Merge pull request #1341 from GSA/shared_memory

use shared memory instead of expiring dict for jobs cache

.ds.baseline

@@ -384,5 +384,5 @@
       }
     ]
   },
-  "generated_at": "2024-09-11T14:31:46Z"
+  "generated_at": "2024-09-26T20:29:19Z"
 }

.github/workflows/checks.yml

@@ -54,7 +54,7 @@ jobs:
     - name: Check for dead code
       run: make dead-code
     - name: Run tests with coverage
-      run: poetry run coverage run --omit=*/notifications_utils/*,*/migrations/* -m pytest --maxfail=10
+      run: poetry run coverage run --omit=*/migrations/* -m pytest --maxfail=10
       env:
         SQLALCHEMY_DATABASE_TEST_URI: postgresql://user:password@localhost:5432/test_notification_api
         NOTIFY_E2E_TEST_EMAIL: ${{ secrets.NOTIFY_E2E_TEST_EMAIL }}

Makefile

@@ -81,7 +81,7 @@ test: ## Run tests and create coverage report
 	poetry run black .
 	poetry run flake8 .
 	poetry run isort --check-only ./app ./tests
-	poetry run coverage run --omit=*/notifications_utils/*,*/migrations/* -m pytest --maxfail=10
+	poetry run coverage run --omit=*/migrations/* -m pytest --maxfail=10
 
 	poetry run coverage report -m --fail-under=95
 	poetry run coverage html -d .coverage_cache

app/aws/s3.py

@@ -1,31 +1,43 @@
 import datetime
 import re
 import time
+from multiprocessing import Manager
 
 import botocore
 from boto3 import Session
-from expiringdict import ExpiringDict
 from flask import current_app
 
-from app import redis_store
 from app.clients import AWS_CLIENT_CONFIG
 from notifications_utils import aware_utcnow
 
 FILE_LOCATION_STRUCTURE = "service-{}-notify/{}.csv"
 
 # Temporarily extend cache to 7 days
 ttl = 60 * 60 * 24 * 7
-JOBS = ExpiringDict(max_len=20000, max_age_seconds=ttl)
+manager = Manager()
+job_cache = manager.dict()
 
 
-JOBS_CACHE_HITS = "JOBS_CACHE_HITS"
-JOBS_CACHE_MISSES = "JOBS_CACHE_MISSES"
-
 # Global variable
 s3_client = None
 s3_resource = None
 
 
+def set_job_cache(job_cache, key, value):
+    job_cache[key] = (value, time.time() + 8 * 24 * 60 * 60)
+
+
+def clean_cache():
+    current_time = time.time()
+    keys_to_delete = []
+    for key, (_, expiry_time) in job_cache.items():
+        if expiry_time < current_time:
+            keys_to_delete.append(key)
+
+    for key in keys_to_delete:
+        del job_cache[key]
+
+
 def get_s3_client():
     global s3_client
     if s3_client is None:
@@ -127,29 +139,29 @@ def get_s3_files():
 
     s3res = get_s3_resource()
     current_app.logger.info(
-        f"JOBS cache length before regen: {len(JOBS)} #notify-admin-1200"
+        f"job_cache length before regen: {len(job_cache)} #notify-admin-1200"
     )
     for object in objects:
         # We put our csv files in the format "service-{service_id}-notify/{job_id}"
         try:
             object_arr = object.split("/")
             job_id = object_arr[1]  # get the job_id
             job_id = job_id.replace(".csv", "")  # we just want the job_id
-            if JOBS.get(job_id) is None:
+            if job_cache.get(job_id) is None:
                 object = (
                     s3res.Object(bucket_name, object)
                     .get()["Body"]
                     .read()
                     .decode("utf-8")
                 )
                 if "phone number" in object.lower():
-                    JOBS[job_id] = object
+                    set_job_cache(job_cache, job_id, object)
         except LookupError:
             # perhaps our key is not formatted as we expected.  If so skip it.
             current_app.logger.exception("LookupError #notify-admin-1200")
 
     current_app.logger.info(
-        f"JOBS cache length after regen: {len(JOBS)} #notify-admin-1200"
+        f"job_cache length after regen: {len(job_cache)} #notify-admin-1200"
     )
 
 
@@ -287,20 +299,6 @@ def get_job_from_s3(service_id, job_id):
     return None
 
 
-def incr_jobs_cache_misses():
-    if not redis_store.get(JOBS_CACHE_MISSES):
-        redis_store.set(JOBS_CACHE_MISSES, 1)
-    else:
-        redis_store.incr(JOBS_CACHE_MISSES)
-
-
-def incr_jobs_cache_hits():
-    if not redis_store.get(JOBS_CACHE_HITS):
-        redis_store.set(JOBS_CACHE_HITS, 1)
-    else:
-        redis_store.incr(JOBS_CACHE_HITS)
-
-
 def extract_phones(job):
     job = job.split("\r\n")
     first_row = job[0]
@@ -333,7 +331,8 @@ def extract_phones(job):
 
 
 def extract_personalisation(job):
-    job = job.split("\r\n")
+
+    job = job[0].split("\r\n")
     first_row = job[0]
     job.pop(0)
     first_row = first_row.split(",")
@@ -348,18 +347,15 @@ def extract_personalisation(job):
 
 
 def get_phone_number_from_s3(service_id, job_id, job_row_number):
-    # We don't want to constantly pull down a job from s3 every time we need a phone number.
-    # At the same time we don't want to store it in redis or the db
-    # So this is a little recycling mechanism to reduce the number of downloads.
-    job = JOBS.get(job_id)
+    job = job_cache.get(job_id)
     if job is None:
         current_app.logger.info(f"job {job_id} was not in the cache")
         job = get_job_from_s3(service_id, job_id)
         # Even if it is None, put it here to avoid KeyErrors
-        JOBS[job_id] = job
-        incr_jobs_cache_misses()
+        set_job_cache(job_cache, job_id, job)
     else:
-        incr_jobs_cache_hits()
+        # skip expiration date from cache, we don't need it here
+        job = job[0]
 
     if job is None:
         current_app.logger.error(
@@ -369,22 +365,17 @@ def get_phone_number_from_s3(service_id, job_id, job_row_number):
 
     # If we look in the JOBS cache for the quick lookup dictionary of phones for a given job
     # and that dictionary is not there, create it
-    if JOBS.get(f"{job_id}_phones") is None:
-        JOBS[f"{job_id}_phones"] = extract_phones(job)
+    if job_cache.get(f"{job_id}_phones") is None:
+        phones = extract_phones(job)
+        set_job_cache(job_cache, f"{job_id}_phones", phones)
 
     # If we can find the quick dictionary, use it
-    if JOBS.get(f"{job_id}_phones") is not None:
-        phone_to_return = JOBS.get(f"{job_id}_phones").get(job_row_number)
-        if phone_to_return:
-            return phone_to_return
-        else:
-            current_app.logger.warning(
-                f"Was unable to retrieve phone number from lookup dictionary for job {job_id}"
-            )
-            return "Unavailable"
+    phone_to_return = phones[job_row_number]
+    if phone_to_return:
+        return phone_to_return
     else:
-        current_app.logger.error(
-            f"Was unable to construct lookup dictionary for job {job_id}"
+        current_app.logger.warning(
+            f"Was unable to retrieve phone number from lookup dictionary for job {job_id}"
         )
         return "Unavailable"
 
@@ -393,13 +384,10 @@ def get_personalisation_from_s3(service_id, job_id, job_row_number):
     # We don't want to constantly pull down a job from s3 every time we need the personalisation.
     # At the same time we don't want to store it in redis or the db
     # So this is a little recycling mechanism to reduce the number of downloads.
-    job = JOBS.get(job_id)
+    job = job_cache.get(job_id)
     if job is None:
         job = get_job_from_s3(service_id, job_id)
-        JOBS[job_id] = job
-        incr_jobs_cache_misses()
-    else:
-        incr_jobs_cache_hits()
+        set_job_cache(job_cache, job_id, job)
 
     # If the job is None after our attempt to retrieve it from s3, it
     # probably means the job is old and has been deleted from s3, in
@@ -414,12 +402,14 @@ def get_personalisation_from_s3(service_id, job_id, job_row_number):
 
     # If we look in the JOBS cache for the quick lookup dictionary of personalisations for a given job
     # and that dictionary is not there, create it
-    if JOBS.get(f"{job_id}_personalisation") is None:
-        JOBS[f"{job_id}_personalisation"] = extract_personalisation(job)
+    if job_cache.get(f"{job_id}_personalisation") is None:
+        set_job_cache(
+            job_cache, f"{job_id}_personalisation", extract_personalisation(job)
+        )
 
     # If we can find the quick dictionary, use it
-    if JOBS.get(f"{job_id}_personalisation") is not None:
-        personalisation_to_return = JOBS.get(f"{job_id}_personalisation").get(
+    if job_cache.get(f"{job_id}_personalisation") is not None:
+        personalisation_to_return = job_cache.get(f"{job_id}_personalisation")[0].get(
             job_row_number
         )
         if personalisation_to_return:

app/celery/tasks.py

@@ -446,6 +446,11 @@ def regenerate_job_cache():
     s3.get_s3_files()
 
 
+@notify_celery.task(name="clean-job-cache")
+def clean_job_cache():
+    s3.clean_cache()
+
+
 @notify_celery.task(name="delete-old-s3-objects")
 def delete_old_s3_objects():
     s3.cleanup_old_s3_objects()

app/config.py

@@ -269,6 +269,11 @@ class Config(object):
                     "expires": 60,
                 },  # Ensure it doesn't run if missed
             },
+            "clean-job-cache": {
+                "task": "clean-job-cache",
+                "schedule": crontab(hour=2, minute=11),
+                "options": {"queue": QueueNames.PERIODIC},
+            },
             "cleanup-unfinished-jobs": {
                 "task": "cleanup-unfinished-jobs",
                 "schedule": crontab(hour=4, minute=5),

tests/app/aws/test_s3.py

@@ -40,6 +40,7 @@ def test_cleanup_old_s3_objects(mocker):
     whereas a 30 day old job ("A") is.
     """
     mocker.patch("app.aws.s3.get_bucket_name", return_value="Bucket")
+
     mock_s3_client = mocker.Mock()
     mocker.patch("app.aws.s3.get_s3_client", return_value=mock_s3_client)
     mock_remove_csv_object = mocker.patch("app.aws.s3.remove_csv_object")
@@ -110,7 +111,6 @@ def test_get_s3_file_makes_correct_call(notify_api, mocker):
 def test_get_phone_number_from_s3(
     mocker, job, job_id, job_row_number, expected_phone_number
 ):
-    mocker.patch("app.aws.s3.redis_store")
     get_job_mock = mocker.patch("app.aws.s3.get_job_from_s3")
     get_job_mock.return_value = job
     phone_number = get_phone_number_from_s3("service_id", job_id, job_row_number)
@@ -175,7 +175,6 @@ def test_get_job_from_s3_exponential_backoff_file_not_found(mocker):
 def test_get_personalisation_from_s3(
     mocker, job, job_id, job_row_number, expected_personalisation
 ):
-    mocker.patch("app.aws.s3.redis_store")
     get_job_mock = mocker.patch("app.aws.s3.get_job_from_s3")
     get_job_mock.return_value = job
     personalisation = get_personalisation_from_s3("service_id", job_id, job_row_number)