KillMS error: "Problem inverting Matrix, saving as errSVDArray_9440 will make it svd-able"

[BiavaNadia]

Running the pipeline with python3 in Bologna and Trieste on different datasets we found a problem in inverting matrix with killMS at step DDS3_full. This happens for two / three subbands (around 120 MHz and at 168 MHz) and the pipeline stops when it encounters this problem. We got around the problem by removing the problematic datasets but noting that they do not create problems in the previous steps. Also one of the two observations had already been calibrated one years ago using python2 without encountering this problem.
Could it be a problem with using python3?

[cyriltasse]

This one is a little bit tough to investigate without reproducing myself... I'll invite you to the dev repo of kMS/DDF, you can have a try with these and if that doesn't solve the problem I'll have to reproduce...

[BiavaNadia]

Ok, thanks