RedisTimeSeries 1.6 ปล่อยตัว: ฟีเจอร์ใหม่สำหรับ High-Performance Time Series

วันนี้ เรายินดีประกาศ General Availability ของ RedisTimeSeries 1.6 บล็อกโพสต์นี้รายละเอียด Major New Features ที่พร้อมใช้งานแล้ว

About RedisTimeSeries

RedisTimeSeries เป็น High-performance, Memory-first Time-series Data Structure สำหรับ Redis RedisTimeSeries รองรับ Time Series Multi-tenancy (สามารถ Hold Many Time Series Simultaneously) และสามารถ Serve Multiple Clients Accessing These Time Series Simultaneously ตอนนี้พร้อมใช้งานเป็นส่วนหนึ่งของ Redis Stack ด้วย

Major New Features in RedisTimeSeries 1.6

Enhanced Querying Capabilities
- Cross-time Series Aggregations
- Filtering Results by Timestamp or Sample Value
- Selecting Which Labels to Retrieve
- Aligning Aggregation Buckets
- Deleting Samples
Improved Performance
Keyspace Notifications
Redis Enterprise Support for Redis on Flash

Enhanced Querying Capabilities

Cross-time Series Aggregations

ก่อน Version 1.6 มีเพียงหนึ่งประเภทของ Aggregation:

For a Single Time Series – Aggregating the Samples over Equidurational Timeframes

ตั้งแต่ Version 1.6 สองประเภท Aggregation ใหม่เป็นไปได้:

For Multiple Time Series – Aggregating Samples with Equal Timestamps over the Set of Time Series
For Multiple Time Series – First, Aggregate Each Time Series over Equidurational Timeframes, Then for Each Timeframe Aggregate the Results over the Set of Time Series

มาสาธิตประเภท Aggregation ใหม่แรก เริ่มสร้าง Two Stocks และเพิ่ม Prices ที่ 3 Timestamps ต่างกัน:

TS.CREATE stock:A LABELS type stock name A
TS.CREATE stock:B LABELS type stock name B
TS.MADD stock:A 1000 100 stock:A 1010 110 stock:A 1020 120
TS.MADD stock:B 1000 120 stock:B 1010 110 stock:B 1020 100

ตอนนี้เราสามารถ Retrieve Maximal Stock Price per Timestamp:

redis:6379> TS.MRANGE - + WITHLABELS FILTER type=stock GROUPBY type REDUCE max

1) 1) "type=stock"
 2) 1) 1) "type"
 2) "stock"
 2) 1) "__reducer__"
 2) "max"
 3) 1) "__source__"
 2) "stock:A,stock:B"
 3) 1) 1) (integer) 1000
 2) 120
 2) 1) (integer) 1010
 2) 110
 3) 1) (integer) 1020
 2) 120

FILTER type=stock

Clause ทำให้เราเหลือเพียง Single Time Series Representing Stock Prices

GROUPBY type REDUCE max

Clause แยก Time Series เป็น Groups with Identical Type Values และแล้วสำหรับแต่ละ Timestamp Aggregate All Series that Share the Same Type Value Using the Max Aggregator

FILTER label=value

Supported by TS.MRANGE and TS.MREVRANGE Additional Filtering Clauses Supported as Well (See Documentation)

ต่อไป เราจะสาธิตประเภท Aggregation ใหม่ที่สอง:

ก่อน สร้าง Two Stocks และเพิ่ม Prices ที่ 9 Different Timestamps

TS.CREATE stock:A LABELS type stock name A
TS.CREATE stock:B LABELS type stock name B
TS.MADD stock:A 1000 100 stock:A 1010 110 stock:A 1020 120
TS.MADD stock:B 1000 120 stock:B 1010 110 stock:B 1020 100
TS.MADD stock:A 2000 200 stock:A 2010 210 stock:A 2020 220
TS.MADD stock:B 2000 220 stock:B 2010 210 stock:B 2020 200
TS.MADD stock:A 3000 300 stock:A 3010 310 stock:A 3020 320
TS.MADD stock:B 3000 320 stock:B 3010 310 stock:B 3020 300

ตอนนี้ สำหรับแต่ละ Stock เราจะ Calculate Average Stock Price per 1000 Millisecond Time Frame และจากนั้น Retrieve Stock with the Maximal Average for that Timeframe:

redis:6379> TS.MRANGE - + WITHLABELS AGGREGATION avg 1000 FILTER type=stock GROUPBY type REDUCE max

1) 1) "type=stock"
 2) 1) 1) "type"
 2) "stock"
 2) 1) "__reducer__"
 2) "max"
 3) 1) "__source__"
 2) "stock:A,stock:B"
 3) 1) 1) (integer) 1000
 2) 110
 2) 1) (integer) 2000
 2) 210
 3) 1) (integer) 3000
 2) 310

GROUPBY label REDUCE reducer

Supported by TS.MRANGE and TS.MREVRANGE Reducer Can Be Sum, Min, or Max

Filtering Results by Timestamp or Sample Value

เมื่อใช้ TS.RANGE, TS.REVRANGE, TS.MRANGE, และ TS.MREVRANGE คุณไม่ได้ต้องการ Retrieve หรือ Aggregate All Samples เสมอ

Using

[FILTER_BY_TS ts...]

You Can Filter the Sample by a List of Specific Timestamps

Using

[FILTER_BY_VALUE min max]

You Can Filter the Sample by Minimal and Maximal Values

Consider, for Example, a Sampled Metric Where the Normal Values Are Between -100 and 100 but the Value 9999 Is Used as an Indication of Bad Measurement

TS.CREATE temp:TLV LABELS type temp location TLV
TS.MADD temp:TLV 1000 30 temp:TLV 1010 35 temp:TLV 1020 9999
temp:TLV 1030 40

Now, Let's Retrieve All Values, Ignore Out-of-range Values:

TS.RANGE temp:TLV - + FILTER_BY_VALUE -100 100

Let's Also Retrieve the Average Value, Ignore Out-of-range Values:

TS.RANGE temp:TLV - + FILTER_BY_VALUE -100 100 AGGREGATION avg 1000

Selecting Which Labels to Retrieve

เมื่อใช้ TS.MRANGE, TS.MREVRANGE, และ TS.MGET เราไม่ได้ต้องการ Values of All Labels Associated with the Matching Time Series เสมอแต่ Only the Values of Selected Labels

SELECTED_LABELS

Allows Selecting Which Labels to Retrieve Given the Following Time Series and Data:

TS.CREATE temp:TLV LABELS type temp location TLV
TS.MADD temp:TLV 1000 30 temp:TLV 1010 35 temp:TLV 1020 9999
temp:TLV 1030 40

To Get All the Labels Associated with Matched Time Series We'll Use

WITHLABELS

redis:6379> TS.MGET WITHLABELS FILTER type=temp

1) 1) "temp:TLV"
 2) 1) 1) "type"
 2) "temp"
 2) 1) "location"
 2) "TLV"
 3) 1) (integer) 1030
 2) 40

But Suppose We Want Only the Location We Can Use

SELECTED_LABELS:

redis:6379> TS.MGET SELECTED_LABELS location FILTER type=temp

1) 1) "temp:TLV"
 2) 1) 1) "location"
 2) "TLV"
 3) 1) (integer) 1030
 2) 40

Aligning Aggregation Buckets

Suppose We Want to Receive the Average Daily Temperature but Our 'Day' Starts at 06:00 In This Case We Would Like to ALIGN the Intervals to 06:00 to 05:59 Instead of to 00:00 to 23:59

When Using TS.RANGE, TS.REVRANGE, TS.MRANGE, and TS.MREVRANGE It Is Now Possible to Align the Aggregation Buckets with the Requested Start, End, or Specific Timestamp – Using

ALIGN

To Demonstrate Alignment Let's Add the Following Data:

TS.CREATE stock:A LABELS type stock name A
TS.MADD stock:A 1000 100 stock:A 1010 110 stock:A 1020 120
TS.MADD stock:A 1030 200 stock:A 1040 210 stock:A 1050 220
TS.MADD stock:A 1060 300 stock:A 1070 310 stock:A 1080 320

Next, We'll Aggregate Without Using

ALIGN

(Which Means Default Alignment: 0)

redis:6379> TS.RANGE stock:A - + AGGREGATION min 20

1) 1) (integer) 1000
 2) 100
2) 1) (integer) 1020
 2) 120
3) 1) (integer) 1040
 2) 210
4) 1) (integer) 1060
 2) 300
5) 1) (integer) 1080
 2) 320

And Now with

ALIGN

redis:6379> TS.RANGE stock:A – + ALIGN 10 AGGREGATION min 20

1) 1) (integer) 990
 2) 100
2) 1) (integer) 1010
 2) 110
3) 1) (integer) 1030
 2) 200
4) 1) (integer) 1050
 2) 220
5) 1) (integer) 1070
 2) 310

Setting

ALIGN

to 10 Means that a Bucket Should Start at Time 10 and All the Buckets (Each with a 20 Milliseconds Duration) Are Aligned Accordingly

When the Start Timestamp for the Range Query Is Explicitly Stated (Not '-') It Is Also Possible to Set ALIGN to that Time by Setting Align to '-' or to '

start

redis:6379> TS.RANGE stock:A 5 + ALIGN – AGGREGATION min 20

1) 1) (integer) 985
 2) 100
2) 1) (integer) 1005
 2) 110
3) 1) (integer) 1025
 2) 200
4) 1) (integer) 1045
 2) 220
5) 1) (integer) 1065
 2) 310

Similarly When the End Timestamp for the Range Query Is Explicitly Stated (Not '+') It Is Also Possible to Set

ALIGN

to that Time by Setting Align to '+' or to 'end'

Deleting Samples

TS.DEL Allows Deleting Samples in a Given Time Series Within Two Timestamps

For Example

TS.DEL stock:A 1020 1050

Will Delete All Samples with Timestamps Between 1020 and 1050 (Inclusive) The Returned Value Is the Number of Samples Deleted

Improved Performance

Many Optimizations Were Implemented and Most Queries Would Now Execute Much Faster Compared to RedisTimeSeries 1.4

The Following Table Details the Number of Queries per Second Achievable on a Single Node for TSBS Queries (Which We Described Here) The Table Lists Only the Subset of the TSBS Queries that Were Supported in Version 1.4

Query type	1.4 queries/sec	v1.6 queries/sec	% change (higher-better)
tsbs-scale100_cpu-max-all-1	1388	1500	8.07%
tsbs-scale100_double-groupby-1	100	108	8.00%
tsbs-scale100_groupby-orderby-limit	793	1282	61.66%
tsbs-scale100_single-groupby-1-1-1	13448	21347	58.74%
tsbs-scale100_single-groupby-1-1-12	2383	3921	64.54%

We Can Observe an Improvement of 8% to 65% in the Number of Queries per Second Compared to RedisTimeSeries 1.4

Keyspace Notifications

Redis Keyspace Notifications Allow Redis Clients to Subscribe to Pub/Sub Channels in Order to Receive Events Affecting the Redis Data Set in Some Way You Can, for Example, Use RedisGears to Trigger a Function with These Notifications

As an Example It Is Possible to Implement a Time Series Predictor or Anomaly Detector that Listens to the Stream of Samples and Generates Real-time Predictions and Warnings Check, for Example, This Test Which Subscribes to Various RedisTimeSeries Commands and Generates Events

Redis Enterprise Support for Redis on Flash

Since Version 1.6 RedisTimeSeries Can Run in a Redis on Flash but It's Important to Note that RoF Is Implemented on the Key Level Namely the Value of the Whole Time Series Lives Either on FLASH or RAM

RedisTimeSeries Is Part of Redis Stack

RedisTimeSeries ตอนนี้เป็นส่วนหนึ่งของ Redis Stack คุณสามารถ Download Latest Redis Stack Server Binaries สำหรับ macOS, Ubuntu หรือ Redhat หรือ Install ด้วย Docker, Homebrew หรือ Linux

Visualize Time Series Data with RedisInsight

RedisInsight เป็น Visual Tool สำหรับ Developers ที่ให้ Excellent Way to Explore Data from RedisTimeSeries During Development Using Redis or Redis Stack

คุณสามารถ Execute Time Series Queries และ Observe Results Directly from Graphical User Interface RedisInsight Can Now Visualize RedisTimeSeries Query Results

In Addition RedisInsight Contains Quick Guides and Tutorials for Learning RedisTimeSeries Interactively

Learn More on RedisTimeSeries on redis.io and developer.redis.com