Bug #55177
closedgcc is not installed on centos 8.stream nodes
0%
Description
2022-04-04T09:35:31.257 INFO:tasks.workunit.cluster1.client.mirror.smithi037.stderr:HEAD is now at b2b4ff400c5 Merge branch 'pacific-backport-march' of https://github.com/adk3798/ceph into wip-yuri2-testing-2022-03-31-1523-pacific 2022-04-04T09:35:31.258 DEBUG:teuthology.orchestra.run.smithi037:> cd -- /home/ubuntu/cephtest/clone.cluster1.client.mirror/qa/workunits && if test -e Makefile ; then make ; fi && find -executable -type f -printf '%P\0' >/home/ubuntu/cephtest/workunits.list.cluster1.client.mirror 2022-04-04T09:35:31.326 INFO:tasks.workunit.cluster1.client.mirror.smithi037.stdout:for d in direct_io fs ; do ( cd $d ; make all ) ; done 2022-04-04T09:35:31.328 INFO:tasks.workunit.cluster1.client.mirror.smithi037.stdout:make[1]: Entering directory '/home/ubuntu/cephtest/clone.cluster1.client.mirror/qa/workunits/direct_io' 2022-04-04T09:35:31.328 INFO:tasks.workunit.cluster1.client.mirror.smithi037.stdout:cc -Wall -Wextra -D_GNU_SOURCE direct_io_test.c -o direct_io_test 2022-04-04T09:35:31.329 INFO:tasks.workunit.cluster1.client.mirror.smithi037.stderr:make[1]: cc: Command not found 2022-04-04T09:35:31.329 INFO:tasks.workunit.cluster1.client.mirror.smithi037.stderr:make[1]: *** [Makefile:6: direct_io_test] Error 127 2022-04-04T09:35:31.330 INFO:tasks.workunit.cluster1.client.mirror.smithi037.stdout:make[1]: Leaving directory '/home/ubuntu/cephtest/clone.cluster1.client.mirror/qa/workunits/direct_io' 2022-04-04T09:35:31.330 INFO:tasks.workunit.cluster1.client.mirror.smithi037.stdout:make[1]: Entering directory '/home/ubuntu/cephtest/clone.cluster1.client.mirror/qa/workunits/fs' 2022-04-04T09:35:31.330 INFO:tasks.workunit.cluster1.client.mirror.smithi037.stdout:cc -Wall -Wextra -D_GNU_SOURCE test_o_trunc.c -o test_o_trunc 2022-04-04T09:35:31.331 INFO:tasks.workunit.cluster1.client.mirror.smithi037.stderr:make[1]: cc: Command not found 2022-04-04T09:35:31.331 INFO:tasks.workunit.cluster1.client.mirror.smithi037.stdout:make[1]: Leaving directory '/home/ubuntu/cephtest/clone.cluster1.client.mirror/qa/workunits/fs' 2022-04-04T09:35:31.332 DEBUG:teuthology.orchestra.run:got remote process result: 2 2022-04-04T09:35:31.333 INFO:tasks.workunit.cluster1.client.mirror.smithi037.stderr:make[1]: *** [Makefile:6: test_o_trunc] Error 127 2022-04-04T09:35:31.333 INFO:tasks.workunit.cluster1.client.mirror.smithi037.stderr:make: *** [Makefile:4: all] Error 2
Looking at ceph-cm-ansible, gcc seems to be there in the list of packages to install:
$ cat roles/testnode/vars/centos_8.yml --- # vars specific to any centos 8.x version [...] packages: [...] # for workunits, - gcc - git
and there is also
$ cat roles/testnode/vars/centos_8_stream.yml --- # vars specific to centos stream version 8.x packages_to_upgrade: - systemd packages: - lvm2
which, for 8.stream, is supposed to be applied in addition to centos_8.yml.
But, only libgcc is actually installed:
$ yum list | grep gcc libgcc.x86_64 8.5.0-10.el8 @baseos csgcca.x86_64 2.1.1-1.el8 epel gcc.x86_64 8.5.0-10.el8 CentOS-AppStream gcc-c++.x86_64 8.5.0-10.el8 CentOS-AppStream [...]
Updated by David Galloway about 2 years ago
I don't think those two packages: [] lists get combined actually. That could be why.
Updated by David Galloway about 2 years ago
Actually... ceph-cm-ansible never ran in that job you linked.
Updated by Ilya Dryomov about 2 years ago
David Galloway wrote:
I don't think those two
packages: []lists get combined actually. That could be why.
Maybe, but that change dates back to mid February. I didn't check but I don't think it's been broken for that long.
Updated by Ilya Dryomov about 2 years ago
David Galloway wrote:
Actually... ceph-cm-ansible never ran in that job you linked.
I thought it was supposed to be run before the fog image is captured?
Do you know when the 8.stream image was last updated?
Updated by David Galloway about 2 years ago
Ilya Dryomov wrote:
David Galloway wrote:
Actually... ceph-cm-ansible never ran in that job you linked.
I thought it was supposed to be run before the fog image is captured?
It still gets run at the beginning of every job. There are tasks that can't really be backed into the FOG image (partitioning of drives, for example).
Do you know when the 8.stream image was last updated?
2022-04-03 12:51:52
So it sounds like maybe we were relying on a bug that was baked into the old image?
Updated by Ilya Dryomov about 2 years ago
David Galloway wrote:
It still gets run at the beginning of every job. There are tasks that can't really be backed into the FOG image (partitioning of drives, for example).
But you said it didn't get run in this case?
2022-04-03 12:51:52
So it sounds like maybe we were relying on a bug that was baked into the old image?
I'm not sure, maybe we just didn't notice it because distros are selected at random and on top of that some distro-specific failures can get masked by other failures.
gcc has always been needed by the workunit task, as alluded to by the "for workunits" comment up there. Needing make, gcc, etc to be installed on test nodes is not a new requirement, I think it is over 10 years old.