Sorry I haven't had a chance to look at this, but I'd guess the stats
package is crashing becuase of the weird reconfigure stuff we do
during the test, and if it exceeds a crash threshold, other CouchDB
components get restarted, and if full commits is off, and then other
tests fail. So basically the bug is in our tests, not the production
code itself.
To test my hypothesis, can we disable the stats stuff for people who
have failures and see if that fixes it?
-Damien
On Oct 10, 2009, at 11:37 AM, Chris Anderson wrote:
On Sat, Oct 10, 2009 at 9:12 AM, Christopher Brown <[email protected]>
wrote:
No one responded to my message about intermittent failures and I'm
concerned about this as well.
Can someone from the core team chime in about this? Is it considered
ok, and is there a plan to eventually fix this?
My gut tells me this has to do with delayed_commits and certain
performance environments. We should probably add an option to the test
runner to put it in full-commit mode. This way when people are running
it on platforms where the timing becomes an issue they can switch to
slow-but-safe, so that all operations are fully synchronous from the
browser down to disk.
If you feel like messing around with this toggle, changing
X-Couch-Full-Commit from false to true at the top of the
rev_stemming.js file will probably fix this.
I don't think intermittent failures are a reason to block 0.10, but I
do think we should clean this up for 0.11. Maybe the simplest thing is
to add a checkbox to the test suite to toggle full-commit.
Chris
Thanks,
Chris
On Sat, Oct 10, 2009 at 4:08 AM, Robert Dionne
<[email protected]> wrote:
Sporadic failures in the stats test are fairly common on my machine,
"reduce" perhaps less often. The stats test fails often enough
that I'm
pleasantly surprised when all the tests run. I believe you're
correct that
the issue with stats is a race condition. I have some time today
and will
take a look at it, it's been bugging me.
On Oct 10, 2009, at 6:28 AM, Jason Davies wrote:
On 8 Oct 2009, at 05:29, Noah Slater wrote:
I would like call a vote for the Apache CouchDB 0.10.0 release,
second
round.
I'm getting intermittent failures on the "reduce" and "stats"
tests. The
reduce test fails with "Assertion failed: db.info().doc_count ==
(i - 1) *
10 * 11 + (j + 1) * 11". The stats test fails with "Assertion
'should
increase the request count for every request' failed: expected
'28', got
'26'".
Platform: Mac OS X 10.6.1, Erlang R13B02 (erts-5.7.3) [source]
[64-bit]
[smp:2:2] [rq:2] [async-threads:0] [hipe] [kernel-poll:false]
I tried recompiling Erlang without hipe and still got the same
results.
I'm guessing they are simply due to race conditions in the tests
and
aren't anything to worry about so I would be a +1 if someone else
could
comment on the nature of these failures.
Thanks,
--
Jason Davies
www.jasondavies.com
--
Christopher Brown, VP of Engineering
Opscode, Inc.
T: (425) 281-0727, E: [email protected]
Twitter, IRC, Github: skeptomai
--
Chris Anderson
http://jchrisa.net
http://couch.io
