[Python-Dev] latest bsddb3 test problems (original) (raw)

Gregory P. Smith greg at electricrain.com
Mon Sep 22 03:50:32 EDT 2003

I'm seeing this fail as well, on Redhat 9. Backing out Greg's changes over the weekend doesn't help... the tests still mark as "passed", even though there's the fairly horrible error/warnings. I'm guessing something that raises an error in a non-main thread doesn't get picked up by regrtest.py

testbsddb3 is failing on pretty much every box I've tried it on here - Solaris and Redhat linux, most running the latest-n-greatest release of DB4.1 (we use it pretty heavily in a variety of internal applications). Solaris 8 (DB 4.1.25. gcc 2.96): DBAgainError: (11, 'Resource temporarily unavailable -- mmap: Resource temporarily unavailable') Solaris 7 (DB 4.1.25, gcc 2.95.2) Crashes with a SIGBUS. The version of bsddb.c from a week ago has a lot of errors, but doesn't crash.

That's sounds like inconsistent behaviour. Look at the diffs. All of the changes since the python 2.3 release to _bsddb.c have been trivial. The _bsddb.c file version is not the culprit.

Redhat 10beta/Rawhide (DB 4.1.25) DBNoMemoryError: (12, 'Cannot allocate memory -- Lock table is out of available locks') alternating with SIGBUS

Redhat 7.1,7.2,7.3 (DB 3.2.9) Crashes with a SIGBUS.

I'm going to check with Spambayes as an actual application and see whether the tests that are failing are at fault, or if we've got something that's just busted. Can anyone report the bsddb3 tests as passing on their system? Note that they're not run by default.

It is passing on the systems that i've tested it on in the past few days: (Using BerkeleyDB 4.0, 4.1 and a prerelease of 4.2).

Linux/i686 Gentoo 1.4 Linux/alpha Debian "testing"

I do see the occasional DBDeadLockError which can be ignored. (Yes, I agree, the test suite needs work to get rid of things like that. if its not going to consider it an error it should at least catch with a comment stating why to avoid the confusion)

NOTE: A weakness of the bsddb tests that I have witnessed causing problems:

It often doesn't wipe its temporary directory before the tests (in theory it should clean up on shutdown, but thats only if all goes well in the tearDown methods). If a previous run was aborted abnormally and left a database environment & db around it can cause problems during future test runs if the testsuite trys to use that environment and db.

I'm at a loss as to whether to delay the release or not...

Anthony

Do the same tests fail similarly when run on python 2.3? I'd imagine so. IMHO, I wouldn't hold python 2.3.1 up due to bsddb test suite issues.

-greg

More information about the Python-Dev mailing list