Amazon Redshift COMPOUND SORTKEY - does insertion order matter? - amazon-redshift

Let's say I've created an empty table in Redshift like this:
CREATE TABLE my_table (
val_1 INT ,
val_2 INT ,
val_3 FLOAT
)
COMPOUND SORTKEY(val_1, val_2)
;
When I first populate the table (let's say with the results of some query), should the records be inserted in the SORTKEY order, using the ORDER BY in the code below:
INSERT INTO my_table
SELECT val_1, val_2, val_3 FROM other_table
ORDER BY val_1, val_2
Or is there no need to do that; i.e. SORTKEY ordering of inserted records is handled physically by Redshift itself? Thx.

Assuming the same behaviour for INSERT INTO as for loading via the COPY command, there is no need to order the records first. According to the AWS docs all the following constraints be fulfilled in order to add the records to sorted region of the table - in your example you have a COMPOUND SORTKEY of 2 columns:
The table uses a compound sort key with only one sort column.
The sort column is NOT NULL.
The table is 100 percent sorted or empty.
All the new rows are higher in sort order than the existing rows, including rows marked for deletion. In this instance, Amazon Redshift uses the first eight bytes of the sort key to determine sort order.

Related

Use index to speed up query using values from different tables

I have a table products, a table orders and a table orderProducts.
Products have a name as a PK (apple, banana, mango) and a price .
orders have a created_at date and an id as a PK.
orderProducts connects orders and products, so they have a product_name and an order_id. Now I would like to show all orders for a given product that happened in the last 24 hours.
I use the following query:
SELECT
orders.id,
orders.created_at,
products.name,
products.price
FROM
orderProducts
JOIN products ON
products.name=orderProducts.product
JOIN orders ON
orders.id=orderProducts.order
WHERE
products.name='banana'
AND
orders.created_at BETWEEN NOW() - INTERVAL '24 HOURS' AND NOW()
ORDER BY
orders.created_at
This works, but I would like to optimize this query with an index. This index would need to first be ordered by
the product name, so it can be filtered
then the created_at of the order in descending order, so it can select only the ones from 24 hours ago
The problem is, that from what I have seen, indexes can only be created on a single table, without the possibility of joining another tables values to it. Since two individual index do not solve this problem either, I was wondering if there was an alternative way to optimize this particular query.
Here are the table scripts:
CREATE TABLE products
(
name text PRIMARY KEY,
price integer,
)
CREATE TABLE orders
(
id SERIAL PRIMARY KEY,
created_at TIMESTAMP DEFAULT NOW(),
)
CREATE TABLE orderProducts
(
product text REFERENCES products(name),
"order" integer REFERENCES orders(id),
)
First of all. Please do not put indices everywhere - that lead to slower changing operations...
As proposed by #Laurenz Albe - do not guess - check.
Other than that. Note that you know product name, price is repeated - so you can query that once. Question if in your case two queries are going to be faster then single one... Check that.
Please read docs. I would try this index:
create index orders_id_created_at on orders(created_at desc, id)
Normally id should go first, since that is unique, however here system should be able to filter out on both predicates - where/join. Just guessing here.
orderProducts I would like to see index on both columns, however for this query only one should be needed. In practice you are going from products to orders, or other way - both paths are possible, that is why I've wrote about indexing both columns. I would use two separate indexes:
create index orderproducts_product_id on orderproducts (product_id) include (order_id);
create index orderproducts_order_id on orderproducts (order_id) include (product_id);
Probably that is not changing much, but... idea is to use only index, but not the table itself.
These rules are important in terms of performance:
Integer index faster than string index, therefore, you should try to make the primary keys always be an integer. Because join the tables uses primary keys too.
If when in where clauses always use two fields then we must create an index for both fields.
Foreign-Keys are not indexed, you must create an index for foreign-key fields manually.
So, recommended table scripts will be are that:
CREATE TABLE products
(
id serial primary key,
name text,
price integer
);
CREATE UNIQUE INDEX products_name_idx ON products USING btree (name);
CREATE TABLE orders
(
id SERIAL PRIMARY KEY,
created_at TIMESTAMP DEFAULT NOW()
);
CREATE INDEX orders_created_at_idx ON orders USING btree (created_at);
CREATE TABLE orderProducts
(
product_id integer REFERENCES products(id),
order_id integer REFERENCES orders(id)
);
CREATE INDEX orderproducts_product_id_idx ON orderproducts USING btree (product_id, order_id);
---- OR ----
CREATE INDEX orderproducts_product_id ON orderproducts (product_id);
CREATE INDEX orderproducts_order_id ON orderproducts (order_id);

Reading an append-only list from PostgreSQL

I would like to implement an append-only list in PostgreSQL. Basically, this is trivial: Create a table, and only ever INSERT into that table.
However, I would like to be able to read that list again, in the order it was created. How can I do this? Is a simple SELECT * FROM MyTable enough? If not, what do I sort by?
Rows in a relational database have no inherent sort order. The only way to get a guaranteed sort order is to use an order by.
You can either create an identity column that is incremented on every insert or a timestamp column that records the precise time a row was inserted (or do both).
e.g.
create table append_only
(
id bigint generated always as identity,
... other columns ...
created_at timestamp default clock_timestamp()
);
Then use that column for an order by. By having both, you can use the id column as a tie breaker when sorting by the timestamp in case two rows were inserted at exactly same microsecond.
You could create column with data type SERIAL(similiar to AUTOINCREMENT/SEQUENCE):
CREATE TABLE myTable(id SERIAL, ...)
SELECT * FROM myTable ORDER BY id;

Compute shared hstore key names in Postgresql

If I have a table with an HSTORE column:
CREATE TABLE thing (properties hstore);
How could I query that table to find the hstore key names that exist in every row.
For example, if the table above had the following data:
properties
-------------------------------------------------
"width"=>"b", "height"=>"a"
"width"=>"b", "height"=>"a", "surface"=>"black"
"width"=>"c"
How would I write a query that returned 'width', as that is the only key that occurs in each row?
skeys() will give me all the property keys, but I'm not sure how to aggregate them so I only have the ones that occur in each row.
The manual gets us most of the way there, but not all the way... way down at the bottom of http://www.postgresql.org/docs/8.3/static/hstore.html under the heading "Statistics", they describe a way to count keys in an hstore.
If we adapt that to your sample table above, you can compare the counts to the # of rows in the table.
SELECT key
FROM (SELECT (each(properties)).key FROM thing1) AS stat
GROUP BY key
HAVING count(*) = (select count(*) from thing1)
ORDER BY key;
If you want to find the opposite (all those keys that are not in every row of your table), just change the = to < and you're in business!

Index method for a column used only for ordering

I have a table product_images with a foreign key product_id and integer field order to manualy set order of product's images. Knowing that the table will be used only like this:
SELECT * FROM product_images
WHERE product_id = ?
ORDER BY "order"
-- what is the optimal index method for product_id and order?
Is that enough?:
CREATE INDEX product_images_unique_order
ON "product_images"("product_id", "order");
SQL Fiddle
Yes, that should do it.
PostgreSQL might decide not to use that index, depending on how many rows you have, how many images any given product_id has, and how scattered about the table all of the rows with the same product_id are, and how wide the rows of the product_images table are; plus many other things.
But by having that index you provide PostgreSQL with the opportunity to use it.

Should this PostgreSQL query use the indexes?

I have two tables:
CREATE TABLE soils (
sample_id TEXT PRIMARY KEY,
project_id TEXT,
technician_id TEXT
);
CREATE INDEX soils_idx
ON soils
USING btree
(sample_id COLLATE pg_catalog."default");
CREATE TABLE assays (
sample_id TEXT PRIMARY KEY,
mo_ppm NUMERIC
);
CREATE INDEX assays_idx
ON assays
USING btree
(sample_id COLLATE pg_catalog."default");
Each table contains about a half million records, and, in reality, about 20 additional columns each, of type TEXT (omitted in the DDL posted above to save time here).
When I perform the query:
EXPLAIN SELECT
s.sample_id, s.project_id, s.technician_id, a.mo_ppm
FROM
soils AS s INNER JOIN assays AS a ON s.sample_id = a.sample_id
I get 2 SEQ SCANs, rather than a lookup to the index. Is that expected behaviour?
Since you have no WHERE conditions, you effectively read the whole table. It's cheaper to run sequential scans and not involve any indexes at all.
Try:
EXPLAIN
SELECT s.sample_id, s.project_id, s.technician_id, a.mo_ppm
FROM soils s
JOIN assays a USING (sample_id)
WHERE <some condition that returns few rows>;
... and an index matching the WHERE condition should be used.
You don't need to define an index on a PRIMARY KEY column. A PK constraint is implemented with a unique index automatically. Your additional index is redundant and of no use.
An index on a foreign key column would be a good idea, but there isn't one in your example, which looks odd. Like the two tables could be combined into one. Probably just over-simplification for the test case.
Finally, for big tables, I would consider using a simple integer primary key instead of text, possibly a serial column. That's typically faster.
Yes, that's expected behaviour. On the other hand it depends on your random_page_cost, seq_page_cost and effective_cache_size settings. Your query doesn't have WHERE clause hence it might be faster to read everything sequentially. You can try to penalise sequential scan:
set enable_seqscan = off;
explain analyse <your query>;
and then compare plan/cost/IO wait (it is not possible to disable seq-scan but it gets very high cost -- ~1e7 (or 1e8)).
If you have SSD and WHERE clause in your query then you can lower random_page_cost to 1.5..2.5 and encourage PG to use index.