-
Notifications
You must be signed in to change notification settings - Fork 3.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
release-22.2: kv: campaign on rejected lease request when leader not live in node liveness #95221
Merged
nvanbenschoten
merged 2 commits into
cockroachdb:release-22.2
from
nvanbenschoten:backport22.2-87244
Jan 26, 2023
Merged
release-22.2: kv: campaign on rejected lease request when leader not live in node liveness #95221
nvanbenschoten
merged 2 commits into
cockroachdb:release-22.2
from
nvanbenschoten:backport22.2-87244
Jan 26, 2023
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The structure of the interface definition lists methods that require a read lock first, without an associated RLocked suffix, which is implied. Pure refactor.
Thanks for opening a backport. Please check the backport criteria before merging:
If some of the basic criteria cannot be satisfied, ensure that the exceptional criteria are satisfied within.
Add a brief release justification to the body of your PR to justify this backport. Some other things to consider:
|
erikgrinaker
approved these changes
Jan 16, 2023
@erikgrinaker , is 22.1 possible? thanks |
See #95237. |
tbg
approved these changes
Jan 24, 2023
…iveness Fixes cockroachdb#84655. Related to cockroachdb#49220. This commit extends the logic introduced in 8aa1c14 to simultaneously campaign for Raft leadership when rejecting a lease request on a Raft follower that observes that the current Raft leader is not live according to node liveness. These kinds of cases are often associated with asymmetric network partitions. In such cases, the Raft leader will be unable to acquire an epoch-based lease until it heartbeats its liveness record. As a result, the range can only regain availability if a different replica acquires the lease. However, the protection added in 8aa1c14 prevents followers from acquiring leases to protect against a different form of unavailability. After this commit, the followers will attempt to acquire Raft leadership when detecting such cases by campaigning. This allows these ranges to recover availability once Raft leadership moves off the partitioned Raft leader to one of the followers that can reach node liveness and can subsequently acquire the lease. Campaigning for Raft leadership is safer than blindly allowing the lease request to be proposed (through a redirected proposal). This is because the follower may be arbitrarily far behind on its Raft log and acquiring the lease in such cases could cause unavailability (the kind we saw in cockroachdb#37906). By instead calling a Raft pre-vote election, the follower can determine whether it is behind on its log without risking disruption. If so, we don't want it to acquire the lease — one of the other followers that is caught up on its log can. If not, it will eventually become leader and can proceed with a future attempt to acquire the lease. The commit adds a test that reproduces the failure mode described in cockroachdb#84655. It creates an asymmetric network partition scenario that looks like: ``` [0] raft leader / initial leaseholder ^ / \ / \ v v [1]<--->[2] raft followers ^ ^ \ / \ / v [3] liveness range ``` It then waits for the raft leader's lease to expire and demonstrates that one of the raft followers will now call a Raft election, which allows it to safely grab Raft leadership, acquire the lease, and recover availability. Without the change, the test failed. Release justification: None. Too risky for the stability period. Potential backport candidate after sufficient baking on master. Release note (bug fix): A bug causing ranges to remain without a leaseholder in cases of asymmetric network partitions has been resolved.
nvanbenschoten
force-pushed
the
backport22.2-87244
branch
from
January 26, 2023 19:28
f846710
to
8bacca2
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Backport 2/2 commits from #87244.
/cc @cockroachdb/release
Fixes #84655.
Related to #49220.
This commit extends the logic introduced in 8aa1c14 to simultaneously campaign for Raft leadership when rejecting a lease request on a Raft follower that observes that the current Raft leader is not live according to node liveness. These kinds of cases are often associated with asymmetric network partitions.
In such cases, the Raft leader will be unable to acquire an epoch-based lease until it heartbeats its liveness record. As a result, the range can only regain availability if a different replica acquires the lease. However, the protection added in 8aa1c14 prevents followers from acquiring leases to protect against a different form of unavailability.
After this commit, the followers will attempt to acquire Raft leadership when detecting such cases by campaigning. This allows these ranges to recover availability once Raft leadership moves off the partitioned Raft leader to one of the followers that can reach node liveness and can subsequently acquire the lease.
Campaigning for Raft leadership is safer than blindly allowing the lease request to be proposed (through a redirected proposal). This is because the follower may be arbitrarily far behind on its Raft log and acquiring the lease in such cases could cause unavailability (the kind we saw in #37906). By instead calling a Raft pre-vote election, the follower can determine whether it is behind on its log without risking disruption. If so, we don't want it to acquire the lease — one of the other followers that is caught up on its log can. If not, it will eventually become leader and can proceed with a future attempt to acquire the lease.
The commit adds a test that reproduces the failure mode described in #84655. It creates an asymmetric network partition scenario that looks like:
It then waits for the raft leader's lease to expire and demonstrates that one of the raft followers will now call a Raft election, which allows it to safely grab Raft leadership, acquire the lease, and recover availability. Without the change, the test failed.
Release justification: None. Too risky for the stability period. Potential backport candidate after sufficient baking on master.
Release note (bug fix): A bug causing ranges to remain without a leaseholder in cases of asymmetric network partitions has been resolved.