Re: Handling requests to an upstream that is unresponsive.

2019-10-01 Thread Sudheer Vinukonda
+1 on the problem :) We’ve often had very similar production issues too! A related issue here that plays a role in all of this mechanism is server session pools and the fact that ATS only marks a server down on a “connect failure”. In particular, when ATS tries to reuse an existing idle connec

Handling requests to an upstream that is unresponsive.

2019-10-01 Thread Alan M. Carroll
I would like to propose a change to how TS handles unresponsive upstream destinations. The current logic is * If there is a round robin, skip elements that are down. * Attempt to connect. * If the connect fails, and the upstream is marked down and within the down server cache time, restrict the

Re: Tentative proposal: TS mutexes should be queues not OS mutexes

2019-10-01 Thread Walt Karas
I assume thread affinity can't literal mean no lock contention. You'd need a lock on the thread run queue wouldn't you? Continuations can't only get queued for the event thread from the event thread. I don't think we can say conclusively that there would be a significant difference due to lock c

ApacheCon North America 2020, project participation

2019-10-01 Thread Rich Bowen
Hi, folks, (Note: You're receiving this email because you're on the dev@ list for one or more Apache Software Foundation projects.) For ApacheCon North America 2019, we asked projects to participate in the creation of project/topic specific tracks. This was very successful, with about 15 projects

Re: [ANNOUNCE] Apache Traffic Server Fall 2019 Summit

2019-10-01 Thread Bryan Call
Here is a link to the map: https://goo.gl/maps/TNgXj499GyB4bJyr5 Also, give me a call if you have problems getting into the building: 408-460-4547 I will update this information on the summit page. -Bryan > On Sep 30, 2019, at 10:23 PM, Masakazu Kitajo

Re: Tentative proposal: TS mutexes should be queues not OS mutexes

2019-10-01 Thread Alan Carroll
Do you have any more specific information on mutex contention? We have in fact already looked at doing this, I think back in 2015 with Dan Xu. The goal there was to have queues with the mutexes to avoid rescheduling. As a precursor Dan did some profiling and the only significant lock contention he