[go-nuts] Re: Go GC time creeping up

2016-12-16 Thread Evan Digby
Hi Dave, Just to follow up as I believe we've resolved this issue. TL;DR: Ever-growing maps with convoluted logic surrounded by locks lead to poor response times. Your response prompted us to not write off what we had assumed was an insignificant increase in heap size (~100mb over 24 hours) b

[go-nuts] Re: Go GC time creeping up

2016-12-12 Thread Evan Digby
Hi Dave, Thanks for the insight. I'll look further into the heap creep. I'll also try to get together a log of a run past 8 hours over the next day or so on a machine we're not regularly restarting the go process on to see if it shows anything different. Thanks again, Evan On Monday, 12 Dec

[go-nuts] Re: Go GC time creeping up

2016-12-12 Thread Dave Cheney
Thanks for posting this. Once your application gets to about 1600-1900mb goal the GC pauses seem to stabilise at 3.3+3.2. It looks like the anount of memory in use is creeping up slowly, from around 850 to 950 over the span of the trace. GC sweep time is proportional to the size of the heap, so