ic-metrics

Allocation-free, dependency-free measurement arithmetic.

The no_std library exports record_sample for consumer-owned count/total fields and MeasurementSummary for samples, total, latest and maximum. MeasurementHistogram adds a fixed-size distribution with caller-selected bounds. Zero is a valid observation; an empty summary has no latest or maximum value. Count and total saturate independently. Consumers establish units, identity and reset boundaries before comparing observations; saturated totals cannot supply exact interval arithmetic.

use ic_metrics::MeasurementSummary;

let mut summary = MeasurementSummary::EMPTY;
assert_eq!(summary.latest(), None);
summary.record(0);
assert_eq!(summary.samples(), 1);
assert_eq!(summary.latest(), Some(0));

Published 0.2.6 adds checked_mean(samples, total) and MeasurementSummary::mean(). Both return Result<Option<u64>, MeasurementMeanError>: empty data is None, measured zero is Some(0), and nonempty unsaturated means use floor division. A nonzero total without samples or either counter at u64::MAX returns a typed error, including an exactly reached cap. The raw-field function supports consumer-owned report shapes without constructing an accumulator.

The application guide is included in packaged crate documentation, with a compiled example of fixed named histograms, separate units and event counts. It covers sample admission, async attribution, resets, reporting and cost qualification; applications own those policies.

For a distribution, construct MeasurementHistogram::new([10, 100]). Its disjoint buckets cover 0..=10 and 11..=100; overflow() counts larger observations. bucket_counts() returns those two counts, and summary() includes every observation. Duplicate or descending bounds return a typed HistogramBoundsError identifying the invalid bound. Construction and recording support constant evaluation. Storage is fixed, recording searches at most the configured number of bounds, and every bucket saturates independently. Buckets describe ranges; they do not provide exact percentiles. Units, thresholds and cumulative export remain consumer-owned. The histogram API is published in 0.2.4.

Known callers are IcyDB, Canic, ic-timers, ic-backup, IC Blob Storage's restoration test probe and Toko Miner's production game-shard action metrics. Their attribution, registries, callback roles, replication, persistence and reporting remain local. ic-backup's host integration measures local durations in nanoseconds and prepared chunk sizes in bytes, with a four-bound byte histogram in the inspected implementation. Its diagnostics establish no IC instruction or cycle cost. The crate has no global consumer registry, stable-memory allocation or endpoints. Consumers own their summary and histogram instances, labels and persistence policy. The extraction contract records scope and evidence.

The published 0.2.0 hard cut removes the ic feature and call_context_instructions. Platform reads belong in consumer adapters: IcyDB and Canic use ic_cdk::api::call_context_instruction_counter(); IC Timers uses its existing ic0::performance_counter(1) binding. Both read counter 1. Consumers own target gating, native handling, call-context identity and attribution. IcyDB's inclusive overlapping spans and Canic's exclusive endpoint accounting remain different consumer contracts.

Verified 0.2.13 publication is bound in the release record. The current handoff distinguishes local source, remote/publication state, locks and owning CI. Consumers exposing MeasurementSummary in public APIs must coordinate their 0.2 dependency identity. Stored data, reports and endpoints are unchanged. Historical reader execution remains in the source-bound evidence and tagged releases; current CI qualifies arithmetic and repository tooling.

Measuring cost

Measure raw canister Wasm bytes, IC instructions and actual cycle charges separately. The shared arithmetic allocates nothing; labels, map updates and report encoding are consumer costs. Build profiles belong to the consumer workspace. The performance audit records historical compiler checks and size/instruction/cycle comparisons for the retired reader test canister, with their workload limits. It does not establish consumer savings.

The separate consumer lookup measurements cover IcyDB and Canic's borrowed-key changes. Their isolated IC fixtures show lower instruction work and cycle charges for repeated keys, but higher costs for first insertions. Use the recorded workload and artifact identities when assessing those consumer changes; upgrading ic-metrics alone does not apply them.

The two-bound histogram cost experiment compares Canic's source-copied recording path with count/total, summary and histogram storage. It measures IC instructions and actual cycle charges in Linux PocketIC, plus raw Wasm bytes and target-specific slot sizes. Its repeated-key, two-bound workload does not establish whole-application cost or histogram adoption. Its original replay inputs are unavailable. The separate durable arithmetic replay provides a checksum-bound bundle of new fixture sources, locks, raw results and measured Wasms. It measures direct arithmetic without Canic lookup or attribution; its observations have their own source and workload identity.

Development

Install rustup, Git, GNU Make, Bash 3.2 or newer and a SHA-256 utility. The pinned toolchain is Rust 1.99.0 and the MSRV is 1.88.0. Install wasm32-unknown-unknown for Wasm checks. make help lists focused commands; select named tests during implementation. Full make ci requires an explicit request outside configured CI.

Prepare the pinned checkout-local Cargo tools during developer setup:

make install-rust-tools
make rust-tools-check
make install-hooks

make fmt sorts every workspace manifest before Rust formatting. make fmt-check checks both without changing files; CI installs the same pinned formatter. Run make install-hooks once per clone and after updating shared tooling. The repository-local pre-commit hook formats the staged snapshot, refreshes selected files, rejects partial staging and preserves unrelated working edits. It refuses to replace existing hook installations. make hook-check exercises these contracts in scratch repositories without creating commits.

See agent rules, host support and the current handoff. Reviewed shared tooling is vendored; normal checks and release scripts need no Shared Tooling sibling checkout.

Prepare the reviewed local executables explicitly:

make install-tools
make tools-check
export PATH="$PWD/.tools/host/bin:$PWD/.tools/ic/bin:$PWD/.tools/rust/bin:$PATH"

The setup installs jq, Mike Farah yq, ripgrep with PCRE2 and cloc plus Quill, ICP CLI, didc, ic-wasm, PocketIC and Binaryen, plus cargo-sort, cargo-sort-derives and candid-extractor, under this checkout's .tools/. Rust tools use exact pins and locked Cargo installation; interrupted setup can leave earlier tools installed and retains build output for inspection. Archive installers retain previous and failed candidates. tools-check is offline; ordinary validation never installs tools. System bootstrap packages and installation behavior are documented in local setup.

The reviewed shared Make include also supplies make cloc for this workspace's Rust runtime/test report and make cloc-tooling for sibling tooling inventory. Reports use the prepared local tools. The tooling inventory separates matching shared snapshots from local code and supporting data; counts guide review and do not establish equivalent behavior or performance.

make check-pins checks dependency and workflow declarations offline using Git and the prepared local parsers. Their single reviewed pin owner is ci/tool-versions.env; IC executable pins live in ci/ic-tools.tsv. Local setup and focused commands are in host support. The library has no Cargo dependencies. Tracked lockfiles preserve workspace identity. Dependency changes must prepare every affected independent workspace graph before release validation. Checks never upgrade dependencies or install tools implicitly.

Releases and adoption

The public repository is dragginzgame/ic-metrics. The changelog preserves the initial 0.1.0 scaffold, the 0.1.1 arithmetic release and later tooling releases. ic-metrics 0.2.13 is published on crates.io with tag v0.2.13. Declare the published release in the consumer's root dependency catalog:

[workspace.dependencies]
ic-metrics = "0.2.13"

Members inherit with ic-metrics = { workspace = true }. Consumers using MeasurementHistogram require ic-metrics = "0.2.4" or a later compatible published minimum in the root catalog. Either mean projection requires at least published 0.2.6. Published consumers must resolve the registry package rather than require a sibling checkout. The qualification issue links each consumer's integration evidence. Native qualification is recorded against its actual revision in the host record.

People and agents contribute through topic branches and pull requests under the contribution rules. An explicit PR request includes the agent's scoped commits, branch push and PR creation. Ordinary fixes and continuation authorize local work. Merging and direct integration-branch pushes require separate authority and required checks.

Explicitly requested make release-patch, release-minor and release-major use the same reviewed workflow, with RELEASE_REMOTE=origin and RELEASE_BRANCH=main and RELEASE_DELIVERY=direct. This repo rejects PR release delivery before dispatch; adopting it requires merged-checkout adapters and their own qualification. They require cargo-edit (cargo set-version) and the prepared pinned Cargo tools and rustfmt, run the complete offline gate, prepare only Cargo metadata and release notes, commit, tag and atomically push. They do not publish the crate. Normal release targets reconcile an interrupted release at its saved commit. Newer committed fixes or a different requested increment then receive fresh preflight and full validation before the next release is prepared. Failed validation logs remain under the Git directory's release-state/validation-failures/; earlier attempts are retained. make release-resume VERSION=X.Y.Z remains available for explicit selection; identity, payload and destination conflicts stop recovery. The 0.2.11 runner also rechecks the index, working payload and exact tag after the final consumer check. Completed direct recovery confirms local and remote tag identity and branch ancestry before reporting success; unavailable observations or conflicts stop without another push. The released Shared Tooling 0.1.25 snapshot also refreshes the matching local tracking ref from confirmed delivery, checking its type under Git's update lock. Unrelated, newer or symbolic refs are preserved; an optional refresh failure reports a fetch remedy without repeating delivery. Native qualification of this consumer batch remains pending in the release record.

For a subsequent committed package release, the maintainer can run:

make publish-check
make publish

publish-check delegates to Cargo's dry run. publish uploads only ic-metrics to crates.io, using the existing package version, locked dependencies and normal Cargo credential setup (cargo login --registry crates-io). Both retain Cargo's package verification and dirty-tree rejection. Publication creates no Git commit or tag, does not bump versions and preserves build artifacts. See Cargo's publication documentation.

cargo package -p ic-metrics --locked --offline --allow-dirty prepares and verifies a local development archive without uploading it. The package includes the canonical MIT license text, source and this README; documentation links resolve to the public repository. The manifest restricts publication to crates.io.