rocksdb: revert the change introduced by dc41731 #10595
Conversation
dc41731 reverts the submodule reference of rocksdb, and point it to an older version which fails to support ppc64le and AArch64. Signed-off-by: Kefu Chai <kchai@redhat.com>
|
@smithfarm, as a part of #10438, we updated the rocksdb submodule and then reverted that change in #10060. in this change, we reverted the revert. probably we can drop the change of rocksdb in #10060, and note it in "Conflicts", as that patch won't apply without #10438. and also AFAICT, no bug fix relies on the ppc64le and aarch64 support, so we don't need to backport #10438, right? |
|
@tchaikov Correct, no jewel bugfixes need it. It's a feature request. @michelmno will be able to explain better, but my understanding is that facebook/rocksdb@7ca731b is needed for Ceph to build for ppc64le and aarch64, so there is an interest in updating the rocksdb submodule to that in the jewel release. |
|
@smithfarm , @tchaikov, I still need the fix of facebook/rocksdb@7ca731b to be able to build ceph jewel branch for ppc64le. |
|
@michelmno i understand, but in general, we don't backport features. |
|
@tchaikov I tend to see facebook/rocksdb@7ca731b as a build/ops fix. Could it be cherry-picked so we pick up just this one commit? (Like you did in #10750 for hammer) |
|
@smithfarm yeah. guess we need to have a tracker ticket for this fix then. @michaelndn could you file one for it? |
|
@michelmno I guess @tchaikov means that no feature like "ppc64(le) support" was contemplated or announced for jewel. The idea here is just to enable the build to complete on this architecture, not to support it. |
|
@smithfarm @michelmno exactly. |
|
as suggested in #10595 (comment) |
|
I just opened ceph/rocksdb#12 - once it is merged, we could bump the rocksdb sha1 in the jewel branch. |
dc41731 reverts the submodule reference of rocksdb, and point it to an
older version which fails to support ppc64le and AArch64.
Signed-off-by: Kefu Chai kchai@redhat.com