Your message is rather dry in the context of the previous long discussions, so it is hard to troubleshoot such a report. Can you provide details what version are you running now? Also, how often it happens, any error messages before it happens? Do you suspend more than one time the transaction?
Might not be the case, but if you use locking functions exposed by htable module, there was a recent fix for them.
Cheers, Daniel
On 08/05/15 23:43, Alex Balashov wrote:
Unfortunately, we do keep running into total deadlocks with this approach under high load. We've had to revert async call processing (with mqueue + rtimer + t_suspend) back to non-async because the deadlocks were growing politically unacceptable.
I never managed to get a backtrace of the state of the processes at the time the deadlock occurs, because the customer would soon restart the proxy when this happened, as it would happen during busy times in the middle of production days. The only core dumps I have are the crashes on shutdown, which strikes me as incidental/irrelevant.