Awesome
Erlang MMAP emmap
This Erlang library implements an ability to use memory map files in the memory of the Erlang virtual machine. It offers three sets of functions to implement:
- Generic read/write access to memory mapped files.
- Persistent atomic integer counters supporting basic arithmetic and logical operators.
This feature is an enhancement of the
Erlang's atomic counters, by adding more
atomic operations (e.g.
xchg
,cas
,and
,or
, andxor
) for the counters, as well as adding counter persistence. - Persistent FIFO queue.
- Persistent storage for fixed-size data blocks.
Authors
Supported Platforms
Linux, MacOS
NOTE: On MacOS emmap:resize/2
is not supported, it will return {error, fixed_size}
.
Basic Usage
The basic usage is
{ok, Mem, _Info} = emmap:open("filename", [read, shared, direct]),
{ok, Binary} = file:pread(Mem, 100, 40),
...
ok = file:close(Mem).
The open options is a list containing zero or more options.
From this point, Mem
can be used either with the file
or with the emmap
functions
interchangeably:
{ok, Binary} = file:pread(Mem, Position, Length)
read Length bytes at Position in the file.ok = file:pwrite(Mem, Position, Binary)
writes to the given position.{ok, Binary} = file:read(Mem, Length)
read 1..Length bytes from current position, or returneof
if pointer is at end of file.{ok, Pos} = file:position(Mem, Where)
see file:position/2 documentation.ok = file:close(Mem)
All read/write functions invoke NIFs that don't call any IO functions but rather access memory
via calls to memcpy(2)
, and persistence is achieved by relying on the OS implementation of
saving dirty memory pages to files.
A memory map can be closed either by calling emmap:close/1
or file:close/1
. When using
the direct
option, and emmap:close/1
is called, the memory map is not immediately closed,
but will get automatically closed when all binaries that reference this memory map are garbage
collected.
Atomic operations on the memory mapped file
The emmap
application offers a way to do atomic add
, sub
, xchg
, cas
as well as bitwise
and
, or
, xor
operations using emmap:patomic_*/3
and emmap:patomic_cas/4
functions.
Effectively this directly changes the content of the underlying memory, is thread-safe, and persistent.
{ok, OldValue} = emmap:patomic_add(Mem, Position, 1).
This approach allows to implement persistent atomic counters that survive node restarts.
Atomic persistent counters
The emmap
application allows a user to maintain atomic persistent counters. This could be
useful for continuous numbering of some events in the system which could be efficiently shared
among Erlang or OS processes in a thread-safe way and at the same time being persistent.
This is a very light-weight approach compared to using mnesia
or other form of persistent
storage.
Here is an example:
F = emmap:open_counters("/tmp/mem.bin", 2),
N1 = emmap:inc_counter(F, 0),
N2 = emmap:inc_counter(F, 0, 5),
N3 = emmap:inc_counter(F, 0),
N4 = emmap:inc_counter(F, 0, 2),
N5 = emmap:set_counter(F, 0, 15),
N6 = emmap:read_counter(F, 0),
emmap:close_counters(F),
io_format("N1=~w, N2=~w, N3=~w, N4=~w, N5=~w, N6=~w\n",
[N1, N2, N3, N4, N5, N6]). % Prints: N1=0, N2=1, N3=6, N4=7, N5=9, N6=15
Shared memory and using mutable binaries
While Erlang goes at length to achieve immutability, sometimes applications might need to have access to mutable memory. This can be accomplished by using the direct shared access to the memory mapped file.
Example:
shell1> {ok, MM, _Info} = emmap:open("/tmp/mem.data", 0, 8, [create, direct, read, write, shared, nolock]).
shell2> {ok, MM, _Info} = emmap:open("/tmp/mem.data", 0, 8, [create, direct, read, write, shared, nolock]).
shell1> emmap:pwrite(MM, 0, <<"test1">>).
shell2> {ok, Bin} = emmap:pread(MM, 0, 5).
{ok, <<"test1">>}
shell1> emmap:pwrite(MM, 0, <<"test2">>).
shell2> Bin.
<<"test2">>
shell1> emmap:pwrite(MM, 0, <<"test3">>).
shell2> Bin.
<<"test3">>
Though this may seem odd that a bound Bin
variable returns a different value when we printed
it in the shell2
the second time, it is the result of opening memory mapped file using the
direct
option. In this case the binaries read from memory map point to the actual memory
in that map rather than being copies of that memory. For some applications, such as when
using that memory to store atomic counters, this property can be very valuable.
Using the option direct
has the effect that the mmap file is not closed until all references
to binaries coming out of read/pread have been garbage collected. This is a consequence of
that such binaries are referring directly to the mmap'ed memory.
When passing auto_unlink
option to emmap:open/4
, the memory mapped file will be
automatically deleted when it is closed.
Shared memory without using mutable binaries
This example preserves the immutability of binaries but allows emmap
to have visibility
of memory changes between Erlang processes and also between OS processes.
$ erl -pa _build/default/lib/emmap/ebin
eshell#1> {ok, F, _} = emmap:open("/tmp/q.bin", 0, 128, [auto_unlink, shared, create, read,
write]).
$ erl -pa _build/default/lib/emmap/ebin
eshell#2> {ok, F, _} = emmap:open("/tmp/q.bin", 0, 128, [auto_unlink, shared, create, read,
write]).
eshell#1> emmap:pwrite(F, 0, <<"abcdefg\n">>).
eshell#2> emmap:pread(F, 0, 8). % Changes in eshell#1 are visible in eshell#2
{ok, <<"abcdefg\n">>}
$ head -1 /tmp/q.bin # They are also visible in another OS process reading from file
abcdefg
Here it is without the shared
option:
$ erl -pa _build/default/lib/emmap/ebin
eshell#1> emmap:close(F).
eshell#1> f(F), {ok, F, _} = emmap:open("/tmp/q.bin", 0, 128, [auto_unlink, create, read,
write]).
^G
--> s % Start a new shell process inside the same Erlang VM
--> c 2 % Connect to the new shell
eshell#2> f(F), {ok, F, _} = emmap:open("/tmp/q.bin", 0, 128, [auto_unlink, create, read,
write]).
^G
--> c 1 % Switch back to the 1st shell
eshell#1> emmap:pwrite(F, 0, <<"1234567\n">>).
^G
--> c 2 % Switch to the 2st shell
eshell#2> emmap:pread(F, 0, 8).
{ok,<<0,0,0,0,0,0,0,0>>} % changes from shell1 are invisible in the shell2 Erlang process
# Run this in another terminal
$ head -1 /tmp/q.bin # returns no data because changes in shell1 are invisible
Persistent FIFO used as a container or guarded by a gen_server process
The emmap_queue
module implements a persistent FIFO queue based on a memory-mapped file.
This means that in-memory operations of enqueuing items are automatically persisted on disk.
A queue used as a container will persistent messages stored in queue on disk, and has constant
time complexity of the push and pop operations. The open/3
is given an initial storage in
bytes, which will automatically grow unless the fixed_size
option is provided, in which case
when the queue becomes full, a push/2
call will return {error, full}
. In the example below
we are using auto_unlink
option which automatically deletes the memory mapped file at the end
of the test case (something you might not want in other cases):
{ok, Q} = emmap_queue:open(Filename, 1024, [auto_unlink]),
ok = emmap_queue:push(Q, a),
ok = emmap_queue:push(Q, {b,1}),
ok = emmap_queue:push(Q, {c,d}),
a = emmap_queue:pop(Q),
{b,1} = emmap_queue:pop(Q),
{c,d} = emmap_queue:pop(Q),
nil = emmap_queue:pop_and_purge(Q).
Use emmap_queue:pop_and_purge/1
to reclaim the space in memory when the queue becomes empty.
When a queue is wrapped in a gen_server
, it is suitable for use in a multi-process use cases.
This is implemented using emmap_queue:start_link/4
,emmap_queue:enqueue/2
, and
emmap_queue:dequeue/1
functions. In the example below we are using the auto_unlink
option
which automatically deletes the memory mapped file at the end of the test case (something you
might not want in other cases):
{ok, Pid} = emmap_queue:start_link(?MODULE, Filename, 1024, [auto_unlink]),
ok = emmap_queue:enqueue(Pid, a),
ok = emmap_queue:enqueue(Pid, {b,1}),
ok = emmap_queue:enqueue(Pid, {c,d}),
a = emmap_queue:dequeue(Pid),
{b,1} = emmap_queue:dequeue(Pid),
{c,d} = emmap_queue:dequeue(Pid),
nil = emmap_queue:dequeue(Pid).
Persistent storage for fixed-size data blocks
The purpose is to store, read, and remove arbitrary data blocks of a fixed size. Each block of data stored has a unique integer address (internally translated into an offset from the beginning of the memory-mapped file). Persistent storage tries to reuse a free block closest to the file start or allocate a new block when needed. The file may be automatically resized (expanded) when required.
To start using a memory-mapped file as a storage call emmap:init_block_storage/2
providing emmap
handler and block size (it will be saved in the storage header).
The new flag fit
added to emmap:open/4 option list. When set and the existing file opened has a size
less than requested region length, the file will be stretched to the given length. If the file size is
greater than requested, with fit flag the mapped region will fit the file size. Without the fit
flag
attempt to map existing file of a different size will result in error.
% open underlying memory-mapped file
{ok, MFile, Info} = emmap:open("storage.bin", 0, 4096, [create, write, fit, shared]),
% init block storage of the fixed block size
ok = emmap:init_block_storage(MFile, 22),
When opening an existing file, it may be possible that it was left in an inconsistent state in
case of abnormal termination of the program modifying it. To ensure consistency, call
emmap:repair_block_storage/1
to check and fix the file at once, or repeatedly call
emmap:repair_block_storage/3
. The latter version with continuation is recommended for relatively
big storages, to avoid long-running NIF calls. The repair operation checks (and fixes)
inconsistency between "free blocks" and "used blocks" masks in the internal tree-like representation.
repair_chunks(MFile, N) ->
repair_chunks(MFile, 0, N).
repair_chunks(_MFile, eof, _) ->
ok;
repair_chunks(MFile, Start, N) ->
Cont = emmap:repair_block_storage(MFile, Start, N),
repair_chunks(MFile, Cont, N).
To read all blocks, use emmap:read_blocks/1
, or emmap:read_blocks/3
for reads with continuation,
limiting the number of blocks read in one shot, to avoid long-running NIF calls.
List = emmap:read_blocks(MFile),
or
read_chunks(MFile, N) ->
read_chunks(MFile, 0, N, []).
read_chunks(_MFile, eof, _, Acc) ->
lists:concat(Acc);
read_chunks(MFile, Start, N, Acc) ->
{L, Cont} = emmap:read_blocks(MFile, Start, N),
read_chunks(MFile, Cont, N, [L | Acc]).
The function emmap:read_block/2
reads the block at the given address, emmap:store_block/2
writes the data block into the storage, and emmap:free_block/2
deletes the block at the given
address.
Addr = emmap:store_block(MFile, Data),
Bytes = emmap:read_block(MFile, Addr),
emmap:free_block(MFile, Addr),
The storage capacity is limited by internal organization. It depends on the number of tree levels, and can be
configured by defining the environment variable BS_LEVELS
. The maximum number of stored blocks is 64 ^ BS_LEVELS
.
The default BS_LEVELS
value is 3, so the default capacity is 64 * 64 * 64 = 262144
blocks.
An attempt to store a block will return {error, full}
when the storage has no free slots.
The result of freeing a block is true
on success, false
if there is no block with the given address
found, or {error, Reason}
for common emmap error cases.
The read_block/2
returns bytes, eof
when no block exists at the given address or common error.