Closed Gbury closed 8 years ago
Seems that not using the shiny --lock=true
, and instead doing something like frogmap --lock=false -- 'froglock cmd'
solves the problem, so I think the cause is that file descriptors opened by frogmap are not cleaned/closed soon enough and the frogmap
call reaches the file descriptors limit (which can be seen with ulimit -{S,H}n
).
So while there is a temporary fix, it might be nice to see how we could clean frogmap
's sub processes file descriptors in time.
I don't remember, but I thought most of IOs were made through with_foo
functions…
Closed with afc19e9e6220799dc73da632faabe35699f016be
While running some benchs lately, i have seen that when doing a lot of tests (i.e about 20000), frogmap seems to freeze after doing 1018 tasks, using internal locking (i.e --lock true option, which is the default). Additionally, when trying to check what happens with
froglock status
, froglock seemed to be frozen too.This might be a very problematic issue.