maria-developers team mailing list archive

[Alexander Barkov <bar@xxxxxxxxxxx>]

Hello Elena,

You wrote on IRC:

On 01/31/2017 08:10 PM, Slack wrote:
> bar: hi! We have this in 10.0 valgrind, what do you think about it?
> http://buildbot.askmonty.org/buildbot/builders/work-amd64-valgrind/builds/9873/steps/test/logs/stdio
> 
> i'm asking you because you fixed a similar-looking problem once:
> https://jira.mariadb.org/browse/MDEV-4890

I'm not very familiar with this part. But IIUC, two things could happen:
- Somebody changed the code in mysqld.cc, so the SSL cleanup routines
  do not get called correctly any more.

- Somebody is wrong inside the library itself.


When did this problem first appear? After some commit?
Or after the SSQL library update/change on the build box?
Is it specific to 10.0?


When I was fixing MDEV-4890, I fixed both:
- added some cleanup function calls
- added the second SSL suppression
The problem was reproducible on my box.


Now I cannot reproduce the problem on my box.

I tried:
cmake . -DCMAKE_BUILD_TYPE=Debug -DWITH_VALGRIND=1
./mtr --valgrind binlog.binlog_max_extension
./mtr --valgrind federated.error_on_close-8313
./mtr --valgrind rpl.rpl_old_master


Note, we have these valgrind suppressions:

> {
>    OpenSSL still reachable.
>    Memcheck:Leak
>    fun:*alloc
>    fun:CRYPTO_malloc
>    fun:sk_new
>    obj:*libssl*
>    fun:SSL_COMP_get_compression_methods
>    fun:SSL_library_init
> }
> 
> 
> {
>    OpenSSL still reachable.
>    Memcheck:Leak
>    fun:*alloc
>    fun:CRYPTO_malloc
>    obj:*libssl*
>    fun:SSL_COMP_get_compression_methods
>    fun:SSL_library_init
> }

Do you know why the problem does not get suppressed?
Perhaps we should try to remove "obj:*libssl*" ?

But before doing this it would be nice to know when the problem first
appeared.