First commit

resources/node/files/scripts/common.sh

@@ -0,0 +1,81 @@
+#!/bin/sh
+
+
+pingSentinel() {
+    svc=${VALKEY_HEADLESS_SERVICE:-"localhost"}
+
+    resp=$(timeout -s 15 $1 \
+        valkey-cli \
+        -h ${svc} \
+        -p $VALKEY_SENTINEL_PORT \
+        ping)
+    ret=${?}
+    echo $resp
+    return ${ret}
+}
+
+getPrimaryInfo() {
+    valkey-cli --csv -h ${VALKEY_HEADLESS_SERVICE} -p ${VALKEY_SENTINEL_PORT} sentinel get-primary-addr-by-name "mymaster"| \
+        awk -F ',' '{ gsub(/"/,"",$0); print $1 " " $2 }'
+    return ${?}
+}
+
+getSentinelMasterName() {
+    getPrimaryInfo | awk -F ' ' '{print $1}'
+}
+
+getFailOverStatus() {
+    # Check if the failover is finished
+    local failoverStatus
+    primaryInfo=$(getPrimaryInfo)
+    primaryHost=$(echo ${primaryInfo} | awk -F ' ' '{print $1}')
+    currentHost=$(hostname -f)
+    if [[ "${primaryHost}" != "${currentHost}" ]]; then
+        echo "Failover finished"
+        return 0
+    else
+        echo "Failover in progress"
+        return 1
+    fi
+}
+
+waitForSentinel() {
+    tout=60
+    while true; do
+        response=$(pingSentinel 5)
+        if [ "${response}" = "PONG" ]; then
+            echo "Sentinel is responding"
+            break
+            return 0
+        fi
+
+        echo "Sentinel is not responding [${response}]"
+        sleep 1
+        tout=$((tout - 1))
+        if [ "${tout}" -le 0 ]; then
+            echo "Sentinel ping timed out"
+            return 124
+        fi
+    done
+    sleep 10
+}
+
+sentinelIsMaster() {
+    # Check if the sentinel is master
+    primaryInfo=$(getPrimaryInfo)
+    primaryHost=$(echo ${primaryInfo} | awk -F ' ' '{print $1}')
+    currentHost=$(hostname -f)
+    if [[ "${primaryHost}" != "${currentHost}" ]]; then
+        echo "Sentinel is not master"
+        return 1
+    else
+        echo "Sentinel is master"
+        return 0
+    fi
+}
+
+sentinelReset() {
+    # Reset the sentinel
+    valkey-cli -h ${VALKEY_HEADLESS_SERVICE} -p ${VALKEY_SENTINEL_PORT} sentinel reset "${1}"
+    return $?
+}

resources/node/files/scripts/liveness-local.sh

@@ -0,0 +1,27 @@
+#!/bin/sh
+
+ping_valkey() {
+    resp=$(timeout -s 15 $1 \
+        valkey-cli \
+        -h localhost \
+        -p $VALKEY_PORT \
+        ping)
+    ret=${?}
+    echo $resp
+    return ${ret}
+}
+
+response=$(ping_valkey 5)
+if [ "$?" -eq "124" ]; then
+  echo "Timed out"
+  exit 1
+fi
+
+firstWord=$(echo $response |  awk 'NR==1 {print $1;}')
+if [ "$response" != "PONG" ] && [ "$firstWord" != "LOADING" ] && [ "$firstWord" != "MASTERDOWN" ]; then
+  echo "Valey is not alive [${response}]"
+  exit 1
+fi
+
+echo "$( date +'[%Y/%m/%d %H:%M:%S]') Valkey is alive"
+exit 0

resources/node/files/scripts/ping-sentinel.sh

@@ -0,0 +1,27 @@
+#!/bin/sh
+
+ping_sentinel() {
+    resp=$(timeout -s 15 $1 \
+        valkey-cli \
+        -h localhost \
+        -p $VALKEY_SENTINEL_PORT \
+        ping)
+    ret=${?}
+    echo $resp
+    return ${ret}
+}
+
+response=$(ping_sentinel 5)
+if [ "${?}" -eq 124 ]; then
+    echo "Sentinel ping timed out"
+    exit 124
+fi
+
+if [ "${response}" != "PONG" ];
+then
+    echo "Sentinel is not responding"
+    exit 1
+fi
+
+echo "$( date +'[%Y/%m/%d %H:%M:%S]') Sentinel is responding"
+exit 0

resources/node/files/scripts/pre-stop-sentinel.sh

@@ -0,0 +1,32 @@
+#!/bin/sh
+
+source /opt/scripts/common.sh
+
+tmout=120
+while true ; do
+    echo "I'm the primary pod and you are stopping me. Starting sentinel failover"
+    echo "Waiting for failover to finish..."
+    fo=$(getFailOverStatus)
+    if [ $? -eq 0 ]; then
+        echo "Primary has been successfully failed over to a different pod."
+        break
+    fi
+
+    sleep 1
+    tmout=$((tmout - 1))
+    if [ "${tmout}" -le 0 ]; then
+        echo "Failover timed out"
+        exit 1
+    fi
+done
+
+echo "Primary has been successfuly failed over to a different pod."
+
+echo "Removing myself from the sentinel configuration"
+sentinelReset $(hostname -f)
+if [ $? -ne 0 ]; then
+    echo "Failed to remove myself from the sentinel configuration"
+    exit 1
+fi
+
+exit 0

resources/node/files/scripts/pre-stop.sh

@@ -0,0 +1,73 @@
+#!/bin/sh
+
+# Run Valkey command
+vcli() {
+    valkey-cli -h 127.0.0.1 -p "${VALKEY_PORT}" "$@"
+    return $?
+}
+
+# Run Sentinel command
+vcli-sentinel() {
+    valkey-cli -h "${VALKEY_HEADLESS_SERVICE}" -p "${VALKEY_SENTINEL_PORT}" sentinel "$@"
+    return $?
+}
+
+getFailOverStatus() {
+    # Check if the failover is finished
+    local failoverStatus
+    primaryInfo=$(vcli-sentinel get-primary-addr-by-name "mymaster")
+    primaryHost=$(echo ${primaryInfo} | awk -F ' ' '{print $1}')
+    currentHost=$(hostname -f)
+    if [[ "${primaryHost}" != "${currentHost}" ]]; then
+        echo "Failover finished"
+        return 0
+    else
+        echo "Failover in progress"
+        return 1
+    fi
+}
+
+getRole() {
+    # Get the role of the current node
+    VALKEY_ROLE=$(vcli role | head -1)
+    echo "${VALKEY_ROLE}"
+}
+
+isPrimary() {
+    # Check if the current node is the primary
+    role=$(getRole)
+    if [ "${role}" = "master" ]; then
+        echo "I'm the master"
+    else
+        echo "I'm not the master, I'm a ${role}"
+    fi
+}
+
+if isPrimary && ! getFailOverStatus; then
+    echo "I'm the primary and you are stopping me, pausing client connections"
+    #Pausing write connections to avoid data loss"
+    vcli CLIENT PAUSE "22000" WRITE
+
+    echo "Starting failover"
+    vcli-sentinel failover "mymaster"
+    echo "Waiting for sentinel to complete failover for up to 120s"
+    tmout=120
+    while true ; do
+        echo "I'm the primary pod and you are stopping me. Starting sentinel failover"
+        getFailOverStatus
+        if [ $? -eq 0 ]; then
+            echo "Primary has been successfully failed over to a different pod."
+            break
+        fi
+        echo "Waiting for failover to finish..."
+
+        sleep 1
+        tmout=$((tmout - 1))
+        if [ "${tmout}" -le 0 ]; then
+            echo "Failover timed out"
+            exit 1
+        fi
+    done
+fi
+
+exit 0

resources/node/files/scripts/readiness-local.sh

@@ -0,0 +1,26 @@
+#!/bin/sh
+
+ping_valkey() {
+    resp=$(timeout -s 15 $1 \
+        valkey-cli \
+        -h localhost \
+        -p $VALKEY_PORT \
+        ping)
+    ret=${?}
+    echo $resp
+    return ${ret}
+}
+
+response=$(ping_valkey 5)
+if [ "$?" -eq "124" ]; then
+  echo "Timed out"
+  exit 1
+fi
+
+if [ "$response" != "PONG" ]; then
+  echo "Valey is not ready [${response}]"
+  exit 1
+fi
+
+echo "$( date +'[%Y/%m/%d %H:%M:%S]') Valkey is ready"
+exit 0

resources/node/files/scripts/start-node.sh

@@ -0,0 +1,99 @@
+#!/bin/sh
+
+pingSentinel() {
+    resp=$(timeout -s 15 $1 \
+        valkey-cli \
+        -h ${VALKEY_HEADLESS_SERVICE} \
+        -p ${VALKEY_SENTINEL_PORT} \
+        ping)
+    ret=${?}
+    echo $resp
+    return ${ret}
+}
+
+getPrimaryInfo() {
+    valkey-cli -t 15 --csv -h ${VALKEY_HEADLESS_SERVICE} -p ${VALKEY_SENTINEL_PORT} sentinel get-primary-addr-by-name "mymaster"| \
+        awk -F ',' '{ gsub(/"/,"",$0); print $1 " " $2 }'
+    return ${?}
+}
+
+waitForSentinel() {
+    tout=60
+    while true; do
+        response=$(pingSentinel 5)
+        if [ "${response}" = "PONG" ]; then
+            echo "Sentinel is responding"
+            break
+            return 0
+        fi
+
+        echo "Sentinel is not responding [${response}]"
+        sleep 1
+        tout=$((tout - 1))
+        if [ "${tout}" -le 0 ]; then
+            echo "Sentinel ping timed out"
+            return 124
+        fi
+    done
+    sleep 10
+}
+
+startValkey() {
+    # Start Valkey
+    echo "Running : [valkey-server ${@}]"
+    valkey-server ${@}
+    ret=${?}
+    if [ "${ret}" -ne 0 ]; then
+        echo "Failed to start Valkey"
+        exit ${ret}
+    fi
+}
+
+setupPrimary=0
+primaryHost=""
+primaryPort=""
+
+
+#if [[ -f /etc/valkey/sentinel.conf ]]; then
+#    primaryHost="$(awk '/monitor/ {print $4}' /etc/valkey/sentinel.conf)"
+#    primaryPort="$(awk '/monitor/ {print $5}' /etc/valkey/sentinel.conf)"
+#    echo "Found previous primary ${primaryHost}:${primaryPort}"
+#fi
+
+if [[ ! -f /etc/valkey/replication.conf ]]; then
+    cp /etc/valkey/replication.conf.orig /etc/valkey/replication.conf
+fi
+
+waitForSentinel
+ret=${?}
+if [ "${ret}" -ne 0 ]; then
+    exit ${ret}
+fi
+
+primaryInfo=$(getPrimaryInfo)
+if [ "${?}" -ne 0 ]; then
+    echo "No primary found, seting up node as primary"
+    startPrimary=1
+else
+    primaryHost=$(echo ${primaryInfo} | awk -F ' ' '{print $1}')
+    primaryPort=$(echo ${primaryInfo} | awk -F ' ' '{print $2}')
+    echo "Primary host is : ${primaryHost}, port: ${primaryPort}"
+    currentHost=$(hostname -f)
+    echo "Current host is : ${currentHost}"
+    if [ "${primaryHost}" != "${currentHost}" ]; then
+        echo "Not the primary, setting up as replica"
+        startPrimary=0
+    else
+        echo "This is the primary"
+        startPrimary=1
+    fi
+fi
+
+if [ "${startPrimary}" -eq 1 ]; then
+    echo "Starting Valkey as primary"
+    cat $1
+    startValkey ${@}
+else
+    echo "Starting Valkey as replica"
+    startValkey ${@} "--replicaof" "${primaryHost}" "${primaryPort}"
+fi

resources/node/files/scripts/startSentinel.sh

@@ -0,0 +1,38 @@
+#!/bin/sh
+
+# Generate sentinel.conf file for Valkey Sentinel
+source /opt/scripts/common.sh
+
+PATH=${PATH}:/
+SENTINEL_MASTER=""
+
+updateSentinelMaster(){
+    cp /etc/valkey/sentinel.conf.orig /etc/valkey/sentinel.conf
+    hostname=$(hostname -f)
+
+    pingSentinel 1
+    if [ $? -ne 0 ]; then
+        SENTINEL_MASTER=$(hostname -f)
+    else
+        SENTINEL_MASTER=$(getSentinelMasterName)
+    fi
+
+    sed -i "s/SENTINEL_MASTER/${SENTINEL_MASTER}/g" /etc/valkey/sentinel.conf
+
+}
+
+
+if [ -f /etc/valkey/sentinel.conf ]; then
+    echo "Sentinel configuration file already exists, starting with this."
+    reps=$(awk -F ':' '/REPLICAS/ {print $2}' /etc/valkey/sentinel.conf)
+    if [ "${reps}" = "${VALKEY_REPLICAS}" ]; then
+        echo "Sentinel configuration file already has the correct number of replicas (${VALKEY_REPLICAS})."
+    else
+        echo "Updating Sentinel configuration file with the correct number of replicas (${VALKEY_REPLICAS})."
+        updateSentinelMaster
+    fi
+else
+    updateSentinelMaster
+fi
+
+valkey-sentinel /etc/valkey/sentinel.conf