1038
1038
* some snapshot we have. Since we examine the procarray with only shared
1039
1039
* lock, there are race conditions: a backend could set its xmin just after
1040
1040
* we look. Indeed, on multiprocessors with weak memory ordering, the
1042
1042
* that such a backend must have held shared ProcArrayLock overlapping our
1043
1043
* own hold of ProcArrayLock, else we would see its xmin update. Therefore,
1044
1044
* any snapshot the other backend is taking concurrently with our scan cannot