So to be more specific :

• dark exptime=20 duplicate=3
• scienceTrace halogen=4 duplicate=1
• dark exptime=20 duplicate=3
• scienceArc krypton=3 duplicate=1
• dark exptime=20 duplicate=3
• scienceArc argon=10 duplicate=1
• dark exptime=20 duplicate=3
• scienceArc xenon=3 duplicate=1

repeat this 5 times.

So, I reduced a LAM krypton exposure, there is still quite a lot of bad features remaining in the spectra, depending on the fiber.
The black curve is the clipped-mean spectra of all fibers.

price when I plot using pfsArm.mask, almost everything get masked, am I misusing something ?

What mask planes do you have, and what are their counts? This might help.

Thanks, it definely helped, I've attached a file with the masks value and their occurence for each fiber.

Just ignore only the (mask & NO_DATA) != 0 pixels, and you'll get a lot more spectral data.

Thanks.
I've also noticed that INTRP flag contains partly junk as well :

INTRP means we interpolated the pixel because it's bad (which makes the image looks nicer). Those pixels should be ignored during the extraction (not because of INTRP, but because of what caused INTRP to get set). If the spectra have bad pixels with INTRP set, it means that there are image pixels in the same row as the INTRP pixels that are also bad but we're not catching.

The conclusion is that we're able to extract spectra, see figures from N2(sn1).
But there are still some junk features in it, those are mostly flagged INTRP.
If you ignore INTRP | NO_DATA, the spectra is cleaner but you lose quite a lot of data points.