There are many breaking changes bundled up in v3, and so this guide serves as a quick overview of what you'll probably need to check out during adoption. This won't focus on every change, just the user facing ones that can affect how your application code interacts with a cache. It also doesn't serve as a guide on how to change things, just pointing out on which areas the new documentation should be consulted on.
In order to reduce a lot of the option parsing involved in Cachex, the options given to Cachex.start_link/2
and Cachex.start/2
have changed pretty drastically. The easiest way to see the differences is to look again at the documentation for these functions, but here's a very high level summary of changes (if these are things you use, please do check the docs):
- The
:commands
option now expects a list ofcommand
records. - The
:default_ttl
,:disable_ode
and:ttl_interval
are now passed as anexpiration
record in the:expiration
option (and have had their names changed). - The
:ets_opts
option has been completely removed. - The
:fallback
option now expects afallback
record. - The
:hooks
option now expects a list ofhook
records. - The
:limit
option now expect alimit
record rather than a%Cachex.Limit{}
(the shorthand of an integer is still valid). - The
:record_stats
option has had the name changed simply to:stats
. - The
:transactions
option has also been renamed to:transactional
.
All of these changes are based around improvements to the internal cache states and should also make it easier to understand the structures being passed around, whereas previously there were a lot of loose Keyword definitions. Please see either the module documentation, or guides for the feature you're using, for examples on how to use the new options.
Fallback caching has changed quite significantly in v3; there is no longer a :fallback
option on calls to get/3
, or any other calls which previously supported it. There are a few reasons for this, but the general one being that it was not always intuitive which calls did in fact support fallbacks. People consistently requested a feature that already existed, which means that it needed to be made more obvious. Due to this there will be a new fetch/4
function in the main interface which replaces the behaviour. Going forward, the root term "fetch" will be used as synonymous to "fallback".
Here is an example of the previous v2.x branch vs. the same behaviour in the v3.x branch:
# v2.x using the `:fallback` option to `get/3`.
Cachex.get(:my_cache, "key", fallback: &String.reverse/1)
# v3.x using the `fetch/4` signature.
Cachex.fetch(:my_cache, "key", &String.reverse/1)
The signature change allows for an explicit function in the API dedicated to lazy evaluation, and should hopefully be more approachable to those new to the project. It should be noted that the function being passed to fetch/4
is optional if you have set a default fallback function in the main cache options.
Hooks have changed pretty drastically, and yet they should be pretty close to what already exists. The main difference as of Cachex v3 is that Hooks are now driven by the behaviour a lot more than previously. When registering a hook on a cache, you now provide a hook
record rather than a struct. These records consist of purely the hook module, the hook state, and an optional name to use to register the hook with. Everything else is now driven by behaviour functions in the module registered. This decision was taken as hooks remain fairly constant for a specific job, and so moving into the module definition makes a lot of sense.
Rather than define the changes here, please see the documentation for Hooks going forward. Generally the options from the old %Cachex.Hook{}
struct have moved to have analogous functions in the module behaviour, but please check to be certain.
Nothing much to say here other than the previous :worker
provision has been renamed over to :cache
as the notion of cache workers is now redundant (and has been for a long time at this point). You can still use the received provision in the same way, it's just a tag change to make it clearer as to what is being delivered.
Cache streams no longer accept the :of
option, as it was oversimplifying exactly what was happening (and thus prone to error). However, you can now pass an ETS match specification as the second argument to filter internally in ETS before entering the stream. This is a small optimization, but also makes the developer think about matching more, rather than not appreciating exactly what was happening.
A happy side effect of this is that a stream will now respect the expiration time of records (at stream creation time), whereas before you could still receive expired records in the stream output (and would, regularly).
The Janitor is enabled by default as of Cachex v3. It seems that explicitly turning it off is preferable to explicitly turning it on for the developer experience, as you tend to assume it's just running by default - and then you panic when nothing is being removed. Minor change, but technically incompatible so worth mentioning.
Previously the :amount
option dictated how much the value should be incremented/decremented by, but at this point it's an extra argument (the third parameter), which will default to 1
.
This is a minor change, but worth mentioning. Going forward set/4
has been replaced with put/4
. This is nothing more than a name change, as @fishcakez rightfully pointed out that put/4
is a better naming convention for Elixir. The old set/4
has been deprecated and simply forwards to put/4
, so you should likely migrate to avoid that extra function hop :).
The format of the map being returned from Cachex.stats/2
has been modified due to some normalization which took place; this will look much clearer and adopts snake_case over camelCase (etc). It also correctly tracks custom invocations at this point, rather than ignoring them (like it did previously).
In earlier versions of Cachex, { :missing, nil }
would be returned to signal that a value did not exist in the cache. This has been removed to simply return { :ok, nil }
because (believe it or not) the overhead of figuring out if something was missing was actually quite large in some cases. If you need the same behaviour, you should avoid setting nil
explicitly in your cache and put something else in instead - that way { :ok, nil }
is semantically the same as { :missing, nil }
.