Reports until 21:03, Tuesday 14 July 2015
H1 CDS (CDS, ISC, SUS)
jeffrey.kissel@LIGO.ORG - posted 21:03, Tuesday 14 July 2015 (19653)
Filter Bank GAIN / OFFSET Ramp
J. Kissel, S. Dwyer, J. Driggers, R. Bork, D. Barker

While trying to recover green alignment this afternoon, we found that guardian transitions were getting stuck because the alignment offset's ramping light would not turn off after the value requested would reach the requested value. We quickly (thanks to Dave and Rolf) that this was because the SUS OFFLOAD python / guardian script that's used to offload the green WFS to the alignment sliders is requesting an offset with a ridiculous about of precision (i.e. to the 16th decimal place, i.e double precision). The internal EPICs ramp doesn't have the same precision, my guess is float, (i.e. to the 8th decimal place). Since EPICs can only make 1e-8 steps, and 1e-16 final offset is requested, he ramp light never turns off.

There are three levels of solutions:
- To fix the problem immediately, write the requested OFFSET value (to a reasonable precision, like 3 decimal places) directly to EPICs via command line, then the ramp light goes off and you can move on with your initial alignment. We've done this
- To fix this problem in the medium term, make sure that all scripts writing EPICs records round off the request to at least floating point precision, if not within-human-reason-precision-for-an-EPICs-record of 1e-3. We will do this.
- To fix this problem long-term, EPICs should be fixed to round off or even truncate any requests to the precision that it can handle. We'll probably never do this.
Images attached to this report