Stay Curious!

How to monitor InnoDB lock waits

This is one in a series of articles on how to use the innotop MySQL and InnoDB monitor. In this article I show how innotop can display locks that are causing a transaction to wait.

First, the background. The output of SHOW ENGINE INNODB STATUS’s transaction section prints InnoDB transactions, including information about the locks they’re waiting for, if any:

---TRANSACTION 0 93789797, ACTIVE 20 sec, process no 9537, OS thread id 38900535
9 starting index read
mysql tables in use 1, locked 1
LOCK WAIT 2 lock struct(s), heap size 320
MySQL thread id 23733, query id 194861215 worker 192.168.0.31 xaprb Updating
update test.test set a = 1 where a = 2
------- TRX HAS BEEN WAITING 20 SEC FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 0 page no 299998 n bits 200 index `PRIMARY` of table `test/test` trx id 0 93789797 lock_mode X locks rec but not gap waiting
Record lock, heap no 77 PHYSICAL RECORD: n_fields 15; compact format; info bits 0
[... a bunch of record dump information omitted here ...]

The important information there is the line beginning RECORD LOCKS. It, together with the line above it, tells what the transaction is waiting for, and for how long.

innotop parses this information and formats it for easier comprehension. It’s possible to see which transactions are waiting by using the “T” key to enter InnoDB Transaction mode, and filtering or sorting away all transactions whose “lock_wait_time” is zero. You’d use a similar set of steps as I described in my previous article on finding who locks a table.

While this technique will show you which transactions are waiting, it won’t show you what they’re waiting for. If you have a high-concurrency environment, or if you just have a lot of table or index scans, you might see lots of transactions waiting for locks. Maybe you have a bunch of things being locked for a long time, in many different tables and indexes (not likely, or your server would probably be totally crippled). Or, more likely, there are just a few log-jams causing many transactions to wait. This could be due to bad database architecture, not a fast enough machine, badly written queries, or many other factors. You should be suspicious of badly written queries and badly designed tables first, unless you already know that’s not the problem. That’s why you need to know what everything is blocked on, and then examine the queries that use it.

For that reason, I created a separate display mode for innotop (new in version 0.1.128), specifically to show information about lock waits. You enter that mode with the “W” key, and it looks like the following screenshot (click through for a larger image):

innotop screenshot

By default it’s a pretty terse display, for the reasons I mentioned above. It doesn’t show a lot of information about the transactions; just the connection ID and some information about what it’s waiting for. If you want to see what the connection is doing, you can toggle back to “T” or “Q” mode and get more details, including running EXPLAIN on the query (more about that in future articles). There is also more information that’s hidden by default. You can press “c” to choose the columns you want to view.

This is one method to identify the resources that are most often locked, but as always, I’m curious to hear other ideas.

Posted on Wed, Aug 2, 2006. Approximately 600 Words.

Databases