[Swift-devel] how to put different wrapper behaviour into production

[Ben Clifford]

Last week or so I made some patches that change wrapper.sh to copy lots of 
stuff around to a(n assumed) worker-local filesystem rather than using the 
shared filesystem.

I don't particularly like this for general use - it means doing more 
steps, and more stuff to go wrong. Most especially, the worker-node-local 
info log files mean that if something goes wrong during execution (as 
often happens) there is a much greater level of difficulty in getting hold 
of those logs to debug.

There are two paths that I see:

i) add a swift runtime option that is passed to the wrapper, to select 
more-worker-node-local or less-worker-node-local behaviour; with one 
wrapper script able to function in both modes.

or

ii) allow the wrapper script to be specified as a runtime option; supply 
the standard wrapper script and the worker-node local script.


Option i leads down a path of perhaps having lots of different options 
passed to the worker. This might be a good thing or might not.

Option ii allows more open ended customisation of the wrapper scripts, but 
is likely to result in people keeping their own versions of the wrapper 
script around which will quickly stagnate and cause problems when they try 
to use.

I'm somewhat inclined towards option ii.

--