Sql-server – Recursive CTE with partition

You say that the hierarchy gets modified. Presumably while this operation is running, there is some amount of blocking which is taking place then?

Even if the hierarchy is changing, are the roots for items changing?

Have you looked at the time it would take to just make the mapping table from root to item and index it?

I would like to see the execution plan to see what is happening - the CTE should get spooled, but as a manually materialized and indexed table it might perform better in the later steps.

Even with heavy activity, it would seem to me that someone has to be blocked if DML operations are changing data which this process is reading.

So I'd strongly consider taking a snapshot of the hierarchy.

In addition, you have a number of other INNER JOINs - you should review whether it is, in fact, the CTEs at all and whether there are any indexes which are missing to make those joins effective. The execution plan should tell you that.

You appear to have quite a few things in the WHERE clause which might help reduce some operations (and determine which indexes might be the best)), but it's hard to tell without looking at the execution plan or the indexes.

First off, you do not want to use char(50). Use varchar(50) or just text. Read more:

• Any downsides of using data type “text” for storing strings?

Assuming the following rules:

• Basic slugs never end with a dash.
• Duplicate slugs are suffixed with a dash and a sequential number (-123).

Note that all of the following methods are subject to a race conditions: concurrent operations might identify the same "free" name for the next slug.
To defend against it, you can impose a UNIQUE constraint on slug and be prepared to repeat an INSERT upon duplicate key violation or you to take out a write lock on the table at the start of the transaction.

If you glue the suffix to the basic slug name with a dash and allow basic slugs to end in separate numbers, the specification is a tiny bit ambiguous (see comments). I suggest a unique delimiter of your choice instead (which is otherwise disallowed).

Efficient rCTE

WITH RECURSIVE
  input AS (SELECT 'news-on-apple'::text AS slug)  -- input basic slug here once
, cte   AS (
   SELECT slug || '-' AS slug  -- append '-' once, if basic slug exists
        , 1 as suffix          -- start with suffix 1
   FROM   article
   JOIN   input USING (slug)

   UNION ALL
   SELECT c.slug, c.suffix + 1  -- increment by 1 ...
   FROM   cte     c
   JOIN   article a ON a.slug = c.slug || c.suffix  -- ... if slug-n already exists
   )
(
SELECT slug || suffix AS slug
FROM   cte
ORDER  BY suffix DESC  -- pick the last (free) one
LIMIT  1
)  -- parentheses required
UNION  ALL  -- if the basic slug wasn't taken, fall back to that
SELECT slug FROM input
LIMIT  1;

Better performance without rCTE

If you worry about thousands of slugs competing for the same slug or generally want to optimize performance, I'd consider a different, faster approach.

WITH input AS (SELECT 'news-on-apple'::text  AS slug
                    , 'news-on-apple-'::text AS slug1)  -- input basic slug here
SELECT i.slug
FROM   input        i
LEFT   JOIN article a USING (slug)
WHERE  a.slug IS NULL  -- doesn't exist yet.

UNION ALL
(  -- parentheses required
SELECT i.slug1 || COALESCE(right(a.slug, length(i.slug1) * -1)::int + 1, 1)
FROM   input        i
LEFT   JOIN article a ON a.slug LIKE (i.slug1 || '%')  -- match up to last "-"
                     AND right(a.slug, length(i.slug1) * -1) ~ '^\d+$' -- suffix numbers only
ORDER  BY right(a.slug, length(i.slug1) * -1)::int DESC
)
LIMIT  1;

• If the basic slug isn't taken yet, the more expensive second SELECT is never executed - same as above, but much more important here. Check with EXPLAIN ANALYZE, Postgres is smart that way with LIMIT queries. Related:

  • Optimize a query on two big tables

• Check for the leading string and the suffix separately, so the LIKE expression can use a basic btree index with text_pattern_ops like

  CREATE INDEX article_slug_idx ON article (slug text_pattern_ops);
  

  Detailed explanation:

  • Pattern matching with LIKE, SIMILAR TO or regular expressions in PostgreSQL

• Convert the suffix to integer before you apply max(). Numbers in text representation don't work.

Optimize performance

To get the optimum, consider storing the suffix separated from the basic slug and concatenate the slug as needed: concat_ws('-' , slug, suffix::text) AS slug

CREATE TABLE article (
   article_id serial PRIMARY KEY
 , title text NOT NULL
 , slug  text NOT NULL
 , suffix int
);

The query for a new slug then becomes:

SELECT slug
    || COALESCE((
          SELECT '-'::text || (max(suffix) + 1)::text
          FROM   article a
          WHERE  a.slug = i.slug), '') As slug
FROM  (SELECT 'news-on-apple'::text AS slug) i  -- input basic slug here

Ideally supported with a unique index on (slug, suffix).

Query for list of slugs

In any version of Postgres you can provide rows in a VALUES expression.

SELECT *
FROM   article
JOIN  (
   VALUES
     ('slug-foo'::text, 1)
     ('slug-bar',7)
   ) u(slug,suffix) USING (slug,suffix);

You can also use IN with a set of row-type expressions Which is shorter:

SELECT *
FROM   article
WHERE (slug,suffix) IN (('slug-foo', 1), ('slug-bar',7));

Details under this related question (as commented below):

• the <set clause>'s <multiple column assignment>

For long lists, the JOIN to a VALUES expression is typically faster.

In Postgres 9.4 (released today!) you can also use the new variant of unnest() to unnest multiple arrays in parallel.

Given an array of basic slugs and a corresponding array of suffixes (as per comment):

SELECT *
FROM   article
JOIN   unnest('{slug-foo,slug-bar}'::text[]
            , '{1,7}'::int[]) AS u(slug,suffix) USING (slug,suffix);