Thread: Re: [HACKERS] The dangers of "-F"

> For instance, if there are assumptions that all data blocks are
> written before this fact is recorded in a log file, then
> "write data blocks" "fsynch" "write log" "fsynch" doesn't break
> that assumption, 
> 
Are we really doing a sync after the pg_log write ? While the sync
after datablock write seems necessary to guarantee consistency,
the sync after log write is actually not necessary to guarantee consistency.
Would it be a first step, to special case the writing to pg_log, as
to not sync (extra switch to backend) ? This would avoid the syncs
for read only transactions, since they don't cause data block writes.

Andreas

> 
> > For instance, if there are assumptions that all data blocks are
> > written before this fact is recorded in a log file, then
> > "write data blocks" "fsynch" "write log" "fsynch" doesn't break
> > that assumption, 
> > 
> Are we really doing a sync after the pg_log write ? While the sync
> after datablock write seems necessary to guarantee consistency,
> the sync after log write is actually not necessary to guarantee consistency.
> Would it be a first step, to special case the writing to pg_log, as
> to not sync (extra switch to backend) ? This would avoid the syncs
> for read only transactions, since they don't cause data block writes.

You are right.  We don't need a sync after the pg_log write.

--  Bruce Momjian                        |  http://www.op.net/~candle maillist@candle.pha.pa.us            |  (610)
853-3000+  If your life is a hard drive,     |  830 Blythe Avenue +  Christ can be your backup.        |  Drexel Hill,
Pennsylvania19026
 

Bruce Momjian wrote:
> 
> >
> > > For instance, if there are assumptions that all data blocks are
> > > written before this fact is recorded in a log file, then
> > > "write data blocks" "fsynch" "write log" "fsynch" doesn't break
> > > that assumption,
> > >
> > Are we really doing a sync after the pg_log write ? While the sync
> > after datablock write seems necessary to guarantee consistency,
> > the sync after log write is actually not necessary to guarantee consistency.
> > Would it be a first step, to special case the writing to pg_log, as
> > to not sync (extra switch to backend) ? This would avoid the syncs
> > for read only transactions, since they don't cause data block writes.
> 
> You are right.  We don't need a sync after the pg_log write.

We need. I agreed with extra switch to backend.

Vadim

> Bruce Momjian wrote:
> > 
> > >
> > > > For instance, if there are assumptions that all data blocks are
> > > > written before this fact is recorded in a log file, then
> > > > "write data blocks" "fsynch" "write log" "fsynch" doesn't break
> > > > that assumption,
> > > >
> > > Are we really doing a sync after the pg_log write ? While the sync
> > > after datablock write seems necessary to guarantee consistency,
> > > the sync after log write is actually not necessary to guarantee consistency.
> > > Would it be a first step, to special case the writing to pg_log, as
> > > to not sync (extra switch to backend) ? This would avoid the syncs
> > > for read only transactions, since they don't cause data block writes.
> > 
> > You are right.  We don't need a sync after the pg_log write.
> 
> We need. I agreed with extra switch to backend.

We need the switch only so was can "guarentee" that we can restore up
until 30 seconds before crash.  Without fsync of pg_log, we are waiting
for the OS to do the sync, and that will add at most another 30 seconds
of open time(OS's sync every 30 seconds, usually).  One nice thing I
think will be than an independent process will be doing the fsync, so no
queries will have to wait for it to happen.

--  Bruce Momjian                        |  http://www.op.net/~candle maillist@candle.pha.pa.us            |  (610)
853-3000+  If your life is a hard drive,     |  830 Blythe Avenue +  Christ can be your backup.        |  Drexel Hill,
Pennsylvania19026