[pve-devel] [PATCH corosync-pve/kronosnet 0/4] cherry-pick bug fixes

[Fabian Grünbichler]

culmination of 4 weeks of triaging together with the respective upstream
devs and endless hours staring at corosync debug traces, this fixes the
following issues:

- knet losing join messages if network is overloaded, pushing corosync
  into a retransmit loop, potentially causing a full-cluster fence event
  with just a single node acting up
- corosync potentially corrupting messages during membership changes

and another one reported by someone else:

- corosync causing high network load by not holding the token in case
  messages are queued for retransmission

all of the fixes are taken from the respective stable queue with
releases slated for later this week.

corosync:

Fabian Grünbichler (2):
  cherry-pick fixes
  bump version to 3.1.5-pve2

 ...cel_hold_on_retransmit-config-option.patch | 132 ++++++++++++++++++
 ...ch-totempg-buffers-at-the-right-time.patch | 113 +++++++++++++++
 debian/changelog                              |   8 ++
 debian/patches/series                         |   2 +
 4 files changed, 255 insertions(+)
 create mode 100644 debian/patches/0003-totem-Add-cancel_hold_on_retransmit-config-option.patch
 create mode 100644 debian/patches/0004-totemsrp-Switch-totempg-buffers-at-the-right-time.patch

kronosnet:

Fabian Grünbichler (2):
  fix #3672: cherry-pick knet fixes
  bump version to 1.22-pve2

 ...eq_num-initialization-race-condition.patch | 53 +++++++++++
 ...or-messages-to-trigger-faster-link-d.patch | 92 +++++++++++++++++++
 debian/changelog                              |  6 ++
 debian/patches/series                         |  3 +-
 4 files changed, 153 insertions(+), 1 deletion(-)
 create mode 100644 debian/patches/0001-host-fix-dst_seq_num-initialization-race-condition.patch
 create mode 100644 debian/patches/0002-udp-use-ICMP-error-messages-to-trigger-faster-link-d.patch

-- 
2.30.2