fabtests/prov/efa: Add remote exit early test #10810
Conversation
| printf("server posts recv\n"); | ||
| } | ||
|
|
||
| ft_start(); |
There was a problem hiding this comment.
why do you need ft_start() and ft_stop()?
There was a problem hiding this comment.
to track the time in between
jiaxiyan
force-pushed
the
remote_exit
branch
from
aeb07c1 to
b8c5596
Compare
| post_writedata = true; | ||
|
|
||
| hints->ep_attr->type = FI_EP_RDM; | ||
| hints->caps |= FI_MSG | FI_RMA; |
There was a problem hiding this comment.
This should be already handled by -o
There was a problem hiding this comment.
I don't see this being set in ft_parse_api_opts
| * either ft_exchange_keys() or ft_sync() | ||
| */ | ||
| if (post_writedata) { | ||
| ret = ft_exchange_keys(&remote); |
There was a problem hiding this comment.
Notice: ft_exchange_keys finally calls ft_sync(), which will call ft_sync_inband(true), so it will cause the receiver to always post an outstanding recv buffer. This is ok for RMA test because the rx buffer post shouldn't matter here, unless we will support FI_RX_CQ_DATA
jiaxiyan
force-pushed
the
remote_exit
branch
from
b8c5596 to
6adb74d
Compare
|
|
||
| @pytest.mark.functional | ||
| def test_remote_exit_early_post_send(cmdline_args): | ||
| for message_size in [65536, 131072, 1048576]: |
There was a problem hiding this comment.
nit: use a fixture as parameters
Add tests to make sure when sender exits early in the middle of send, receiver polls cq and handle errors gracefully. Add options for receiver posts receive, sender posts tagged send, and sender posts fi_writedata. Signed-off-by: Jessie Yang <jiaxiyan@amazon.com>
jiaxiyan
force-pushed
the
remote_exit
branch
from
6adb74d to
23da0fa
Compare
|
This should be pure EFA change and don't need to check Intel CI failure. |
Add tests to make sure when sender exits early in the middle of send, receiver polls cq and handle errors gracefully.
Add options for receiver posts receive, sender post tagged send, and sender posts fi_writedata.