File 47.patch of Package cluster-glue.40043
From 27789f6d4c4b5bcb76a38861109b982eb06022d8 Mon Sep 17 00:00:00 2001
From: Eloi da Silva <etch.linux@gmail.com>
Date: Wed, 2 Jul 2025 15:49:27 +0000
Subject: [PATCH 1/2] stonith/external/ec2: add function ec2_retry and use it
in awscli and imds requests
---
lib/plugins/stonith/external/ec2 | 58 +++++++++++++++++++++++++-------
1 file changed, 46 insertions(+), 12 deletions(-)
diff --git a/lib/plugins/stonith/external/ec2 b/lib/plugins/stonith/external/ec2
index bdbceefa..d7f6065f 100755
--- a/lib/plugins/stonith/external/ec2
+++ b/lib/plugins/stonith/external/ec2
@@ -56,6 +56,9 @@ unknown_are_stopped=0
action_default="reset" # Default fence action
ec2_tag_default="Name" # EC2 Tag containing the instance's uname
ec2_profile_default="default" # EC2 Profile containing the AWS's profile
+ec2_dmi_file="/sys/devices/virtual/dmi/id/board_asset_tag" # Only supported on nitro-based instances.
+ec2_retry_retries=4 # Default retries used by ec2_retry function
+ec2_retry_sleep=3 # Default sleep time used by ec2_retry function
sleep_time="1"
@@ -179,17 +182,48 @@ EOF
exit 0;
}
+function ec2_retry() {
+ local retries=$ec2_retry_retries sleep=$ec2_retry_sleep
+ local tries=$(($retries + 1))
+ local result rc
+
+ for try in $(seq $tries) ;do
+ result=$(eval "${@}")
+ rc=$?
+ [ $rc -eq 0 ] && break
+ sleep $sleep
+ done
+
+ if [ $rc -ne 0 ]; then
+ ha_log.sh err "$@ failed $tries tries"
+ exit 1
+ fi
+
+ echo "$result"
+ return $rc
+}
+
function is_instance_running()
{
local token
local myinstance
local mystatus
-
- # get session token, required for IMDSv2
- token="$(curl -H "X-aws-ec2-metadata-token-ttl-seconds: 21600" -X PUT http://169.254.169.254/latest/api/token)"
+ local ec2_dmi_file_content
- # get my instance id
- myinstance="$(curl -H "X-aws-ec2-metadata-token: $token" http://169.254.169.254/latest/meta-data/instance-id)"
+ if [ -r "$ec2_dmi_file" ] && [ -s "$ec2_dmi_file" ]; then
+ ec2_dmi_file_content="$(cat "$ec2_dmi_file")"
+ case "$ec2_dmi_file_content" in
+ i-0*) myinstance="$ec2_dmi_file_content" ;;
+ esac
+ fi
+
+ if [ -z "$myinstance" ] ;then
+ # get session token, required for IMDSv2
+ token="$(ec2_retry curl -H \"X-aws-ec2-metadata-token-ttl-seconds: 21600\" -X PUT http://169.254.169.254/latest/api/token)"
+
+ # get my instance id
+ myinstance="$(ec2_retry curl -H \"X-aws-ec2-metadata-token: $token\" http://169.254.169.254/latest/meta-data/instance-id)"
+ fi
# check my status.
# When the EC2 instance be stopped by the "aws ec2 stop-instances" , the stop processing of the OS is executed.
@@ -216,7 +250,7 @@ function instance_for_port()
local instance=""
# Look for port name -n in the INSTANCE data
- instance=`aws ec2 describe-instances $options --filters "Name=tag-value,Values=${port}" "Name=tag-key,Values=${ec2_tag}" --query 'Reservations[*].Instances[*].InstanceId' `
+ instance=`ec2_retry aws ec2 describe-instances $options --filters \"Name=tag-value,Values=${port}\" \"Name=tag-key,Values=${ec2_tag}\" --query 'Reservations[*].Instances[*].InstanceId' `
if [ -z $instance ]; then
instance_not_found=1
@@ -228,7 +262,7 @@ function instance_for_port()
function instance_on()
{
- aws ec2 start-instances $options --instance-ids $instance
+ ec2_retry aws ec2 start-instances $options --instance-ids $instance
}
function instance_off()
@@ -237,7 +271,7 @@ function instance_off()
# nothing to do
ha_log.sh info "Assuming unknown instance $instance is already off"
else
- aws ec2 stop-instances $options --instance-ids $instance --force
+ ec2_retry aws ec2 stop-instances $options --instance-ids $instance --force
fi
}
@@ -251,7 +285,7 @@ function instance_status()
if [ "$unknown_are_stopped" = 1 -a $instance_not_found ]; then
ha_log.sh info "$instance stopped (unknown)"
else
- status=`aws ec2 describe-instances $options --instance-ids $instance --query 'Reservations[*].Instances[*].State.Name' `
+ status=`ec2_retry aws ec2 describe-instances $options --instance-ids $instance --query 'Reservations[*].Instances[*].State.Name' `
rc=$?
fi
ha_log.sh info "status check for $instance is $status"
@@ -261,8 +295,8 @@ function instance_status()
function monitor()
{
- # Is the device ok?
- aws ec2 describe-instances $options --filters "Name=tag-key,Values=${ec2_tag}" | grep INSTANCES &> /dev/null
+ # Is the device ok?
+ ec2_retry aws ec2 describe-instances $options --filters \"Name=tag-key,Values=${ec2_tag}\" | grep INSTANCES &> /dev/null
}
TEMP=`getopt -o qVho:e:p:n:t:U --long version,help,action:,port:,option:,profile:,tag:,quiet,unknown-are-stopped \
@@ -409,7 +443,7 @@ case $action in
;;
gethosts|hostlist|list)
# List of names we know about
- a=`aws ec2 describe-instances $options --filters "Name=tag-key,Values=${ec2_tag}" --query 'Reservations[*].Instances[*].Tags[?Key==\`'${ec2_tag}'\`].Value' | sort -u`
+ a=$(ec2_retry aws ec2 describe-instances $options --filters \"Name=tag-key,Values=${ec2_tag}\" --query 'Reservations[*].Instances[*].Tags[?Key==\`'${ec2_tag}'\`].Value' |sort -u)
echo $a
;;
stat|status)
From c1297e2a51f8dae2454e9144b78b1b89108a4777 Mon Sep 17 00:00:00 2001
From: Eloi da Silva <etch.linux@gmail.com>
Date: Fri, 11 Jul 2025 16:54:11 +0000
Subject: [PATCH 2/2] ec2: add logging if fail to get instance id from local
file and fallback to IMDS
---
lib/plugins/stonith/external/ec2 | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/lib/plugins/stonith/external/ec2 b/lib/plugins/stonith/external/ec2
index d7f6065f..30971521 100755
--- a/lib/plugins/stonith/external/ec2
+++ b/lib/plugins/stonith/external/ec2
@@ -195,7 +195,7 @@ function ec2_retry() {
done
if [ $rc -ne 0 ]; then
- ha_log.sh err "$@ failed $tries tries"
+ ha_log.sh err "command=$@ ; result=$result ; failed $tries tries"
exit 1
fi
@@ -210,6 +210,7 @@ function is_instance_running()
local mystatus
local ec2_dmi_file_content
+ # get my instance id from local file
if [ -r "$ec2_dmi_file" ] && [ -s "$ec2_dmi_file" ]; then
ec2_dmi_file_content="$(cat "$ec2_dmi_file")"
case "$ec2_dmi_file_content" in
@@ -217,7 +218,10 @@ function is_instance_running()
esac
fi
+ # if not $myinstance then fallback to ec2 imds to get instance id
if [ -z "$myinstance" ] ;then
+ ha_log.sh info "Unable to get Instance ID from local file, then fallback to EC2 IMDS"
+
# get session token, required for IMDSv2
token="$(ec2_retry curl -H \"X-aws-ec2-metadata-token-ttl-seconds: 21600\" -X PUT http://169.254.169.254/latest/api/token)"
