@Jesper, curious how you determined that? Is it in the spec? or the source? Or is this a common GC pattern that you presume is being used?
On Thursday, April 20, 2017 at 7:01:03 AM UTC-7, Jesper Louis Andersen wrote: > > A somewhat common culprit seems to be the following case: > > 1. In order for the GC to switch from marking to sweeping, it needs all > cores to agree. This requires a "barrier" in the system and thus we have to > wait on all CPUs. > 2. The barrier check happens on a function call. > 3. A CPU core is currently executing a long-running operation inside a > function, but is never calling another function in doing so. Usually there > is a loop involved. > 4. The system is now practically stalled. > > There may be other reasons, but the above scheme is somewhat common. A > workaround is to make sure you break long-running loops every N iterations > and coerce the system to consider a garbage collection by returning from a > function, say. Work is being done to automate this test so low latency can > be achieved automatically without you having to intervene[0] > > To check for this assumption, set GODEBUG=gctrace=1 and look at the pauses > of the different phases. It should tell you if this is the culprit or not. > It could easily be something else, for instance pressure on the GC itself > due to heavy allocation. > > [0] This will come at a slight cost to throughput, but it is probably a > valid trade-off to make in a modern world of highly parallelized systems. > > > On Thu, Apr 20, 2017 at 3:49 PM Lee Armstrong <les...@gmail.com > <javascript:>> wrote: > >> See attached graph which shows the GC pauses of an application we have. >> >> I am frequently seeing pauses of 1-1.5 seconds. This is using Go 1.8.1 >> and have a large map that is frequently accessed and items are removed and >> added to it. These can be of some size. >> >> Is there a way to get these pauses down at all? Would forcing a GC() >> after removing a batch of elements help at all? >> >> Alongside the pauses I see some substantial CPU usage showing up in >> traces for the GC scan. >> >> Thanks in advance! >> >> -- >> You received this message because you are subscribed to the Google Groups >> "golang-nuts" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to golang-nuts...@googlegroups.com <javascript:>. >> For more options, visit https://groups.google.com/d/optout. >> > -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
