On Sun, Mar 6, 2016 at 9:41 PM, Alex Rousskov <rouss...@measurement-factory.com> wrote: > On 03/06/2016 11:16 AM, Kinkie wrote: > >> As I cannot reproduce the issue it's really hard to do a root cause analysis. > > Agreed. However, if you can make SBuf/MemBlob use memory pools before > those pools are initialized, then we do have a bug worth fixing (even if > Yuri does not suffer from that specific bug). It may be possible to > easily reproduce _that_ bug by adding a few static SBuf variables > directly to squid.h.
old_api should be already doing that by rounding strings' sizes up to SmallestStringBeforeMemIsInitialized (16kb + 4 bytes), I'm working on the assumption that for some reason this mechanism is failing either at startup or at shutdown time. >> At this point I'm thinking about reinstrumenting MemBlob not to use >> mem/old_api but to use its own pools. > > Why would that make any difference? Whatever pools MemBlob is using, > they must be initialized prior to use. Thus, it is probably best to fix > "old_api" initialization than to duplicate that [buggy] code. .. or to get out from there at all. Having the MemBlob pools be static members of the MemBlob class will guarantee proper initialization before even the first MemBlob is instantiated, won't it? -- Francesco _______________________________________________ squid-dev mailing list squid-dev@lists.squid-cache.org http://lists.squid-cache.org/listinfo/squid-dev