On 26/10/2022 18.18, Alex Bennée wrote:
Daniel P. Berrangé <[email protected]> writes:
CC'ing Marc-André as original author of the change
On Tue, Oct 25, 2022 at 01:57:23PM +0100, Alex Bennée wrote:
Juan Quintela <[email protected]> writes:
Previous commit removed the creation of the fifo. Without it, I get
random failure during tests with high load, please consider
reintroduce it.
My guess is that there is a race between the two socats when we leave
them to create the channel, better return to the previous behavior.
I can't reproduce the problem when I run ./test-io-channel-command
test alone, I need to do the make check. And any (unrelated) change
can make it dissapear.
I was chasing a similar problem with this test although I don't see it
timeout while running (I don't think our unit tests time out). I'm
provisionally queuing this to testing/next unless anyone objects.
It won't build on Win32 since that platform lacks mkfifo.
The test normally works since socat will call mknod to create
the fifo.
I think the problem is that we have a race condition where the
client socat runs before the server socat, and so won't see the
fifo. This will be where high load triggers problems.
Ok I shall drop the patch from testing/next - we need a better solution.
Could we maybe at least revert the patch that introduced the problem? ...
the failing test is annoying ...
Thomas
