Tom Lane wrote:
> Andrew Dunstan <andrew@dunslane.net> writes:
>
>> We've seen similar things with other Windows file operations, IIRC. What
>> bothers me is that the problem might be precisely because the 1 second
>> sleep between the CloseHandle() call and the CreateFileMapping() call
>> might not be enough due to system load, so repeating the cycle without
>> increasing the sleep period will just repeat the error.
>>
>
> What system load? This is only called after all the backends are dead.
> And surely one CreateFileMapping syscall per second does not materially
> contribute to any load that is being caused by something else.
>
>
>
I didn't say Postgres was creating the load. In the case where this has
been happening for my client, there is an Apache server which can chew
up the machine mightily. I don't have any evidence that just repeating
the cycle a few times won't work, but neither do you have any that it
will, and I don't think the extra code complexity will be terribly
great. If it were more than a few extra lines I'd probably agree with you.
cheers
abdrew