The WS2812 RGB LED atop the Bowl of Fire Floodlight …
… failed in the usual way after a bit over a year of constant use.
I’d done an unusually good job of epoxying the ersatz heatsink in place:
I wrapped the bulb in a towel with only the heatsink sticking out, whacked the side of the heatsink parallel to the glass with a plastic-face hammer, and it popped right off:
The missing piece of the epoxy ring turned out to be a divot ripped out of the glass, which I didn’t notice until I’d chipped the fragment off the aluminum, so no pictures.
Given the relative strengths of epoxy and glass, pulling a divot seems impossible, but folks doing 3D printing on glass platforms have been reporting exactly that failure for years. The prevailing theory seems to involve small scratches and defects in the glass surface, with subsequent weakening, and stresses applied to the epoxy perpendicular to the glass surface pulling the cracks apart.
Replacing the RGB LED will require drilling it out and probably a complete rewiring, because I seem to have made liberal use of epoxy inside the heatsink and brass tube.
The Smell of Molten Projects in the Morning 
An old article on hide glue (probably in Fine Woodworking) mentioned that the weakest variety of such was used to work glass in just that way. IIRC, as the glue cools and contracts, it takes a divot with it. Never tried it myself; I don’t use hide glue.
OK, I’ll bite: why would anyone want to pull a divot out of glass? I’m sure it has something to do with art, but … [grin]
Not my cup of H2SO4, but yeah, ART. For the
boredinterested, look up “glue glass chipping”. A couple of examples looked all right.boggle
I never knew that was a thing and, it seems, nobody who’s had a 3D printed object rip a chunk out of a glass platform does, either. Ya learn something new every day around here …
Thanks!