Develop Reference

What is the recommended way to instantiate and pass around a redis client with FastAPI

I'm using FastAPI with Redis. My app looks something like this
from fastapi import FastAPI
import redis
# Instantiate redis client
r = redis.Redis(host='localhost', port=6379, db=0, decode_responses=True)
# Instantiate fastapi app
app = FastAPI()
#app.get("/foo/")
async def foo():
x = r.get("foo")
return {"message": x}
#app.get("/bar/")
async def bar():
x = r.get("bar")
return {"message": x}
Is it bad practice to create r as a module-scoped variable like this? If so what are the drawbacks?
In Tiangolo's tutorial on setting up a SQL database connection he uses a dependency, which I guess in my case would look something like this
from fastapi import Depends, FastAPI
import redis
# Instantiate fastapi app
app = FastAPI()
# Dependency
def get_redis():
return redis.Redis(host='localhost', port=6379, db=0, decode_responses=True)
#app.get("/foo/")
async def foo(r = Depends(get_redis)):
x = r.get("foo")
return {"message": x}
#app.get("/bar/")
async def bar(r = Depends(get_redis)):
x = r.get("bar")
return {"message": x}
I'm a bit confused as to which of these methods (or something else) would be preferred and why.

Depends will evaluate every time your function got a request, so your second example will create a new connection for each request. As #JarroVGIT said, we can use connection pooling to maintain the connection from FastAPI to Redis and reduce open-closing connection costs.
Usually, I create a different file to define the connection. Let's say we have config/db.py:
import redis
def create_redis():
return redis.ConnectionPool(
host='localhost',
port=6379,
db=0,
decode_responses=True
)
pool = create_redis()
Then in the main.py
from fastapi import Depends, FastAPI
import redis
from config.db import pool
app = FastAPI()
def get_redis():
# Here, we re-use our connection pool
# not creating a new one
return redis.Redis(connection_pool=pool)
#app.get("/items/{item_id}")
def read_item(item_id: int, cache = Depends(get_redis)):
status = cache.get(item_id)
return {"item_name": status}
#app.put("/items/{item_id}")
def update_item(item_id: int, cache = Depends(get_redis)):
cache.set(item_id, "available")
return {"status": "available", "item_id": item_id}
Usually, I also split the dependencies file like the doc so we can call it from our routing module, but for simplicity, I will leave it like this.
You can check this repo to experiment by yourself. It has more comprehensive code and I have already created several scenarios that might help you understand the difference. And it will also cover how your first example may block other endpoints.

In your second example, every time your creating new redis instance and one time it reach max connection limit. If you put code like this that much more clean and re-usable,
from fastapi import FastAPI
import redis
class AppAPI(FastAPI):
def __init__(self):
self.redis_client = redis.Redis(host='localhost', port=6379, db=0, decode_responses=True)
#app.get("/foo/")
async def foo():
x = self.redis_client.get("foo")
return {"message": x}

https://github.com/redis/redis-py#connection-pools. You can define the pool at module level and import it wherever needed. All Redis connection will be created out of the pool.
pool = redis.ConnectionPool(host='localhost', port=6379, db=0)
r = redis.Redis(connection_pool=pool)

Cache or Reuse Mongodb connection in AWS lambda using Python

I'm building a serverless application using Python and Mongodb. In documentation I found that I need to write db connection outside handler function. I have used Mangum python package as adapter to handle API gateway.
from fastapi import FastAPI, Body, status, Depends
from mangum import Mangum
from motor.motor_asyncio import AsyncIOMotorClient
from fastapi.responses import JSONResponse
from app.utility.config import MONGODB_URL, MAX_CONNECTIONS_COUNT, MIN_CONNECTIONS_COUNT, MAX_DB_THREADS_WAIT_COUNT, MAX_DB_THREAD_QUEUE_TIMEOUT_COUNT
application= FastAPI()
client = AsyncIOMotorClient(str(MONGODB_URL),
maxPoolSize=MAX_CONNECTIONS_COUNT,
minPoolSize=MIN_CONNECTIONS_COUNT,
waitQueueMultiple = MAX_DB_THREADS_WAIT_COUNT,
waitQueueTimeoutMS = MAX_DB_THREAD_QUEUE_TIMEOUT_COUNT )
async def get_database() -> AsyncIOMotorClient:
return client
#application.post("/createStudent")
async def create_student(student = Body(...), db: AsyncIOMotorClient = Depends(get_database)):
new_student = await db["college"]["students"].insert_one(student)
created_student = await db["college"]["students"].find_one({"_id": new_student.inserted_id})
return JSONResponse(status_code=status.HTTP_201_CREATED, content=created_student)
#application.post("/createTeacher")
async def create_teacher(teacher = Body(...), db: AsyncIOMotorClient = Depends(get_database)):
new_teacher = await db["college"]["students"].insert_one(teacher)
created_teacher = await db["college"]["students"].find_one({"_id": new_teacher.inserted_id})
return JSONResponse(status_code=status.HTTP_201_CREATED, content=created_teacher)
handler = Mangum(application)
For every API request, new connection is created. How to cache db so that new request uses old connection. Every time new request is created so that lambda compute time is increased dramatically after db hits max connection limit.
There are examples for node js but for python I could not find anywhere

How to use Uvicorn with asyncio.create_task() to put task in background?

Let's say I have a web app driven by Uvicorn server, the app implements GraphQL API with a mutation that starts long calculations on the server-side and a query endpoint that checks the status of the server. Let's say we want to know how many tasks are running in the background.
I have a simplified code, which does not work:
import asyncio
import logging
import time
import ariadne
import ariadne.asgi
import uvicorn
import starlette as sl
import starlette.applications
query_t = ariadne.QueryType()
mutation_t = ariadne.MutationType()
FIFO = []
async def long_task():
print('Starting long task...')
global FIFO
FIFO = [1, *FIFO]
# mock long calc with 5sec sleep
time.sleep(5)
FIFO.pop()
print('Long task finished!')
#mutation_t.field('longTask')
async def resolve_long_task(_parent, info):
print('Start to resolve long task...')
asyncio.create_task(long_task())
print('Resolve finished!')
return {}
#query_t.field('ping')
async def resolve_ping(_parent, info):
return f'FIFO has {len(FIFO)} elements'
def main():
schema_str = ariadne.gql('''
type Mutation{
longTask: longTaskResponse
}
type longTaskResponse {
message: String
}
type Query {
ping: String
}
''')
schema = ariadne.make_executable_schema(schema_str, query_t, mutation_t)
gql_app = ariadne.asgi.GraphQL(schema)
app = sl.applications.Starlette(routes=[sl.routing.Mount('/graphql', gql_app)])
uvicorn.run(app,
host='0.0.0.0',
port=9002,
log_level='error')
if __name__ == '__main__':
main()
After running
$ python main.py
I send a mutation in the GraphQL GUI in the first tab:
mutation longTaskQueue{
longTask {
message
}
}
In the second tab, I try to retrieve the length of the FIFO:
query ping {
ping
}
It seems that it's possible to run 2 long_task, but ping is waiting until all long_task will be finished. My general question is how to run heavy code in the background and do not block GQL API?

After many attempts I made it, now I can put many tasks in the background and track their amount (API is not freezing on one long task). What is happening is
the blocking computations are run in a pool:
import asyncio
import logging
import time
import ariadne
import ariadne.asgi
import uvicorn
import starlette as sl
import starlette.applications
from concurrent.futures import ProcessPoolExecutor, ThreadPoolExecutor
query_t = ariadne.QueryType()
mutation_t = ariadne.MutationType()
FIFO = []
async def long_task():
print('Starting long task...')
global FIFO
FIFO = [1, *FIFO]
# mock long calc with 5sec sleep
time.sleep(5)
FIFO.pop()
print('Long task finished!')
def run(corofn, *args):
loop = asyncio.new_event_loop()
try:
coro = corofn(*args)
asyncio.set_event_loop(loop)
return loop.run_until_complete(coro)
finally:
loop.close()
#mutation_t.field('longTask')
async def resolve_long_task(_parent, info):
loop = asyncio.get_event_loop()
executor = ThreadPoolExecutor(max_workers=5)
loop.set_default_executor(ProcessPoolExecutor())
print('Start to resolve long task...')
loop.run_in_executor(executor, run, long_task)
print('Resolve finished!')
return {}
#query_t.field('ping')
async def resolve_ping(_parent, info):
return f'FIFO has {len(FIFO)} elements'
def main():
schema_str = ariadne.gql('''
type Mutation{
longTask: longTaskResponse
}
type longTaskResponse {
message: String
}
type Query {
ping: String
}
''')
schema = ariadne.make_executable_schema(schema_str, query_t, mutation_t)
gql_app = ariadne.asgi.GraphQL(schema)
app = sl.applications.Starlette(routes=[sl.routing.Mount('/graphql', gql_app)])
uvicorn.run(app,
host='0.0.0.0',
port=9002,
log_level='error')
if __name__ == '__main__':
main()
The solution inspired by this answer.

Flask Celery task locking

I am using Flask with Celery and I am trying to lock a specific task so that it can only be run one at a time. In the celery docs it gives a example of doing this Celery docs, Ensuring a task is only executed one at a time. This example that was given was for Django however I am using flask I have done my best to convert this to work with Flask however I still see myTask1 which has the lock can be run multiple times.
One thing that is not clear to me is if I am using the cache correctly, I have never used it before so all of it is new to me. One thing from the doc's that is mentioned but not explained is this
Doc Notes:
In order for this to work correctly you need to be using a cache backend where the .add operation is atomic. memcached is known to work well for this purpose.
Im not truly sure what that means, should i be using the cache in conjunction with a database and if so how would I do that? I am using mongodb. In my code I just have this setup for the cache cache = Cache(app, config={'CACHE_TYPE': 'simple'}) as that is what was mentioned in the Flask-Cache doc's Flask-Cache Docs
Another thing that is not clear to me is if there is anything different I need to do as I am calling my myTask1 from within my Flask route task1
Here is an example of my code that I am using.
from flask import (Flask, render_template, flash, redirect,
url_for, session, logging, request, g, render_template_string, jsonify)
from flask_caching import Cache
from contextlib import contextmanager
from celery import Celery
from Flask_celery import make_celery
from celery.result import AsyncResult
from celery.utils.log import get_task_logger
from celery.five import monotonic
from flask_pymongo import PyMongo
from hashlib import md5
import pymongo
import time
app = Flask(__name__)
cache = Cache(app, config={'CACHE_TYPE': 'simple'})
app.config['SECRET_KEY']= 'super secret key for me123456789987654321'
######################
# MONGODB SETUP
#####################
app.config['MONGO_HOST'] = 'localhost'
app.config['MONGO_DBNAME'] = 'celery-test-db'
app.config["MONGO_URI"] = 'mongodb://localhost:27017/celery-test-db'
mongo = PyMongo(app)
##############################
# CELERY ARGUMENTS
##############################
app.config['CELERY_BROKER_URL'] = 'amqp://localhost//'
app.config['CELERY_RESULT_BACKEND'] = 'mongodb://localhost:27017/celery-test-db'
app.config['CELERY_RESULT_BACKEND'] = 'mongodb'
app.config['CELERY_MONGODB_BACKEND_SETTINGS'] = {
"host": "localhost",
"port": 27017,
"database": "celery-test-db",
"taskmeta_collection": "celery_jobs",
}
app.config['CELERY_TASK_SERIALIZER'] = 'json'
celery = Celery('task',broker='mongodb://localhost:27017/jobs')
celery = make_celery(app)
LOCK_EXPIRE = 60 * 2 # Lock expires in 2 minutes
#contextmanager
def memcache_lock(lock_id, oid):
timeout_at = monotonic() + LOCK_EXPIRE - 3
# cache.add fails if the key already exists
status = cache.add(lock_id, oid, LOCK_EXPIRE)
try:
yield status
finally:
# memcache delete is very slow, but we have to use it to take
# advantage of using add() for atomic locking
if monotonic() < timeout_at and status:
# don't release the lock if we exceeded the timeout
# to lessen the chance of releasing an expired lock
# owned by someone else
# also don't release the lock if we didn't acquire it
cache.delete(lock_id)
#celery.task(bind=True, name='app.myTask1')
def myTask1(self):
self.update_state(state='IN TASK')
lock_id = self.name
with memcache_lock(lock_id, self.app.oid) as acquired:
if acquired:
# do work if we got the lock
print('acquired is {}'.format(acquired))
self.update_state(state='DOING WORK')
time.sleep(90)
return 'result'
# otherwise, the lock was already in use
raise self.retry(countdown=60) # redeliver message to the queue, so the work can be done later
#celery.task(bind=True, name='app.myTask2')
def myTask2(self):
print('you are in task2')
self.update_state(state='STARTING')
time.sleep(120)
print('task2 done')
#app.route('/', methods=['GET', 'POST'])
def index():
return render_template('index.html')
#app.route('/task1', methods=['GET', 'POST'])
def task1():
print('running task1')
result = myTask1.delay()
# get async task id
taskResult = AsyncResult(result.task_id)
# push async taskid into db collection job_task_id
mongo.db.job_task_id.insert({'taskid': str(taskResult), 'TaskName': 'task1'})
return render_template('task1.html')
#app.route('/task2', methods=['GET', 'POST'])
def task2():
print('running task2')
result = myTask2.delay()
# get async task id
taskResult = AsyncResult(result.task_id)
# push async taskid into db collection job_task_id
mongo.db.job_task_id.insert({'taskid': str(taskResult), 'TaskName': 'task2'})
return render_template('task2.html')
#app.route('/status', methods=['GET', 'POST'])
def status():
taskid_list = []
task_state_list = []
TaskName_list = []
allAsyncData = mongo.db.job_task_id.find()
for doc in allAsyncData:
try:
taskid_list.append(doc['taskid'])
except:
print('error with db conneciton in asyncJobStatus')
TaskName_list.append(doc['TaskName'])
# PASS TASK ID TO ASYNC RESULT TO GET TASK RESULT FOR THAT SPECIFIC TASK
for item in taskid_list:
try:
task_state_list.append(myTask1.AsyncResult(item).state)
except:
task_state_list.append('UNKNOWN')
return render_template('status.html', data_list=zip(task_state_list, TaskName_list))
Final Working Code
from flask import (Flask, render_template, flash, redirect,
url_for, session, logging, request, g, render_template_string, jsonify)
from flask_caching import Cache
from contextlib import contextmanager
from celery import Celery
from Flask_celery import make_celery
from celery.result import AsyncResult
from celery.utils.log import get_task_logger
from celery.five import monotonic
from flask_pymongo import PyMongo
from hashlib import md5
import pymongo
import time
import redis
from flask_redis import FlaskRedis
app = Flask(__name__)
# ADDING REDIS
redis_store = FlaskRedis(app)
# POINTING CACHE_TYPE TO REDIS
cache = Cache(app, config={'CACHE_TYPE': 'redis'})
app.config['SECRET_KEY']= 'super secret key for me123456789987654321'
######################
# MONGODB SETUP
#####################
app.config['MONGO_HOST'] = 'localhost'
app.config['MONGO_DBNAME'] = 'celery-test-db'
app.config["MONGO_URI"] = 'mongodb://localhost:27017/celery-test-db'
mongo = PyMongo(app)
##############################
# CELERY ARGUMENTS
##############################
# CELERY USING REDIS
app.config['CELERY_BROKER_URL'] = 'redis://localhost:6379/0'
app.config['CELERY_RESULT_BACKEND'] = 'mongodb://localhost:27017/celery-test-db'
app.config['CELERY_RESULT_BACKEND'] = 'mongodb'
app.config['CELERY_MONGODB_BACKEND_SETTINGS'] = {
"host": "localhost",
"port": 27017,
"database": "celery-test-db",
"taskmeta_collection": "celery_jobs",
}
app.config['CELERY_TASK_SERIALIZER'] = 'json'
celery = Celery('task',broker='mongodb://localhost:27017/jobs')
celery = make_celery(app)
LOCK_EXPIRE = 60 * 2 # Lock expires in 2 minutes
#contextmanager
def memcache_lock(lock_id, oid):
timeout_at = monotonic() + LOCK_EXPIRE - 3
print('in memcache_lock and timeout_at is {}'.format(timeout_at))
# cache.add fails if the key already exists
status = cache.add(lock_id, oid, LOCK_EXPIRE)
try:
yield status
print('memcache_lock and status is {}'.format(status))
finally:
# memcache delete is very slow, but we have to use it to take
# advantage of using add() for atomic locking
if monotonic() < timeout_at and status:
# don't release the lock if we exceeded the timeout
# to lessen the chance of releasing an expired lock
# owned by someone else
# also don't release the lock if we didn't acquire it
cache.delete(lock_id)
#celery.task(bind=True, name='app.myTask1')
def myTask1(self):
self.update_state(state='IN TASK')
print('dir is {} '.format(dir(self)))
lock_id = self.name
print('lock_id is {}'.format(lock_id))
with memcache_lock(lock_id, self.app.oid) as acquired:
print('in memcache_lock and lock_id is {} self.app.oid is {} and acquired is {}'.format(lock_id, self.app.oid, acquired))
if acquired:
# do work if we got the lock
print('acquired is {}'.format(acquired))
self.update_state(state='DOING WORK')
time.sleep(90)
return 'result'
# otherwise, the lock was already in use
raise self.retry(countdown=60) # redeliver message to the queue, so the work can be done later
#celery.task(bind=True, name='app.myTask2')
def myTask2(self):
print('you are in task2')
self.update_state(state='STARTING')
time.sleep(120)
print('task2 done')
#app.route('/', methods=['GET', 'POST'])
def index():
return render_template('index.html')
#app.route('/task1', methods=['GET', 'POST'])
def task1():
print('running task1')
result = myTask1.delay()
# get async task id
taskResult = AsyncResult(result.task_id)
# push async taskid into db collection job_task_id
mongo.db.job_task_id.insert({'taskid': str(taskResult), 'TaskName': 'myTask1'})
return render_template('task1.html')
#app.route('/task2', methods=['GET', 'POST'])
def task2():
print('running task2')
result = myTask2.delay()
# get async task id
taskResult = AsyncResult(result.task_id)
# push async taskid into db collection job_task_id
mongo.db.job_task_id.insert({'taskid': str(taskResult), 'TaskName': 'task2'})
return render_template('task2.html')
#app.route('/status', methods=['GET', 'POST'])
def status():
taskid_list = []
task_state_list = []
TaskName_list = []
allAsyncData = mongo.db.job_task_id.find()
for doc in allAsyncData:
try:
taskid_list.append(doc['taskid'])
except:
print('error with db conneciton in asyncJobStatus')
TaskName_list.append(doc['TaskName'])
# PASS TASK ID TO ASYNC RESULT TO GET TASK RESULT FOR THAT SPECIFIC TASK
for item in taskid_list:
try:
task_state_list.append(myTask1.AsyncResult(item).state)
except:
task_state_list.append('UNKNOWN')
return render_template('status.html', data_list=zip(task_state_list, TaskName_list))
if __name__ == '__main__':
app.secret_key = 'super secret key for me123456789987654321'
app.run(port=1234, host='localhost')
Here is also a screen shot you can see that I ran myTask1 two times and myTask2 a single time. Now I have the expected behavior for myTask1. Now myTask1 will be run by a single worker if another worker attempt to pick it up it will just keep retrying based on whatever i define.

In your question, you point out this warning from the Celery example you used:
In order for this to work correctly you need to be using a cache backend where the .add operation is atomic. memcached is known to work well for this purpose.
And you mention that you don't really understand what this means. Indeed, the code you show demonstrates that you've not heeded that warning, because your code uses an inappropriate backend.
Consider this code:
with memcache_lock(lock_id, self.app.oid) as acquired:
if acquired:
# do some work
What you want here is for acquired to be true only for one thread at a time. If two threads enter the with block at the same time, only one should "win" and have acquired be true. This thread that has acquired true can then proceed with its work, and the other thread has to skip doing the work and try again later to acquire the lock. In order to ensure that only one thread can have acquired true, .add must be atomic.
Here's some pseudo code of what .add(key, value) does:
1. if <key> is already in the cache:
2. return False
3. else:
4. set the cache so that <key> has the value <value>
5. return True
If the execution of .add is not atomic, this could happen if two threads A and B execute .add("foo", "bar"). Assume an empty cache at the start.
Thread A executes 1. if "foo" is already in the cache and finds that "foo" is not in the cache, and jumps to line 3 but the thread scheduler switches control to thread B.
Thread B also executes 1. if "foo" is already in the cache, and also finds that "foo" is not in the cache. So it jumps to line 3 and then executes line 4 and 5 which sets the key "foo" to the value "bar" and the call returns True.
Eventually, the scheduler gives control back to Thread A, which continues executing 3, 4, 5 and also sets the key "foo" to the value "bar" and also returns True.
What you have here is two .add calls that return True, if these .add calls are made within memcache_lock this entails that two threads can have acquired be true. So two threads could do work at the same time, and your memcache_lock is not doing what it should be doing, which is only allow one thread to work at a time.
You are not using a cache that ensures that .add is atomic. You initialize it like this:
cache = Cache(app, config={'CACHE_TYPE': 'simple'})
The simple backend is scoped to a single process, has no thread-safety, and has an .add operation which is not atomic. (This does not involve Mongo at all by the way. If you wanted your cache to be backed by Mongo, you'd have to specify a backed specifically made to send data to a Mongo database.)
So you have to switch to another backend, one that guarantees that .add is atomic. You could follow the lead of the Celery example and use the memcached backend, which does have an atomic .add operation. I don't use Flask, but I've does essentially what you are doing with Django and Celery, and used the Redis backend successfully to provide the kind of locking you're using here.

I also found this to be a surprisingly hard problem. Inspired mainly by Sebastian's work on implementing a distributed locking algorithm in redis I wrote up a decorator function.
A key point to bear in mind about this approach is that we lock tasks at the level of the task's argument space, e.g. we allow multiple game update/process order tasks to run concurrently, but only one per game. That's what argument_signature achieves in the code below. You can see documentation on how we use this in our stack at this gist:
import base64
from contextlib import contextmanager
import json
import pickle as pkl
import uuid
from backend.config import Config
from redis import StrictRedis
from redis_cache import RedisCache
from redlock import Redlock
rds = StrictRedis(Config.REDIS_HOST, decode_responses=True, charset="utf-8")
rds_cache = StrictRedis(Config.REDIS_HOST, decode_responses=False, charset="utf-8")
redis_cache = RedisCache(redis_client=rds_cache, prefix="rc", serializer=pkl.dumps, deserializer=pkl.loads)
dlm = Redlock([{"host": Config.REDIS_HOST}])
TASK_LOCK_MSG = "Task execution skipped -- another task already has the lock"
DEFAULT_ASSET_EXPIRATION = 8 * 24 * 60 * 60 # by default keep cached values around for 8 days
DEFAULT_CACHE_EXPIRATION = 1 * 24 * 60 * 60 # we can keep cached values around for a shorter period of time
REMOVE_ONLY_IF_OWNER_SCRIPT = """
if redis.call("get",KEYS[1]) == ARGV[1] then
return redis.call("del",KEYS[1])
else
return 0
end
"""
#contextmanager
def redis_lock(lock_name, expires=60):
# https://breadcrumbscollector.tech/what-is-celery-beat-and-how-to-use-it-part-2-patterns-and-caveats/
random_value = str(uuid.uuid4())
lock_acquired = bool(
rds.set(lock_name, random_value, ex=expires, nx=True)
)
yield lock_acquired
if lock_acquired:
rds.eval(REMOVE_ONLY_IF_OWNER_SCRIPT, 1, lock_name, random_value)
def argument_signature(*args, **kwargs):
arg_list = [str(x) for x in args]
kwarg_list = [f"{str(k)}:{str(v)}" for k, v in kwargs.items()]
return base64.b64encode(f"{'_'.join(arg_list)}-{'_'.join(kwarg_list)}".encode()).decode()
def task_lock(func=None, main_key="", timeout=None):
def _dec(run_func):
def _caller(*args, **kwargs):
with redis_lock(f"{main_key}_{argument_signature(*args, **kwargs)}", timeout) as acquired:
if not acquired:
return TASK_LOCK_MSG
return run_func(*args, **kwargs)
return _caller
return _dec(func) if func is not None else _dec
Implementation in our task definitions file:
#celery.task(name="async_test_task_lock")
#task_lock(main_key="async_test_task_lock", timeout=UPDATE_GAME_DATA_TIMEOUT)
def async_test_task_lock(game_id):
print(f"processing game_id {game_id}")
time.sleep(TASK_LOCK_TEST_SLEEP)
How we test against a local celery cluster:
from backend.tasks.definitions import async_test_task_lock, TASK_LOCK_TEST_SLEEP
from backend.tasks.redis_handlers import rds, TASK_LOCK_MSG
class TestTaskLocking(TestCase):
def test_task_locking(self):
rds.flushall()
res1 = async_test_task_lock.delay(3)
res2 = async_test_task_lock.delay(5)
self.assertFalse(res1.ready())
self.assertFalse(res2.ready())
res3 = async_test_task_lock.delay(5)
res4 = async_test_task_lock.delay(5)
self.assertEqual(res3.get(), TASK_LOCK_MSG)
self.assertEqual(res4.get(), TASK_LOCK_MSG)
time.sleep(TASK_LOCK_TEST_SLEEP)
res5 = async_test_task_lock.delay(3)
self.assertFalse(res5.ready())
(as a goodie there's also a quick example of how to setup a redis_cache)

With this setup, you should still expect to see workers receiving the task, since the lock is checked inside of the task itself. The only difference will be that the work won't be performed if the lock is acquired by another worker.
In the example given in the docs, this is the desired behavior; if a lock already exists, the task will simply do nothing and finish as successful. What you want is slightly different; you want the work to be queued up instead of ignored.
In order to get the desired effect, you would need to make sure that the task will be picked up by a worker and performed some time in the future. One way to accomplish this would be with retrying.
#task(bind=True, name='my-task')
def my_task(self):
lock_id = self.name
with memcache_lock(lock_id, self.app.oid) as acquired:
if acquired:
# do work if we got the lock
print('acquired is {}'.format(acquired))
return 'result'
# otherwise, the lock was already in use
raise self.retry(countdown=60) # redeliver message to the queue, so the work can be done later

How to trigger a function after return statement in Flask

I have 2 functions.
1st function stores the data received in a list and 2nd function writes the data into a csv file.
I'm using Flask. Whenever a web service has been called it will store the data and send response to it, as soon as it sends response it triggers the 2nd function.
My Code:
from flask import Flask, flash, request, redirect, url_for, session
import json
app = Flask(__name__)
arr = []
#app.route("/test", methods=['GET','POST'])
def check():
arr.append(request.form['a'])
arr.append(request.form['b'])
res = {'Status': True}
return json.dumps(res)
def trigger():
df = pd.DataFrame({'x': arr})
df.to_csv("docs/xyz.csv", index=False)
return
Obviously the 2nd function is not called.
Is there a way to achieve this?
P.S: My real life problem is different where trigger function is time consuming and I don't want user to wait for it to finish execution.

One solution would be to have a background thread that will watch a queue. You put your csv data in the queue and the background thread will consume it. You can start such a thread before first request:
import threading
from multiprocessing import Queue
class CSVWriterThread(threading.Thread):
def __init__(self, *args, **kwargs):
threading.Thread.__init__(self, *args, **kwargs)
self.input_queue = Queue()
def send(self, item):
self.input_queue.put(item)
def close(self):
self.input_queue.put(None)
self.input_queue.join()
def run(self):
while True:
csv_array = self.input_queue.get()
if csv_array is None:
break
# Do something here ...
df = pd.DataFrame({'x': csv_array})
df.to_csv("docs/xyz.csv", index=False)
self.input_queue.task_done()
time.sleep(1)
# Done
self.input_queue.task_done()
return
#app.before_first_request
def activate_job_monitor():
thread = CSVWriterThread()
app.csvwriter = thread
thread.start()
And in your code put the message in the queue before returning:
#app.route("/test", methods=['GET','POST'])
def check():
arr.append(request.form['a'])
arr.append(request.form['b'])
res = {'Status': True}
app.csvwriter.send(arr)
return json.dumps(res)

P.S: My real life problem is different where trigger function is time consuming and I don't want user to wait for it to finish execution.
Consider using celery which is made for the very problem you're trying to solve. From docs:
Celery is a simple, flexible, and reliable distributed system to process vast amounts of messages, while providing operations with the tools required to maintain such a system.
I recommend you integrate celery with your flask app as described here. your trigger method would then become a straightforward celery task that you can execute without having to worry about long response time.

Im actually working on another interesting case on my side where i pass the work off to a python worker that sends the job to a redis queue. There are some great blogs using redis with Flask , you basically need to ensure redis is running (able to connect on port 6379)
The worker would look something like this:
import os
import redis
from rq import Worker, Queue, Connection
listen = ['default']
redis_url = os.getenv('REDISTOGO_URL', 'redis://localhost:6379')
conn = redis.from_url(redis_url)
if __name__ == '__main__':
with Connection(conn):
worker = Worker(list(map(Queue, listen)))
worker.work()
In my example I have a function that queries a database for usage and since it might be a lengthy process i pass it off to the worker (running as a seperate script)
def post(self):
data = Task.parser.parse_args()
job = q.enqueue_call(
func=migrate_usage, args=(my_args),
result_ttl=5000
)
print("Job ID is: {}".format(job.get_id()))
job_key = job.get_id()
print(str(Job.fetch(job_key, connection=conn).result))
if job:
return {"message": "Job : {} added to queue".format(job_key)}, 201
Credit due to the following article:
https://realpython.com/flask-by-example-implementing-a-redis-task-queue/#install-requirements

You can try use streaming. See next example:
import time
from flask import Flask, Response
app = Flask(__name__)
#app.route('/')
def main():
return '''<div>start</div>
<script>
var xhr = new XMLHttpRequest();
xhr.open('GET', '/test', true);
xhr.onreadystatechange = function(e) {
var div = document.createElement('div');
div.innerHTML = '' + this.readyState + ':' + this.responseText;
document.body.appendChild(div);
};
xhr.send();
</script>
'''
#app.route('/test')
def test():
def generate():
app.logger.info('request started')
for i in range(5):
time.sleep(1)
yield str(i)
app.logger.info('request finished')
yield ''
return Response(generate(), mimetype='text/plain')
if __name__ == '__main__':
app.run('0.0.0.0', 8080, True)
All magic in this example in genarator where you can start response data, after do some staff and yield empty data to end your stream.
For details look at http://flask.pocoo.org/docs/patterns/streaming/.

You can defer route specific actions with limited context by combining after_this_request and response.call_on_close. Note that request and response context won't be available but the route function context remains available. So you'll need to copy any request/response data you'll need into local variables for deferred access.
I moved your array to a local var to show how the function context is preserved. You could change your csv write function to an append so you're not pushing data endlessly into memory.
from flask import Flask, flash, request, redirect, url_for, session
import json
app = Flask(__name__)
#app.route("/test", methods=['GET','POST'])
def check():
arr = []
arr.append(request.form['a'])
arr.append(request.form['b'])
res = {'Status': True}
#flask.after_this_request
def add_close_action(response):
#response.call_on_close
def process_after_request():
df = pd.DataFrame({'x': arr})
df.to_csv("docs/xyz.csv", index=False)
return response
return json.dumps(res)