One caching API, Multiple backends

Instead of interacting via empty interface, what are your thoughts on using []byte? The user would take on the onus of [de]serialization and the skip the guessing game around type

Cool project. I'll consider using it

https://github.com/adelowo/onecache/blob/ae397c7322a160a3f0e9ffba976417260bd732ec/filesystem/fs.go#L117

I was going to submit a pull request but could not figure out exactly how to fix it given some issues I was running into getting it set up on GoLand so I could both edit and debug it (hard to edit and debug an external library in my IDE w/o renaming all the import paths.)

Anyway, without this is makes error handling harder than it needs to be.

You should be using lock.Lock() for writes (Set, Delete, etc.) and lock.RLock() for reads (Get)

See https://github.com/adelowo/onecache/blob/master/memory/memory.go#L24 and https://github.com/adelowo/onecache/blob/master/memory/memory.go#L41

Otherwise use a regular sync.Mutex()

It will be good if you will delete expired items in background. If client will add all items with ExpiresAt and read only several items you will have storage that will grow infinitely.

In get call we are getting panic error randomly, here is the stack trace

goroutine 1549 [running]: runtime.throw(0xc451b2, 0x21) /usr/local/go/src/runtime/panic.go:617 +0x72 fp=0xc000b16600 sp=0xc000b165d0 pc=0x42bd92 runtime.mapaccess2_faststr(0xb51c20, 0xc00017f6b0, 0xc0009bc460, 0x49, 0xc00017f688, 0xc0001c8700) /usr/local/go/src/runtime/map_faststr.go:116 +0x4a9 fp=0xc000b16670 sp=0xc000b16600 pc=0x412589 github.com/adelowo/onecache/memory.(*InMemoryStore).Delete(0xc00017f680, 0xc00029a400, 0x40, 0x49, 0xc0001c87d8) /jenkins/workspace/Vega_1.0.0_Nightly_Build/pkg/mod/github.com/adelowo/[email protected]/memory/memory.go:84 +0x70 fp=0xc000b166b0 sp=0xc000b16670 pc=0x59f330 github.com/adelowo/onecache/memory.(*InMemoryStore).Get(0xc00017f680, 0xc00029a400, 0x40, 0x10, 0x10, 0xb22820, 0xb, 0xc000b16768) /jenkins/workspace/Vega_1.0.0_Nightly_Build/pkg/mod/github.com/adelowo/[email protected]/memory/memory.go:73 +0x16c fp=0xc000b16708 sp=0xc000b166b0 pc=0x59f20c

Added quote in below code snipet where its failing

func (i *InMemoryStore) Get(key string) ([]byte, error) { i.lock.RLock()

item := i.data[i.keyfn(key)]
if item == nil {
	i.lock.RUnlock()
	return nil, onecache.ErrCacheMiss
}

if item.IsExpired() {
	i.lock.RUnlock()
	**

i.Delete(key)

** return nil, onecache.ErrCacheMiss }

i.lock.RUnlock()
return copyData(item.Data), nil

}

This is needed since in some places the user cannot configure a few things e.g cache key ( #19 ). we default to an internal implementation based on the adapter.

See #20

In the memory store) -- The items are accessed directly in a test suite without making use of the lock, hence there's a race condition on running tests with -race

There is also something similar in the fs store

Previously stored the serialized data for each key so it needed to be deserialized to check if it had expired. ie. GC needed to deserialize the whole corpus to do its job.

This change updates the code to store *onecache.Item. The ExpiresAt attribute can then be used directly to check expires.

During Set calls the data is copied before save and during Get calls the data is copied before return. This allows the caller to modify their data without issue, as they were allowed to before.

https://github.com/adelowo/onecache/blob/master/utils.go#L31-L34

If the type is a []byte, the serializer should return it as is rather than try encoding it with encoding/gob. I have no idea if encoding the bytes would cause an issue but it seems like a nice addition to have in the codebase

There are a few changes I would like to add to the package.

• [x] Introduction of options so we can have something like NewWithOpts ( This would fix #19 anyways )

• [x] Removal of Extend and Get ( Too much of a breaking change ??? )

• ~~[ ] Allow for multiple servers for the redis adapter . ( Nice to have )~~