fabtest/benchmarks: Process retcode with ft_exit_code for rma_pingpong#12161
Conversation
| if (inject_size_set) | ||
| inject_size = opts.inject_size; | ||
| if (inject_size_set && inject_size < opts.inject_size) { | ||
| FT_ERR("Provider does not support inject size %zu (max size %zu)", opts.inject_size, inject_size); |
There was a problem hiding this comment.
So if user call fi_rma_bw -o write -j 32, it will fail in the same way as fi_setopt(FI_OPT_INJECT_RMA_SIZE ... ) because such inject size cannot be satisfied anyway right? If so then the setopt call is not redundant... But I agree that this check is better than fi_setopt
There was a problem hiding this comment.
Turns out this setopt is needed to disable inject for test_fabtests_performance.
Will remove this commit.
| ret = ft_getinfo(hints, &fi); | ||
| if (ret) | ||
| return ret; | ||
|
|
||
| if (oob_sock >= 0 && opts.dst_addr) { | ||
| ret = ft_sock_sync(oob_sock, 0); | ||
| if (ret) | ||
| return ret; | ||
| } | ||
|
|
||
| ret = ft_getinfo(hints, &fi); | ||
| if (ret) | ||
| return ret; | ||
|
|
There was a problem hiding this comment.
This sync before fi_getinfo() is intentional. It makes sure that address resolution can work as expected for all providers. Some providers require the server be ready before the client can resolve the address via fi_getinfo(). See L1390 for when the server makes the sync call.
alekswn
changed the title
alekswn
force-pushed
the
wide-wqe-fabtest
branch
from
779fdc6 to
5d639fa
Compare
alekswn
force-pushed
the
wide-wqe-fabtest
branch
from
5d639fa to
e93e678
Compare
alekswn
changed the title
Signed-off-by: Alexey Novikov <nalexey@amazon.com>
alekswn
force-pushed
the
wide-wqe-fabtest
branch
from
e93e678 to
06dd3ad
Compare
|
bot:aws:retest |
|
@j-xiong Could you restart Intel CI? |
|
CI restarted. Still in the queue so it may take some time to get the status updated. |
3 participants
No description provided.