#91 [2.6] Missed signal

[Brian J. Watson]

When demo-proclb is run repeatedly on the most recent
2.6.10 kernel (post 1.9.0), eventually one of the
children misses the SIGUSR1 that's used to tell them to
start processing records. It might take awhile for this
to happen; I usually leave the following running overnight:

# while [ 1 ]; do ./demo-proclb -n 16 datafile; done

This is almost certainly a 2.6 kernel bug, because it
does not happen when I run the same command overnight
on the same test cluster with the latest 2.4 kernel
(post 1.2.2).

Brian

[Roger Tsang]

Possibly fixed in 1.9.6 with refactoring of ssi_wait/signal/broadcast_event()