跳转至主要内容

可变流

This type of stream is only available in Timeplus Enterprise, with high performance query and UPSERT (UPDATE or INSERT).

As the name implies, the data in the stream is mutable. Value with the same primary key(s) will be overwritten.

The primary use case of mutable streams is serving as the lookup/dimensional data in Streaming JOIN, supporting millions or even billions of unique keys. You can also use mutable streams as the "fact table" to efficiently do range queries or filtering for denormalized data model, a.k.a. OBT (One Big Table).

Learn more about why we introduced Mutable Streams by checking this blog.

语法​

CREATE MUTABLE STREAM [IF NOT EXISTS] stream_name (
    <col1> <col_type>,
    <col2> <col_type>,
    <col3> <col_type>,
    <col4> <col_type>
    INDEX <index1> (col3)
    FAMILY <family1> (col3,col4)
    )
PRIMARY KEY (col1, col2)
SETTINGS
    logstore_retention_bytes=..,
    logstore_retention_ms=..,
    shards=..

Learn more

示例​

Create a mutable stream​

Create the stream with the following SQL:

CREATE MUTABLE STREAM device_metrics
(
  device_id string,
  timestamp datetime64(3),
  batch_id uint32,
  region string,
  city string,
  lat float32,
  lon float32,
  battery float32,
  humidity uint16,
  temperature float32
)
PRIMARY KEY (device_id, timestamp, batch_id)

备注：

  • The compound primary key is a combination of device_id, timestamp and the batch_id. Data with exactly the same value for those 3 columns will be overridden.
  • Searching data with any column in the primary key is very fast.
  • By default there is only 1 shard and no extra index or optimization.

Load millions of rows​

You can use CREATE RANDOM STREAM and a Materialized View to generate data and send to the mutable stream. But since we are testing massive historical data with duplicated keys, we can also use INSERT INTO .. SELECT to load data.

INSERT INTO device_metrics
SELECT
    'device_' || to_string(floor(rand_uniform(0, 2400))) AS device_id,
    now64(9) AS timestamp,
    floor(rand_uniform(0, 50)) AS batch_id,
    'region_'||to_string(rand()%5) AS region,
    'city_'||to_string(rand()%10) AS city,
    rand()%1000/10 AS lat,
    rand()%1000/10 AS lon,
    rand_uniform(0,100) AS battery,
    floor(rand_uniform(0,80)) AS humidity,
    rand_uniform(0,100) AS temperature,
    now64() AS _tp_time
FROM numbers(50_000_000)

Depending on your hardware and server configuration, it may take a few seconds to add all data.

0 rows in set. Elapsed: 11.532 sec. Processed 50.00 million rows, 400.00 MB (4.34 million rows/s., 34.69 MB/s.)

查询​

When you query the mutable stream, Timeplus will read all historical data without any duplicated primary key.

SELECT count() FROM table(device_metrics)

Sample output:

┌─count()─┐
│  120000 │
└─────────┘

1 row in set. Elapsed: 0.092 sec.

You can filter data efficiently with any part of the primary key:

SELECT count() FROM table(device_metrics) WHERE batch_id=5

Sample output:

┌─count()─┐
│    2400 │
└─────────┘

1 row in set. Elapsed: 0.078 sec. Processed 120.00 thousand rows, 480.00 KB (1.54 million rows/s., 6.15 MB/s.)

Another example:

SELECT * FROM table(device_metrics) WHERE device_id='device_1' AND timestamp>now()-1h

Sample output:

┌─device_id─┬───────────────timestamp─┬─batch_id─┬─region───┬─city───┬──lat─┬──lon─┬───battery─┬─humidity─┬─temperature─┬────────────────_tp_time─┐
│ device_1  │ 2024-07-10 11:38:14.878 │        0 │ region_1 │ city_1 │ 21.1 │ 21.1 │  81.35298 │       41 │    81.35298 │ 2024-07-10 11:38:14.880 │
│ device_1  │ 2024-07-10 11:38:14.878 │       49 │ region_3 │ city_3 │ 33.3 │ 33.3 │ 62.507397 │       79 │   62.507397 │ 2024-07-10 11:38:14.880 │
└───────────┴─────────────────────────┴──────────┴──────────┴────────┴──────┴──────┴───────────┴──────────┴─────────────┴─────────────────────────┘

50 rows in set. Elapsed: 0.015 sec.

You can also query the mutable stream in the streaming SQL.

SELECT .. FROM mutable_stream

This will query all existing data and accept new incoming data.

Mutable stream can also be used in JOINs.

高级设置​

Retention Policy for Streaming Storage​

Like normal streams in Timeplus, mutable streams use both streaming storage and historical storage. New data are added to the streaming storage first, then continuously write to the historical data with deduplication/merging process. When you create the mutable stream, you can configure the maximum size of the streaming storage or Time-To-Live (TTL).

For example, if you want to keep up to 8GB or half an hour data in the streaming storage, you can add the following settings in the DDL:

CREATE MUTABLE STREAM ..
(
  ..
)
PRIMARY KEY ..
SETTINGS
    logstore_retention_bytes=8589934592, -- 8GB
    logstore_retention_ms=1800000; -- half an hour

Secondary Index​

Regardless of whether you choose a single column or multiple columns as the primary key(s), Timeplus will build an index for those columns. Queries with filtering on these columns will leverage the index to boost performance and minimize data scanning.

For other columns, if they are frequently filtered, you can also define secondary indexes for them.

例如：

CREATE MUTABLE STREAM device_metrics
(
  device_id string,
  timestamp datetime64(3),
  batch_id uint32,
  region string,
  city string,
  ..
  index sidx1 (region)
  index sidx2 (city)
)
PRIMARY KEY (device_id, timestamp, batch_id)

When you query data with filters on those columns, Timeplus will automatically leverage the indexed data to improve query performance.

Column Family​

For One-Big-Table(OBT) or extra wide table with dozens or even hundreds of columns, it's not recommended to run SELECT * FROM .., unless you need to export data.

More commonly, you need to query a subset of the columns in different use cases. For those columns which are commonly queried together, you can define column families to group them, so that data for those columns will be saved together in the same file. Properly defining column families can optimize the disk i/o and avoid reading unnecessary data files.

Please note, one column can appear in up to one column family. The columns as primary keys are in a special column family. There should be no overlap for the column families or primary keys.

Taking the previous device_metrics as an example, the lat and lon are commonly queried together. You can define a column family for them.

CREATE MUTABLE STREAM device_metrics
(
  device_id string,
  timestamp datetime64(3),
  batch_id uint32,
  region string,
  city string,
  lat float32,
  lon float32,
  battery float32,
  humidity uint16,
  temperature float32,
  FAMILY cf1 (lat,lon)
)
PRIMARY KEY (device_id, timestamp, batch_id)

Multi-shard​

Another optimization is to create multiple shards to partition the data when it scales. For example, to create 3 shards for device_metrics:

CREATE MUTABLE STREAM device_metrics
(
  device_id string,
  timestamp datetime64(3),
  batch_id uint32,
  region string,
  city string,
  lat float32,
  lon float32,
  battery float32,
  humidity uint16,
  temperature float32
)
PRIMARY KEY (device_id, timestamp, batch_id)
SETTINGS shards=3

Performance Tuning​

If you are facing performance challenges with massive data in mutable streams, please consider adding secondary indexes, column families and use multiple shards.

key_space_full_scan_threads​

Additionally, you can configure the number of threads for full-scan of the key space at the query time using the key_space_full_scan_threads setting, e.g.:

SELECT * FROM table(a_mutable_stream) WHERE num=166763.6691744028
SETTINGS key_space_full_scan_threads=8;
最后 由 Jove Zhong 于 更新