Currently, in response to memory pressure events, we drop all WebGLContexts with context-lost. What we should consider is doing a first round of memory freeing in response to memory pressure events, where we drop all caches and all non-required memory buffers. That is, we should do a pass where we try to benignly free up memory before doing a second 'critical memory event' pass where we break anything. Concept: Memory pressure event: Drop cached memory JS content memory pressure event? (bug 813530) GC/CC Lose old/unused WebGL contexts? [1] Critical memory pressure event: Lose WebGL contexts This could allow us to reduce memory usage without always killing the WebGLContext. This could allow webgl content to continue running, with only a hiccup while we tear down caches and GC. If we still haven't freed enough[2] after the low memory event, we could trigger a critical memory event and axe webgl contexts. (Restoring GL data can be very expensive for an app, even when implemented properly) [1] If a webgl context hasn't been used in the last, say, 10m, we might want to consider these 'expendable' for the purposes of freeing memory. [2] This is probably the big questionmark here. Maybe we could check our memory usage before and after the low-memory event, and only trigger critical-memory if we don't manage to free much. (Perhaps as a percent of total usage?)

Flagging this as p2; it's a good idea to be more fine-grained for memory, though it mainly matters for mobile. There we have other work to do to get to significant viability before this becomes an issue. I do like losing old/unused WebGL contexts first though; unused + in inactive tabs could be good.