[erlang-questions] disconnected nodes
Wed Feb 19 17:48:52 CET 2014
Just a heads up for anyone following this / people from the future who
It's not clear what exactly caused the EBADFs and EINVALs, but we were
running R15B03 and it seems this was yet another instance of the old
'schedulers not being woken up' issue, as discussed e.g. here:
Essentially, after some time, only one scheduler (out of 16!) would be
utilised and had an ever-growing insane run queue.
Setting +sws proposal and +swt very_low seems to have resolved
everything and now the run queue distribution is completely uniform. We
will also be migrating to R16B ASAP.
More information about the erlang-questions