Re: 2.4DEVEL -> 2.4STABLE?

Well throw my previous short-term fix out. I think this is a lot better.

1) the patch does UFS only. if it's ok then I'll do the other FS's in the
weekend / early next week.
2) both -S options still run a separate check after the swap rebuild. It
should ALWAYS be ok with the new code. (I think we should get rid of that
check or perhaps ifdef it. However for now it's a useful assert() if someone
is forcing a check or testing new check related code.
3) it won't assert on an auto-check. This is by design - if squid is brought
up automatically, the last thing a sysadmin is likely to want is squid
asserting again and again.
4) -S reportonly reads the swap.state, skipping of mis sized files. They
don't get added to the in-memory state though.
5) -S force, or a DIRTY flag cause missing or mis-sized files to not be
added to the in-memory state. mis-sized files are removed except as per 4)
6) we don't check for files that shouldn't be there - should we do so? what
happens if a file exists, isn't in the bitmap and we create a new entry over
that file?
7) I haven't added a mis-sized file check to rebuildfromDirectory (yet). It
may even be there now - I will look into this in the weekend. We don't need
to worry about missing files though :-]

I have tested all the permutations I could think of (and create)
-S reportonly
-S force
missing files
files to big/small

I haven't tested things like
-S reportonly with two files for the same URL and the first encountered
reference exists and the second is missing...

However the logic of the code should work in those cases as well. As soon as
we have a swap file number we check the file on disk. (during dirty startups
or -S startups)

performance - about 1/2 a clean startup for me. (2000 entries/sec on my P133
w/64Mb RAM).

Comments/Abuse etc please

Rob

----- Original Message -----
From: "Adrian Chadd" <adrian@creative.net.au>
To: <squid-dev@squid-cache.org>
Sent: Friday, October 20, 2000 6:38 PM
Subject: Re: 2.4DEVEL -> 2.4STABLE?

> On Fri, Oct 20, 2000, Henrik Nordstrom wrote:
> > Adrian Chadd wrote:
> >
> > > uhm .. to make this clearer - the current code doesn't delete the bad
> > > swapfiles (or perform any other corrective action) so it sits in a
> > > tight loop.
> >
> > Only if you don't use the -N option I presume..
>
> Ok. So writing email at awake+30 seconds is a bad thing. At least
> I don't commit code at the same time.
>
> Robert - how about writing up a patch to implement your doublecheck
> ideas ?
>
>
>
> Adrian
>
> --
> Adrian Chadd The Law of Software Development and
> <adrian@creative.net.au> Envelopment at MIT:
> "Every program in development at MIT expands until it can read mail."
>