On 10/23/2014 08:59 AM, Fujii Masao wrote: Sounds reasonable, for back-branches. Although I'm still worried we might miss some corner-case unless we go with a more wholesale solution.
Don't really want to be the intruder here, but isn't that the simple patch attached?
That's not right. Should check *after* the write if the segment was completed, and close it if so. Like the attached.
There is still a small window between XLogWalRcvFlush and XLogArchiveForceDone in XLogWalRcvWrite if the standby crashes exactly between them.
Yeah. I think we can live with that.
Just tested this patch with the combo pg_switch_xlog() + stop/immediate and the apparition of the .ready files is not fixed.
