Skip to content

I would like my cached methods to never return expired values and block wait instead. #264

Open

Description

@Wim-De-Clercq

Hello,

I have a multiproces/thread scenario where my cached function can be called at the same time.

I end up falling into the scenario of

function for this key against the cache. When the dogpile
lock cannot be acquired, the method will block if no
previous value is available, until the lock is released and
a new value available. If a previous value
is available, that value is returned immediately without blocking.

So what I have is an expired value in the cache.
Thread 1 retrieves it, notices it's expired, starts creating a new value, and will return the new value.
Meanwhile; Thread 2 retrieves it, notices it's expired, notices thread 1 is busy creating and returns the old value.

But the old value is entirely useless for my case. It's expired for a reason. Is there a way to configure the cache region to never return expired values and just block/wait for creation instead. Just like a cache miss?

(simplified) Code to reproduce my issue:

import os
import time
from multiprocessing import Process

from dogpile.cache import make_region

cache_region = make_region()
cache_region.configure(
    "dogpile.cache.dbm",
    expiration_time=1,
    arguments={
        "filename": "/tmp/cache.dbm",
    },
)


@cache_region.cache_on_arguments()
def get_value():
    print(os.getpid(), "Making new value...")
    time.sleep(2)
    return f"Got value {time.time():.0f}"


def run():
    print(os.getpid(), get_value())


print("3 sync calls in a row (value should be made once):")
run()
run()
run()

processes = []
for i in range(2):
    process = Process(target=run)
    processes.append(process)

time.sleep(2)
print()
print("2 async calls after cache expired (value should be made once) and be different from the previous:")
for process in processes:
    process.start()
for process in processes:
    process.join()

Example output:

3 sync calls in a row (value should be made once):
103954 Making new value...
103954 Got value 1740579012
103954 Got value 1740579012
103954 Got value 1740579012

2 async calls after cache expired (value should be made once) and be different from the previous:
103978 Got value 1740579012
103977 Making new value...
103977 Got value 1740579016

In the above output, the line 103978 Got value 1740579012 is bad for me. That should never return an expired value.

Activity

  1. zzzeek commented on Feb 26, 2025

    @zzzeek
    Member

    Meanwhile; Thread 2 retrieves it, notices it's expired, notices thread 1 is busy creating and returns the old value.

    well that's the entire secret sauce of "dogpile cache" :)

    There's a way out for this one though it's not as flexible as it should be, but you can add a region invalidation strategy that considers any timeout to be hard invalidated. can confirm this works with your clear test case, thanks for that

    from dogpile.cache.region import DefaultInvalidationStrategy
    
    class CustomInvalidationStrategy(DefaultInvalidationStrategy):
        def __init__(self, cache_region):
            super().__init__()
            self.region = cache_region
    
        def is_hard_invalidated(self, timestamp: float) -> bool:
            return time.time() - timestamp > self.region.expiration_time
    
    
    cache_region = make_region()
    cache_region.configure(
        "dogpile.cache.redis",
        expiration_time=1,
        arguments={
            "host": "localhost",
            "redis_expiration_time": 10,
            "distributed_lock": True,
            "thread_local_lock": False,
            "lock_timeout": 30,
        },
        region_invalidator=CustomInvalidationStrategy(cache_region)
    )
  2. Wim-De-Clercq commented on Feb 26, 2025

    @Wim-De-Clercq
    Author

    Thanks for your swift reply. (I changed my sample code to dbm in the meantime, thinking it would be easier to run in case you didnt run redis, but anyway doesn't matter)

    I did indeed look at the InvalidationStrategy, but didn't see a decent way to get the expiration time. I did not think to just pass the region to a custom constructor there 🤦

    Extra question with this solution though. Would there be a way to figure out the correct expiration time if it's overridden on the decorator?

    cache_region.configure(
        expiration_time=5,
        ...
    
    
    @cache_region.cache_on_arguments(expiration_time=1)
    ...
  3. zzzeek commented on Feb 26, 2025

    @zzzeek
    Member

    not at the moment hence not as flexible as it should be

  4. jvanasco commented on Sep 12, 2025

    @jvanasco
    Member

    IIRC, I've implemented this in the past with a custom ProxyBackend. Within the custom ProxyBackend, you can analyze the timestamp in the metadata payload and return NO_VALUE for expired timestamps. As far as dogpile.cache's logic is concerned, this is a cache miss. The expiry in the metadata should reflect the override.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions