feat: [IC-1992] Implement `do_split_subnet` endpoint by kpop-dfinity · Pull Request #8692 · dfinity/ic

kpop-dfinity · 2026-02-05T10:56:11Z

Introduce a new registry canister endpoint split_subnet which when executed, as the name suggest, will instruct the specified subnet to split into two.

When executed the following mutations will be made to the registry:

Create a new subnet
Update the subnets list entry
Change the original subnet membership
Create/update catch up package contents for both subnets
Reroute the routing table
Update the canister migrations entry

Note

For now the endpoint is not enabled as there is still work to be done on the consensus side.

…8673) Support `--since DATETIME` and `--until DATETTIME` arguments for ` //ci/githubstats:query`. For example: ``` bazel run //ci/githubstats:query -- top 10 impact --since '2026-01-30 12:23' --verbose ... AND (bt.first_start_time >= '2026-01-30 12:23:00+00:00'::timestamptz) ... ``` --------- Co-authored-by: IDX GitHub Automation <infra+github-automation@dfinity.org>

…til arguments (#8674) I occasionally want to see if a specific commit fixed an error so I want to see the test statistics since the time of that commit. To not have to figure out the date and time of a commit manually this extends the `--since` and `--until` flags to also accept git commits from which the time will then be taken. For example to see if bbb6cb0 has the effect we hoped for you could run: ``` $ bazel run //ci/githubstats:query -- top 10 impact --since bbb6cb0 --verbose ... WHERE ... AND (bt.first_start_time >= '2026-01-30 12:23:19+00:00'::timestamptz) ... ... ``` --------- Co-authored-by: IDX GitHub Automation <infra+github-automation@dfinity.org>

Update the proposal descriptions based on several rules: - Match [NNS Dapp](https://github.com/dfinity/nns-dapp/blob/main/frontend/src/lib/i18n/en.governance.json) - Use imperative form - they describe actions, e.g. "A proposal to change ..." => "Change", "Updates ..." => "Update ..." - Did not use the ManageNetworkEconomics text from the NNS Dapp, but kept the rewritten one to avoid listing every parameter - they easily get out-dated.

…8667) # Why When maturity disbursement feature was added, it used the same minimum for the spawning neurons. For spawning, the minimum requirement has 2 purposes: (1) avoid spawning too many neurons given the same amount of maturity, i.e. avoid DoS (2) the resulting neuron should always have >= 1 ICP stake (minimum stake) with worst maturity modulation (-5%). For maturity disbursement, (2) isn't needed anymore, and the 1/0.95 value is more confusing than 1, without a good reason. # What Check the maturity disbursement >= 1 ICP, directly, without applying worst maturity modulation. Note that since we are lowering the minimum requirement, all maturity disbursement requests that previously succeed will still succeed, while some requests (`[1, 1 / 0.95)`) that previously fail will now succeed. This should not cause confusing behavior in the frontends, and thus frontend changes can be released independently from this.

The struct was deprecated a long time ago (it's a combination of ICP and cycles; but transferring ICP via canister calls was never implemented; so it was eventually replaced by `Cycles` fields in most places). Following recent changes that added (and populated and persisted) a `Cycles` field next to every `Funds` field, it is now safe to drop the type from the Replicated State and from checkpoints. (Dropping it from the canonical state is independent and also going on in parallel.)

This will stop encoding Funds fields in Request and Response messages in the canonical state tree.

This PR introduces the required changes in the orchestrator to support recalled GuestOS versions as part of SEV recovery. In the upgrade loop, if the upgrade version is recalled according to the latest registry version, we return an error early, but still set the subnet assignment, which allows to provision SSH keys and continue with a normal recovery. Most of the modified logic lies in the registry replicator. Indeed, shortly after booting, the latest registry version available locally might not yet reflect the latest version of the registry canister. Thus, if the GuestOS version was recalled at the latest version in the canister, the node might not yet see it and still proceed with the upgrade. We propose to first check if the replicator _has replicated all versions which were certified before its initialization_, which is done asynchronously. This guarantees that if the GuestOS version was recalled _before the node rebooted_, then the node will not upgrade to it. As a safeguard to not stay stuck forever, if the replicator fails to replicate all versions which were certified before its initialization after 30 minutes, we will proceed with trusting our own view of the registry to determine whether the target version is recalled or not. Also, this change does not target the NNS to avoid delaying NNS Recovery. The NNS should not benefit from SEV anyways. P.S.: A lot of the changed lines mock the `RegistryHelper` to test the implementation even though recalled replica versions are not supported yet. Once they are, a lot of the mocking code can be deleted. They are marked with `TODO(NODE-1754)` in comments. --------- Co-authored-by: IDX GitHub Automation <infra+github-automation@dfinity.org>

Fix for the autofix job [failing](https://github.com/dfinity/ic/actions/runs/21682746474/job/62521450291#step:5:288) on external PRs. --------- Co-authored-by: IDX GitHub Automation <infra+github-automation@dfinity.org>

Co-authored-by: IDX GitHub Automation <infra+github-automation@dfinity.org>

…::list_state_hashes_to_certify (#8678) This PR introduces a new `StateHashMetadata` struct returned by `StateManager::list_state_hashes_to_certify`. The motivation for the change is the function `StateManager::list_state_hashes_to_certify` returns a pair of height and hash and extending the pair into a tuple (to include extra stuff in follow-up PRs) is not a good practice.

…Rs (#8689) Co-authored-by: IDX GitHub Automation <infra+github-automation@dfinity.org>

alin-at-dfinity

LGTM, thank you!

This adds correct workflow permissions and introduces a dry-run mode. Successful run here: https://github.com/dfinity/ic/actions/runs/21712164393/job/62618331297 Crate publishing based on this example: https://github.com/dfinity/ic-bn-lib/blob/main/.github/workflows/publish-ic-bn-lib.yml

We already check (and actually panic) if we attempt to call `SystemState::push_input()` with a message addressed to a different canister. So the `Callback::originator` field is entirely redundant.

@AleDema

…nit test (#8656) Following @AleDema's [suggestion](9d8ad1b43#r176197856), this PR uses `matches!` instead of an equality check for the new subnet assignment in upgrade loop unit tests. Indeed, since `SubnetAssignment::Unknown != SubnetAssignment::Unknown` ([source](https://sourcegraph.com/github.com/dfinity/ic@0e4b3d10a063292b9432ba529381b0fff7f8a166/-/blob/rs/orchestrator/src/orchestrator.rs?L60)), `assert_ne!(new_subnet_assignment, SubnetAssignment::Unknown)` would always pass, even if the assignment was `Unknown`. Using pattern matching fixes this. The original comment suggested to preserve the reflexive property, i.e. have that `SubnetAssignment::Unknown == SubnetAssignment::Unknown`. Though I think that from a conceptual point of view, two unknown assignments should not be considered equal, as they could be unknown for different reasons (similar to `NaN != NaN` for floating point numbers).

The pre-generated dev launch measurement should match the actual dev SEV launch measurement. This fixes a bug where a dev SEV node is unable to join a dev testnet because the blessed version contained a "prod" measurement. Fix: correct the `vcpus` used in the pre-generated dev launch measurement to match the dev defaults (in deployment.json.template).

Last follow up PR that replaces `local_test_on_nns_subnet` with `state_machine_test_on_nns_subnet`. Results in significant speed up in 3 out of 4 tests (smaller speed up in the last one) that were using the proc macro `parameterized_upgrades`.

Another improvement in the SNS integration tests replacing the older `local_test_on_sns_subnet` with the more efficient `state_machine_test_on_sns_subnet`. For tests that are using it, the speed up ranges from roughly 30% up to 50%. Note that there is still one file (`rs/sns/integration_tests/src/neuron.rs`) where the older function is used as it was not a simple string replace to make it work. It can be replaced in another PR in the future.

Following up to #8679, performing a similar speed up in Root canister's integration tests. The results are even more dramatic in this case, a ~5x speed up, roughly going from ~25s to about ~5s for most tests.

Update unreleased_changelog.md (if there are behavior changes, even if they are non-breaking). - Done
Are there BREAKING changes? - no, the endpoint was unimplemented until now
Is a data migration needed? - no
Security review? - design has been reviewed, implementation review is scheduled to happen soon

Co-authored-by: Pierugo Pace <pierugo.pace@dfinity.org>

github-actions Bot added the feat label Feb 5, 2026

basvandijk and others added 14 commits February 5, 2026 12:08

chore: Bump certification version to V23 (#8675)

59d02e0

This will stop encoding Funds fields in Request and Response messages in the canonical state tree.

fix: allow the autofix job to git push to external PRs (#8683)

eb93e28

Fix for the autofix job [failing](https://github.com/dfinity/ic/actions/runs/21682746474/job/62521450291#step:5:288) on external PRs. --------- Co-authored-by: IDX GitHub Automation <infra+github-automation@dfinity.org>

feat(proposal-cli): add support for dogecoin watchdog (#8621)

19b0a5c

Co-authored-by: IDX GitHub Automation <infra+github-automation@dfinity.org>

fix: generate-config-fixtures and lock-generate to push to external P…

68ed0ac

…Rs (#8689) Co-authored-by: IDX GitHub Automation <infra+github-automation@dfinity.org>

Fix protobuf

5b9f12e

fix tests

935d976

Merge branch 'master' into kpop/feat/nns_changes

6478e8f

kpop-dfinity changed the title ~~feat: [IC-1992] Create a registry canister endpoint~~ feat: [IC-1992] Create a registry canister endpoint for subnet splitting Feb 5, 2026

kpop-dfinity requested review from alin-at-dfinity and jasonz-dfinity February 5, 2026 12:30

kpop-dfinity added the CI_ALL_BAZEL_TARGETS Runs all bazel targets label Feb 5, 2026

alin-at-dfinity reviewed Feb 6, 2026

View reviewed changes

jasonz-dfinity reviewed Feb 7, 2026

View reviewed changes

Comment thread rs/nns/governance/src/proposals/execute_nns_function.rs Outdated

Comment thread rs/nns/governance/src/proposals/execute_nns_function.rs

jasonz-dfinity reviewed Feb 9, 2026

View reviewed changes

Comment thread rs/registry/canister/canister/canister.rs Outdated

Comment thread rs/registry/canister/src/mutations/do_split_subnet.rs Outdated

Comment thread rs/registry/canister/src/mutations/do_split_subnet.rs Outdated

Comment thread rs/registry/canister/src/mutations/do_split_subnet.rs

jasonz-dfinity reviewed Feb 9, 2026

View reviewed changes

Comment thread rs/registry/canister/src/mutations/do_split_subnet.rs

cgundy and others added 7 commits February 20, 2026 10:19

chore: [MR-539] Drop redundant Callback::originator field (#8420)

e34b12f

We already check (and actually panic) if we attempt to call `SystemState::push_input()` with a message addressed to a different canister. So the `Callback::originator` field is entirely redundant.

perf: Speed up root canister integration tests (#8688)

b62ed16

Following up to #8679, performing a similar speed up in Root canister's integration tests. The results are even more dramatic in this case, a ~5x speed up, roughly going from ~25s to about ~5s for most tests.

Base automatically changed from kpop/feat_nns_interface_changes to master March 13, 2026 10:23

Merge branch 'master' into kpop/feat/nns_changes

6e29bf0

kpop-dfinity marked this pull request as ready for review March 13, 2026 10:31

github-actions Bot added the @governance label Mar 13, 2026

.

f63738c

kpop-dfinity requested review from jasonz-dfinity and pierugo-dfinity March 17, 2026 09:47

jasonz-dfinity reviewed Mar 19, 2026

View reviewed changes

kpop-dfinity and others added 4 commits March 20, 2026 13:13

Reviewer's comments

8131808

Merge branch 'master' into kpop/feat/nns_changes

5d3d89a

clippy

a62205d

clippy

511b7b4

kpop-dfinity removed the CI_ALL_BAZEL_TARGETS Runs all bazel targets label Mar 20, 2026

kpop-dfinity added 3 commits March 20, 2026 17:45

Merge branch 'master' into kpop/feat/nns_changes

6680a7f

Merge branch 'master' into kpop/feat/nns_changes

95ffcab

Merge branch 'master' into kpop/feat/nns_changes

b83940b

kpop-dfinity requested a review from jasonz-dfinity March 24, 2026 08:29

pierugo-dfinity reviewed Mar 24, 2026

View reviewed changes

kpop-dfinity and others added 2 commits March 25, 2026 16:57

Apply suggestions from code review

04a346c

Co-authored-by: Pierugo Pace <pierugo.pace@dfinity.org>

Pierugo's comments

151a19c

jasonz-dfinity approved these changes Mar 25, 2026

View reviewed changes

Comment thread rs/registry/canister/src/mutations/do_split_subnet.rs Outdated

Merge branch 'master' into kpop/feat/nns_changes

6d72ec2

pierugo-dfinity approved these changes Mar 26, 2026

View reviewed changes

flags

3c090ec

kpop-dfinity enabled auto-merge March 26, 2026 08:33

kpop-dfinity added this pull request to the merge queue Mar 26, 2026

Merged via the queue into master with commit e0a18b9 Mar 26, 2026
38 checks passed

kpop-dfinity deleted the kpop/feat/nns_changes branch March 26, 2026 10:26

Conversation

kpop-dfinity commented Feb 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Note

Uh oh!

alin-at-dfinity left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants

kpop-dfinity commented Feb 5, 2026 •

edited

Loading