[erlang-questions] Hot code loading and OTP

Jay Nelson jay@REDACTED
Fri Sep 21 17:19:56 CEST 2007


A friend who hadn't seen erlang before just finished reading Joe's  
book.  We were discussing hot code loading and he didn't realize that  
prefixing a function call with a Module: caused a load of code from  
disk.  I attempted to search through the book and find the  
description, but was unable to find an explicit reference.  Was this  
accidentally left out?

My real question has to deal with OTP and relup / appup code  
loading.  Does the use of my error logging functions such as  
my_logger:log_error(Msg) interfere with an orderly code upgrade?   
Should I eschew all uses of Module:Function (I guess by using import,  
unless there is some other mechanism to avoid triggering an unwanted  
code load) so that code is only loaded on signal from an OTP upgrade?

I was just going to try an appup for the first time when it occurred  
to me that the logger would get upgraded before the rest of the  
system was even copied to disk properly depending on how I install  
the software...

As I write this I think I've realized how OTP achieves the code  
upgrade in an orderly fashion:

1) The OTP release system is guided by the current release version  
and directory.
2) Any reference to M,F,A or M:F checks the initial location on disk  
from which it was loaded -- which is determined by the release and  
module-vsn.num directory name in the lib directory.
3) Appup and relup change the release number and vsn.num for all  
modules on disk.
4) The triggering switch is controlled and causes a lockstep upgrade  
by switching the code loader to be using the release directories.

Just be careful not to replace any beam file in place in a previous  
release directory or you will get an uncontrolled code upgrade.

Is that all correct?

jay




More information about the erlang-questions mailing list