From 326c3251e74f7ea5156ace251fdcc1d49f1084c3 Mon Sep 17 00:00:00 2001
From: Denis O <denis.o@linux.com>
Date: Fri, 1 Nov 2024 15:04:12 +0000
Subject: [PATCH] Add provider cache offline errors handling  (#3527)

* add offline errors handling

* Improved handling of offline errors

* Cleanup

* Add test for handling offline errors

* Linter fixes

* Linter fixes

* imports update
---
 terraform/cache/handlers/provider.go      | 27 ++++++++++++++++-
 terraform/cache/handlers/provider_test.go | 36 +++++++++++++++++++++++
 2 files changed, 62 insertions(+), 1 deletion(-)
 create mode 100644 terraform/cache/handlers/provider_test.go

diff --git a/terraform/cache/handlers/provider.go b/terraform/cache/handlers/provider.go
index 0b76002722..168d671dd6 100644
--- a/terraform/cache/handlers/provider.go
+++ b/terraform/cache/handlers/provider.go
@@ -4,6 +4,7 @@ package handlers
 import (
 	"context"
 	liberrors "errors"
+	"strings"
 	"syscall"
 
 	"github.com/gruntwork-io/terragrunt/terraform/cache/models"
@@ -31,6 +32,15 @@ var availablePlatforms []*models.Platform = []*models.Platform{
 	{OS: "windows", Arch: "amd64"},
 }
 
+var offlineErrors = []error{
+	syscall.ECONNREFUSED,
+	syscall.ECONNRESET,
+	syscall.ECONNABORTED,
+	syscall.EHOSTUNREACH,
+	syscall.ENETUNREACH,
+	syscall.ENETDOWN,
+}
+
 // ProviderHandlers is a slice of ProviderHandler.
 type ProviderHandlers []ProviderHandler
 
@@ -120,7 +130,7 @@ func (handler *CommonProviderHandler) DiscoveryURL(ctx context.Context, registry
 
 	urls, err := DiscoveryURL(ctx, registryName)
 	if err != nil {
-		if !liberrors.As(err, &NotFoundWellKnownURL{}) && !liberrors.Is(err, syscall.ECONNREFUSED) {
+		if !IsOfflineError(err) {
 			return nil, err
 		}
 
@@ -134,3 +144,18 @@ func (handler *CommonProviderHandler) DiscoveryURL(ctx context.Context, registry
 
 	return urls, nil
 }
+
+// IsOfflineError returns true if the given error is an offline error and can be use default URL.
+func IsOfflineError(err error) bool {
+	if liberrors.As(err, &NotFoundWellKnownURL{}) {
+		return true
+	}
+
+	for _, connErr := range offlineErrors {
+		if liberrors.Is(err, connErr) || strings.Contains(err.Error(), connErr.Error()) {
+			return true
+		}
+	}
+
+	return false
+}
diff --git a/terraform/cache/handlers/provider_test.go b/terraform/cache/handlers/provider_test.go
new file mode 100644
index 0000000000..c91732400c
--- /dev/null
+++ b/terraform/cache/handlers/provider_test.go
@@ -0,0 +1,36 @@
+package handlers_test
+
+import (
+	"errors"
+	"syscall"
+	"testing"
+
+	"github.com/gruntwork-io/terragrunt/terraform/cache/handlers"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestIsOfflineError(t *testing.T) {
+	t.Parallel()
+	testCases := []struct {
+		err      error
+		expected bool
+		desc     string
+	}{
+		{syscall.ECONNREFUSED, true, "connection refused"},
+		{syscall.ECONNRESET, true, "connection reset by peer"},
+		{syscall.ECONNABORTED, true, "connection aborted"},
+		{syscall.ENETUNREACH, true, "network is unreachable"},
+		{errors.New("get \"https://registry.terraform.io/.well-known/terraform.json\": dial tcp: lookup registry.terraform.io on 185.12.64.1:53: dial udp 185.12.64.1:53: connect: network is unreachable"), true, "network is unreachable"},
+		{errors.New("get \"https://registry.terraform.io/.well-known/terraform.json\": read tcp 10.10.230.10:58328->10.245.10.15:443: read: connection reset by peer"), true, "network is unreachable"},
+		{errors.New("random error"), false, "a random error that should not be offline"},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.desc, func(t *testing.T) {
+			t.Parallel()
+			result := handlers.IsOfflineError(tc.err)
+			assert.Equal(t, tc.expected, result, "Expected result for %v is %v", tc.desc, tc.expected)
+		})
+	}
+}