index xyz is not a btree

Started by David Parkeralmost 21 years ago3 messagesgeneral
Jump to latest
#1David Parker
dparker@tazznetworks.com

We are getting the "index xyz is not a btree" error pretty frequently in
our build system (in which we build several databases). This is with
8.0.1 on Solaris 9/intel. During a database build we create the
database, import metadata, import some application data, then finally
run a vacuum on it.

The error only happens intermittently, but every occurrence is
consistent:

1) always the same index - "attribute_pkey" on our "attirbute" table
- one created implicitly by a primary key constraint
2) the error happens during a vacuum command

From what I have found in the archives the consensus is that this error

comes from a bad disk or a bad build of postgres, but I don't know what
exactly I would have to screw up in configure to get a "bad build" of
postgres (everything seems to be working fine with it, otherwise).

When this occurs we can generally just re-run the build, and it finishes
fine, so maybe it is some kind of disk weirdness, but I have not seen
this with any of our 7.4.5 builds (which may or may not mean
anything....)

Any suggestions about where to look in the source code for clues or
specific debugging I can turn on would be appreciated. Thanks!

- DAP
------------------------------------------------------------------------
----------
David Parker Tazz Networks (401) 709-5130

#2Tom Lane
tgl@sss.pgh.pa.us
In reply to: David Parker (#1)
Re: index xyz is not a btree

"David Parker" <dparker@tazznetworks.com> writes:

The error only happens intermittently, but every occurrence is
consistent:

1) always the same index - "attribute_pkey" on our "attirbute" table
- one created implicitly by a primary key constraint
2) the error happens during a vacuum command

Hmm ... that seems a bit too consistent for a hardware issue doesn't it?

Have you looked at the index file with pg_filedump, or even just od?

regards, tom lane

#3David Parker
dparker@tazznetworks.com
In reply to: Tom Lane (#2)
Re: index xyz is not a btree

No, I haven't tried that - I wasn't aware of pg_filedump. What's tricky
is that when the build detects a failure, it deletes the output
directory (so we don't get left with a half-built db) so I'll have to
fiddle with it....

I just downloaded pg_filedump from the redhat site, so I'll play around
with it, and report back if/when I find anything.

Thanks.

- DAP

Show quoted text

1) always the same index - "attribute_pkey" on our "attirbute"
table
- one created implicitly by a primary key constraint
2) the error happens during a vacuum command

Hmm ... that seems a bit too consistent for a hardware issue
doesn't it?

Have you looked at the index file with pg_filedump, or even just od?

regards, tom lane