Re: [Log4cpp-devel] RE: Log4cpp on OpenVMS

[Tony C. <dra...@as...>]

Hi,

A correction to my earlier's posting.

fsync() actually returns only after the I/O completes.
There's an alternative sync() which returns before the I/O completes.

I think O_SYNC and fsync() should be pretty much the same, with fsync()
doing the file I/O synchronization explicitely.

Tony

Bastiaan Bakker wrote:
> 
> Tony Cheung wrote:
> 
> >Hi Steve and others,
> >
> >First, the changes (if made) will be applied to OpenVMS compilations
> >only. I am not sure if other platforms require the changes.
> >
> >For the disk caching, in order allow other users to "tail -f" on the
> >application log real-time. The OS needs to flush the changes to the disk
> >after each line of the logging. The caching should only be applicable to
> >within a single line of logging, otherwise there would be a delay for
> >the last few lines of logging to show up. BTW, the fsync() returns
> >before the I/O completes.
> >
> >I agree it's a tradeoff between real-time feedback or better disk
> >performance.
> >
> >Ideas?
> >
> >Tony
> >
> 
> This feature may be more generally useful as well. In combination with a
> write cache disabled disk it can provide 'extra safe' logging. Of course
> log4j/log4cpp are still 'non reliable, best effort'  loggers, but this
> may help a bit.
> So I  suggest to add this as a runtime option to FileAppender. Now for
> the implementation,  doe anyone have a good opion on whether to use
> fdatasync() or the O_SYNC open flag?
> 
> Bastiaan