|
 |
79554c |
From a05eb8673bd1d5d3d41f2ed39df2650b19681d08 Mon Sep 17 00:00:00 2001
|
|
|
79554c |
From: Damien Ciabrini <dciabrin@redhat.com>
|
|
|
79554c |
Date: Fri, 3 Mar 2017 15:31:30 +0100
|
|
|
79554c |
Subject: [PATCH] galera: fix the first bootstrap when cluster has no data
|
|
|
79554c |
|
|
|
79554c |
The resource agent selects the first node to go into Master state
|
|
|
79554c |
based on the biggest commit version found on each node. If case no
|
|
|
79554c |
data were written yet into the galera cluster, the current node is
|
|
|
79554c |
selected as a "fallback" node to bootstrap the cluster.
|
|
|
79554c |
|
|
|
79554c |
The way the "fallback" node is selected is wrong because every node
|
|
|
79554c |
takes a different decision, and this ultimately yields to 3
|
|
|
79554c |
single-node galera clusters being started. To fix that, let the
|
|
|
79554c |
"fallback" node be the last one in the wsrep_cluster_address, so that
|
|
|
79554c |
the selection algorithm yields coherent results across nodes.
|
|
|
79554c |
---
|
|
|
79554c |
heartbeat/galera | 14 ++++++++++++--
|
|
|
79554c |
1 file changed, 12 insertions(+), 2 deletions(-)
|
|
|
79554c |
|
|
|
79554c |
diff --git a/heartbeat/galera b/heartbeat/galera
|
|
|
79554c |
index decbaa2..475a8ba 100755
|
|
|
79554c |
--- a/heartbeat/galera
|
|
|
79554c |
+++ b/heartbeat/galera
|
|
|
79554c |
@@ -451,14 +451,24 @@ pcmk_to_galera_name()
|
|
|
79554c |
detect_first_master()
|
|
|
79554c |
{
|
|
|
79554c |
local best_commit=0
|
|
|
79554c |
- local best_node="$NODENAME"
|
|
|
79554c |
local last_commit=0
|
|
|
79554c |
local missing_nodes=0
|
|
|
79554c |
local nodes=""
|
|
|
79554c |
local nodes_recovered=""
|
|
|
79554c |
+ local all_nodes
|
|
|
79554c |
+ local best_node_gcomm
|
|
|
79554c |
+ local best_node
|
|
|
79554c |
+
|
|
|
79554c |
+ all_nodes=$(echo "$OCF_RESKEY_wsrep_cluster_address" | sed 's/gcomm:\/\///g' | tr -d ' ' | tr -s ',' ' ')
|
|
|
79554c |
+ best_node_gcomm=$(echo "$all_nodes" | sed 's/^.* \(.*\)$/\1/')
|
|
|
79554c |
+ best_node=$(galera_to_pcmk_name $best_node_gcomm)
|
|
|
79554c |
+ if [ -z "$best_node" ]; then
|
|
|
79554c |
+ ocf_log error "Could not determine initial best node from galera name <${best_node_gcomm}>."
|
|
|
79554c |
+ return
|
|
|
79554c |
+ fi
|
|
|
79554c |
|
|
|
79554c |
# avoid selecting a recovered node as bootstrap if possible
|
|
|
79554c |
- for node in $(echo "$OCF_RESKEY_wsrep_cluster_address" | sed 's/gcomm:\/\///g' | tr -d ' ' | tr -s ',' ' '); do
|
|
|
79554c |
+ for node in $all_nodes; do
|
|
|
79554c |
local pcmk_node=$(galera_to_pcmk_name $node)
|
|
|
79554c |
if [ -z "$pcmk_node" ]; then
|
|
|
79554c |
ocf_log error "Could not determine pacemaker node from galera name <${node}>."
|