[Tarantool-patches] [PATCH v2 2/3] fio: close unused descriptors automatically
Vladislav Shpilevoy
v.shpilevoy at tarantool.org
Fri Mar 20 03:00:36 MSK 2020
Indeed. It is cheaper and simpler to use cdata here.
On 19/03/2020 23:53, Igor Munkin wrote:
> Vlad,
>
> As you mentioned to me, the doc[1] specify 'userdata' (i.e. opaque type)
> as a return type for <fio.open>. Current implementation returns a table,
> so as we discussed there is nothing preventing you from dropping all
> hacks you used in the patch and implement file handle object via cdata
> or userdata.
See new patch below.
====================
fio: close unused descriptors automatically
Fio.open() returned a file descriptor, which was not closed
automatically after all its links were nullified. In other words,
GC didn't close the descriptor.
This was not really useful, because after fio.open() an exception
may appear, and user needed to workaround this to manually call
fio_object:close(). Also this was not consistent with io.open().
Now fio.open() object closes the descriptor automatically when
GCed.
Closes #4727
@TarantoolBot document
Title: fio descriptor is closed automatically by GC
fio.open() returns a descriptor which can be closed manually by
calling :close() method, or it will be closed automatically, when
it has no references, and GC deletes it.
:close() method existed always, auto GC was added just now.
Keep in mind, that the number of file descriptors is limited, and
they can end earlier than GC will be triggered to collect not
used descriptors. It is always better to close them manually as
soon as possible.
diff --git a/src/lua/fio.lua b/src/lua/fio.lua
index 4692e1026..d3c257b88 100644
--- a/src/lua/fio.lua
+++ b/src/lua/fio.lua
@@ -5,11 +5,16 @@ local ffi = require('ffi')
local buffer = require('buffer')
local fiber = require('fiber')
local errno = require('errno')
+local schedule_task = fiber._internal.schedule_task
ffi.cdef[[
int umask(int mask);
char *dirname(char *path);
int chdir(const char *path);
+
+ struct fio_handle {
+ int fh;
+ };
]]
local const_char_ptr_t = ffi.typeof('const char *')
@@ -160,7 +165,22 @@ fio_methods.stat = function(self)
return internal.fstat(self.fh)
end
-local fio_mt = { __index = fio_methods }
+fio_methods.__serialize = function(self)
+ return {fh = self.fh}
+end
+
+local fio_mt = {
+ __index = fio_methods,
+ __gc = function(obj)
+ if obj.fh >= 0 then
+ -- FFI GC can't yield. Internal.close() yields.
+ -- Collect the garbage later, in a worker fiber.
+ schedule_task(internal.close, obj.fh)
+ end
+ end,
+}
+
+ffi.metatype('struct fio_handle', fio_mt)
fio.open = function(path, flags, mode)
local iflag = 0
@@ -202,10 +222,13 @@ fio.open = function(path, flags, mode)
if err ~= nil then
return nil, err
end
-
- fh = { fh = fh }
- setmetatable(fh, fio_mt)
- return fh
+ local ok, res = pcall(ffi.new, 'struct fio_handle', fh)
+ if not ok then
+ internal.close(fh)
+ -- This is OOM.
+ return error(res)
+ end
+ return res
end
fio.pathjoin = function(...)
diff --git a/test/app/gh-4727-fio-gc.result b/test/app/gh-4727-fio-gc.result
new file mode 100644
index 000000000..d793eded3
--- /dev/null
+++ b/test/app/gh-4727-fio-gc.result
@@ -0,0 +1,104 @@
+-- test-run result file version 2
+test_run = require('test_run').new()
+ | ---
+ | ...
+fiber = require('fiber')
+ | ---
+ | ...
+fio = require('fio')
+ | ---
+ | ...
+--
+-- gh-4727: fio handler GC.
+--
+flags = {'O_CREAT', 'O_RDWR'}
+ | ---
+ | ...
+mode = {'S_IRWXU'}
+ | ---
+ | ...
+filename = 'test4727.txt'
+ | ---
+ | ...
+fh1 = nil
+ | ---
+ | ...
+fh2 = nil
+ | ---
+ | ...
+-- Idea of the test is that according to the Open Group standard,
+-- open() always returns the smallest available descriptor. This
+-- means, that in 'open() + close() + open()' the second open()
+-- should return the same value as the first call, if no other
+-- threads/fibers managed to interfere. Because of the
+-- interference the sequence may need to be called multiple times
+-- to catch a couple of equal descriptors.
+test_run:wait_cond(function() \
+ collectgarbage('collect') \
+ local f = fio.open(filename, flags, mode) \
+ fh1 = f.fh \
+ f = nil \
+ collectgarbage('collect') \
+-- GC function of a fio object works in a separate fiber. Give it \
+-- time to execute. \
+ fiber.yield() \
+ f = fio.open(filename, flags, mode) \
+ fh2 = f.fh \
+ f = nil \
+ collectgarbage('collect') \
+ fiber.yield() \
+ return fh1 == fh2 \
+end) or {fh1, fh2}
+ | ---
+ | - true
+ | ...
+
+-- Ensure, that GC does not break anything after explicit close.
+-- Idea of the test is the same as in the previous test, but now
+-- the second descriptor is used for something. If GC of the first
+-- fio object is called even after close(), it would close the
+-- same descriptor, already used by the second fio object. And it
+-- won't be able to write anything. Or will write, but to a
+-- totally different descriptor created by some other
+-- fiber/thread. This is why read() is called on the same file
+-- afterwards.
+f = nil
+ | ---
+ | ...
+test_run:wait_cond(function() \
+ f = fio.open(filename, flags, mode) \
+ fh1 = f.fh \
+ f:close() \
+ f = fio.open(filename, flags, mode) \
+ fh2 = f.fh \
+ return fh1 == fh2 \
+end)
+ | ---
+ | - true
+ | ...
+collectgarbage('collect')
+ | ---
+ | - 0
+ | ...
+fiber.yield()
+ | ---
+ | ...
+f:write('test')
+ | ---
+ | - true
+ | ...
+f:close()
+ | ---
+ | - true
+ | ...
+f = fio.open(filename, flags, mode)
+ | ---
+ | ...
+f:read()
+ | ---
+ | - test
+ | ...
+f:close()
+ | ---
+ | - true
+ | ...
diff --git a/test/app/gh-4727-fio-gc.test.lua b/test/app/gh-4727-fio-gc.test.lua
new file mode 100644
index 000000000..d0ab585ca
--- /dev/null
+++ b/test/app/gh-4727-fio-gc.test.lua
@@ -0,0 +1,60 @@
+test_run = require('test_run').new()
+fiber = require('fiber')
+fio = require('fio')
+--
+-- gh-4727: fio handler GC.
+--
+flags = {'O_CREAT', 'O_RDWR'}
+mode = {'S_IRWXU'}
+filename = 'test4727.txt'
+fh1 = nil
+fh2 = nil
+-- Idea of the test is that according to the Open Group standard,
+-- open() always returns the smallest available descriptor. This
+-- means, that in 'open() + close() + open()' the second open()
+-- should return the same value as the first call, if no other
+-- threads/fibers managed to interfere. Because of the
+-- interference the sequence may need to be called multiple times
+-- to catch a couple of equal descriptors.
+test_run:wait_cond(function() \
+ collectgarbage('collect') \
+ local f = fio.open(filename, flags, mode) \
+ fh1 = f.fh \
+ f = nil \
+ collectgarbage('collect') \
+-- GC function of a fio object works in a separate fiber. Give it \
+-- time to execute. \
+ fiber.yield() \
+ f = fio.open(filename, flags, mode) \
+ fh2 = f.fh \
+ f = nil \
+ collectgarbage('collect') \
+ fiber.yield() \
+ return fh1 == fh2 \
+end) or {fh1, fh2}
+
+-- Ensure, that GC does not break anything after explicit close.
+-- Idea of the test is the same as in the previous test, but now
+-- the second descriptor is used for something. If GC of the first
+-- fio object is called even after close(), it would close the
+-- same descriptor, already used by the second fio object. And it
+-- won't be able to write anything. Or will write, but to a
+-- totally different descriptor created by some other
+-- fiber/thread. This is why read() is called on the same file
+-- afterwards.
+f = nil
+test_run:wait_cond(function() \
+ f = fio.open(filename, flags, mode) \
+ fh1 = f.fh \
+ f:close() \
+ f = fio.open(filename, flags, mode) \
+ fh2 = f.fh \
+ return fh1 == fh2 \
+end)
+collectgarbage('collect')
+fiber.yield()
+f:write('test')
+f:close()
+f = fio.open(filename, flags, mode)
+f:read()
+f:close()
More information about the Tarantool-patches
mailing list