summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorKonrad Bańka <k.banka@samsung.com>2021-01-25 07:44:49 +0100
committerKonrad Bańka <k.banka@samsung.com>2021-01-25 10:37:24 +0100
commite1d62ce450daaba624e5ff73fec4dbafd1af8b89 (patch)
tree58feb0b49e2b05e82716b3f21e54a2358a1aee52
parent8ab4fba7165b3b9418730cb1cdd424346925a93f (diff)
[COMMON][ETCD] Skip startup self-discovery for etcd nodes
Current startup script of etcd checks whether all assumed other nodes are already running, before proceeding. This check, however, also includes checking localhost, but due to using headless service statefulset pod DNS discovery, it doesnt succeed immediately. In some deployments k8s DNS server may be laggy, thus failing startup script to finish before liveness check. This patch fixes such failures of 1 pod etcd clusters, and improves startup time for any size ones. Signed-off-by: Konrad Bańka <k.banka@samsung.com> Issue-ID: OOM-2668 Change-Id: I2f9263a0f4964b0a495631775d0cbbceef25e85b
-rw-r--r--kubernetes/common/etcd/templates/statefulset.yaml4
1 files changed, 4 insertions, 0 deletions
diff --git a/kubernetes/common/etcd/templates/statefulset.yaml b/kubernetes/common/etcd/templates/statefulset.yaml
index f5592bd252..e39b8c4ca2 100644
--- a/kubernetes/common/etcd/templates/statefulset.yaml
+++ b/kubernetes/common/etcd/templates/statefulset.yaml
@@ -133,6 +133,10 @@ spec:
# we should wait for other pods to be up before trying to join
# otherwise we got "no such host" errors when trying to resolve other members
for i in $(seq 0 $((${INITIAL_CLUSTER_SIZE} - 1))); do
+ if [ "${SET_NAME}-${i}" == "${HOSTNAME}" ]; then
+ echo "Skipping self-checking"
+ continue
+ fi
while true; do
echo "Waiting for ${SET_NAME}-${i}.${SERVICE_NAME} to come up"
ping -W 1 -c 1 ${SET_NAME}-${i}.${SERVICE_NAME} > /dev/null && break