[MDEV-11877] Regression: MariaDB 10.0.26 failed to build on PowerPC (ppc32) in Debian Created: 2017-01-23  Updated: 2020-09-02  Resolved: 2020-09-02

Status: Closed
Project: MariaDB Server
Component/s: N/A
Affects Version/s: 10.0.26, 10.0.27, 10.0.28, 10.0.29
Fix Version/s: N/A

Type: Bug Priority: Minor
Reporter: Otto Kekäläinen Assignee: VicenČ›iu Ciorbaru
Resolution: Not a Bug Votes: 0
Labels: None

Attachments: File mariadb-10.0_10.0.29-0+deb8u1_powerpc-20170121-2137.gz    

 Description   

It was reported in Debian that 10.0.26 in Jessie failed to build on PowerPC. As other PowerPC issues were fixed, we assumed wrongly this one was fixed too, but when 10.0.29 was uploaded to Jessie, the failure can still be seen.

See https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=832931 for longer discussion.

The failures happens when tests are about to start:

Creating var directory '/«PKGBUILDDIR»/builddir/mysql-test/var'...
 
Checking supported features...
 
MariaDB Version 10.0.29-MariaDB-0+deb8u1
 
 - SSL connections supported
 
Sphinx 'indexer' binary not found, sphinx suite will be skipped
 
Using suites: main-,archive-,binlog-,csv-,federated-,funcs_1-,funcs_2-,handler-,heap-,innodb-,innodb_fts-,innodb_zip-,maria-,multi_source-,optimizer_unfixed_bugs-,parts-,percona-,perfschema-,plugins-,roles-,rpl-,sys_vars-,unit-,vcol-,connect,oqgraph,sequence,spider,spider/bg,sql_discovery,metadata_lock_info,query_response_time
 
Collecting tests...
 
Installing system database...
 
mysql-test/CMakeFiles/test-force.dir/build.make:49: recipe for target 'mysql-test/CMakeFiles/test-force' failed
 
make[5]: *** [mysql-test/CMakeFiles/test-force] Terminated
 
E: Caught signal ‘Terminated’: terminating immediately
 
debian/rules:129: recipe for target 'override_dh_auto_test' failed
 
make[1]: *** [override_dh_auto_test] Terminated
 
CMakeFiles/Makefile2:8965: recipe for target 'mysql-test/CMakeFiles/test-force.dir/all' failed
 
make[4]: *** [mysql-test/CMakeFiles/test-force.dir/all] Terminated
 
CMakeFiles/Makefile2:8973: recipe for target 'mysql-test/CMakeFiles/test-force.dir/rule' failed
 
make[3]: *** [mysql-test/CMakeFiles/test-force.dir/rule] Terminated
 
debian/rules:210: recipe for target 'build-arch' failed
 
make: *** [build-arch] Terminated
 
Makefile:2885: recipe for target 'test-force' failed
 
make[2]: *** [test-force] Terminated
 
Build killed with signal TERM after 150 minutes of inactivity

Full log attached.

Debian build logs visible at https://buildd.debian.org/status/logs.php?pkg=mariadb-10.0&arch=powerpc

Note that 10.0.x in Sid built OK on PowerPC. This issue is visible only in recent Jessie builds. It might have something to do with libc and jemalloc versions in Debian Jessie. Also note that Jessie only gets security updates at this point, and not all security builds are public, thus for example the attached build log isn't visible in the build summary.

 Comments   
Comment by Daniel Black [ 2017-01-26 ]

Given the debian bug is blocked by debian bug 832931 - incorrect jemalloc size what behaviour is required?

I've also had a jemalloc 64k page size running on a 4k page size ppc64le kernel with mariadb and it also crashed.

Having gdb installed and the error logs in the mysql-test/var directory would help.

In MDEV-10984 svoj mentions fixes that I assume are these ones in 10.2 https://github.com/MariaDB/server/commits/10.2?author=svoj Nov 25, 2016 (referencing MDEV-11296)

Comment by Otto Kekäläinen [ 2017-01-27 ]

Related PR at https://github.com/ottok/mariadb-10.1/pull/8

Comment by Vicențiu Ciorbaru [ 2017-03-28 ]

Alright, the hang seems to be related to jemalloc. It happens when we are freeing a large chunk of memory, spanning multiple system pages. From previous investigations I believe this to be the cause. I don't have an appropriate fix at the moment, other than not using jemalloc on Power, which seems like a bad compromise. Need to investigate if newer versions of JeMalloc fix this.

Comment by Daniel Black [ 2020-09-02 ]

Not so much not a bug, but not our bug.

jemalloc-5.0.0 https://github.com/jemalloc/jemalloc/commit/da4cff0279b2e8f2b0482ae961f2e2f63662342d included a fix which is now in Debian buster.

Generated at Thu Feb 08 07:53:21 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.