I've been trying to sort out TINKERPOP-2550 with kerberos and recently haven't been hassled by those sorts of errors in travis where they've been occurring with some regularity. So that's good. I've also been testing and re-testing around the deadlock issues I'd alluded to prior to code freeze. Travis jobs seem much more stable on 3.4-dev, but since I've only just started seeing master start to shape up the same way, I'd say I still need to keep testing that branch.
I'm also noticing that the build on master is almost twice as long as 3.4-dev. I don't know why and I can't say with certainty when that started to happen. My benchmarking effort didn't show a slow down with driver/server operations so that finding is somewhat at odds with the build itself. There are more tests on master but I didn't think there were significantly more to make up for this difference. I'll be looking into this issue further as well. On Sat, Apr 24, 2021 at 5:47 AM Stephen Mallette <[email protected]> wrote: > Another code freeze week and this time a big one for 3.5.0. All PRs of a > blocking nature were merged so that's good. I'm still finding Travis a bit > finicky with kerberos even after a fair bit of effort on > > https://issues.apache.org/jira/browse/TINKERPOP-2504 > > but that's always been an issue so I'm not thinking that a blocker. I > expect to close that issue and open a new one specific to kerberos unless. > I'm still examining some issues with: > > https://issues.apache.org/jira/browse/TINKERPOP-2550 > > despite it being closed. Found another interesting deadlock that can occur > on close where a synchronized method was calling itself in low resource > environments like travis. I'm going to watch travis for more hangs and run > more tests this week.If you'd like to help test this, you can run the > docker build with --cpus=0.9 (smart HadoopMarch idea there) or go to the > TestClientFactory and set the workerPoolSize on the Cluster in the build() > method to 1 and then run the build. Or alternatively, just write your own > little tests with a Cluster you construct with that setting. I'd be happy > to hear if anyone managed to do the latter actually and if they ran to some > success or failure and what kind of test they wrote in the process. I think > the problem at this point tend to generate from sessionless requests and > open/close situations but any sort of testing is helpful. > > Other than that, let's focus on testing and documentation review this week > and then head on to release. Thanks! >
