Jeff, Nigel,
We do get time a test takes to run, as a part of the output of the test
runs (at least on CentOS runs as I checked here [1]).
To avoid a DHT problem ;) it maybe better to take this sorted list and
assign tests in a cyclic fashion so that all chunks relatively take the
same amount of time to complete, than it being skewed due to the hash?
Shyam
[1] Test run chosen at random:
https://build.gluster.org/job/centos6-regression/1350/consoleFull
On 01/10/2017 05:26 PM, Jeff Darcy wrote:
With regard to assigning files to chunks, I suggest we start by using an algorithm similar to that we use in DHT.
hash=$(cat $filename | md5sum) # convert from hex to decimal?
chunk=$((hash % number_of_chunks))
if [ x"$chunk" = x"$my_chunk_id" ]; then
bash $filename # ...and so on
fi
This is completely automatic, robust as the test set or directory structure changes (or as the number of workers changes), and should give us an approximately equal distribution among chunks.
_______________________________________________
Gluster-devel mailing list
Gluster-devel@xxxxxxxxxxx
http://www.gluster.org/mailman/listinfo/gluster-devel
_______________________________________________
Gluster-devel mailing list
Gluster-devel@xxxxxxxxxxx
http://www.gluster.org/mailman/listinfo/gluster-devel
