Re: [HACKERS] Fairly serious bug induced by latest guc enum changes

Tom Lane wrote:
> Magnus Hagander <magnus@hagander.net> writes:
>> Tom Lane wrote:
>>> Hmm ... or at least more or less fixed. Seems like there's no provision
>>> to close and reopen the file if enableFsync changes. Not sure if that's
>>> worth worrying about.
>
>> We didn't have that before either, did we?
>
> No, so I think it's a pre-existing bug.

Ok, at least I'm reading the code right.

>> We close it when the sync bit
>> changes, but not just if we change say between fsync() and fdatasync().
>> Is there any actual reason we'd want to close it?
>
> The point is that if you turn the fsync GUC on or off while using a wal
> sync mode that requires supplying an option flag to open(), then really
> you ought to close the WAL file and re-open it with the new correct
> option flags. The fact that we're not doing that implies that the
> effects of a change in fsync might not fully take effect until the next
> WAL segment is started. Whether this is worth fixing isn't real clear.

What scenario does it actually happen in, though? Doesn't the check:
if (get_sync_bit(sync_method) != get_sync_bit(new_sync_method))

take care of that? If the sync bit changed, we close the file?

Or are you talking about changing the variable "fsync"? If so, doesn't
"fsync=off" also change the behavior of other parts of the code, so it's
not just WAL, which means it'd be pretty unsafe *anyway* unless you
actually "sync" the disks, and not just fsync?

//Magnus

//Magnus

--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers