Merge pull request #1478 from GSA/main

12/12/2024 Production Deploy

app/celery/provider_tasks.py

@@ -1,6 +1,5 @@
 import json
 import os
-import random
 from datetime import timedelta
 
 from botocore.exceptions import ClientError
@@ -30,7 +29,8 @@ DELIVERY_RECEIPT_DELAY_IN_SECONDS = 30
 @notify_celery.task(
     bind=True,
     name="check_sms_delivery_receipt",
-    max_retries=72,
+    max_retries=48,
+    default_retry_delay=300,
 )
 def check_sms_delivery_receipt(self, message_id, notification_id, sent_at):
     """
@@ -62,10 +62,7 @@ def check_sms_delivery_receipt(self, message_id, notification_id, sent_at):
                 carrier=carrier,
                 provider_response=provider_response,
             )
-            base_delay = 3600  # one hour
-            jitter = random.randint(-1200, +1200)  # nosec B311
-            retry_delay = base_delay + jitter
-            raise self.retry(countdown=retry_delay, exc=ntfe)
+            raise self.retry(exc=ntfe)
         except ClientError as err:
             # Probably a ThrottlingException but could be something else
             error_code = err.response["Error"]["Code"]
@@ -80,10 +77,7 @@ def check_sms_delivery_receipt(self, message_id, notification_id, sent_at):
                 carrier=carrier,
                 provider_response=provider_response,
             )
-            base_delay = 3600  # one hour
-            jitter = random.randint(-1200, +1200)  # nosec B311
-            retry_delay = base_delay + jitter
-            raise self.retry(countdown=retry_delay, exc=err)
+            raise self.retry(exc=err)
 
     if status == "success":
         status = NotificationStatus.DELIVERED

app/clients/cloudwatch/aws_cloudwatch.py

@@ -158,7 +158,7 @@ class AwsCloudwatchClient(Client):
                 message["delivery"].get("phoneCarrier", "Unknown Carrier"),
             )
 
-        if time_now > (created_at + timedelta(hours=73)):
+        if time_now > (created_at + timedelta(hours=3)):
             # see app/models.py Notification. This message corresponds to "permanent-failure",
             # but we are copy/pasting here to avoid circular imports.
             return "failure", "Unable to find carrier response."

app/config.py

@@ -167,7 +167,7 @@ class Config(object):
     current_minute = (datetime.now().minute + 1) % 60
 
     CELERY = {
-        "worker_max_tasks_per_child": 2000,
+        "worker_max_tasks_per_child": 500,
         "broker_url": REDIS_URL,
         "broker_transport_options": {
             "visibility_timeout": 310,

migrations/versions/0413_add_message_id.py

@@ -0,0 +1,28 @@
+"""
+
+Revision ID: 0413_add_message_id
+Revises: 412_remove_priority
+Create Date: 2023-12-11 11:35:22.873930
+
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "0413_add_message_id"
+down_revision = "0412_remove_priority"
+
+
+def upgrade():
+    op.add_column("notifications", sa.Column("message_id", sa.Text))
+    op.create_index(
+        "ix_notifications_message_id",
+        "notifications",
+        ["message_id"],
+        unique=False,
+    )
+
+
+def downgrade():
+    op.drop_index("ix_notifications_message_id", table_name="notifications")
+    op.drop_column("notifications", "message_id")