Sample random revisions from big namespaces

SQL

# http://jan.kneschke.de/projects/mysql/order-by-rand/
# https://stackoverflow.com/questions/249301/simple-random-samples-from-a-sql-database
# https://mariadb.com/kb/en/library/data-sampling-techniques-for-efficiently-finding-a-random-row/#case-consecutive-auto_increment-without-gaps-10-rows
 
### All non-deleted revisions in the article namespace
# We lose 30-40% of rows as either duplicates or to the WHERE clause, so we need to bump up the limit by about 60%
# Downloaded on run 384406
/*
SELECT @min := MIN(rev_id), @max := MAX(rev_id) FROM revision;
SELECT 
	DISTINCT revision.rev_id, revision.rev_comment_id, revision.rev_timestamp, 
    revision.rev_minor_edit, revision.rev_parent_id, revision.rev_len, page.* 
FROM revision
JOIN page on revision.rev_page = page.page_id
JOIN (SELECT FLOOR(@min + (@max - @min + 1) * RAND(1)) AS rev_id FROM revision LIMIT 16000) AS random USING (rev_id)
WHERE rev_deleted = 0 AND page_namespace = 0
LIMIT 10000;
*/

### All non-deleted revisions in the article-talk namespace
# 29, 33, 30, 37, 26, 31 of 500 = 6.2%, so maybe 200,000 for 10,000
# Downloaded on run 384578
/*
SELECT @min := MIN(rev_id), @max := MAX(rev_id) FROM revision;
SELECT 
	DISTINCT revision.rev_id, revision.rev_comment_id, revision.rev_timestamp, 
    revision.rev_minor_edit, revision.rev_parent_id, revision.rev_len, page.* 
FROM revision
JOIN page on revision.rev_page = page.page_id
JOIN (SELECT FLOOR(@min + (@max - @min + 1) * RAND(5)) AS rev_id FROM revision LIMIT 200000) AS random USING (rev_id)
WHERE rev_deleted = 0 AND page_namespace = 1
LIMIT 10000;
*/

### All non-deleted revisions in the user namespace
# 20, 22, 22, 15, 12 = 3.64%, so maybe 300,000 for 10,000
# Downloaded run 384592
/*
SELECT @min := MIN(rev_id), @max := MAX(rev_id) FROM revision;
SELECT 
	DISTINCT revision.rev_id, revision.rev_comment_id, revision.rev_timestamp, 
    revision.rev_minor_edit, revision.rev_parent_id, revision.rev_len, page.* 
FROM revision
JOIN page on revision.rev_page = page.page_id
JOIN (SELECT FLOOR(@min + (@max - @min + 1) * RAND(1)) AS rev_id FROM revision LIMIT 300000) AS random USING (rev_id)
WHERE rev_deleted = 0 AND page_namespace = 2
LIMIT 10000;
*/

### All non-deleted revisions in the user-talk namespace
# 40, 38, 35, 35, 40 = 7.52% so 200,000 for 10,000 should be OK
# Downloaded run 384595
/*
SELECT @min := MIN(rev_id), @max := MAX(rev_id) FROM revision;
SELECT 
	DISTINCT revision.rev_id, revision.rev_comment_id, revision.rev_timestamp, 
    revision.rev_minor_edit, revision.rev_parent_id, revision.rev_len, page.* 
FROM revision
JOIN page on revision.rev_page = page.page_id
JOIN (SELECT FLOOR(@min + (@max - @min + 1) * RAND(1)) AS rev_id FROM revision LIMIT 200000) AS random USING (rev_id)
WHERE rev_deleted = 0 AND page_namespace = 3
LIMIT 10000;
*/
SQL

Checking query status...
