os/blestore/NVMEDevice: fix the I/O logic for read #13971
Conversation
optimistyzy
force-pushed
the
0315_1
branch
4 times, most recently
from
4d3d127
to
ed7b015
Compare
|
This patch is not ready for review, need to be tested. Thanks. |
liewegas
changed the title
|
I have test this pr based on commit f46b327 of master branch. root@node3:ceph(master)$ ceph -s |
|
@liupan1111 @liuhongtong @yuyuyu101 I think that this patch can solve the rand read crash related issue |
src/os/bluestore/NVMEDevice.cc
Outdated
| while (t->return_code > 0) | ||
| ioc.cond.wait(l); | ||
| while (t->return_code > 0) { | ||
| ioc.aio_wait(); |
There was a problem hiding this comment.
this isn't a safe check. we may enter the loop while return_code == 0, but before wait other thread change this and signal. at last we are stuck into wait infinitely!
There was a problem hiding this comment.
I re-visited the code, and think that it is a safe check and will not cause the issue you mentioned.
| std::unique_lock<std::mutex> l(ctx->lock); | ||
| ctx->cond.notify_all(); | ||
| // read submitted by AIO | ||
| if(!task->return_code) { |
There was a problem hiding this comment.
why could judge whether is a AIO read by return_code?
There was a problem hiding this comment.
From the code. For AIO read, we set the initialized value of return_code to 0.
optimistyzy
force-pushed
the
0315_1
branch
4 times, most recently
from
cdf40c7
to
dceb911
Compare
Aio_submit will submit both aio_read/write, and also there are synchronized read and random read, so we need to handle the read I/O completion in a correct way. Since random read has its own ioc, so the num_reading for ioc will be at most 1, which will be easy to handle in io_complete. And we need only to differentiate whethere it is an aio_read. Also fix the exception logic in command send, make the style consistent. Signed-off-by: optimistyzy <optimistyzy@gmail.com>
Aio_submit will submit both aio_read/write, and also there
are synchronized read and random read, so we need to
handle the read I/O completion in a correct way.
Also fix the exception logic in command send, make the style
consistent.
Signed-off-by: optimistyzy optimistyzy@gmail.com