Nikita Popov
Nikita Popov
@rcanavan Apcu 5.1.11 is now released with a probable fix for this issue. Could you please check whether it works for you as well?
@remicollet yeah, I changed it already
The issue from #259 was also related to cache expunge on full cache, so looks like we're still dealing with the same problem area... As @dugwood mentioned, it would be...
@rcanavan The start time is reset whenever the cache is (explicitly or automatically) cleared.
@dugwood @rcanavan I've just checked the glibc rwlock implementation to find out what exactly it is FUTEX_WAITing for. As @rcanavan already suspected, this is not the pid. In fact, the...
@rcanavan Assuming that this is really the same issue as in #259, those values look sensible. The problem sequence is a wlock wlock wunlock wunlock on one thread. The second...
The reason for the -3 readers would be that the rwlock implementation has overflow protection for the number of readers. So when __writer=0, __nr_readers=-1 trying to acquire an rlock will...
I've added checks for failed wlock operations in https://github.com/krakjoe/apcu/commit/885ea551d7ee548189c868e58873b2610d483043. If a wlock fails we print a warning and abort the operation. That will at least prevent us from corrupting the...
@lichunqiang Which version are you currently using? I'd suggest trying the 5.1.11 release first (which definitely fixes a apcu_store deadlock). The additional changes on master may help with debugging and...
Does anybody here still experience deadlock issues with APCu 5.1.11? If so, I'd suggest trying the new 5.1.12 release as well.