Subject | Re: [firebird-support] Re: How can a pointer page vanish? |
---|---|
Author | Helen Borrie |
Post date | 2007-10-09T05:19:03Z |
At 02:18 PM 9/10/2007, you wrote:
delete-stumps and cruft relating to rollbacks, and flag them OK to
go. But it won't touch anything on pages that remain "interesting"
in some way. The Oldest Transaction (=="Oldest interesting") tells
you that there are no transactions older (lower numbered) that have
an interest in any pages. That Oldest Transaction is interested in
some pages that were touched after it began but we can't tell what
other, more recent inactive transactions are "interesting".
I see you restored the DB at the end of August and there have been 11
more transactions than are accounted for by Next
Transaction....wonder what they were? I've seen higher generation
numbers than 25 million, too.
It's all a bit academic until Vlad or Ann eyeballs the problem in
terms of the last encountered problem and the numbers...the one where
the engine thought it had maxed out on record_ids...maybe it maxed
out on something else that it doesn't know how to report on. It's a
bit like picking over crow's entrails, isn't it? The index stats for
that table could be useful, if they were accessible, since indexes
use pointer pages too.
./heLen
>According to our DBA a sweep is being run nightly.Sweep will do what sweep can...it will find any obsolete recversions,
>on the database how could then "The gap" be over a million?
delete-stumps and cruft relating to rollbacks, and flag them OK to
go. But it won't touch anything on pages that remain "interesting"
in some way. The Oldest Transaction (=="Oldest interesting") tells
you that there are no transactions older (lower numbered) that have
an interest in any pages. That Oldest Transaction is interested in
some pages that were touched after it began but we can't tell what
other, more recent inactive transactions are "interesting".
I see you restored the DB at the end of August and there have been 11
more transactions than are accounted for by Next
Transaction....wonder what they were? I've seen higher generation
numbers than 25 million, too.
It's all a bit academic until Vlad or Ann eyeballs the problem in
terms of the last encountered problem and the numbers...the one where
the engine thought it had maxed out on record_ids...maybe it maxed
out on something else that it doesn't know how to report on. It's a
bit like picking over crow's entrails, isn't it? The index stats for
that table could be useful, if they were accessible, since indexes
use pointer pages too.
./heLen