Disabling requests-cache is not thread safe. Discussion of the future of _restclient's usage of requests-cache about hydrotools HOT 6 CLOSED

FYI, a better option for using requests-cache is using CachedSession directly, instead of patching with install_cache(). It makes it more explicit what you are and aren't caching, and doesn't affect downstream requests calls. It's mostly thread-safe, except for cache_disabled(), as you noted; instead, when you want to make a non-cached request, you can just use a regular requests.Session.

Also, the library is being maintained again (with a fairly large release last week), and issues and PRs are welcome.

from hydrotools.

With reference to CacheControl, the default cache is a dict. They also have a FileCache that will persist requests using pickle and a RedisCache. I'm wondering about performance using the FileCache and portability using the RedisCache.

from hydrotools.

With reference to CacheControl, the default cache is a dict. They also have a FileCache that will persist requests using pickle and a RedisCache. I'm wondering about performance using the FileCache and portability using the RedisCache.

Thanks for doing a little digging... Having not done that so far myself, the issues that you mentioned are problematic and do not sound like they would support our needs. Specifically a FileCache does not support our needs nor an in-RAM solution like Redis. Additionally, I am not overly keen in bringing in another dependency just to fix this one issue. Plus with the status of requests-cache:

Noticed last commit was 14-Aug-2019. Which is almost 1.5 years ago as of today.

#14 may deem it necessary to develop our own cache sub-package. For reference, this is something @jarq6c and @aaraney have previously discussed, but never formally tabled. At face value it seems that we should be able to learn from the implementations of both CacheControl and requests-cache and sort of merge them. I think having something a little more general, as we've discussed prior, that permits caching things in general via HD5 as well as SQlite would be a requirement for such a feature.

With all of the being said, I don't want this issue to dissolve into another subject, but just speak generally and voice the potential feature addition to address this bug.

from hydrotools.

The disadvantages of a hypothetical evaluation_tools._cache package are that someone has to write and maintain it. We may also end up reinventing the wheel to some extent. On the other hand, perhaps there is a way to design it such that wheel reinventing is minimized and we're really just writing wrappers around existing libraries.

The potential advantages of such a package go beyond its primary use to cache NWIS and NWM data. We could unify caching and storing of data through a singular interface. This might allow data/computational scientists using evaluation_tools to speed up development of long workflows by storing interim data. This interface could be incorporated into an event_detection_service that caches events. If an object store (or cloud bucket) became available we could use this package to automatically push data from any other evaluation_tools subpackage.

I'm imagining a subpackage that takes evaluation_tools canonical pandas.DataFrames or geopandas.GeoDataFrames and automagically caches and/or stores the data in a variety of formats.

from hydrotools.

@jarq6c its ironic you mention an event cache, I may or may not have had the need to implement:

class EventCache(object):
    """
        Event cache manager
        FIXME current multiprocessing works because of a guarantee of process order,
        but this may need to be better protected from reader/writer issues
    """

I would encourage thinking about the needs of caching and what exactly is being cached. A dataframe cache with clear semantics isn't too far fetched.

from hydrotools.

Because of #78, multiprocessing usage should become more obsolete in the context of retrieving data. With that in mind, I am going to close this and we can reopen in the future to continue discussion.

from hydrotools.