>>>>>Hi,
>>>>>Why last few records (newly added before ssytem crash) still could be lost even I have FLUSH after END TRANSACTION in every transaction?
>>>>>
>>>>>Thank you
>>>>
>>>>I believe one possibility is that the OS has a write-cache enabled. I am not sure whether in this case a FLUSH would force a write.
>>>
>>>Have a look at MSKB #Q281281. It provides some interesting information on the operation of FLUSH vis-a-vis the OS.
>>
>>Thanks.
>>
>>Hmmm... "they cache the FLUSH commands and then return operation to the application" seems dangerous to me.
>
>Yes. It does also say (those OSes) "...continue to issue FLUSH commands in the background.".
>Nevertheless, I would like to see two additional things for VFP, and I will be writing a Wish List entry for them:
>1) An option on the FLUSH command to stipulate that the OS is to CONFIRM FLUSH SUCCESS before returning control to VFP;
>2) Improved discussion of the FLUSH command in the VFP Command Reference section of the documentation. It has has only a word or two changed since VFP 2.5 (at least) and the advent of buffering, datasessions, differing handling by different OSes, etc. needs to be described in detail.
>
>cheers

I agree wholeheartedly with both of your suggestions, Jim. Also an option to actually initiate the OS-level flush would be nice, as opposed to waiting for the next one to roll around. I think Hilmar's way of putting it was a very polite understatement. To put it bluntly: A FLUSH that does not flush is virtually useless.

Mike