Pythian Group

We hit an ORA-01450 error today trying to do online rebuild for an index in an unusable state. This was a non-unique index on a fairly large column — VARCHAR2(800 CHAR).

SQL> alter index i1 rebuild online;
alter index i1 rebuild online
*
ERROR at line 1:
ORA-00604: error occurred at recursive SQL level 1
ORA-01450: maximum key length (3215) exceeded

It rang a bell. I remembered that I encountered this issue a while ago, but I couldn’t recall the details. I know that it has nothing to do with the actual data size — it’s an error that can occur during index creation. A single index block must be able to fit at least two index entries, so the maximum index key size defined by the block size and overhead.

The key length is calculated as:

key length = sum of all column lengths
           + number of columns
           + 2 (key length)
           + 6 (restricted ROWID)
           + 1 (ROWID field length)

If I didn’t miss anything, the key length for my index should be 800 + 1 + 2 + 6 + 1 = 810. Hold on . . . we have character length semantic here — let’s check the database character set:

Read the rest of this entry . . .

So, how does one gather statistics on indexes? With InnoDB it’s one thing, with MyISAM it’s another thing. This post however, will focus on MyISAM. There are various ways, each having their own drawbacks and positive sides.

What about ANALYZE TABLE? It’s great — very fast in InnoDB. But with MyISAM a lock occurs while scanning the table.

OPTIMIZE TABLE? Still relatively quick with MyISAM, unless the table in question is huge. InnoDB requires an ALTER TABLE — it might take forever, depending on the server configuration, as OPTIMIZE TABLE for InnoDB maps to a ALTER TABLE tableName ENGINE=InnoDB

We all know how ANALYZE TABLE, REPAIR TABLE, and OPTIMIZE TABLE work with MyISAM. However, there’s a less explored way, for MyISAM — the myisam_stats_method.

Read the rest of this entry . . .