tests/sys/aio: Fix vectored_big_iovcnt flaky test failure

[olivier]

Summary

The vectored_big_iovcnt test was failing intermittently with "aio short
write (16384)".

The test creates an MD device with a fixed size of GLOBAL_MAX (16384
bytes) but calculates the buffer size as 512 * (max_buf_aio + 1). When
max_buf_aio exceeds 31, this results in a buffer larger than the MD
device capacity, causing the write operation to be truncated to the
device size and triggering a test failure

Test Plan

We had this quiet rare flaky behaviour with this test, and the error message was this one:
aio short write (16384)
So, once improved this error message, the new error message was:
aio short write: got 16384, expected: 131584 (max_buf_aio=256, iovcnt=257)

To reproduce easly:

$ sudo sysctl vfs.aio.max_buf_aio=256
vfs.aio.max_buf_aio: 16 -> 256
$sudo kyua test sys/aio/aio_test:vectored_big_iovcnt

But remember: There is nothing in the standard aio tests that is modifing the default value.

[asomers]

sys/aio/aio_test.c

848

This part should be unnecessary, since Kyua runs each test case in a fresh tempdir. Why did you find it necessary?

1834

Lowering max_buf_aio here will prevent overshooting the device capacity. But it will also prevent the test from exceeding the kernel's internal limit, defeating the point of the test. Instead of reducing max_buf_aio, it would be better to increase the md's device size.

[olivier]

sys/aio/aio_test.c

848

Because I was seeing this in my jenking error report:

Files left in work directory after failure: mdunit_link

But you’ve right, I need to remove it because it is useless since kyua remove all the tempdir at the end, and less code it better.

1834

OK, will propose a new version!

[olivier]

Increase the MD device size to support larger (non default) vfs.aio.max_buf_aio

[olivier]

Added new version more simpler: Just increase the MD size

[asomers]

The MD_LEN variable is now badly named. I think it's ok to leave its value the same. Most test cases don't require 1 MB of I/O. But you should at least change that name.

[olivier]

In D52914#1209913, @asomers wrote:

The MD_LEN variable is now badly named. I think it's ok to leave its value the same. Most test cases don't require 1 MB of I/O. But you should at least change that name.

So, let’s use DEVICE_IO_LEN then (to clarifies that this constant represents the I/O buffer length used for device tests (16 KB), not the size of the MD device itself which is now 1 MB) ?

[asomers]

In D52914#1209926, @olivier wrote:

In D52914#1209913, @asomers wrote:

The MD_LEN variable is now badly named. I think it's ok to leave its value the same. Most test cases don't require 1 MB of I/O. But you should at least change that name.

So, let’s use DEVICE_IO_LEN then (to clarifies that this constant represents the I/O buffer length used for device tests (16 KB), not the size of the MD device itself which is now 1 MB) ?

Ok, that should work.

[olivier]

Rename MD_LEN to DEVICE_IO_LEN