[
https://issues.apache.org/jira/browse/IMPALA-11640?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Joe McDonnell resolved IMPALA-11640.
------------------------------------
Fix Version/s: Impala 4.2.0
Resolution: Fixed
> Build fails on Ubuntu 18/20 when using shared libraries
> -------------------------------------------------------
>
> Key: IMPALA-11640
> URL: https://issues.apache.org/jira/browse/IMPALA-11640
> Project: IMPALA
> Issue Type: Bug
> Components: Backend
> Affects Versions: Impala 4.2.0
> Reporter: Joe McDonnell
> Assignee: Joe McDonnell
> Priority: Major
> Fix For: Impala 4.2.0
>
>
> When building on Ubuntu 18 or Ubuntu 20 with shared libraries (-so), the
> build fails because the unifiedbetests binary fails to run as part of the
> validate-unified-backend-test-filters.py invocation:
> {noformat}
> 16:39:22 F1005 23:39:22.237543 88570 unwind_safeness.cc:76] Check failed:
> !error failed to find symbol dlopen:
> /home/ubuntu/Impala/be/build/release/kudu_util/libkudu_util.so: undefined
> symbol: dlopen
> 16:39:22 FAILED: Unified backend test executable returned an error when trying
> 16:39:22 to list tests.
> 16:39:22 Command: /home/ubuntu/Impala/bin/run-binary.sh
> /home/ubuntu/Impala/be/build/release//service/unifiedbetests
> --gtest_list_tests
> 16:39:22 Return Code: -6
> 16:39:22 stdout:
> 16:39:22
> 16:39:22 stderr:
> 16:39:22 None{noformat}
> When building locally, other binaries also fail to execute with the same
> message.
> One theory is that the code in unwind_safeness.cc has never worked, but
> Ubuntu 16.04 is impacted by a glibc bug that prevents it from setting an
> error ([https://sourceware.org/bugzilla/show_bug.cgi?id=19509]). Newer Ubuntu
> correctly report the error, which leads to the failure.
> One option is to change unwind_safeness.cc to tolerate missing
> dlopen/dlclose. Impala doesn't ship using shared libraries. If the
> unwind_safeness.cc variables that contain dlopen/dlclose are actually used
> after a failure to resolve dlopen/dlclose, then it would result in a SIGSEGV
> and it would be very obvious.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
