bugfix: refactor io to avoid output leak in execsync of cri

[YaoZengzeng]

Signed-off-by: YaoZengzeng [email protected]

Ⅰ. Describe what this PR did

Now pouch will close and remove all the IO when exec finished.

Actually it will lead to a tricky bug.

Because there are so many levels of buffer in the path of exec. So when exec finished, some output of the exec may still be in one of the middle buffer. And if IO is closed now, we won't be able to get them.

This is why some of the CRI CI test case failed sometimes.

So we should let IO close itself other than triggered by the signal that exec finished.

Ⅱ. Does this pull request fix one issue?

Ⅲ. Describe how you did it

The IO path as follows:

fifo -> ring buffer -> backend

Before this PR, when exec finished, it will close IO including close ring and backend, but there may be still some data left in the fifo or ring.

So what this PR do is : the close of ring buffer is triggered by the data copying between fifo and ring.

And when the exec finished , it will close IO which include waiting for the ring to flush the data to the backend and finally close the backend.

Ⅳ. Describe how to verify it

Ⅴ. Special notes for reviews

[pouchrobot]

ping @YaoZengzeng

CI fails according integration system.
Please refer to the CI failure Details button to corresponding test, and update your PR to pass CI.

If this is flaky test, welcome to track this with profiling an issue.

build url: https://travis-ci.org/alibaba/pouch/builds/383092229
build duration: 583s

[pouchrobot]

ping @YaoZengzeng
Conflict happens after merging a previous commit.
Please rebase the branch against master and push it back again. Thanks a lot.

[pouchrobot]

ping @YaoZengzeng
Conflict happens after merging a previous commit.
Please rebase the branch against master and push it back again. Thanks a lot.

[YaoZengzeng]

The message of conflict is annoying, close it!