DataLoader implements a counter which can be used against Field Resolver functions. It calls a batch function after the number of calls to Load values reaches the loaders capacity.
The following terms are used throughout the documentation:
- Element - Refers to the item to be resolved. This can be a record from a database table or GraphQL type.
Use the DataLoader when fetching a known number of elements, where each element has a field resolver associated with it that hits a database or has some other time consuming operation to resolve the data. This is typically useful when making index type queries where n number of root elements is requested and each root element has an associated model to be fetched.
For example, for a users
table which contains:
- first_name
- last_name
- status_id (foreign key - status table)
and a status
table which contains:
- id
- value (string)
Performing a query like:
{
users(num: 10) {
first_name
status {
value
}
}
}
where the users
resolver returns an array of users (10 in this case). This
will typically result in 1 call to return all 10 users, and 10 calls to resolve
the status
field for each user.
Use the DataLoader by setting its capacity to n (10 in this case) and
providing a batch loader function which accepts the keys and should return
n number of status
records. The result of which is a single call to
the database to return the status
elements after number of calls to Load()
has hit the set capacity.
Note that the capacity also acts as a floor. In instances where at least n calls are known, all n+1 calls are executed depending on the strategy used.
Internally, the DataLoader waits for the Load()
function to be called n times,
where n is the initial DataLoader capacity. The Load()
function returns a
Thunk, or ThunkMany function, which block when called until the number of calls
to Load equal the loaders capacity. Each call to the returned Thunk function
then returns the values for the keys it is attached to.
DataLoader is the basic interface for the library. It contains a provided strategy and cache strategy.
NewDataLoader(int, BatchFunction, func(int, BatchFunction) Strategy, ...Option) DataLoader
NewDataLoader returns a new instance of a DataLoader tracking to the capacity
provided and using the provided execution and cache strategy. The second argument
should return a strategy which accepts a capacity value and the BatchFunction
Load(context.Context, Key) Thunk
Returns a Thunk for the specified keys. Internally Load adds the
provided keys to the keys array and returns a callback function which when
called returns the values for the provided keys. Load does not block callers.
LoadMany(context.Context, ...Key) ThunkMany
Returns a ThunkMany for the specified keys. Internally LoadMany adds the
provided keys to the keys array and returns a callback function which when
called returns the values for the provided keys. LoadMany does not block
callers.
The options include:
WithCache(Cache) Option
WithCache sets the provided cache strategy on the loader
WithTracer(Cache) Option
WithTracer sets the provided tracer on the loader
Strategy is a interface to be used by implementors to hold and track data.
Load(context.Context, Key) Thunk
Load should return the Thunk function linked to the provided key. Load should
not reference a cache nor should it block.
LoadMany(context.Context, ...Key) ThunkMany
LoadMany should return a ThunkMany function linked to the provided keys.
LoadMany should not reference a cache nor should it block.
LoadNoOp(context.Context) ResultMap
LoadNoOp should not block the caller nor return values to the caller. It is
called when a value is retrieved from the cache and it's responsibility is to
increment the internal loads counter.
The sozu strategy batches all calls to the batch function, including n+1 calls. Since the strategy returns a Thunk or ThunkMany, calling Load or LoadMany before performing another long running process will allow the batch function to run concurrently to any other operations.
NewSozuStrategy(Options) func(int, BatchFunction) Strategy
NewSozuStrategy returns a function which returns a new instance of a sozu
strategy for the provided capacity.
The Options values include:
WithTimeout(time.Duration) Option
WithTimeout sets the configured timeout on the strategy. Default to 16 milliseconds
The standard strategy batches the first calls to the batch function, all subsequent callers call the batch function directly. Since the strategy returns a Thunk or ThunkMany, calling Load or LoadMany before performing another long running process will allow the batch function to run concurrently to any other operations.
NewStandardStrategy(Options) func(int, BatchFunction) Strategy
NewStandardStrategy returns a function which returns a new instance of the
standard strategy for the provided capacity.
The Options include:
WithTimeout(time.Duration) Option
WithTimeout sets the configured timeout on the strategy. Default to 16 milliseconds
The once strategy does not track calls to load and is useful for single calls to the batch function. Passing
InBackground: true
to the options allows the batch function to be called in the background, otherwise the batch function is called when Thunk, or ThunkMany, is called. It is useful for keeping a consistent API across an application and for automatically performing data queries in a background go routine.
NewOnceStrategy(Options) func(int, BatchFunction) Strategy
NewOnceStrategy returns a functions which returns an instance of the once
strategy ignoring the provided capacity value.
The Options include:
WithInBackground() Option
WithInBackground enables the batch function to execute in background on calls to
Load/LoadMany
ResultMap acts as a wrapper around a basic
map[string]Result
and provides accessor functions that tell the callers about the map or its state. ResultMap is not go routine safe.When creating a new instance of the result map, a capacity must be provided.
NewResultMap(int) ResultMap
NewResultMap returns a new instance of a result map with the set capacity
Set(string, Result)
Set sets the result in the result map
GetValue(Key) Result
GetValue returns the stored value for the provided key.
GetValueForString(String) Result
GetValue returns the stored value for the provided string value.
Length() int
Length returns the number of results it contains.
Keys() []string
Keys returns the keys used to identify the data within this result map.
Key is an interface each element's identifier must implement. Each Key must be unique (therefore a good candidate is the primary key for a database entry).
String() string
String should return a string representation of the unique identifier. The
resulting string is used in the ResultMap
to map the element to it's
identifier.
Raw() interface{}
Raw should return the underlying value of the key. Examples are: int
, string
.
Keys wraps an array of keys and provides a way of tracking keys to be resolved by the batch function. It also provides methods to tell its state. Keys is not go routine safe.
NewKeys(int) Keys
NewKeys returns a new key store with it's length set to the capacity. If the
capacity is known to be exact provide the exact value. Otherwise adding a buffer
can be useful to prevent unnecessary memory growth.
NewKeysWith(key ...Key) Keys
NewKeysWith returns a Keys array with the provided keys.
Append(...Key)
Append adds one or more keys to the internal array.
Capacity() int
Capacity returns the set capacity for the key array.
Length() int
Length returns the number of keys in the array.
ClearAll()
ClearAll deletes all the stored keys from the key array.
Keys() []interface{}
Keys returns a unique array of interface{} types for each key after calling each
keys Raw()
method
IsEmpty() bool
IsEmpty returns true if there are no keys in the keys array.
Cache provides an interface for caching strategies. The library provides a no-op caching strategy which does not store any values.
NewNoOpCache() Cache
NewNoOpCache returns an instance of a no operation cache. Internally it doesn't
store any values and all it's getter methods return nil or false.
SetResult(context.Context, Key, Result)
SetResult adds a value to the cache. The cache should store the value based on
it's implementation.
SetResultMap(context.Context, ResultMap)
SetResultMap saves all the elements in the provided ResultMap
GetResult(context.Context, Key) Result
GetResult should return the result matching the key or nil if none are found.
GetResultMap(context.Context, ...Key) ResultMap
GetResultMap returns a result map which contains the values for only the
provided keys.
Delete(context.Context, Key) bool
Delete removes the value for the provided key and returns true if successful.
ClearAll(context.Context) bool
ClearAll removes all values from the cache and returns true if successfully
cleared
Tracer provides an interface used by the DataLoader for tracing requests through the application. Tracing occurs on calls to
Load
,LoadMany
and when theBatchFunction
gets called.
NewNoOpTracer() Tracer
NewNoOpTracer returns an instance of a blank tracer that does not output anything.
NewOpenTracingTracer() Tracer
NewOpenTracingTracer returns an instance of a tracer which conforms to the open
tracing standard.
Load(context.Context, Key) (context.Context, LoadFinishFunc)
Load performs tracing around calls to the Load
function. It returns a context
with tracing information and a finish function which ends the tracing.
LoadMany(context.Context, Key) (context.Context, LoadManyFinishFunc)
LoadMany performs tracing around calls to the LoadMany
function. It returns a
context with tracing information and a finish function which ends the tracing.
Batch(context.Context) (context.Context, BatchFinishFunc)
Batch performs tracing around calls to the Batch
function. It returns a context
with the tracing information attached to it and a finish function which ends the
tracing.
LoadFinishFunc(Result)
LoadFinishFunc ends tracing started by Load
and gets passed the resolved
result for the queried key.
LoadManyFinishFunc(ResultMap)
LoadFinishFunc ends tracing started by LoadMany
and gets passed the resolved
result map for the queried keys.
BatchFinishFunc(ResultMap)
BatchFinishFunc ends tracing started by Batch
and gets passed the resolved
result map for the key or keys.
Counter provides an interface used to atomically count and track a value and to track if it's reached a certain set count. Internally it is used by strategies to track the number of calls to
Load
andLoadMany
.
NewCounter(int) Counter
NewCounter returns a new instance of a counter for the provided capacity. Once
the capacity is reached, the counter is considered complete.
Increment() bool
Increment increases the counter by one and returns true if the counter has hit
it's capacity.
ResetCounter()
ResetCounter sets the counter back to 0 but keeps the original capacity.
Both the Standard
and Sozu
strategies allow for concurrent operations before
executing the returned Thunk function and both ensure that other callers are not
blocked when waiting for data to be read. This allows certain resolvers that use
a shared strategy to block some of the time while allowing other resolvers to
receive their data and continue their operations.
The standard strategy initially calls the batch function when one of two conditions are met:
- The number of calls the Load or LoadMany equals the capacity of the loader
- The timeout set has been reached (default to 16 milliseconds)
Subsequent calls to Load()
or LoadMany()
will execute the batch function for
each caller.
The standard strategy is useful for instance when n+1 calls are lower than the capacity. For instance, it could be known that the app will need to resolve at least 10 keys, but if there are more it is significantly less than 10 more (e.g. 2 more calls for a total of 12).
The sozu strategy initially calls the batch function when one of two conditions are met:
- The number of calls the Load or LoadMany equals the capacity of the loader
- The timeout set has been reached (default to 16 milliseconds)
Subsequent calls to Load()
or LoadMany()
will call the batch function when
one of two conditions are met:
- The number of calls the Load or LoadMany equals the capacity of the loader
- The timeout set has been reached (default to 16 milliseconds)
The sozu strategy is useful for instance when known calls to load will be called in a defined and equal group. For instance, the initial call can be to resolve 10 keys. Then it should be known that if there are more keys, that there will be 10 more keys resulting in the batch function being called twice, each time with 10 keys.
The once strategy initially calls the batch function under one of two conditions:
- If
InBackground
option is false, it executes the batch function then Thunk is invoked - If
InBackground
option is true, it executes the batch function in a background go routine. Calls to the Thunk function will block until the go routine completes.
The once strategy is useful when wanting to execute a batch function a single time either directly, and keeping a consistent application api aligning with the other functions. Or when wanting to call the batch function in the background while performing some other time consuming operation and not wanting to handle the actual go routine.
- Set a max duration that a call to
Load(Key)
can block. Start from the initial call toLoad(Key)
. - Caching approach/interface
- Tests!!
- LoadMany - ability to call load with multiple keys
- Examples
- Logging
- Tracing