UCP/WIREUP/IB: Fix error message when FLID is not available #10515
Conversation
src/uct/ib/base/ib_iface.c
Outdated
| return 0; | ||
| if (!iface->config.flid_enabled) { | ||
| flid_info_str = "disabled"; | ||
| } else if ((uct_ib_iface_gid_extract_flid(&iface->gid_info.gid)) == 0 || |
There was a problem hiding this comment.
| } else if ((uct_ib_iface_gid_extract_flid(&iface->gid_info.gid)) == 0 || | |
| } else if ((uct_ib_iface_gid_extract_flid(&iface->gid_info.gid) == 0) || |
yosefe
force-pushed
the
topic/ucp-wireup-ib-fix-error-message-when
branch
from
5873820 to
4b5f309
Compare
src/ucp/wireup/select.c
Outdated
| @@ -592,7 +592,7 @@ static UCS_F_NOINLINE ucs_status_t ucp_wireup_select_transport( | |||
| } | |||
|
|
|||
| is_reachable = 0; | |||
|
|
|||
| ucs_snprintf_safe(info_str, info_str_size, "unreachable"); | |||
There was a problem hiding this comment.
maybe use snprintf here for consistency then? In this function only snprintf is used
|
@yosefe do you think we still need to be able to print an info string from
|
src/ucp/wireup/select.c
Outdated
| @@ -592,7 +592,7 @@ static UCS_F_NOINLINE ucs_status_t ucp_wireup_select_transport( | |||
| } | |||
|
|
|||
| is_reachable = 0; | |||
|
|
|||
| ucs_snprintf_safe(info_str, info_str_size, "unreachable"); | |||
There was a problem hiding this comment.
AFAIR we currently enable info_str to be NULL and its value is checked inside uct_iface_fill_info_str_buf.
I actually don't remember why we didn't pass an info str buffer to all ucp_wireup_add_X_lanes functions, but maybe now it can be removed (as I mentioned in the general comment) and info_str can be defined inside ucp_wireup_select_transport.. wdyt?
There was a problem hiding this comment.
removed info_str from parameters and moved it inside select_transport
Also fix RC info message for a corrupted address
Removed it from select.c:2513
We still need to reset uct_info string to "not available" in UCP before going over remote addresses, since ucp_wireup_is_reachable() can return 0 also if tl name is different - without calling uct_iface_is_reachable_v2. Added a gtest that UCT is_reachable always fills the info string |
Before:
[1740477778.122527] [swx-ucx-ibr-03:1950035:0] select.c:641 UCX ERROR no active messages transport to <no debug data>: self/memory - Destination is unreachable, rc_verbs/mlx5_0:1 - Destination is unreachableAfter:
[1740487396.304113] [swx-ucx-ibr-03:1951997:0] select.c:642 UCX ERROR no active messages transport to swx-ucx-ibr-01:3618365: self/memory - unreachable, rc_verbs/mlx5_0:1 - different subnet prefix 0xfec0000000000002/0xfec0000000000001 and FLID is not available