5
4

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 5 years have passed since last update.

InfluxDB v1.2のサブクエリを試す

Posted at

先日、InfluxDB v1.2がリリースされ、サブクエリが使えるようになったので、早速試してみました。

[InfluxDB 1.2 released with subqueries and 50% better write throughput on larger hardware]
(https://www.influxdata.com/influxdb-1-2-released-with-subqueries-and-50-better-write-performance-on-larger-hardware/)

サブクエリとは

v1.1までのInfluxDBでは、SQLライクに複数のフィールドの値を計算できるものの、集計関数の中には単一の値しか書くことが出来ませんでした。

o select value1 * 100 from ... # OK

x select mean(value1 * 100) from ...     # 関数の中には計算式を書けない
o select mean(value1) * 100 from ...     # これはOK

x select mean(value1 * value2) from ...
o select mean(value1) * mean(value2) from ... # 関数の結果を計算はできるがやりたいのはコレジャナイ

上記のような計算を行う場合、従来は(value1 * value2)のクエリの計算結果を一旦measurementとして保存し、その結果に対して再度クエリをかけて平均を取る、ということを行う必要がありました。しかしv1.2でサポートされたサブクエリを使うと、同等のことが1回のクエリで可能になります。

select mean(value) from (select value1 * value2 from ...)

使用例その1: 加重平均

まずは上記で例に挙げたような加重平均を試してみます。サンプルデータは下記のとおり。(表示を見やすくするため、precision rfc3339としています)

> select * from sample_data
name: sample_data
time       count host  value
----       ----- ----  -----
2017-01-26T00:00:00Z 33    host1 0.1686618
2017-01-26T00:00:00Z 10    host2 0.772844364
2017-01-26T00:01:00Z 60    host1 0.025078533
2017-01-26T00:01:00Z 86    host2 0.589876001
2017-01-26T00:02:00Z 28    host1 0.795133127
2017-01-26T00:02:00Z 41    host2 0.801407917
2017-01-26T00:03:00Z 91    host1 0.081822447
2017-01-26T00:03:00Z 90    host2 0.765902936
2017-01-26T00:04:00Z 61    host1 0.51287518
2017-01-26T00:04:00Z 38    host2 0.53270741
2017-01-26T00:05:00Z 51    host1 0.060094054
2017-01-26T00:05:00Z 79    host2 0.128055132
2017-01-26T00:06:00Z 72    host1 0.190071427
2017-01-26T00:06:00Z 35    host2 0.005363927
2017-01-26T00:07:00Z 25    host1 0.211173586
2017-01-26T00:07:00Z 61    host2 0.194961101
2017-01-26T00:08:00Z 41    host1 0.211522929
2017-01-26T00:08:00Z 16    host2 0.022634187
2017-01-26T00:09:00Z 75    host1 0.131138756
2017-01-26T00:09:00Z 89    host2 0.819108491

countを、各valueに対する重み値とすると、重み付けした値(value * count)は問題なく計算できます。

> select host, value * count as weighed_value from sample_data
name: sample_data
time                 host  weighed_value
----                 ----  -------------
2017-01-26T00:00:00Z host1 5.5658394
2017-01-26T00:00:00Z host2 7.72844364
2017-01-26T00:01:00Z host1 1.50471198
2017-01-26T00:01:00Z host2 50.729336086
2017-01-26T00:02:00Z host1 22.263727556000003
2017-01-26T00:02:00Z host2 32.857724597
2017-01-26T00:03:00Z host1 7.445842677000001
2017-01-26T00:03:00Z host2 68.93126424
2017-01-26T00:04:00Z host1 31.28538598
2017-01-26T00:04:00Z host2 20.24288158
2017-01-26T00:05:00Z host1 3.064796754
2017-01-26T00:05:00Z host2 10.116355427999999
2017-01-26T00:06:00Z host1 13.685142743999998
2017-01-26T00:06:00Z host2 0.187737445
2017-01-26T00:07:00Z host1 5.27933965
2017-01-26T00:07:00Z host2 11.892627161
2017-01-26T00:08:00Z host1 8.672440089
2017-01-26T00:08:00Z host2 0.362146992
2017-01-26T00:09:00Z host1 9.8354067
2017-01-26T00:09:00Z host2 72.900655699

ここから、各時刻における全ホスト間の平均値は、元のクエリをfrom句にいれて、以下のように計算できます。

> select mean(*) from (select host, value * count as weighed_value from sample_data) where time >= '2017-01-26T00:00:00Z' and time < '2017-01-26T00:10:00Z' group by time(1m)
name: sample_data
time                 mean_weighed_value
----                 ------------------
2017-01-26T00:00:00Z 6.64714152
2017-01-26T00:01:00Z 26.117024033000003
2017-01-26T00:02:00Z 27.560726076500004
2017-01-26T00:03:00Z 38.1885534585
2017-01-26T00:04:00Z 25.76413378
2017-01-26T00:05:00Z 6.590576090999999
2017-01-26T00:06:00Z 6.936440094499999
2017-01-26T00:07:00Z 8.5859834055
2017-01-26T00:08:00Z 4.5172935405
2017-01-26T00:09:00Z 41.368031199499995

それぞれの時刻において、host1とhost2の平均値となっています。
また、各ホスト毎に10分間の平均値も、下記のようにして計算できます。

> select mean(*) from (select host, value * count as weighed_value from sample_data) where time >= '2017-01-26T00:00:00Z' and time < '2017-01-26T00:10:00Z' group by time(10m),host
name: sample_data
tags: host=host1
time                 mean_new_value
----                 --------------
2017-01-26T00:00:00Z 10.860263353000002

name: sample_data
tags: host=host2
time                 mean_new_value
----                 --------------
2017-01-26T00:00:00Z 27.5949172868

使用例その2: 累積値

ネットワークのトラフィック等をSNMPで取得すると累積値として結果が得られますが、これもInfluxDBでは扱いづらいデータでした。

> select * from sample_data
name: sample_data
time                 host  value
----                 ----  -----
2017-01-25T23:59:00Z host1 0
2017-01-25T23:59:00Z host2 0
2017-01-26T00:00:00Z host1 0.380175335
2017-01-26T00:00:00Z host2 0.654674526
2017-01-26T00:01:00Z host1 0.548469287
2017-01-26T00:01:00Z host2 1.518483451
2017-01-26T00:02:00Z host1 0.912088251
2017-01-26T00:02:00Z host2 1.618768565
2017-01-26T00:03:00Z host1 1.024910481
2017-01-26T00:03:00Z host2 1.661952017
2017-01-26T00:04:00Z host1 1.194558728
2017-01-26T00:04:00Z host2 2.532767878
2017-01-26T00:05:00Z host1 1.518409392
2017-01-26T00:05:00Z host2 3.402324839
2017-01-26T00:06:00Z host1 1.603926464
2017-01-26T00:06:00Z host2 4.058450398
2017-01-26T00:07:00Z host1 2.06296154
2017-01-26T00:07:00Z host2 4.794882977
2017-01-26T00:08:00Z host1 2.408185365
2017-01-26T00:08:00Z host2 4.902904745
2017-01-26T00:09:00Z host1 3.351145834
2017-01-26T00:09:00Z host2 4.915808337

derivative関数を使うと差分を計算できますが、その結果をさらに集計する、という事が従来は出来ませんでした。

> select derivative(value,1m) from sample_data group by host
name: sample_data
tags: host=host1
time                 derivative
----                 ----------
2017-01-26T00:00:00Z 0.380175335
2017-01-26T00:01:00Z 0.16829395199999997
2017-01-26T00:02:00Z 0.36361896400000004
2017-01-26T00:03:00Z 0.11282223000000002
2017-01-26T00:04:00Z 0.16964824700000003
2017-01-26T00:05:00Z 0.32385066399999984
2017-01-26T00:06:00Z 0.08551707200000003
2017-01-26T00:07:00Z 0.45903507599999993
2017-01-26T00:08:00Z 0.34522382500000015
2017-01-26T00:09:00Z 0.942960469

name: sample_data
tags: host=host2
time                 derivative
----                 ----------
2017-01-26T00:00:00Z 0.654674526
2017-01-26T00:01:00Z 0.8638089250000001
2017-01-26T00:02:00Z 0.10028511400000006
2017-01-26T00:03:00Z 0.043183451999999845
2017-01-26T00:04:00Z 0.8708158610000001
2017-01-26T00:05:00Z 0.8695569609999998
2017-01-26T00:06:00Z 0.6561255589999999
2017-01-26T00:07:00Z 0.7364325790000006
2017-01-26T00:08:00Z 0.10802176799999952
2017-01-26T00:09:00Z 0.012903591999999797

複数ホストがある中で、最大、最小の値を取得したい場合、サブクエリを使うと以下のように書くことが出来ます。

> select max(derivative) as max, min(derivative) as min from (select derivative(value,1m) from sample_data group by host) where time >= '2017-01-26T00:00:00Z' and time < '2017-01-27T00:00:00Z' group by time(1m) fill(none)
name: sample_data
time                 max                 min
----                 ---                 ---
2017-01-26T00:00:00Z 0.654674526         0.380175335
2017-01-26T00:01:00Z 0.8638089250000001  0.16829395199999997
2017-01-26T00:02:00Z 0.36361896400000004 0.10028511400000006
2017-01-26T00:03:00Z 0.11282223000000002 0.043183451999999845
2017-01-26T00:04:00Z 0.8708158610000001  0.16964824700000003
2017-01-26T00:05:00Z 0.8695569609999998  0.32385066399999984
2017-01-26T00:06:00Z 0.6561255589999999  0.08551707200000003
2017-01-26T00:07:00Z 0.7364325790000006  0.45903507599999993
2017-01-26T00:08:00Z 0.34522382500000015 0.10802176799999952
2017-01-26T00:09:00Z 0.942960469         0.012903591999999797

また、各ホスト毎に、10分間の最大値、最小値も取得可能です。

> select max(derivative) as max, min(derivative) as min from (select derivative(value,1m) from sample_data group by host) where time >= '2017-01-26T00:00:00Z' and time < '2017-01-27T00:00:00Z' group by time(10m),host fill(none)
name: sample_data
tags: host=host1
time                 max         min
----                 ---         ---
2017-01-26T00:00:00Z 0.942960469 0.08551707200000003

name: sample_data
tags: host=host2
time                 max                min
----                 ---                ---
2017-01-26T00:00:00Z 0.8708158610000001 0.012903591999999797
5
4
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
5
4

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?