[Sprint] sprint-loop-47

bash/daml-export

@@ -93,11 +93,24 @@ function exportRange {
     [ -n "$DAML_TOKEN" ] && args+=(--access-token-file "$DAML_TOKEN")
     java "${args[@]}"
     local export_exit=$?
-    if [ $export_exit -eq 0 ]; then
-      touch "$dir/export.good"
-    else
+    if [ $export_exit -ne 0 ]; then
+      # SUR-3644: surface the Java export failure to the caller so the main
+      # loop skips correction/build instead of reporting false success.
+      echo "REDO $dir"
+      return "$export_exit"
+    fi
+    # SUR-3644: Java reported success, but the build step depends on the
+    # exported sources existing. Treat missing artifacts as a failed export
+    # rather than touching export.good and proceeding on partial output.
+    local missing=()
+    [ -r "$dir/daml.yaml" ] || missing+=("daml.yaml")
+    [ -r "$dir/Export.daml" ] || missing+=("Export.daml")
+    if [ "${#missing[@]}" -gt 0 ]; then
       echo "REDO $dir"
+      log::error "daml-export: export of $dir is missing required files: ${missing[*]}"
+      return 1
     fi
+    touch "$dir/export.good"
     sleep 1
   else
     echo "Skipping complete export from $start to $stop"
@@ -183,6 +196,9 @@ CUR_INT=$(hex_to_dec "$START_OFFSET")
 STOP_INT=$(hex_to_dec "$STOP_OFFSET")
 
 count=0
+# SUR-3644: track whether any range failed to export so the script can exit
+# non-zero overall instead of masking the failure behind a clean build wait.
+EXPORT_FAILED=false
 RUNNING_PROCS=()
 RUNNING_OUTPUT_DIRS=()
 # TO_BUILD removed (SUR-1871): the previous queue accumulated every
@@ -277,12 +293,23 @@ while [ "$CUR_INT" -le "$STOP_INT" ]; do
 
   verifyExport "$OUTPUT_DIR"
   if [ $? -ge 2 ]; then
-    exportRange "$CUR_HEX" "$NEXT_HEX" "$OUTPUT_DIR" "$BASE_DIR"
+    # SUR-3644: gate correction/build on a successful export. A failed export
+    # (Java error or missing artifacts) must not be corrected/built.
+    if exportRange "$CUR_HEX" "$NEXT_HEX" "$OUTPUT_DIR" "$BASE_DIR"; then
+      export_ok=true
+    else
+      export_ok=false
+    fi
+  else
+    export_ok=true
   fi
 
   refresh_running_procs
 
-  if [ "${#RUNNING_PROCS[@]}" -lt "$MAX_PARALLEL" ]; then
+  if [ "$export_ok" != true ]; then
+    log::error "daml-export: skipping correction/build for $OUTPUT_DIR after failed export"
+    EXPORT_FAILED=true
+  elif [ "${#RUNNING_PROCS[@]}" -lt "$MAX_PARALLEL" ]; then
     correct_export "$OUTPUT_DIR"
     build "$OUTPUT_DIR" &
     RUNNING_PROCS+=($!)
@@ -335,6 +362,12 @@ done < <(find "$TARGET_DIR" -name export.good -print | sort)
 wait_for_running_procs
 WAIT_STATUS=$?
 NORMAL_COMPLETION=true
+# SUR-3644: a failed export earlier in the run must not be masked by a clean
+# build wait — fail loudly so CI/automation does not proceed on missing or
+# partial artifacts.
+if [ "$EXPORT_FAILED" = true ]; then
+  exit 1
+fi
 if [ "$WAIT_STATUS" -ne 0 ]; then
   exit "$WAIT_STATUS"
 fi

bash/minikube-test-environment

@@ -59,6 +59,7 @@ function start_or_create_minikube {
         --kubernetes-version "$KUBERNETES_VERSION" \
         --driver="$MINIKUBE_DRIVER" \
         --force-systemd=true \
+        "--cni=$CNI" \
         "--nodes=$NODES" \
         "--memory=$NODE_MEMORY"
     fi
@@ -68,6 +69,7 @@ function start_or_create_minikube {
       --kubernetes-version "$KUBERNETES_VERSION" \
       --driver="$MINIKUBE_DRIVER" \
       --force-systemd=true \
+      "--cni=$CNI" \
       "--nodes=$NODES" \
       "--memory=$NODE_MEMORY"
   fi

bash/replace-validator

@@ -119,6 +119,26 @@ function waitForPodTermination() {
   done
 }
 
+# Pre-flight check that every requested node has a pod in the populated
+# map. Emits a clear error naming the missing node and selector and returns
+# non-zero so `main` can bail out BEFORE any key copy or label mutation,
+# rather than aborting on an opaque `nounset` error when dereferencing a
+# missing associative-array element. The map is passed by nameref. (SUR-3649)
+function validateNodesMapped() {
+  local -n _map=${1:?}
+  local label=${2:?}
+  shift 2
+  local node
+  local rc=0
+  for node in "$@"; do
+    if [ -z "${_map[$node]+set}" ]; then
+      log::error "replace-validator: no pod found for node '$node' with selector '$label'"
+      rc=1
+    fi
+  done
+  return "$rc"
+}
+
 function main() {
   set -euo pipefail
   local -A node2pod=()
@@ -128,6 +148,9 @@ function main() {
   local -A pod2node=()
   # shellcheck disable=SC2153
   mapPods "$LABEL" node2pod pod2node
+  # SUR-3649: fail cleanly before any destructive action when a requested
+  # node has no matching pod, instead of crashing on nounset below.
+  validateNodesMapped node2pod "$LABEL" "$FROM_NODE" "$TRGT_NODE"
   local from_pod=${node2pod[$FROM_NODE]}
   local to_pod=${node2pod[$TRGT_NODE]}
 

tests/daml-export.bats

@@ -249,6 +249,44 @@ EOF
   rm -f "$log_file"
 }
 
+@test "daml-export exits non-zero and skips correction/build when java export fails (SUR-3644)" {
+  tmp=$(mktemp -d)
+  stub_bin=$(mktemp -d)
+  log_file=$(mktemp)
+  write_daml_stub "$stub_bin"
+  printf '%s\n' '#!/usr/bin/env bash' 'exit 1' >"$stub_bin/java"
+  chmod +x "$stub_bin/java"
+
+  run env PATH="$stub_bin:$PATH" DAML_STUB_LOG="$log_file" DAML_EXPORT_SOURCE_ONLY= \
+    bash "$DAML_EXPORT" -d "$tmp" -b 0 -e 1 -s 1 -P 5
+  [ "$status" -ne 0 ]
+  [[ "$output" == *"REDO"* ]]
+  # correct_export must not run on a failed export.
+  [[ "$output" != *"Correct "* ]]
+  # No build was scheduled: the daml stub never logged a start.
+  run ! grep -q "^start:" "$log_file"
+  rm -rf "$tmp" "$stub_bin"
+  rm -f "$log_file"
+}
+
+@test "daml-export exits non-zero when java succeeds but artifacts are missing (SUR-3644)" {
+  tmp=$(mktemp -d)
+  stub_bin=$(mktemp -d)
+  log_file=$(mktemp)
+  write_daml_stub "$stub_bin"
+  printf '%s\n' '#!/usr/bin/env bash' 'exit 0' >"$stub_bin/java"
+  chmod +x "$stub_bin/java"
+
+  run env PATH="$stub_bin:$PATH" DAML_STUB_LOG="$log_file" DAML_EXPORT_SOURCE_ONLY= \
+    bash "$DAML_EXPORT" -d "$tmp" -b 0 -e 1 -s 1 -P 5
+  [ "$status" -ne 0 ]
+  [[ "$output" == *"missing required files"* ]]
+  # No build was scheduled: the daml stub never logged a start.
+  run ! grep -q "^start:" "$log_file"
+  rm -rf "$tmp" "$stub_bin"
+  rm -f "$log_file"
+}
+
 @test "daml-export does not schedule duplicate builds for same output dir (SUR-3352)" {
   tmp=$(mktemp -d)
   stub_bin=$(mktemp -d)

tests/minikube-test-environment.bats

@@ -60,3 +60,15 @@ teardown() {
   [ "$status" -eq 0 ]
   grep -qx 'stop' "$MINIKUBE_ARGV_LOG"
 }
+
+@test "create passes the default CNI to minikube start (SUR-3648)" {
+  run env "PATH=$PATH" "MINIKUBE_ARGV_LOG=$MINIKUBE_ARGV_LOG" "$MK_ENV" create
+  grep -qx 'start' "$MINIKUBE_ARGV_LOG"
+  grep -qx -- '--cni=calico' "$MINIKUBE_ARGV_LOG"
+}
+
+@test "create honors a CNI override on minikube start (SUR-3648)" {
+  run env "PATH=$PATH" "MINIKUBE_ARGV_LOG=$MINIKUBE_ARGV_LOG" CNI=cilium "$MK_ENV" create
+  grep -qx 'start' "$MINIKUBE_ARGV_LOG"
+  grep -qx -- '--cni=cilium' "$MINIKUBE_ARGV_LOG"
+}

tests/replace-validator.bats

@@ -59,6 +59,64 @@ EOF
   [ "$status" -eq 0 ]
 }
 
+@test "validateNodesMapped errors and exits non-zero when FROM_NODE has no pod (SUR-3649)" {
+  TEST_SCRIPT=$(mktemp)
+  cat >"$TEST_SCRIPT" <<EOF
+set -e
+source '$REPO_ROOT/bash/includer.sh'
+@include log
+
+eval "\$(awk '/^function validateNodesMapped\\(/,/^}\$/' '$TARGET')"
+
+declare -A node2pod=([node-B]=example-pod-2)
+validateNodesMapped node2pod app=foo node-A node-B
+echo "should-not-reach"
+EOF
+  run bash "$TEST_SCRIPT"
+  rm -f "$TEST_SCRIPT"
+  [ "$status" -ne 0 ]
+  [[ "$output" == *"no pod found for node 'node-A'"* ]]
+  [[ "$output" == *"app=foo"* ]]
+  [[ "$output" != *"should-not-reach"* ]]
+}
+
+@test "validateNodesMapped errors when TRGT_NODE has no pod (SUR-3649)" {
+  TEST_SCRIPT=$(mktemp)
+  cat >"$TEST_SCRIPT" <<EOF
+set -e
+source '$REPO_ROOT/bash/includer.sh'
+@include log
+
+eval "\$(awk '/^function validateNodesMapped\\(/,/^}\$/' '$TARGET')"
+
+declare -A node2pod=([node-A]=example-pod-1)
+validateNodesMapped node2pod app=foo node-A node-B
+EOF
+  run bash "$TEST_SCRIPT"
+  rm -f "$TEST_SCRIPT"
+  [ "$status" -ne 0 ]
+  [[ "$output" == *"no pod found for node 'node-B'"* ]]
+}
+
+@test "validateNodesMapped succeeds when both nodes are mapped (SUR-3649)" {
+  TEST_SCRIPT=$(mktemp)
+  cat >"$TEST_SCRIPT" <<EOF
+set -e
+source '$REPO_ROOT/bash/includer.sh'
+@include log
+
+eval "\$(awk '/^function validateNodesMapped\\(/,/^}\$/' '$TARGET')"
+
+declare -A node2pod=([node-A]=example-pod-1 [node-B]=example-pod-2)
+validateNodesMapped node2pod app=foo node-A node-B
+echo "validated-ok"
+EOF
+  run bash "$TEST_SCRIPT"
+  rm -f "$TEST_SCRIPT"
+  [ "$status" -eq 0 ]
+  [[ "$output" == *"validated-ok"* ]]
+}
+
 @test "replace-validator declares @include commands" {
   run grep -E '^@include[[:space:]]+commands\b' "$TARGET"
   [ "$status" -eq 0 ]