fix(localdns): reduce startup polling interval

parts/linux/cloud-init/artifacts/localdns.sh

@@ -61,11 +61,35 @@ CURL_COMMAND="curl -s http://${LOCALDNS_NODE_LISTENER_IP}:8181/ready"
 NETWORKCTL_RELOAD_CMD="networkctl reload"
 
 START_LOCALDNS_TIMEOUT=10
+LOCALDNS_PID_POLL_INTERVAL_SECONDS=0.1
+LOCALDNS_READY_POLL_INTERVAL_SECONDS=0.1
+LOCALDNS_READY_TIMEOUT_SECONDS=60
 
 # DNS health check timeout.
 DNS_HEALTH_CHECK_TIMEOUT=2
 DNS_HEALTH_CHECK_TRIES=2
 
+# Convert a wall-clock timeout budget into a poll count for the configured interval.
+calculate_max_poll_attempts() {
+    local timeout_duration=$1
+    local poll_interval_seconds=$2
+
+    awk -v timeout="${timeout_duration}" -v interval="${poll_interval_seconds}" '
+        BEGIN {
+            if (timeout < 0 || interval <= 0) {
+                exit 1
+            }
+
+            if (timeout == 0) {
+                print 0
+                exit 0
+            }
+
+            printf "%d\n", int((timeout / interval) + 0.999999)
+        }
+    '
+}
+
 # Function definitions used in this file.
 # functions defined until "${__SOURCED__:+return}" are sourced and tested in -
 # spec/parts/linux/cloud-init/artifacts/localdns_spec.sh.
@@ -419,14 +443,21 @@ start_localdns() {
     ${COREDNS_COMMAND} &
 
     # Wait until the PID file is created.
-    local elapsed=0
+    local attempts=0
+    local max_attempts
+    max_attempts=$(calculate_max_poll_attempts "${START_LOCALDNS_TIMEOUT}" "${LOCALDNS_PID_POLL_INTERVAL_SECONDS}") || {
+        echo "Invalid localdns PID poll interval configuration."
+        return 1
+    }
+
     while [ ! -f "${LOCALDNS_PID_FILE}" ]; do
-        sleep 1
-        elapsed=$((elapsed + 1))
-        if [ "$elapsed" -ge "$START_LOCALDNS_TIMEOUT" ]; then
+        if [ "$attempts" -ge "$max_attempts" ]; then
             echo "Timed out waiting for CoreDNS to create PID file at ${LOCALDNS_PID_FILE}."
             return 1
         fi
+
+        sleep "${LOCALDNS_PID_POLL_INTERVAL_SECONDS}"
+        attempts=$((attempts + 1))
     done
 
     COREDNS_PID="$(cat ${LOCALDNS_PID_FILE})"
@@ -436,26 +467,24 @@ start_localdns() {
 
 # Wait for localdns to be ready to serve traffic.
 wait_for_localdns_ready() {
-    local maxattempts=$1
-    local timeout_duration=$2
-    declare -i attempts=0
-    local starttime=$(date +%s)
+    local timeout_duration=$1
+    local starttime
+    local currenttime
+    local elapsedtime
+
+    starttime=$(date +%s)
 
     echo "Waiting for localdns to start and be able to serve traffic."
     until [ "$($CURL_COMMAND)" = "OK" ]; do
-        if [ $attempts -ge $maxattempts ]; then
-            echo "Localdns failed to come online after $maxattempts attempts."
-            return 1
-        fi
         # Check for timeout based on elapsed time.
         currenttime=$(date +%s)
         elapsedtime=$((currenttime - starttime))
         if [ $elapsedtime -ge $timeout_duration ]; then
             echo "Localdns failed to come online after $timeout_duration seconds (timeout)."
             return 1
         fi
-        sleep 1
-        ((attempts++))
+
+        sleep "${LOCALDNS_READY_POLL_INTERVAL_SECONDS}"
     done
     echo "Localdns is online and ready to serve traffic."
     return 0
@@ -1059,7 +1088,7 @@ fi
 start_localdns || exit $ERR_LOCALDNS_FAIL
 
 # Wait to direct traffic to localdns until it's ready.
-wait_for_localdns_ready 60 60 || exit $ERR_LOCALDNS_FAIL
+wait_for_localdns_ready "${LOCALDNS_READY_TIMEOUT_SECONDS}" || exit $ERR_LOCALDNS_FAIL
 
 # Disable DNS from DHCP and point the system at localdns.
 # --------------------------------------------------------------------------------------------------------------------

spec/parts/linux/cloud-init/artifacts/localdns_spec.sh

@@ -707,6 +707,23 @@ EOF
             The status should be failure
             The output should include "Timed out waiting for CoreDNS to create PID file"
         End
+
+        It 'should poll for the PID file every 0.1 seconds'
+            MOCK_SCRIPT="./mock-coredns.sh"
+        cat > "$MOCK_SCRIPT" <<EOF
+#!/bin/bash
+exit 0
+EOF
+            chmod +x "$MOCK_SCRIPT"
+            COREDNS_COMMAND="$MOCK_SCRIPT"
+            START_LOCALDNS_TIMEOUT=1
+            sleep() {
+                echo "sleep called with: $1"
+            }
+            When call start_localdns
+            The status should be failure
+            The output should include "sleep called with: 0.1"
+        End
     End
 
 
@@ -716,35 +733,50 @@ EOF
     Describe 'wait_for_localdns_ready'
         setup() {
             Include "./parts/linux/cloud-init/artifacts/localdns.sh"
+            DATE_SEQUENCE_FILE="/tmp/localdns-date-sequence-$$"
+            SLEEP_LOG_FILE="/tmp/localdns-sleep-log-$$"
+        }
+        cleanup() {
+            rm -f "$DATE_SEQUENCE_FILE" "${DATE_SEQUENCE_FILE}.next" "$SLEEP_LOG_FILE"
         }
         BeforeEach 'setup'
+        AfterEach 'cleanup'
     #------------------------- wait_for_localdns_ready -----------------------------------------------------------
         It 'should return success if localdns is ready'
             CURL_COMMAND="echo OK"
-            MAX_ATTEMPTS=100
             TIMEOUT=5
-            When call wait_for_localdns_ready $MAX_ATTEMPTS $TIMEOUT
+            When call wait_for_localdns_ready $TIMEOUT
             The status should be success
             The output should include "Waiting for localdns to start and be able to serve traffic."
             The output should include "Localdns is online and ready to serve traffic."
         End
 
-        It 'should return failure if localdns is not ready, after timeout'
+        It 'should return failure if localdns is not ready after the wall-clock timeout'
             CURL_COMMAND="echo NOTOK"
-            MAX_ATTEMPTS=1000
             TIMEOUT=2
-            When call wait_for_localdns_ready $MAX_ATTEMPTS $TIMEOUT
-            The status should be failure
-            The output should include "Localdns failed to come online after ${TIMEOUT} seconds (timeout)."
-        End
+            EXPECTED_SLEEP_LOG=$(printf '0.1\n0.1\n')
+            cat > "$DATE_SEQUENCE_FILE" <<EOF
+100
+100
+101
+102
+EOF
+            date() {
+                local current_time
 
-        It 'should return failure if localdns is not ready, after max attempts'
-            CURL_COMMAND="echo NOTOK"
-            MAX_ATTEMPTS=2
-            TIMEOUT=50
-            When call wait_for_localdns_ready $MAX_ATTEMPTS $TIMEOUT
+                current_time=$(head -n 1 "$DATE_SEQUENCE_FILE")
+                tail -n +2 "$DATE_SEQUENCE_FILE" > "${DATE_SEQUENCE_FILE}.next"
+                mv "${DATE_SEQUENCE_FILE}.next" "$DATE_SEQUENCE_FILE"
+
+                echo "$current_time"
+            }
+            sleep() {
+                echo "$1" >> "$SLEEP_LOG_FILE"
+            }
+            When call wait_for_localdns_ready $TIMEOUT
             The status should be failure
-            The output should include "Localdns failed to come online after ${MAX_ATTEMPTS} attempts."
+            The output should include "Localdns failed to come online after ${TIMEOUT} seconds (timeout)."
+            The contents of file "$SLEEP_LOG_FILE" should eq "$EXPECTED_SLEEP_LOG"
         End
     End