lidavidm commented on pull request #12442:
URL: https://github.com/apache/arrow/pull/12442#issuecomment-1081956258
And the client socket is created when the worker is created - at which point
we don't know whether we're using IPv4 or IPv6 yet. It seems we use
`UCX_TCP_AF_PRIO` to choose between these. Setting `UCX_TCP_AF_PRIO=inet6`
prevents the segfault, but we still get a different error:
```
[ RUN ] UcxDataTest.TestDoGetInts
[1648564705.771134] [partita:19997:0] ucp_context.c:1779 UCX INFO UCP
version is 1.12 (release 0)
[1648564705.795517] [partita:19997:0] ucp_worker.c:1867 UCX INFO
ep_cfg[0]: rma(cuda_copy/cuda); am(cuda_copy/cuda);
[1648564705.795792] [partita:19997:0] parser.c:1916 UCX INFO
UCX_* env variables: UCX_LOG_LEVEL=info UCX_TCP_AF_PRIO=inet6
WARNING: Logging before InitGoogleLogging() is written to STDERR
W20220329 10:38:25.796393 19997 ucx_server.cc:299] [server] Listening on
ucx://[::1]:52523
W20220329 10:38:25.796595 19997 ucx_server.cc:302] [server] Listening on
::1:52523
[1648564705.797632] [partita:19997:0] ucp_context.c:1779 UCX INFO UCP
version is 1.12 (release 0)
[1648564705.805506] [partita:19997:0] ucp_worker.c:1867 UCX INFO
ep_cfg[0]: rma(cuda_copy/cuda); am(cuda_copy/cuda);
W20220329 10:38:25.805656 19997 ucx_client.cc:128] Connecting to ::1:52523
W20220329 10:38:25.806041 19997 ucx_client.cc:143] Connected to
local:::1:56388;remote:::1:52523
#
# UCP endpoint
#
# peer: ::1:52523
# lane[0]: cm tcp
#
# am_send: 0..<egr/bcopy>..(inf)
#
# rma_bw: mds rndv_rkey_size 9
#
[1648564705.806244] [partita:19997:a] wireup_cm.c:559 UCX DIAG
client ep 0x7f8b18328130 connect to ::1:52523 failed: device lo is not enabled,
enable it in UCX_NET_DEVICES or use corresponding ip address
[1648564705.806265] [partita:19997:a] wireup_cm.c:145 UCX DIAG
client ep 0x7f8b18328130 failed to connect to ::1:52523 using tcp cms
[1648564705.806302] [partita:19997:a] uct_cm.c:100 UCX DIAG
resolve callback failed with error: Destination is unreachable
[1648564705.812082] [partita:19997:0] ucp_ep.c:1226 UCX DIAG ep
0x7f8b18328130: error 'Destination is unreachable' on CM lane will not be
handled since no error callback is installed
/home/lidavidm/Code/upstream/flight-ucx/cpp/src/arrow/flight/test_definitions.cc:142:
Failure
Failed
'client_->GetFlightInfo(descr, &info)' failed with IOError:
ucp_request_check_status: UCX error -6: UCS_ERR_UNREACHABLE Destination is
unreachable
/home/lidavidm/Code/upstream/flight-ucx/cpp/src/arrow/flight/transport/ucx/ucx_internal.cc:560
CompleteRequestBlocking("ucp_am_send_nbx", request)
/home/lidavidm/Code/upstream/flight-ucx/cpp/src/arrow/flight/transport/ucx/ucx_internal.cc:1120
impl_->SendFrame(FrameType::kHeaders, buffer->data(), buffer->size())
/home/lidavidm/Code/upstream/flight-ucx/cpp/src/arrow/flight/transport/ucx/ucx_client.cc:563
driver->StartCall(kMethodGetFlightInfo)
[1648564705.812775] [partita:19997:0] flush.c:28 UCX ERROR req
0x5628b6f87880: error during flush: Endpoint timeout, flush comp 0x5628b6f87920
count reduced to 0
/home/lidavidm/Code/upstream/flight-ucx/cpp/src/arrow/flight/test_definitions.cc:128:
Failure
Failed
'client_->Close()' failed with IOError: ucp_am_send_nbx: UCX error -80:
UCS_ERR_ENDPOINT_TIMEOUT Endpoint timeout
/home/lidavidm/Code/upstream/flight-ucx/cpp/src/arrow/flight/transport/ucx/ucx_internal.cc:560
CompleteRequestBlocking("ucp_am_send_nbx", request)
/home/lidavidm/Code/upstream/flight-ucx/cpp/src/arrow/flight/transport/ucx/ucx_client.cc:550
connections_.front().Close()
/home/lidavidm/Code/upstream/flight-ucx/cpp/src/arrow/flight/client.cc:616
transport_->Close()
[ FAILED ] UcxDataTest.TestDoGetInts (45 ms)
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
