Not the answer you need?
Register and ask your own question!
Many Forum changes were implemented on Tue 22 Sep. Read about new Ranks, Scoring, and Reactions.
Email [email protected] for any comments or concerns.

Realtime bulk imports on large table with keys

gwgrahamgwgraham EntrantInactive User Role Beginner
Hi. I'm trying to optimise large imports into a large table with keys. Here is the story so far. Can anyone suggest ways of creating a scalable solution?


I want to allow periodic real time imports of up to 100,000 rows into a table with more than 2,000,000 rows. The table has 2 text indexes.

I use a 'load data infile' which is quick. Or was, until the number of rows in the table grew and the inserts slowed down because of index rebuilding (I assume).

I then started disabling and enabling the keys before and after the import, and ran the enable part off as a background process so the user didn't have to wait for it.

However, 'enable keys' locks the table, which makes selects and other operations on that table impossible during the process (currently about 2 mins).

This is a problem. Can anyone suggest how I can get around it. Either using my current method, or using a different method altogether which would be more scaleable.

The table is mysam with fulltext indexes on 3 varchar columns.


My system variables look like this:

[mysqld]
set-variable = max_connections=500
safe-show-database
ft_stopword_file=''
ft_min_word_len=3
default-character-set=utf8
init-connect='SET NAMES utf8'
query_cache_type=1
query_cache_limit=2M
query_cache_size=64M
bulk_insert_buffer_size=64M
key_buffer_size=128M
log-slow-queries
delay-key-write=ON

Comments

  • vgattovgatto Contributor Current User Role Beginner
    It might be worth a try to have an online and offline version of this table. You load data into the offline version, rebuild the index and then swap the two tables. That way, you always have one table available and unlocked for queries.

    It's not uncommon to store the columns you have your fulltext index on in their own table, after they've been preprocessed a little bit. This can help the swap case out a little bit by reducing the size of the two fulltext search tables by storing the columns not involved in your fulltext index somewhere else. This also helps if the users need to see their data immediately in a list view before it is available through search.

    If you really want this problem to go away forever, you should consider using Apache Lucene rather than using MySQL as your search engine.
Sign In or Register to comment.

MySQL, InnoDB, MariaDB and MongoDB are trademarks of their respective owners.
Copyright ©2005 - 2020 Percona LLC. All rights reserved.