fix: retry schema-migration sfn on Batch.AWSBatchException (#7165)

Co-authored-by: nayib-jose-gloria <[email protected]>
chanzuckerberg · Jun 6, 2024 · a6c5645 · a6c5645
1 parent df9b3e5
commit a6c5645
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 12 deletions.
diff --git a/.happy/terraform/modules/schema_migration/main.tf b/.happy/terraform/modules/schema_migration/main.tf
@@ -234,7 +234,7 @@ resource aws_sfn_state_machine sfn_schema_migration {
       },
       "Next": "SpanCollections",
       "Retry": [ {
-          "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException"],
+          "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException", "Batch.AWSBatchException"],
           "IntervalSeconds": 2,
           "MaxAttempts": 7,
           "BackoffRate": 5
@@ -303,7 +303,7 @@ resource aws_sfn_state_machine sfn_schema_migration {
             },
             "Next": "DatasetsExists",
             "Retry": [ {
-                "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException"],
+                "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException", "Batch.AWSBatchException"],
                 "IntervalSeconds": 2,
                 "MaxAttempts": 7,
                 "BackoffRate": 5
@@ -366,7 +366,7 @@ resource aws_sfn_state_machine sfn_schema_migration {
             },
             "End": true,
             "Retry": [ {
-                "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException"],
+                "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException", "Batch.AWSBatchException"],
                 "IntervalSeconds": 2,
                 "MaxAttempts": 7,
                 "BackoffRate": 5
@@ -439,7 +439,7 @@ resource aws_sfn_state_machine sfn_schema_migration {
                   },
                   "Next": "StepFunctionsStartExecution",
                   "Retry": [ {
-                      "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException"],
+                      "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException", "Batch.AWSBatchException"],
                       "IntervalSeconds": 2,
                       "MaxAttempts": 7,
                       "BackoffRate": 5
@@ -578,7 +578,7 @@ resource aws_sfn_state_machine sfn_schema_migration {
         }
       },
       "Retry": [ {
-        "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException"],
+        "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException", "Batch.AWSBatchException"],
         "IntervalSeconds": 2,
         "MaxAttempts": 7,
         "BackoffRate": 5

diff --git a/.happy/terraform/modules/sfn/main.tf b/.happy/terraform/modules/sfn/main.tf
@@ -84,7 +84,7 @@ resource "aws_sfn_state_machine" "state_machine" {
           }
         ],
         "Retry": [ {
-            "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException"],
+            "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException", "Batch.AWSBatchException"],
             "IntervalSeconds": 2,
             "MaxAttempts": 7,
             "BackoffRate": 5
@@ -144,7 +144,7 @@ resource "aws_sfn_state_machine" "state_machine" {
         },
         "Resource": "arn:aws:states:::aws-sdk:batch:registerJobDefinition",
         "Retry": [ {
-            "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException"],
+            "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException", "Batch.AWSBatchException"],
             "IntervalSeconds": 2,
             "MaxAttempts": 7,
             "BackoffRate": 5
@@ -179,7 +179,7 @@ resource "aws_sfn_state_machine" "state_machine" {
         "ResultPath": null,
         "TimeoutSeconds": ${local.timeout},
         "Retry": [ {
-            "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException"],
+            "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException", "Batch.AWSBatchException"],
             "IntervalSeconds": 2,
             "MaxAttempts": 7,
             "BackoffRate": 5
@@ -223,7 +223,7 @@ resource "aws_sfn_state_machine" "state_machine" {
                   }
                 },
                 "Retry": [ {
-                    "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException"],
+                    "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException", "Batch.AWSBatchException"],
                     "IntervalSeconds": 2,
                     "MaxAttempts": 7,
                     "BackoffRate": 5
@@ -271,7 +271,7 @@ resource "aws_sfn_state_machine" "state_machine" {
                   }
                 },
                 "Retry": [ {
-                    "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException"],
+                    "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException", "Batch.AWSBatchException"],
                     "IntervalSeconds": 2,
                     "MaxAttempts": 7,
                     "BackoffRate": 5
@@ -340,7 +340,7 @@ resource "aws_sfn_state_machine" "state_machine" {
         },
         "Resource": "arn:aws:states:::aws-sdk:batch:deregisterJobDefinition",
         "Retry": [ {
-            "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException"],
+            "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException", "Batch.AWSBatchException"],
             "IntervalSeconds": 2,
             "MaxAttempts": 7,
             "BackoffRate": 5
@@ -355,7 +355,7 @@ resource "aws_sfn_state_machine" "state_machine" {
         },
         "Resource": "arn:aws:states:::aws-sdk:batch:deregisterJobDefinition",
         "Retry": [ {
-            "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException"],
+            "ErrorEquals": ["AWS.Batch.TooManyRequestsException", "Batch.BatchException", "Batch.AWSBatchException"],
             "IntervalSeconds": 2,
             "MaxAttempts": 7,
             "BackoffRate": 5