999, max, time, stderr, errors, gc: #, max ms, sum ms, sdv ms, mb
Results:
op rate : NaN [insert:NaN]
partition rate : NaN [insert:NaN]
row rate : NaN [insert:NaN]
latency mean : NaN [insert:NaN]
latency median : 0.0 [insert:0.0]
latency 95th percentile : 0.0 [insert:0.0]
latency 99th percentile : 0.0 [insert:0.0]
latency 99.9th percentile : 0.0 [insert:0.0]
latency max : 0.0 [insert:0.0]
Total partitions : 0 [insert:0]
Total errors : 0 [insert:0]
total gc count : 0
total gc mb : 0
total gc time (s) : 0
avg gc time(ms) : NaN
stdev gc time(ms) : 0
Total operation time : 00:00:00
Improvement over 24 threadCount: -100%
Sleeping for 15s
Running with 54 threadCount
Running [insert] with 54 threads for 100 iteration
type, total ops, op/s, pk/s, row/s, mean, med, .95, .99, .999, max, time, stderr, errors, gc: #, max ms, sum ms, sdv ms, mb
Results:
op rate : NaN [insert:NaN]
partition rate : NaN [insert:NaN]
row rate : NaN [insert:NaN]
latency mean : NaN [insert:NaN]
latency median : 0.0 [insert:0.0]
latency 95th percentile : 0.0 [insert:0.0]
latency 99th percentile : 0.0 [insert:0.0]
latency 99.9th percentile : 0.0 [insert:0.0]
latency max : 0.0 [insert:0.0]
Total partitions : 0 [insert:0]
Total errors : 0 [insert:0]
total gc count : 0
total gc mb : 0
total gc time (s) : 0
avg gc time(ms) : NaN
stdev gc time(ms) : 0
Total operation time : 00:00:00
Improvement over 36 threadCount: NaN%
Sleeping for 15s
id, type, total ops, op/s, pk/s, row/s, mean, med, .95, .99, .999, max, time, stderr, errors, gc: #, max ms, sum ms, sdv ms, mb
4 threadCount, insert, 20, 97, 486, 1943, 38.0, 24.9, 150.4, 150.4, 150.4, 150.4, 0.2, 0.00000, 0, 0, 0, 0, 0, 0
4 threadCount, total, 20, 97, 486, 1943, 38.0, 24.9, 150.4, 150.4, 150.4, 150.4, 0.2, 0.00000, 0, 0, 0, 0, 0, 0
8 threadCount, insert, 20, 263, 1317, 5270, 44.3, 47.8, 75.5, 75.5, 75.5, 75.5, 0.1, 0.00000, 0, 0, 0, 0, 0, 0
8 threadCount, total, 20, 263, 1317, 5270, 44.3, 47.8, 75.5, 75.5, 75.5, 75.5, 0.1, 0.00000, 0, 0, 0, 0, 0, 0
16 threadCount, insert, 20, 259, 1296, 5185, 83.1, 85.4, 122.9, 122.9, 122.9, 122.9, 0.1, 0.00000, 0, 0, 0, 0, 0, 0
16 threadCount, total, 20, 259, 1296, 5185, 83.1, 85.4, 122.9, 122.9, 122.9, 122.9, 0.1, 0.00000, 0, 0, 0, 0, 0, 0
24 threadCount, insert, 20, 4137, 20685, 82738, 134.0, 132.2, 159.2, 159.2, 159.2, 159.2, 0.0, -0.00000, 0, 0, 0, 0, 0, 0
24 threadCount, total, 20, 4137, 20685, 82738, 134.0, 132.2, 159.2, 159.2, 159.2, 159.2, 0.0, -0.00000, 0, 0, 0, 0, 0, 0
36 threadCount, insert, 0, NaN, NaN, NaN, NaN, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.00000, 0, 0, 0, 0, 0, 0
36 threadCount, total, 0, NaN, NaN, NaN, NaN, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.00000, 0, 0, 0, 0, 0, 0
54 threadCount, insert, 0, NaN, NaN, NaN, NaN, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.00000, 0, 0, 0, 0, 0, 0
54 threadCount, total, 0, NaN, NaN, NaN, NaN, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.00000, 0, 0, 0, 0, 0, 0
END
[root@cs7ds4001 tools]#
④ insertとクエリを混ぜ混ぜで実行
threadの指定数の多重度で合計100回実行
# cassandra-stress user profile=stest02.yaml ops\(insert=1,select1=1\) n=100 -rate threads=1
Connected to cluster: Test Cluster
Datatacenter: DC1; Host: localhost/127.0.0.1; Rack: RAC1
Created schema. Sleeping 1s for propagation.
Sleeping 2s...
Running [insert, select1] with 1 threads for 100 iteration
type, total ops, op/s, pk/s, row/s, mean, med, .95, .99, .999, max, time, stderr, errors, gc: #, max ms, sum ms, sdv ms, mb
Generating batches with [5..5] partitions and [20..20] rows (of [20..20] total rows in the partitions)
insert, 17, 73, 366, 1465, 9.7, 8.2, 22.3, 22.3, 22.3, 22.3, 0.2, 0.00000, 0, 0, 0, 0, 0, 0
select1, 15, 83, 83, 83, 3.7, 2.7, 11.3, 11.3, 11.3, 11.3, 0.2, 0.00000, 0, 0, 0, 0, 0, 0
total, 32, 138, 431, 1529, 6.9, 5.7, 15.7, 22.3, 22.3, 22.3, 0.2, 0.00000, 0, 0, 0, 0, 0, 0
Results:
op rate : 138 [insert:73, select1:83]
partition rate : 431 [insert:366, select1:83]
row rate : 1529 [insert:1465, select1:83]
latency mean : 6.9 [insert:9.7, select1:3.7]
latency median : 5.7 [insert:8.2, select1:2.7]
latency 95th percentile : 15.7 [insert:22.3, select1:11.3]
latency 99th percentile : 22.3 [insert:22.3, select1:11.3]
latency 99.9th percentile : 22.3 [insert:22.3, select1:11.3]
latency max : 22.3 [insert:22.3, select1:11.3]
Total partitions : 100 [insert:85, select1:15]
Total errors : 0 [insert:0, select1:0]
total gc count : 0
total gc mb : 0
total gc time (s) : 0
avg gc time(ms) : NaN
stdev gc time(ms) : 0
Total operation time : 00:00:00
END
#