Re: [PATCH v3 9/9] iotests: add nbd-reconnect-on-open test

[Eric Blake <eblake@redhat.com>]

On Mon, Sep 06, 2021 at 10:06:54PM +0300, Vladimir Sementsov-Ogievskiy wrote:
> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
> ---
>  .../qemu-iotests/tests/nbd-reconnect-on-open  | 71 +++++++++++++++++++
>  .../tests/nbd-reconnect-on-open.out           | 11 +++
>  2 files changed, 82 insertions(+)
>  create mode 100755 tests/qemu-iotests/tests/nbd-reconnect-on-open
>  create mode 100644 tests/qemu-iotests/tests/nbd-reconnect-on-open.out

I'm less confident in my review of the python code, but...

> 
> diff --git a/tests/qemu-iotests/tests/nbd-reconnect-on-open b/tests/qemu-iotests/tests/nbd-reconnect-on-open
> new file mode 100755
> index 0000000000..7ee9bce947
> --- /dev/null
> +++ b/tests/qemu-iotests/tests/nbd-reconnect-on-open
> @@ -0,0 +1,71 @@

> +
> +def create_args(open_timeout):
> +    return ['--image-opts', '-c', 'read 0 1M',
> +            f'driver=nbd,open-timeout={open_timeout},'
> +            f'server.type=unix,server.path={nbd_sock}']
> +
> +
> +def check_fail_to_connect(open_timeout):
> +    log(f'Check fail to connect with {open_timeout} seconds of timeout')
> +
> +    start_t = time.time()
> +    qemu_io_log(*create_args(open_timeout))
> +    delta_t = time.time() - start_t
> +
> +    max_delta = open_timeout + 0.2

Is this fractional delay going to bite us on heavily-loaded CI machines?

> +    if open_timeout <= delta_t <= max_delta:
> +        log(f'qemu_io finished in {open_timeout}..{max_delta} seconds, OK')
> +    else:
> +        note = 'too early' if delta_t < open_timeout else 'too long'
> +        log(f'qemu_io finished in {delta_t:.1f} seconds, {note}')
> +
> +
> +qemu_img_create('-f', iotests.imgfmt, disk, '1M')
> +
> +# Start NBD client when NBD server is not yet running. It should not fail, but
> +# wait for 5 seconds for the server to be available.
> +client = qemu_io_popen(*create_args(5))
> +
> +time.sleep(1)
> +qemu_nbd('-k', nbd_sock, '-f', iotests.imgfmt, disk)
> +
> +# client should succeed
> +log(client.communicate()[0], filters=[iotests.filter_qemu_io])
> +
> +# Server was started without --persistent flag, so it should be off now. Let's
> +# check it and it the same time check that with open-timeout=0 client fails

check it and at

> +# immediately.
> +check_fail_to_connect(0)
> +
> +# Check that we will fail after non-zero timeout if server is still unavailable
> +check_fail_to_connect(1)
> diff --git a/tests/qemu-iotests/tests/nbd-reconnect-on-open.out b/tests/qemu-iotests/tests/nbd-reconnect-on-open.out
> new file mode 100644
> index 0000000000..a35ae30ea4
> --- /dev/null
> +++ b/tests/qemu-iotests/tests/nbd-reconnect-on-open.out
> @@ -0,0 +1,11 @@
> +read 1048576/1048576 bytes at offset 0
> +1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
> +
> +Check fail to connect with 0 seconds of timeout
> +qemu-io: can't open: Failed to connect to 'TEST_DIR/PID-nbd-sock': No such file or directory
> +
> +qemu_io finished in 0..0.2 seconds, OK
> +Check fail to connect with 1 seconds of timeout
> +qemu-io: can't open: Failed to connect to 'TEST_DIR/PID-nbd-sock': No such file or directory
> +
> +qemu_io finished in 1..1.2 seconds, OK

Overall, the test looks like a nice demonstration of the feature: you
are showing that the client can now start before the server, and that
the retry for N seconds is handled gracefully both by the creation of
the server and by the expiration of the retry timeout.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org