self-test errors do not break the build

Bug #1791018 reported by Christian Ehrhardt 
12
This bug affects 1 person
Affects Status Importance Assigned to Milestone
mysql-5.7 (Ubuntu)
Triaged
Low
Lena Voytek

Bug Description

Hi,
I was debugging an issue showing up in autopkgtest "upstream" and after a while found that this could have been detected at build time.

For the question "should this never have successfully built?" I forked this bug here.
You might have had reasons - so I'll leave answering if that is intentional or shall be changed to the usual maintainers.

Build log:
https://launchpadlibrarian.net/385184440/buildlog_ubuntu-cosmic-ppc64el.mysql-5.7_5.7.23-0ubuntu2_BUILDING.txt.gz

Contains (among others) this:
main.sp w1 [ retry-fail ]
        Test ended at 2018-08-24 17:23:08

CURRENT_TEST: main.sp
--- /<<PKGBUILDDIR>>/builddir/mysql-test/r/sp.result 2018-08-24 20:22:01.346049858 +0300
+++ /<<PKGBUILDDIR>>/builddir/mysql-test/var/1/log/sp.reject 2018-08-24 20:23:08.283072785 +0300
@@ -5982,9 +5982,9 @@
 INSERT INTO t3 VALUES (1, 3.4), (1, 2), (1, 0.9), (2, 8), (2, 7)|
 SELECT SUM(f2), bug25373(f1) FROM t3 GROUP BY bug25373(f1) WITH ROLLUP|
 SUM(f2) bug25373(f1)
-6.300000071525574 1
+6.299999833106995 1
 15 2
-21.300000071525574 NULL
+21.299999833106995 NULL
 DROP FUNCTION bug25373|
 DROP TABLE t3|
 DROP DATABASE IF EXISTS mysqltest1|

mysqltest: Result content mismatch

safe_process[24766]: Child process: 24767, exit: 1

 - the logfile can be found in '/<<PKGBUILDDIR>>/builddir/mysql-test/var/log/main.sp/sp.log'

Test main.sp has failed 2 times, no more retries!

Revision history for this message
Christian Ehrhardt  (paelzer) wrote :

Reference: wondered while debugging for bug 1791010

Revision history for this message
Lars Tangvald (lars-tangvald) wrote :

During build, passing the mtr suite is only enforced for i386 and amd64 architectures, because upstream has limited support for others, so the test suites tended to be unstable or failing "reliably" (particularly due to floating point differences as in this case).

We actually release arm64 packages for 8.0 now, and fixing tests there also made the situation better on pcc64el, which might have made it temporarily green for autopkgtest only to have it regress again since it receives very little upstream testing.

Previously, autopkgtest for ppc64el was marked as expected failure

Revision history for this message
Otto Kekäläinen (otto) wrote :

Christian, as Lars writes, Oracle only supports amd64 and i386.
https://salsa.debian.org/mariadb-team/mysql/commit/5d6a44b949d1014332720fb30768047e5b1d3007

If you want to run on ppc64el, I warmly recommend MariaDB. MariaDB officially supports ppc64el and IBM is a even a sponsor of the MariaDB Foundation: https://mariadb.org/

Changed in mysql-5.7 (Ubuntu):
status: New → Triaged
importance: Undecided → Low
Revision history for this message
Lars Tangvald (lars-tangvald) wrote :

I did a test at https://launchpad.net/~mysql-ubuntu/+archive/ubuntu/mysql-5.7/+packages by enforcing the test suite on all architectures, and it's actually green. Was going to suggest we enable it for arm64, but maybe we should consider the rest as well.

Revision history for this message
Andreas Hasenack (ahasenack) wrote :

Do we still want to have other arches failing the build if tests do not pass?

Revision history for this message
Christian Ehrhardt  (paelzer) wrote :

This has fallen through the cracks.
How about retrying and reconsidering to enable that now that it is 2022?

Revision history for this message
Lena Voytek (lvoytek) wrote :

I will try to get this enabled again, thanks!

Changed in mysql-5.7 (Ubuntu):
assignee: nobody → Lena Voytek (lvoytek)
Revision history for this message
Lena Voytek (lvoytek) wrote :

Finally getting back to this. I ran an initial test to see how the tests look now in a ppa: https://launchpad.net/~lvoytek/+archive/ubuntu/mysql-testsuite-check

Looks like, as of 8.0.33, armhf still fails most tests due to the lack of cycle timer. Meanwhile s390x is failing a few tests. All other architectures look good though

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.