At least on HPUX, select with delay.tv_sec = 0 and delay.tv_usec = 1000000

does not lead to a one-second delay, but to an immediate EINVAL failure.
This causes CHECKPOINT to crash with s_lock_stuck much too quickly :-(.
Fix by breaking down the requested wait div/mod 1e6.

diff --git a/src/backend/storage/buffer/s_lock.c b/src/backend/storage/buffer/s_lock.c
index 4cbc0ca55128f1f84215a5fd3372d09b9c5036a5..ae9e7c542d820e307b7ef0fcd667673f83e8c661 100644 (file)
--- a/src/backend/storage/buffer/s_lock.c
+++ b/src/backend/storage/buffer/s_lock.c
@@ -91,8 +91,8 @@ s_lock_sleep(unsigned spins, int timeout, int microsec,
 
        if (microsec > 0)
        {
-               delay.tv_sec = 0;
-               delay.tv_usec = microsec;
+               delay.tv_sec = microsec / 1000000;
+               delay.tv_usec = microsec % 1000000;
        }
        else
        {