kv: don't use LegacyTimestamp in Liveness records#56190
kv: don't use LegacyTimestamp in Liveness records#56190nvb wants to merge 2 commits intocockroachdb:masterfrom
Conversation
6db235a to
c2bb8c9
Compare
andreimatei
left a comment
There was a problem hiding this comment.
Reviewable status:
complete! 1 of 0 LGTMs obtained (waiting on @andreimatei)
|
This appears to be running into issues with the |
|
maybe #55520 helps
…On Wed, Nov 4, 2020 at 4:09 PM Nathan VanBenschoten < ***@***.***> wrote:
This appears to be running into issues with the acceptance/version-upgrade
test. I think this is because master still reports a v20.2.0 build tag,
so we still start with a v20.1 cluster in that test. I'll need to fix that
first before merging the rest of this.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#56190 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAC4C4LD3XRHORW4MCGZGUTSOG7J7ANCNFSM4TG4AYCQ>
.
|
|
It does! Thanks for the pointer. |
c2bb8c9 to
3bb19a3
Compare
|
I just pushed a |
3bb19a3 to
75d2fd0
Compare
This was breaking cockroachdb#56190. In that PR, we change the proto encoding of Liveness. We intended to make this safe in 9c0dce9 by using the existing marshalled proto when performing cputs. Unfortunately, that change wasn't quite enough, as it didn't ensure that the local raw byte representation of the proto would get updated on a failed cput. This could lead to an infinite loop of failed liveness updates, as we saw in `acceptance/version-upgrade`. This PR will make that change safe, but unfortunately, we'll have to wait another release for it.
|
Thanks @lucy-zhang! Your change worked. Unfortunately, it revealed that the node liveness code wasn't as ready as it thought it was for variable proto encoding. So we're going to have to land #56477 in v21.1 and hold off on this PR until v21.2. That's a shame, but nothing depends on this change so we can just wait it out. |
This was breaking cockroachdb#56190. In that PR, we change the proto encoding of Liveness. We intended to make this safe in 9c0dce9 by using the existing marshalled proto when performing cputs. Unfortunately, that change wasn't quite enough, as it didn't ensure that the local raw byte representation of the proto would get updated on a failed cput. This could lead to an infinite loop of failed liveness updates, as we saw in `acceptance/version-upgrade`. This PR will make that change safe, but unfortunately, we'll have to wait another release for it.
56477: liveness: replace liveness on raw encoding change r=nvanbenschoten a=nvanbenschoten This was breaking #56190. In that PR, we change the proto encoding of Liveness. We intended to make this safe in 9c0dce9 by using the existing marshalled proto when performing cputs. Unfortunately, that change wasn't quite enough, as it didn't ensure that the local raw byte representation of the proto would get updated on a failed cput. This could lead to an infinite loop of failed liveness updates, as we saw in `acceptance/version-upgrade`. This PR will make that change safe, but unfortunately, we'll have to wait another release for it. Co-authored-by: Nathan VanBenschoten <nvanbenschoten@gmail.com>
Completes migration started in 9c0dce9.