refactor: Hold the whole consumed-cycles total in SubnetMetrics

[mraszyk]

SubnetMetrics::consumed_cycles_by_canisters holds only the canisters' part of
the total; every consumer adds the subnet-level aggregate on top, recomputing
that part each time it reads. That is fine for a consumer that only ever looks
at a committed state -- the certified tree and the gauge -- but it makes the
combined value ill-defined mid-round: the canisters' part is as of the last
commit, while the subnet-level accumulators move as the round proceeds. A
delete_canister adds the deleted canister's consumption to
consumed_cycles_by_deleted_canisters at once, so between that and the next
commit the two parts count the same canister twice.

Hold the whole total instead. consumed_cycles_by_canisters becomes
consumed_cycles_total_including_canisters, still transient, now covering the
subnet-level aggregate as well. The field is private, with a getter of the same
name and a single writer, SubnetMetrics::refresh_consumed_cycles, which takes
the canisters' part and adds consumed_cycles_total() on top: callers hand over
that part only, so the sum lives in one place and none of them can get it wrong.
The certified state tree at /subnet/<subnet_id>/metrics (from certification
version V29) and the replicated_state_consumed_cycles_since_replica_started
gauge both read the getter, so they cannot drift and any future consumer gets a
self-consistent value whenever it reads.

ReplicatedState::refresh_consumed_cycles, renamed from
refresh_consumed_cycles_by_canisters now that it publishes more than the
canisters' part, hands the canister states' sum over on every
commit_and_certify, and new_from_checkpoint does the same on load, so a
replica restarting from a checkpoint agrees with one that keeps running --
which consumed_cycles_total_is_the_same_across_a_restart pins, comparing a
live state against one reloaded from the same canisters and subnet metrics.

The certified encoding and the state hash are unchanged: the byte-exact
expectations in encoding/tests/compatibility.rs and the V29 hash in
state_manager/src/tree_hash.rs still hold, with their fixtures now handing
their canisters' part to refresh_consumed_cycles rather than writing the
aggregate themselves. test_traverse_subnet_metrics_includes_canister_consumed_cycles_at_v29
now also consumes subnet-level cycles, so the total it pins covers both parts
rather than only the canisters' one. The gauge does now depend on the refresh
having run, which in production it has, as commit_and_certify enqueues the
observation after it; the scheduler metrics tests, which never commit a state,
refresh explicitly through the new observe_state_metrics helper.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[Copilot]

Pull request overview

Copilot reviewed 12 out of 12 changed files in this pull request and generated 1 comment.

[zeropath-ai]

✅ No security or compliance issues detected. Reviewed everything up to 0878752.

Security Overview

• 🔎 Scanned files: 12 changed file(s)
• 🔗 Scan Link: https://zeropath.com/app/repositories/f5e9d2bb-29a4-4b64-902a-391d65fd8b2d?scanId=41682ee0-6c2c-4f8e-9fb9-433bcac4158c&codeScanTypes=PrScan&tab=issues

Detected Code Changes

Change Type

Relevant files

Refactor

► rs/canonical_state/src/encoding/tests/compatibility.rs     Update commentary and value semantics for subnet metrics in compatibility tests ► rs/canonical_state/src/encoding/types.rs     Refactor consumed cycles total computation to use new including-canisters field ► rs/canonical_state/src/lazy_tree_conversion.rs     Adjust blob description to reflect new consumed cycles handling ► rs/canonical_state/src/traversal.rs     Update test to use refreshed consumed cycles total including canisters ► rs/execution_environment/src/scheduler/tests/metrics.rs     Introduce observe_state_metrics helper to refresh and record metrics ► rs/replicated_state/src/metadata_state.rs     Rename and refactor consumed cycles fields and related docs to include total including canisters ► rs/replicated_state/src/metadata_state/proto.rs     Align proto to new consumed_cycles_total_including_canisters field ► rs/replicated_state/src/replicated_state.rs     Derive total including canisters from canisters during load/commit; rename refresh method ► rs/state_manager/src/lib.rs     Refresh consumed cycles on state before hashing during checkpoint load ► rs/state_manager/src/tree_hash.rs     Initialize refreshed consumed cycles for tests