Possible glitch with prof_accum ouput (or maybe pprof)
i at eivanov.com
Wed Jan 15 01:12:16 PST 2014
Jason, thanks a lot!
On Tue, Jan 14, 2014 at 10:11 PM, Jason Evans <jasone at canonware.com> wrote:
> On Jan 14, 2014, at 7:07 AM, Evgeniy Ivanov <i at eivanov.com> wrote:
>> When I use prof_accum I get stacks like this one:
>> 0: 0 [100: 1228800] @ 0x7fe9ed4c6c88 0x400b37 0x400b59 0x400b69
>> 0x400b7e 0x400c04 0x7fe9ec9a8d1d 0x4008e9
>> 0: 0 [100: 60211200] @ 0x7fe9ed4c6c88 0x400b07 0x400b6e 0x400b7e
>> 0x400c04 0x7fe9ec9a8d1d 0x4008e9
>> 0: 0 [100: 1228800] @ 0x7fe9ed4c6c88 0x400b37 0x400b79 0x400c04
>> 0x7fe9ec9a8d1d 0x4008e9
>> 100 function calls and bytes allocated are correct, but first counters
>> are "0: 0". I don't know, what they count, but these zeroes make pprof
>> to ignore these lines.
> The zeros are the current allocations/bytes counters; the corresponding cumulative counters are tracking allocations that no longer exist. In order to make pprof use the cumulative counters rather than the current counters, you need to specify --alloc_space or --alloc_objects (--inuse_space is the default).
More information about the jemalloc-discuss