View Source ExBuffer behaviour (ExBuffer v0.4.0)
An ExBuffer
is a process that maintains a collection of items and flushes
them once certain conditions have been met.
An ExBuffer
can flush based on a timeout, a maximum length (item count), a
maximum byte size, or a combination of the three. When multiple conditions are
used, the ExBuffer
will flush when the first condition is met.
ExBuffer
also includes a number of helpful tools for testing and debugging.
Summary
Functions
Lazily chunks an enumerable based on ExBuffer
flush conditions.
Lazily chunks an enumerable based on ExBuffer
flush conditions and raises an ArgumentError
with invalid options.
Dumps the contents of the given ExBuffer
to a list, bypassing a flush
callback and resetting the buffer.
Flushes the given ExBuffer
, regardless of whether or not the flush conditions
have been met.
Returns information about the given ExBuffer
.
Starts an ExBuffer
process linked to the current process.
Types
@type error() ::
:invalid_callback | :invalid_jitter | :invalid_limit | :invalid_partitioner
Errors returned by ExBuffer
functions.
Callbacks
Invoked to flush an ExBuffer
.
The first argument (data
) is a list of items inserted into the ExBuffer
and the
second argument (opts
) is a keyword list of flush options. See the :flush_callback
and :flush_meta
options for ExBuffer.start_link/2
for more information.
This callback can return any term as the return value is disregarded by the ExBuffer
.
This callback is required.
@callback handle_size(item :: term()) :: non_neg_integer()
Invoked to determine the size of an inserted item.
The only argument (item
) is any term that was inserted into the ExBuffer
.
This callback must return a non-negative integer representing the item's byte size.
This callback is optional. See the :size_callback
option for ExBuffer.start_link/2
for information about the default implementation.
Functions
@spec chunk( Enumerable.t(), keyword() ) :: {:ok, Enumerable.t()} | {:error, error()}
Lazily chunks an enumerable based on ExBuffer
flush conditions.
This function currently supports length and size conditions. If multiple
conditions are specified, a chunk is emitted once the first condition is
met (just like an ExBuffer
process).
While this function is useful in it's own right, it's included primarily as
another way to synchronously test applications that use ExBuffer
.
Options
An enumerable can be chunked with the following options:
:max_length
- A non-negative integer representing the maximum allowed length (item count) of a chunk. By default, there is no max length. (Optional):max_size
- A non-negative integer representing the maximum allowed size (in bytes) of a chunk. The:size_callback
option determines how item size is computed. By default, there is no max size. (Optional):size_callback
- The function that will be invoked to deterime the size of an item. For more information, seeExBuffer.start_link/2
. (Optional)
Warning
Including neither
:max_length
nor:max_size
is permitted but will result in a single chunk being emitted. One can achieve a similar result in a more performant way usingStream.into/2
. In that same vein, including only a:max_length
condition makes this function a less performant version ofStream.chunk_every/2
. This function is optimized for chunking by either size or size and count. Any other chunking strategy can likely be achieved in a more efficient way using other methods.
Examples
iex> enum = ["foo", "bar", "baz", "foobar", "barbaz", "foobarbaz"]
...> {:ok, enum} = ExBuffer.chunk(enum, max_length: 3, max_size: 10)
...> Enum.into(enum, [])
[["foo", "bar", "baz"], ["foobar", "barbaz"], ["foobarbaz"]]
iex> enum = ["foo", "bar", "baz"]
...> {:ok, enum} = ExBuffer.chunk(enum, max_size: 8, size_callback: &(byte_size(&1) + 1))
...> Enum.into(enum, [])
[["foo", "bar"], ["baz"]]
iex> ExBuffer.chunk(["foo", "bar", "baz"], max_length: -5)
{:error, :invalid_limit}
@spec chunk!( Enumerable.t(), keyword() ) :: Enumerable.t()
Lazily chunks an enumerable based on ExBuffer
flush conditions and raises an ArgumentError
with invalid options.
For more information on this function's usage, purpose, and options, see ExBuffer.chunk!/2
.
Examples
iex> ["foo", "bar", "baz", "foobar", "barbaz", "foobarbaz"]
...> |> ExBuffer.chunk!(max_length: 3, max_size: 10)
...> |> Enum.into([])
[["foo", "bar", "baz"], ["foobar", "barbaz"], ["foobarbaz"]]
iex> ["foo", "bar", "baz"]
...> |> ExBuffer.chunk!(max_size: 8, size_callback: &(byte_size(&1) + 1))
...> |> Enum.into([])
[["foo", "bar"], ["baz"]]
iex> ExBuffer.chunk!(["foo", "bar", "baz"], max_length: -5)
** (ArgumentError) invalid limit
@spec dump( PartitionSupervisor.name(), keyword() ) :: {:ok, list()} | {:error, :invalid_partition | :not_found}
Dumps the contents of the given ExBuffer
to a list, bypassing a flush
callback and resetting the buffer.
While this functionality may occasionally be desriable in a production environment, it is intended to be used primarily for testing and debugging.
Options
An ExBuffer
can be dumped with the following options:
:partition
- A non-negative integer representing the specific partition index to dump. By default, this function dumps all partitions and concatenates the results together. (Optional)
Examples
iex> ExBuffer.insert(ExBuffer, "foo")
iex> ExBuffer.insert(ExBuffer, "bar")
iex> ExBuffer.dump(ExBuffer)
{:ok, ["foo", "bar"]}
iex> ExBuffer.insert(ExBuffer, "foo")
iex> ExBuffer.insert(ExBuffer, "bar")
iex> ExBuffer.dump(ExBuffer, partition: 0)
{:ok, ["foo"]}
@spec flush( GenServer.server(), keyword() ) :: :ok | {:error, :invalid_partition | :not_found}
Flushes the given ExBuffer
, regardless of whether or not the flush conditions
have been met.
While this functionality may occasionally be desriable in a production environment, it is intended to be used primarily for testing and debugging.
Options
An ExBuffer
can be flushed with the following options:
:async
- A boolean representing whether or not the flush will be asynchronous. By default, this value istrue
. (Optional):partition
- A non-negative integer representing the specific partition index to flush. By default, this function flushes all partitions. (Optional)
Example
iex> ExBuffer.insert(ExBuffer, "foo")
iex> ExBuffer.insert(ExBuffer, "bar")
...>
...> # Invokes flush callback on ["foo"] and then on ["bar"]
iex> ExBuffer.flush(ExBuffer)
:ok
iex> ExBuffer.insert(ExBuffer, "foo")
iex> ExBuffer.insert(ExBuffer, "bar")
...>
...> # Invokes flush callback on ["foo"]
iex> ExBuffer.flush(ExBuffer, partition: 0)
:ok
@spec info( GenServer.server(), keyword() ) :: {:ok, list()} | {:error, :invalid_partition | :not_found}
Returns information about the given ExBuffer
.
This function returns a map per partition with the following keys:
:length
- The number of items in theExBuffer
partition.:max_length
- The maximum length of theExBuffer
partition after applying the:jitter_rate
.:max_size
- the maximum byte-size of theExBuffer
partition after applying the:jitter_rate
.:next_flush
- The amount of time (in ms) until the next scheduled flush of theExBuffer
partition (ornil
if theExBuffer
was started without a time limit).:partition
- The index of theExBuffer
partition.:size
- The byte-size of theExBuffer
partition.:timeout
- The maximum amount of time (in ms) allowed between flushes of theExBuffer
partition after applying the:jitter_rate
.
While this functionality may occasionally be desriable in a production environment, it is intended to be used primarily for testing and debugging.
Options
The information about an ExBuffer
can be retrieved with the following options:
:partition
- A non-negative integer representing the specific partition index to retrieve information for. By default, this function retrieves information for all partitions. (Optional)
Examples
iex> ExBuffer.insert(ExBuffer, "foo")
iex> {:ok, [%{length: length}, %{}]} = ExBuffer.info(ExBuffer)
iex> length
1
iex> ExBuffer.insert(ExBuffer, "foo")
iex> {:ok, [%{length: length, partition: 0}]} = ExBuffer.info(ExBuffer, partition: 0)
iex> length
1
@spec insert(GenServer.server(), term()) :: :ok | {:error, :not_found}
Inserts the given item into the given ExBuffer
based on the partitioner that the
given ExBuffer
was started with.
Example
iex> ExBuffer.insert(ExBuffer, "foo")
:ok
@spec insert_batch(GenServer.server(), Enumerable.t(), keyword()) :: {:ok, non_neg_integer()} | {:error, :not_found}
Inserts the given batch of items into the given ExBuffer
based on the partitioner that
the given ExBuffer
was started with.
All items in the batch are inserted into the same partition.
Options
A batch of items can be inserted into an ExBuffer
with the following options:
:safe_flush
- A boolean denoting whether or not to flush "safely". By default, this value istrue
, meaning that, if a flush condition is met while inserting items, theExBuffer
partition will synchronously flush before continuing to insert items. If this value isfalse
, all items will be inserted before checking if any flush conditions have been met. Afterwards, if a flush condition has been met, theExBuffer
partition will be flushed asynchronously.
Example
iex> ExBuffer.insert_batch(ExBuffer, ["foo", "bar", "baz"])
{:ok, 3}
@spec start_link( module() | nil, keyword() ) :: Supervisor.on_start()
Starts an ExBuffer
process linked to the current process.
The first argument (module
) is optional. It is intended to be used when calling
this function from a module that implements the ExBuffer
behaviour. When a module
is passed, it may interact with the options that were passed in:
If the module implements the
handle_flush/2
callback, it will override the:flush_callback
option.If the module implements the
handle_size/1
callback, it will override the:size_callback
option.If a
:name
option is not present, the module name will be used.
Options
An ExBuffer
can be started with the following options:
:flush_callback
- The function that will be invoked to handle a flush. This function should expect two parameters: a list of items and a keyword list of flush opts. The flush opts include the size and length of the buffer at the time of the flush, the partition index of the flushed buffer, and any provided metadata (see:flush_meta
for more information). This function can return any term as the return value is not used by theExBuffer
. (Required)buffer_timeout
- A non-negative integer representing the maximum time (in ms) allowed between flushes of theExBuffer
. Once this amount of time has passed, theExBuffer
will be flushed. By default, anExBuffer
does not have a timeout. (Optional):flush_meta
- A term to be included in the flush opts under themeta
key. By default, this value will benil
. (Optional):jitter_rate
- A float between 0 and 1 that is used to offset the limits ofExBuffer
partitions. Limits are decreased by a random rate between 0 and this value. By default, no jitter is applied to anExBuffer
.:max_length
- A non-negative integer representing the maximum allowed length (item count) of theExBuffer
. Once the limit is hit, theExBuffer
will be flushed. By default, anExBuffer
does not have a max length. (Optional):max_size
- A non-negative integer representing the maximum allowed size (in bytes) of theExBuffer
. Once the limit is hit (or exceeded), theExBuffer
will be flushed. The:size_callback
option determines how item size is computed. By default, anExBuffer
does not have a max size. (Optional):name
- The registered name for theExBuffer
. This must be either an atom or a:via
tuple. By default (when an implementation module is not used), the name of anExBuffer
isExBuffer
. (Optional):partitioner
- The strategy for assigning items to a partition. The partitioner can be either:rotating
or:random
. The former assigns items to partitions in a round-robin fashion and the latter assigns items randomly. By default, anExBuffer
uses a:rotating
partition. (Optional):partitions
- The number of partitions for theExBuffer
. By default, anExBuffer
has 1 partition. (Optional):size_callback
- The function that will be invoked to determine the size of an item. This function should expect a single parameter representing an item and should return a single non-negative integer representing that item's byte size. The defaultExBuffer
size callback isKernel.byte_size/1
(:erlang.term_to_binary/1
is used to convert non-bitstring inputs to binary if necessary). (Optional)
Additionally, an ExBuffer can also be started with any GenServer
options.