[kronosnet/kronosnet] 48f506: [stats] show loopback links as connected

Branch: refs/heads/master Home: https://github.com/kronosnet/kronosnet Commit: 48f5064c561f29ff84da905ff9ed8ad1b1e841c8 https://github.com/kronosnet/kronosnet/commit/48f5064c561f29ff84da905ff9ed8… Author: Fabio M. Di Nitto <fdinitto(a)redhat.com> Date: 2018-01-08 (Mon, 08 Jan 2018) Changed paths: M libknet/transport_loopback.c Log Message: ----------- [stats] show loopback links as connected Signed-off-by: Fabio M. Di Nitto <fdinitto(a)redhat.com> Commit: d2a6344f7ccddc3b0074f2e5085a95e39034d6a5 https://github.com/kronosnet/kronosnet/commit/d2a6344f7ccddc3b0074f2e5085a9… Author: Fabio M. Di Nitto <fdinitto(a)redhat.com> Date: 2018-01-08 (Mon, 08 Jan 2018) Changed paths: M libknet/threads_common.c Log Message: ----------- [PMTUd] only notify PMTUd when pmtud_running and don't force a run to avoid lock fighting When we request a writelock, we should notify PMTUd to abort and cond_signal only if PMTUd is running. Also, don't request an immediate rerun as we cannot guarantee that the scheduling is favourable to our request and we could end up waiting forever. Signed-off-by: Fabio M. Di Nitto <fdinitto(a)redhat.com> Commit: f636c48ac3cf03f370b734298c87020920280704 https://github.com/kronosnet/kronosnet/commit/f636c48ac3cf03f370b734298c870… Author: Fabio M. Di Nitto <fdinitto(a)redhat.com> Date: 2018-01-08 (Mon, 08 Jan 2018) Changed paths: M libknet/threads_pmtud.c Log Message: ----------- [PMTUd] reset value or suffer the pains of DEADLOCKING! errno is per thread, not per function Signed-off-by: Fabio M. Di Nitto <fdinitto(a)redhat.com> Commit: 06e628fb797e1ef592dcb882ffbd524318ea8f81 https://github.com/kronosnet/kronosnet/commit/06e628fb797e1ef592dcb882ffbd5… Author: Fabio M. Di Nitto <fdinitto(a)redhat.com> Date: 2018-01-08 (Mon, 08 Jan 2018) Changed paths: M libknet/threads_pmtud.c Log Message: ----------- [PMTUd] drop (now) unnecessary and dangerous usleep prior to all threads being able to notify PMTUd of EMSGSIZE errors, we had this random usleep in there to have time to collect data. It was working at the time, but it's a bad idea. On super large clusters (>66 nodes) with 4 links on each node, when applying heavy load (cpghum on all nodes at once), the average latency between nodes can increase so much that the PMTUd thread usleep could literally block corosync for seconds at a time. Drop the usleep and live happily ever after Signed-off-by: Fabio M. Di Nitto <fdinitto(a)redhat.com> Commit: f495c6c83beb7b729207201d19c7db975927bb8b https://github.com/kronosnet/kronosnet/commit/f495c6c83beb7b729207201d19c7d… Author: Fabio M. Di Nitto <fdinitto(a)redhat.com> Date: 2018-01-08 (Mon, 08 Jan 2018) Changed paths: M libknet/threads_common.c M libknet/threads_pmtud.c M libknet/transport_loopback.c Log Message: ----------- Merge pull request #111 from kronosnet/fixes Fixes Compare: https://github.com/kronosnet/kronosnet/compare/b2c0f90a08df...f495c6c83beb