quota deadlock on 6.1-RC1

[Scott Long]

Chris wrote:

> On 05/05/06, Mark Linimon <linimon at lonesome.com> wrote:
> 
>> Make Jakubik wrote:
>>
>> > FreeBSD users now demand stability and performance, as opposed to an
>> > influx of new bells and whistles just before the release [...] I fully
>> > understand that this is a volunteer project [...]
>>
>> I'm sorry, but the former statement proves the latter false.
>>
>> Let's try to do our Semi-Annual Refresher Course On Open Source 
>> Development:
>>
>> The developers (at least 99% of them) work for free.  In their own spare
>> time.  Their motivations vary but I don't believe any of those include
>> being in a position to feel they need to respond to demands.  That's not
>> a positive motivator.  If they wanted to be in that position, they could
>> just stay at their $realjobs for those extra hours.
>>
>> Part of their shared goals, however, is to turn out the best system that
>> they possibly can, in the hopes that people will find it useful and want
>> to contribute back to it.  However, there are no guarantees involved,
>> implicit or explicit.  (If you want to compare and contrast to how much
>> "guarantee" you get from closed-source development, please pull out a 
>> copy
>> of your EULA.  They barely even "guarantee" that there are bits on the 
>> CD.)
>>
>> There's a long process where the developers try to agree on what features
>> need to be included and what bugs need to be fixed.  From the standpoint
>> of the people who attempt to coordinate this process, in technical jargon
>> the process is know as "herding cats".  I would be able to serve as an
>> expert witness in court about this.  (Side note: some of the cats hiss,
>> bite, and scratch; very few, if any, have _any_ interest in being 
>> herder.)
>>
>> There are always tradeoffs between stability and features.  During the
>> 5.X cycle we managed in some degree to de-optimize both: we had features
>> that were only available in an "experimental" branch that some people
>> considered critical (wireless, anyone?) while that "experimental" branch
>> was unsuitable for production use.  The idea was that we would hold on
>> to declaring 5.X "STABLE" until all the major bugs were fixed.
>>
>> And as a consequence, we didn't release for -- what, 2 years?
>>
>> So we've thrown out the idea of "wait until every possible bug is fixed."
>> It leads to rare releases, and larger code chaos, larger instability, and
>> allows FreeBSD's detractors to sniff "well, they're never going to 
>> release
>> anything again."  (Notice how the "BSD is dying" crowd on Slashdot has
>> been a lot quieter since we released 6.0?)
>>
>> So now what we're doing is trying to come up with more regular (not on
>> absolute deadline) releases, with smaller feature sets, to enable smaller
>> sets of new code to be debugged simultaneously.
>>
>> The features that some users see as critical, others don't.  (I don't 
>> have
>> quotas enabled; I have disabled soft-updates on the theory that as a 
>> single
>> user I can trade longer startup time for possibly greater error 
>> recovery).
>>
>> > I wish i was a good Unix C programmer myself, so i could contribute 
>> in a
>> > more direct way
>>
>> And there's the rub.  The people who _are_ good Unix/C programmers are 
>> the
>> ones doing the work, and as such, feel that they have the final say about
>> what goes and what doesn't.  While I hope that each of them will 
>> listen to
>> what individual users are saying, and take good suggestions to heart, the
>> fact remains that they are under no _obligation_ to do so.
>>
>> Finally, as has been pointed out already, the interactions between
>> quotas, soft-updates, and the rest of the system have problems that are
>> probably going to be fairly difficult to isolate and test.  Thus, they
>> could take days, weeks, or months.  If we were to hold the release until
>> these were fixed, basically our last 2 months of QA would be out the
>> window.  This is not a way to keep developers happy; at some point they
>> will tire of the inability to work on the things that they find 
>> interesting,
>> and wander off to find something else more fun to do.
>>
>> Summary: it's too bad that there are regressions, but sometime they're
>> a fact of life.  If these regressions are in areas that are critical for
>> a certain user, that user should just skip 6.1 and wait for 6.2.  The
>> stability gains that 6.1 have over 6.0 in so many other areas means that
>> it's time for 6.1 to go out the door, because for the majority of users
>> it's going to be a big win.
>>
>> As always, we welcome test cases on e.g. non-critical systems, earlier
>> in the release process (or between releases), where there's enough time
>> to thoroughly test that any proposed fix doesn't cause more problems than
>> it cures.  Within a few days of release, that simply isn't the case right
>> now.
>>
>> mcl
>> _______________________________________________
> 
> 
> nice post and ultimately you are correct, I personally would prefere a
> higher gap between releases for the following reasons tho.
> 
> 1 - it does increase stability if the extra time is spent fixing bugs
> and testing the fixes.
> 

No, actually it doesn't.  The real push to get bugs fixed doesn't happen
until about 2 weeks into the start of the release cycle.  It doesn't 
matter if it's been 2 months or 2 years since the last release; letting 
it sit longer absolutely does NOT result in fewer bugs.  in fact, it
results in MORE bugs, because more happens in the longer gap that needs
to be fixed.

> 2 - its easier for administration, upgrading freebsd every 2 to 3
> months isnt ideal for administrators.  I know releases are supported
> for much longer but given what kris told me recently that ports are
> only supported on the very latest release I found that a bit worrying.

There is nothing forcing users to upgrade at every release.  For my
company's product, I just updated us to 6.1, and I don't expect to do
a full update again until 6.3 or 6.4, other than some small targetted
micro-updates as needed.  For my production mail and firewall servers,
they are running RELENG_6_0, and I probably won't upgrade them until
6.2 or 6.3.

> 
> 3 - it raises profile for the release, the euphoria surrounding a
> release is much less if there is one every other month.

It was never, ever suggested that there be one release per month.  It
was planned that there be a release every 18 weeks.  There are snapshot
builds every month, and that is done to make sure that the releases
scripts continue to work, and give the adventurous users something to
play with.

> 
> In terms of a new major version ie. 7.x over 6.x I think that should
> only be released when their is something major changed like 5.x over
> 4.x had a new filesystem smp support etc. and I happen to think
> because 5.1 and 5.2 werent STABLE the 5.3 STABLE release turned out
> very stable at least in my experience and was more stable then 6.0.

5.1 and 5.2 were not given the 'STABLE' nomenclature.  As for 5.3 being
"more stable than 6.0", that is a very subjective statement.  I can
give you a list of at least 500 things that were fixed in between 5.3
and 6.0.  Unfortunately, a few things regressed, as often happens.

> What 
> was major difference between 6.0 and 5.4?  if I understand it
> correctly their has been a lot of work done fixing problems that
> existed in 5.x but no new feature to shout about.  ULE which was
> unfinished in 5.x still remains unfinished in 6.x and I wonder if will
> be a unfinished feature in 7.x.

You obviously never read the release notes, sorry.  MPSAFE VFS was a BIG
feature for 6.0.  Given that your discussion seems to be headed away 
from researched facts, I'll just stop here.  I won't even recognize your
statement about adding Java to the base system.

Scott