cmake: identify the possible incompatibility of rocksdb and tcmalloc #17788
Conversation
so it checks $ENV{GPERF_ROOT} first.
Signed-off-by: Kefu Chai <kchai@redhat.com>
Signed-off-by: Kefu Chai <kchai@redhat.com>
Signed-off-by: Kefu Chai <kchai@redhat.com>
Signed-off-by: Kefu Chai <kchai@redhat.com>
we've updated the rockdb wrapper on ceph side to be compatible with the latest version of rocksdb upstream. so ceph is not compatible with older version of rocksdb. Signed-off-by: Kefu Chai <kchai@redhat.com>
|
When trying to build that branch, I get this error: Is that expected with these changes? If so, the error message is not particularly helpful. What does gperf-tools have to do with this? FWIW, I have gperftools-libs-2.5.93-1.fc26.x86_64, so I should be OK, right? |
yes.
i am all ears.
see the commit message of 94fc1f6.
no, you need to either upgrade your gperftools to the latest master or to downgrade it to 2.5 (without .93). |
|
@jtlayton gperftools-libs contains libtcmalloc and this issue is a problem with the interaction of certain versions of rocksdb and tcmalloc. As for gperftools-libs-2.5.93-1.fc26.x86_64 that's the same version I'm running and it's definitely not OK. The reason this has remained undetected since @tchaikov updated rocksdb is that most OS are unaffected (I believe f26 is the only OS so far identified as having this problem). unittest_rocksdb_option fails in a similar fashion on f26 and yet that unit test has been passing consistently on jenkins systems and presumably others as well. |
the commit d406f228 in gperf implements a c11 feature used by a recent change in rocksdb: 16e03882, which uses aligned_alloc(). and 16e03882 in rocksdb was merged after v5.7 was tagged, while 16e03882 in gperf was merged after v2.6.1 was tagged. because aligned_alloc() is not implemented by tcmalloc until the not-yet-released 2.6.2, if we call aligned_alloc() in an application linked against tcmalloc, what gets called will be the glibc's aligned_alloc(). but if we free() the memory chunk allocated by aligned_alloc(), the tcmalloc's implementation kicks in, then InvalidFree() is called, because the memory chunk being freed was allocated by tcmalloc. in short, "mixing allocators", quote from Dan Mick. in rocksdb, aligned_alloc() is used if _ISOC11_SOURCE is defined, this makes sense, because aligned_alloc() is a C11 function. we could avoid using it by not defining _ISOC11_SOURCE. but as long as _GNU_SOURCE is defined, glibc defines _ISOC11_SOURCE. and libstdc++ requires _GNU_SOURCE, because it uses a fair amount of GNU extensions. Fixes: http://tracker.ceph.com/issues/21422 Signed-off-by: Kefu Chai <kchai@redhat.com>
tchaikov
force-pushed
the
wip-cmake-rocksdb-tcmalloc
branch
from
61ee757
to
ff8d6a7
Compare
|
@jtlayton i put some details in the ticket, HTH. w.r.t. to "If so, the error message is not particularly helpful." i will send a mail to ceph-devel to remind the fc25 users. i agree that the error message can be improved. if anybody can help with it. it will be much appreciated. |
|
How about "use gperf-tools v2.6 or later" ? Your message says to use v2.5, but that's what I have and it doesn't work. Ahh, nm...I see the patch now. Crap, then there is no working package for recent fedora releases. Can you open a bug for them to get the packages there updated? This is likely to break a lot of people that are building on Fedora. |
|
I'm not saying this is a great idea, but we could maybe also make a local patch to rocksdb to not use alloc_aligned based on the version of libtcmalloc |
|
@tchaikov The error message in |
it's a follow-up of ceph#17788 Signed-off-by: Kefu Chai <kchai@redhat.com>
|
Looks like Fedora 26 will release the fix under 2.6.1-5.fc26 so this change is will still prevent Ceph from being built on Fedora 26 since it expect 2.6.2. [1] https://bugzilla.redhat.com/show_bug.cgi?id=1494309 |
|
@dillaman thanks for the heads-up. i will figure out a way to detect the missing |
No description provided.