7 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
Gabor Vincze

Nice post, Petya. Looking forward to read the second part

1
Joshua Dickerson

What is the performance impact from writing every query to disk? I see the daemon does sampling which I was about to ask about while reading the article.

0
Kyle Oppenheim

@Joshua, writing every query to disk can be very expensive. At Groupon, we write slow logs to a different disk array on our large databases. So this wasn’t as scary as it may be in your situation. We are also using the slow log rate limiting feature in Percona Server — https://www.percona.com/doc/percona-server/5.5/diagnostics/slow_extended_55.html.

Rotating the large slow logs requires some special consideration. We will cover that in part 2.

0
Erno Erdelyi

Hi Peter,

Congratulation to the nice post and good luck for the show on next week!

0
vishnu rao

nice post.
waiting for part 2.

0
Peter Colclough

Peter,

e are building a dev system that does something similar, and I have had a lot of success writing sql queries (taken through Proxy at this stage) to a fifo buffer, and reading via a small seperate app from that, passing down the line to a RabbitMQ server. This appears to have minimal impact on the actual DB system, and gets around the IO issues on the server box.

may help… may not…

Peter C

0
Arthur Wilson

Really interesting post Peter – can’t say I understood all of it!

What’s the link to part two? Thanks

0