Clustering ingestion-time partitioned tables - google-cloud-platform

How should I go about clustering an ingestion-time partitioned table?
The examples mentioned in https://cloud.google.com/bigquery/docs/creating-clustered-tables create the destination clustered table as partitioned on a column. What if I want the new clustered table to be ingestion-time partitioned as well (similar to the source)? A work around is to copy each partition separately, for example:
bq query --nouse_legacy_sql --time_partitioning_type DAY --clustering_fields customer_id --destination_table='mydataset.myclusteredtable$20181001' 'SELECT * FROM `mydataset.mytable` where _PARTITIONTIME = timestamp("2018-10-01")'
But that's too much of a hassle.

Related

With SQL or Python how can I find out if a table is part of a sharded set of tables (in BigQuery)?

I want to find out what my table sizes are (in BigQuery).
However I want to sum up the size of of all tables that belong to a specific set of sharded tables.
So I need to find metadata that shows that a table is part of a set of sharded tables.
So I can do: How to get BigQuery storage size for a single table
select
sum(size_bytes)/pow(2, 30) as size_gb
from
<your_dataset>.__TABLES__
But here I can't see if the table is part of a set of sharded set of tables.
This is what my Google Analytics sharded tables look like in BQ:
So somewhere must be metadata that indicates that tables with for example name ga_sessions_20220504 belong to a sharded set ga_sesssions_
Where/how can I find that metadata?
I think you are exploring the right query, most of the time, I use the following query to drill down on shards & it's sizes
SELECT
project_id,
dataset_id,
table_id,
array_reverse(SPLIT(table_id, '_'))[OFFSET(0)] AS shard_pt,
DATE(TIMESTAMP_MILLIS(creation_time)) creation_dt,
ROUND(size_bytes/POW(1024, 3), 2) size_in_gb
FROM
`<project>.<dataset>.__TABLES__`
WHERE
table_id LIKE 'ga_sessions_%'
ORDER BY
4 DESC
Result (on some random GA dataset I have access to FYI)
There is no metadata on Sharded tables via SQL.
Tables being displayed as Sharded in BigQuery UI happens when you do the following ->
Create 2 or more tables that have the following characteristics:
exist in the same dataset
have the exact same table schema
the same prefix
have a suffix of the form _YYYYMMDD (eg. 20210130)
These are something of a legacy feature, they were more commonly used with bigquery’s legacy SQL.
This blog was very insightful on this:
https://mark-mccracken.medium.com/bigquery-date-sharding-vs-date-partitioning-cee3754f7900

how to create partition and cluster on an existing table in big query?

In SQL Server , we can create index like this. How do we create the index after the table already exists? What is the syntax of create clusted index in bigquery?
CREATE INDEX abcd ON `abcd.xxx.xxx`(columnname )
In big query, we can create table like below. But how to create partition and cluster on an existing table?
CREATE TABLE rep_sales.orders_tmp PARTITION BY DATE(created_at) CLUSTER BY created_at AS SELECT * FROM rep_sales.orders
As #Sergey Geron mentioned in the comments, BigQuery doesn’t support indexes. For more information, please refer to this doc.
An existing table cannot be partitioned but you can create a new partitioned table and then load the data into it from the unpartitioned table.
As for clustering of tables, BigQuery supports changing an existing non-clustered table to a clustered table and vice versa. You can also update the set of clustered columns of a clustered table. This method of updating the clustering column set is useful for tables that use continuous streaming inserts because those tables cannot be easily swapped by other methods.
You can change the clustering specification in the following ways:
Call the tables.update or tables.patch API method.
Call the bq command-line tool's bq update command with the --clustering_fields flag.
Note: When a table is converted from non-clustered to clustered or the clustered column set is changed, automatic re-clustering only works from that time onward. For example, a non-clustered 1 PB table that is converted to a clustered table using tables.update still has 1 PB of non-clustered data. Automatic re-clustering only applies to any new data committed to the table after the update.

Bulk upload to Amazon Redshift

I need to insert data on daily basis to AWS Redshift.
The requirement is to analyze only the daily batch inserted to Redshift. Redshift cluster is used by BI tools for analytics.
Question:
What are the best practices to "renew" the data set on a daily basis?
My concern is it is a quite heavy operation and performance will be poor but at the same time it is a quite common situation and I believe it was done before by multiple organization.
If the data is on S3, why not create an EXTERNAL TABLE over it. Then if the query speed over external table it not enough you can load it using CREATE TABLE AS SELECT statement into a temporary table, and once loaded, rename to a name your usual table name.
Sketched SQL:
CREATE EXTERNAL TABLE external_daily_batch_20190422 (
<schema ...>
)
PARTITIONED BY (
<if anything to partition on>
)
ROW FORMAT SERDE <data format>
LOCATION 's3://my-s3-location/2019-04-22';
CREATE TABLE internal_daily_batch_temp
DISTKEY ...
SORTKEY ...
AS
SELECT * from external_daily_batch_20190422;
DROP TABLE IF EXISTS internal_daily_batch__backup CASCADE;
ALTER TABLE internal_daily_batch rename to internal_daily_batch__backup;
ALTER TABLE internal_daily_batch_temp rename to internal_daily_batch;
Incremental load not possible?
By the way, is all of your 10TB of data mutable? Isn't incremental update possible?

Creating Biq Query View With Partitioning

I have a table in big query with 1 GB size. I create a view from this table with partitioning on created_at(timestamp) column. The view is useful for me but I want to write a query using created_at column. When I use this column , does the query run for whole data of view or run for only partitioned values? I want to limit usage of table like 500 MB. Is it possible with views by using partitioning column in where clause?
You can create new partitioned tables (here is the documentation) and copy the data into them.
To query a partitioned table you can use _PARTITIONTIME, for example:
SELECT
[COLUMN]
FROM
[DATASET].[TABLE]
WHERE
_PARTITIONTIME BETWEEN TIMESTAMP('2017-01-01') AND TIMESTAMP('2017-03-01')
Unless you're using actual BigQuery partitioned tables (there is no such thing as partitioned views) you'll be charged for all the data in the columns you access.

Azure SQL Data Warehouse CTAS statistics

Does the "Create table as" function in SQL Data Warehouse create statistics in the background, or do they have to manually be created (as I would when I do a normal "Create table" statement?)
As of the current version, you always have to create column-level statistics on tables, irrespective of whether it was created with a normal CREATE TABLE or the CTAS CREATE TABLE AS... command. It's also good practice to create stats for columns used in JOINs, WHERE clauses, GROUP BY, ORDER BY and DISTINCT clauses.
Regarding tables created with CTAS, the database engine has a correct idea of how many rows are in the table as listed in sys.partitions, but not at the column-level statistics level. For tables created by CREATE TABLE this defaults to 1,000 rows. For the example below, the first table was created with a CTAS and has 208 rows, the second table with an ordinary CREATE TABLE and INSERT from the first table and also has 208 rows, but sys.partitions believes it to have 1,000 eg
Creating any column-level statistics manually will correct this number.
In summary, always manually create statistics against important columns irrespective of how the table was created.