Re: Data loss, vacuum, transaction wrap-around

On Sun, 20 Feb 2005 pgsql(at)mohawksoft(dot)com wrote:

> > On Sun, 20 Feb 2005 pgsql(at)mohawksoft(dot)com wrote:
> >
> >> > On Sat, Feb 19, 2005 at 18:04:42 -0500,
> >> >>
> >> >> Now, lets imagine PostgreSQL is being developed by a large company.
> >> QA
> >> >> announces it has found a bug that will cause all the users data to
> >> >> disappear if they don't run a maintenence program correctly.
> >> Vacuuming
> >> >> one
> >> >> or two tables is not enough, you have to vacuum all tables in all
> >> >> databases.
> >> >
> >> > Except that Postgres isn't a large company and doing the work of
> >> > back patching and testing old versions will be done instead of
> >> > more important work.
> >>
> >> PostgreSQL is an open source project that plays with the big guys. Look
> >> at
> >> the Linux kernel. Imagine their file system guys thinking this way.
> >> Linux
> >> would still be Linus' hobbie.
> >
> > So, you are certain that every Linux file system bug has been patched all
> > the way back to say kernel version 1.0 then? Do you have any evidence of
> > this claim?
>
> No one is suggesting back to version 1.0, but critical data loss bugs that
> are present and relvent in used prior versions are fixed.

I still doubt your claim about patching and youhaven't given any evidence,
but let's just make the assumption it's true because otherwise even trying
to hold a discussion is fruitless.

I also dispute your claim based on the backpatching claim that Linux would
be Linus' hobby if the file system guys thought this way. Given that
"stable" Linux branches often aren't, if there weren't aggregators who
provide upgrades that are at least supposedly tested and reasonably stable
and sometimes did their own back patching, Linux wouldn't have the sort of
success it does because people would have to do alot more choosing between
getting bug fix X and doing huge amounts of tests to make sure nothing
else is broken. Thus, I believe you are greatly overstating the effect
that your first claim has towards your second to the point of making an
invalid argument.

Personally, I'd in general wish that 8.0 got a fix for this because that
way we could (after sufficient testing) push an 8.0 version that we
considered stable to suggest people move to. However, I don't have a
whole lot of time to do such a patch nor to do "sufficient" testing, and
I'm not arrogant enough to believe I can order around volunteers and
companies I'm not a customer of.

> >> >> This bug would get marked as a critical error and a full scale effort
> >> >> would be made to contact previous users to upgrade or check their
> >> >> procedures.
> >> >
> >> > I don't think all commercial companies would do that. I doubt that
> >> even
> >> > most of them would.
> >>
> >> Database companies? You bet they would.
> >
> > Do you have any evidence or are you merely spouting an opinion as fact?
> >
> With Oracle and DB2, yes I have some personal experience.

My last company's experience with Oracle support still leaves me
questioning that claim. They basically got "don't do that then or move to
the newest major revision" when they had a construct which caused the
server to stop responding. It's not the same conditions (although I
believe the DBA did reload from backup because noone could guarantee that
there couldn't possibly have been dataloss), but it's certainly not
indicative of the sort of "full scale efforts" you're describing.