# wasmCloud v2.12.2

2026-11-17 · [Original release notes](https://github.com/wasmCloud/wasmCloud/releases/tag/v2.12.2) · [Full summary](https://ratatosk.io/en/releases/wasmcloud/v2.12.2)

Security 8 · Breaking changes 8 · Important fixes 14 · Enhancements 4 · Dependency updates 10 · Other changes 26

## Security (8)

### Plugin volume refusals excluded from `host_path_denied`

- Affects: setups using plugin volumes and the `host_path_denied` metric

A refused plugin volume could previously increase the metric, making its count include refusals as well as workload policy denials.

> * fix: plugin volume follow-ups from #5673 by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5995

### `reserved_paths` reports plugin resolution errors

- Affects: plugin configurations in `wash host`

When the plugin list failed to merge or bindings did not resolve, `reserved_paths` silently omitted plugin wasm files or binding credential files from reserved paths. Those failures now return an error instead of leaving the paths unreserved.

> * fix(wash-runtime): register plugin volumes only once a plugin loads by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5997

### wash-runtime rejects invalid PostgreSQL `path` point counts

- Affects: PostgreSQL `path` values decoded by wash-runtime (Inferred)

Previously, a negative point count could panic, and a large count could abort the host while allocating, when wash-runtime decoded a PostgreSQL `path`.

> * fix(wash-runtime): harden postgres geometry and fixed-width values by @vados-cosmonic in https://github.com/wasmCloud/wasmCloud/pull/5953

### Plugin volume protection now follows opened directories

- Affects: host component plugins configured with volumes

Previously, `wash host` computed plugin-volume protection separately from the directories plugins actually opened. The loader now protects opened directories through the shared engine policy; volumes unused by any `volumeMount` are not opened or exposed to workloads and are named in a warning.

> * fix: plugin volume follow-ups from #5673 by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5995

### Go HTTP/1 server crash vulnerability fixed

- Affects: setups using the `runtime-operator` or `runtime-gateway` components
- [GO-2026-6859](https://osv.dev/vulnerability/GO-2026-6859)

An HTTP/1 server vulnerability in `net/http` could crash either module and was reachable through the gateway's public HTTP/1 listener. The builder images and toolchains now use Go 1.27.4, which govulncheck reports as vulnerability-free for both modules.

> * fix: bump Go to 1.27.4 and validate tools-bump release tags by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5996

### PostgreSQL `lseg` decoding fixed in wash-runtime

- Affects: wash-runtime's PostgreSQL capability for `lseg` and `lseg[]` queries

A workload with this capability could trigger the old failure by querying an `lseg` cast, without table access: no `lseg` value decoded successfully, and a negative point count could panic while a huge count could abort the host during memory reservation. Valid `lseg` and `lseg[]` values now decode from four coordinates; other encoded lengths return an error.

> * fix(wash-runtime): decode postgres lseg columns without a point count by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5998

### wash-runtime rejects PostgreSQL `line` and `line[]` values

- Affects: workloads using wash-runtime's PostgreSQL capability (Inferred)

Before this change, a workload with the PostgreSQL capability could trigger a panic or host-process abort by querying an `lseg`, even without table access.

> * fix(wash-runtime): decode postgres lseg columns without a point count by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5998

### wash-runtime PostgreSQL p2 decode-error panic fixed

- Affects: wash-runtime PostgreSQL p2 binding

When `into_result_row` encountered a column decode error, the binding panicked because it used `Row::get`. It now returns `query-error::unexpected` instead of panicking.

> * fix(wash-runtime): decode postgres lseg columns without a point count by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5998

## Breaking changes (8)

### Linux process-memory guard now counts page-table memory

- Affects: Linux hosts using the `max-process-memory` guard

Previously, the guard omitted page-table memory, so a dense pool could be killed by the kernel while the guard still read below its limit. The revised reading can reach the configured limit earlier and refuse further starts or growth, leaving dense pools with fewer warm instances at that limit.

Evidence: “The reading goes up for every Linux host, by its page tables, which are usually small beside its anonymous memory. On a dense pool it's about 10%, so a host sized to its limit holds about 10% fewer warm instances than main claims.” https://github.com/wasmCloud/wasmCloud/pull/5935

> * fix(wash-runtime): count page tables in the Linux process footprint by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5935

### wash-runtime derived memory pool bounds

- Affects: wash-runtime setups using derived memory pools

Previously, a 2 MiB layout on macOS could derive 33.5 million memories, whose 64.5 TiB stack pool could not fit in the address space and caused the engine to fail to build. Now, containers below about 500 MiB on Linux or 800 MiB on macOS at Wasmtime's layout derive fewer memories (5,242 at 256 MiB), with a floor of 1,000.

Evidence: “A new bound caps the count at 1% of the memory limit, at 512 bytes a memory. At wasmtime's layout it binds only below about 500 MiB of memory (Linux) or 800 MiB (macOS). **Behaviour change:** a container smaller than that derives fewer memories (5,242 at 256 MiB), never below the 1,000 floor.” https://github.com/wasmCloud/wasmCloud/pull/5917

> * bench: add cargo xtask scale-density presets and bound the derived pool by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5917

### `WorkloadDeployment` condition updates batched per reconciliation

- Affects: runtime-operator `WorkloadDeployment` resources

During rollout reconciliation, stale cache state could cause duplicate ReplicaSets or Workloads, including after a create succeeded but its status write failed. The controller now checks the API reader before treating a missing current ReplicaSet as deleted and uses stable rollout and replica-slot names with ownership, desired-spec, and deletion checks to prevent duplicates.

Evidence: “- Opt the WorkloadDeployment controller into batching ordered condition status changes. Keep existing defaults for other controllers and preserve hard error and explicit skip boundaries.” https://github.com/wasmCloud/wasmCloud/pull/5957

> * fix(runtime-operator): avoid cache-lag rollout churn and empty status writes by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5957

### Per-workload quota lifetime now follows active connections

- Affects: runtime per-workload quotas for HTTP or socket connections

Previously, stopped workloads' quota entries stayed in an idle cache, and HTTP or socket connection slots did not keep quotas alive. Now, a live connection keeps its quota registered even if its client is replaced or the workload stops; after all slots and handles are gone, a later quota starts its grant and refusal counts at zero.

Evidence: “The counts cover this quota's current lifetime. Once a guest holds no + /// slot and every handle to its quota is dropped, the registry forgets it, + /// and the next [`QuotaRegistry::for_guest`] starts again from zero.” https://github.com/wasmCloud/wasmCloud/pull/5980

> * fix(runtime): release workload quotas with their last user by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5980

### Process-memory guard keeps refusals active during unreadable accounting

- Affects: setups using the process-memory guard

Previously, after `MAX_FAILED_READS` consecutive failed samples, the guard returned to normal and stopped enforcing the limit while readings remained unavailable. Now, an unreadable initial footprint or that run of failed samples keeps new starts and growth refused until a valid reading returns; the last measured pressure state is retained.

Evidence: “If the initial footprint is unavailable, or [`MAX_FAILED_READS`] periodic +//! samples fail consecutively, new starts and growth are refused until a valid +//! reading returns. The last measured pressure state is retained; availability +//! of accounting is exposed separately by [`ProcessMemoryGuard::is_readable`].” https://github.com/wasmCloud/wasmCloud/pull/5935

> * fix(wash-runtime): count page tables in the Linux process footprint by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5935

### wash-runtime enforces matching PostgreSQL types for fixed-width values

- Affects: wash-runtime PostgreSQL bindings for `Uuid`, `Point`, `Box`, `Lseg` and `PgLsn` values (Inferred)

Bindings to a different PostgreSQL type now fail with a type error, even when the types have the same width.

> * fix(wash-runtime): harden postgres geometry and fixed-width values by @vados-cosmonic in https://github.com/wasmCloud/wasmCloud/pull/5953

### Outbound HTTP semaphore accessor replaced with connection-slot API

- Affects: Rust integrations using `outbound_http_permits()` (Inferred)

The pooled client previously used `outbound_http_permits()` directly as an `Arc<Semaphore>`. The returned slot from `acquire_outbound_http(wait)` keeps the workload quota registered until the connection closes.

Evidence: “Replace the public raw HTTP semaphore accessor with an async connection-slot API that preserves the quota lifetime.” https://github.com/wasmCloud/wasmCloud/pull/5980

> * fix(runtime): release workload quotas with their last user by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5980

### Plugin-volume validation rejects invalid mount definitions and `emptyDir`

- Affects: plugin-volume configurations with non-absolute or dotted mount paths, undeclared mount names, duplicate volume names or mount paths, or `emptyDir: {}`

Mount paths must now be absolute and contain no `.` or `..`; mounts must name declared volumes, with unique volume names and mount paths. `emptyDir: {}` was reported as an unknown field and is now refused as an unsupported volume.

Evidence: “`to_spec_unresolved`, and so `wash config validate`, checks plugin volumes: an absolute mount path with no `.`/`..`, mounts naming declared volumes, unique volume names and mount paths.” https://github.com/wasmCloud/wasmCloud/pull/5995

> * fix: plugin volume follow-ups from #5673 by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5995

## Important fixes (14)

### Runtime-generated UUIDv7 IDs now use OS-seeded state

- Affects: setups using runtime-generated UUIDv7 IDs for hosts, components, contexts, or stores

Previously, seeding global `fastrand` could make IDs repeat across processes, while hosts starting on the same node and clock tick could generate identical IDs, including the default host ID in NATS control subjects. Application seeding can no longer steer ID generation, though the first ID on each process and thread may block on `getrandom` during early boot until the kernel entropy pool is ready.

> * fix(runtime): isolate UUIDv7 PRNGs with OS-provided seeds by @vados-cosmonic in https://github.com/wasmCloud/wasmCloud/pull/5945

### Duplicate child creation in runtime-operator fixed

- Affects: `WorkloadDeployment` resources managed by runtime-operator (Inferred)

A newly created `ReplicaSet` missing from the cache could be treated as deleted, causing replacements on each reconciliation; a successful create followed by a failed status write could also produce duplicates. The controller now checks ownership and desired state and uses stable names to prevent duplicates in these cases, while preserving existing random-name children.

> * fix(runtime-operator): avoid cache-lag rollout churn and empty status writes by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5957

### Second `wash` provenance filename added to GitHub Releases

- Affects: tools that consume `wash` binary provenance from GitHub Releases

Previously, provenance was attached only as `wash.sigstore.json`, so tools that look for `.intoto.jsonl`—including OpenSSF Scorecard’s Signed-Releases check—did not count it. The new `wash.intoto.jsonl` file contains the same provenance bundle as `wash.sigstore.json`.

> * ci(release): attach binary provenance as wash.intoto.jsonl by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5942

### Linux linear-memory pool sizing now reflects host address space

- Affects: 64-bit Linux hosts using derived linear-memory pool sizing

Previously, the fixed 128 TiB assumption derived too few memories for 4-level aarch64 address spaces and overstated address space on 3-level aarch64 kernels. Linux now sizes from the top of the `[stack]` mapping in `/proc/self/maps`: 2^47 on x86_64, 2^48 on 4-level aarch64, and 2^39 on 3-level aarch64; it falls back to the constant elsewhere or when the mapping cannot be read or is outside the stated bounds.

> * perf(wash-runtime): read the user address space from the kernel by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5934

### Automatic memory pool allocation fallback detection fixed

- Affects: automatically sized memory pools with counts above the fallback floor

When a pool mapping allocation failed on Windows, the previous error check tried to match rustix `NOMEM` on every platform. The check now uses rustix `NOMEM` only on Unix and recognizes Windows mapping failures as `std::io::Error`s for fallback handling.

> * fix(wash-runtime): check rustix errno for pool OOM only on unix by @vados-cosmonic in https://github.com/wasmCloud/wasmCloud/pull/5949

### Component-plugin type-only imports no longer require providers

- Affects: host component plugins with nonempty imports containing only structural types, including record-only interfaces

Previously, a plugin could fail to load when a native advertised an imported record-only interface but declined its binding; without that native, declaring `wasmcloud:host/workload-call` could turn those records into an empty workload callback and advertise them as a capability. These imports are now omitted from workload-callback classification, while native claim and grant checks remain unchanged.

> * fix(wash-runtime): resolve type imports without capability providers by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5932

### wash-runtime `lseg` and `lseg[]` parameter encoding fixed

- Affects: wash-runtime PostgreSQL support for `lseg` and `lseg[]` parameters (Inferred)

When wash-runtime sent these parameter types, the `LineString` encoder produced a `path` that PostgreSQL refused; they are now encoded as four endpoint `float8` coordinates.

> * fix(wash-runtime): harden postgres geometry and fixed-width values by @vados-cosmonic in https://github.com/wasmCloud/wasmCloud/pull/5953

### wash-runtime rejects PostgreSQL `line` and `line[]` parameters

- Affects: wash-runtime PostgreSQL support for `line` and `line[]` parameters (Inferred)

When encoding these parameters, the previous path treated them as a two-point `LineString`, although PostgreSQL represents a line with coefficients of `Ax + By + C = 0`. Encoding them now returns an error stating that `line & line[] are not supported`.

> * fix(wash-runtime): harden postgres geometry and fixed-width values by @vados-cosmonic in https://github.com/wasmCloud/wasmCloud/pull/5953

### runtime-operator stops sending empty status patches

- Affects: Ready Workloads managed by runtime-operator (Inferred)

Previously, reconciliations sent empty status patches when computed status had not changed, consuming API capacity for each Ready Workload. Actual status changes still produce patches.

> * fix(runtime-operator): avoid cache-lag rollout churn and empty status writes by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5957

### `Host` finalization no longer blocks replacement `Host` reconciliation

- Affects: runtime-operator `Host` finalization and `Workload` cleanup (Inferred)

Previously, when a `Host` finalizer occupied the controller’s only worker while deleting many `Workload`s, a replacement `Host` stayed unready until cleanup finished. Cleanup can now run alongside replacement `Host` reconciliation, with two workers sharing a limit of 24 concurrent deletes; deletes check identity and assignment on conflicts, and failures retain the finalizer for retry.

> * fix(runtime-operator): let replacement hosts reconcile during cleanup by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5960

### Workload mount paths validated at API admission

- Affects: Workloads and workload templates that use `volumeMounts`

Previously, the API admitted relative paths and paths containing `.` or `..`, but the host refused to start those workloads. The API now rejects them when applied; on Kubernetes 1.30+, an update may leave an existing invalid value unchanged, while older clusters reject any update to an object with such a value.

> * fix: plugin volume follow-ups from #5673 by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5995

### Host controller concurrency tuning honored

- Affects: Host GroupKind or manager-wide `MaxConcurrentReconciles` concurrency settings

Previously, concurrency was hard-coded at two, so configured worker counts had no effect. A Host GroupKind value now takes precedence over the manager-wide value; otherwise, the manager-wide value applies, with a minimum of two workers.

> * fix(runtime-operator): honor manager tuning for Host workers, keeping at least two by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5961

### Workload placement responds to Host availability

- Affects: unplaced runtime-operator Workloads with a matching schedulable Host, based on `hostSelector`, Environment, and any pinned-host ID

Previously, after no suitable Host was available, a Workload could remain unplaced until the 30-second reconcile interval or about 60 seconds through the ReplicaSet grace period, while its HTTP endpoint returned 404 `no workload bound to host`. A Host placement-change event now wakes eligible Workloads for reconciliation; heartbeat status writes do not trigger the watch.

> * fix(runtime-operator): place waiting Workloads when a Host is schedulable by @officialasishkumar in https://github.com/wasmCloud/wasmCloud/pull/5922

### Plugin volume reservations now follow plugin lifetime

- Affects: host component plugins with volume directories

If a plugin failed after opening its volumes—for example, because a port was already held or a component did not compile—or was dropped, its directories could stay closed to workloads, including when an engine was reused without it. Reservations are now made only after plugin construction completes and are released when the plugin is dropped; directories shared by plugins stay reserved until all of them are dropped.

> * fix(wash-runtime): register plugin volumes only once a plugin loads by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5997

## Enhancements (4)

- New `process_memory.readable` gauge for the process-memory guard
- Opt-in allocator selection for `wash` and `scale_density` (To use: The `mimalloc` or `jemalloc` Cargo feature for `wash`, or the corresponding feature for the `wash-runtime` `scale_density` benchmark)
- Optional healthy-object polling intervals in runtime-operator
- Host-directory mounts for component plugins (To use: Configure the plugin’s `volumes` and `volume_mounts` declarations)

## Dependency updates (10)

### zstd dependencies updated

- [RUSTSEC-2026-0391](https://osv.dev/vulnerability/RUSTSEC-2026-0391), [RUSTSEC-2026-0398](https://osv.dev/vulnerability/RUSTSEC-2026-0398), [RUSTSEC-2026-0403](https://osv.dev/vulnerability/RUSTSEC-2026-0403)

The lockfile now pins `zstd` to 0.13.4 and `zstd-safe` to 7.4.0; Wasmtime is unchanged.

> * fix(deps): update zstd to clear RUSTSEC-2026-0391, 0398 and 0403 by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5941

### Updated `quinn-udp` and runtime-operator OpenTelemetry dependencies

- [GHSA-q7gw-24fr-5p69](https://github.com/advisories/GHSA-q7gw-24fr-5p69), [RUSTSEC-2026-0538](https://osv.dev/vulnerability/RUSTSEC-2026-0538)

`quinn-udp` is now 0.5.16. The runtime-operator’s `otel`, `otel/sdk`, `otel/trace`, and `otel/metric` modules are now 1.47.0, and `otel/log` is added at 1.47.0.

> * fix(deps): bump quinn-udp and otel sdk past RUSTSEC-2026-0538 and GHSA-q7gw-24fr-5p69 by @Aditya1404Sal in https://github.com/wasmCloud/wasmCloud/pull/5982

### `rustix` updated to 1.1.6

- [RUSTSEC-2026-0486](https://osv.dev/vulnerability/RUSTSEC-2026-0486)

Only the `rustix` entries change in the workspace and `examples/grpc-hello-world` lockfiles.

> * fix(security): resolve open code scanning alerts by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5991

7 more, none mentioning security

- `* ci: meet OSPS Baseline maturity level 3 by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5936`
- `* fix(deps): update wasi-gfx-runtime by @MendyBerger in https://github.com/wasmCloud/wasmCloud/pull/5954`
- `* fix(deps): update wasi-gfx-runtime by @MendyBerger in https://github.com/wasmCloud/wasmCloud/pull/5959`
- `* chore: upgrade NATS images to 2.12.15 by @ricochet in https://github.com/wasmCloud/wasmCloud/pull/5955`
- `* chore(deps): Bump the all-github-actions group across 4 directories with 12 updates by @dependabot[bot] in https://github.com/wasmCloud/wasmCloud/pull/5988`
- `* chore(deps): Bump the all-go group across 1 directory with 4 updates by @dependabot[bot] in https://github.com/wasmCloud/wasmCloud/pull/5983`
- `* chore(deps): Bump the all-docker group across 3 directories with 3 updates by @dependabot[bot] in https://github.com/wasmCloud/wasmCloud/pull/5967`

## Other changes (26)

- Benchmark-host tooling adds EC2 arm64 support
- Benchmark publishing adds the `million_warm` preset in its own `density-max` group
- Benchmark, comparison, and k6bench steps wait for the shared `bench-box` lock
- OpenSSF project badges added to the README
- CI workflows updated to `kind-action` v1.15.2
- Repository security checks expanded
- Audit CI installs `osv-scanner` v2.7.0 through the Go module proxy
- Benchmark scale-density presets and pool-count cap
- `max-scale` preset for `scale_density` benchmarks
- Workload components boxed to reduce heap use
- Boxed `InstanceJob::Guest` to reduce channel-slot memory
- `wasi:random/random` now uses the thread generator
- Component plugin maps shared across stores
- Context IDs now reuse store IDs
- Rust CodeQL job Cargo settings updated
- Benchmark result publishing and reporting fixes
- Code-scanning remediation deadlines by severity
- Benchmark pre-flight check now fails on leftover processes
- `logind` provisioning tag for a systemd-logind drop-in
- Compile caching for components without a digest
- Controller lint cleanup reuses a test constant and extracts ReplicaSet deletion into a helper
- Dependency-review license gate allows `LicenseRef-scancode-google-patent-license-golang`
- Short-lived tokens for `tools-bump` pull requests
- Postgres conversion test failures identify cases by type
- `wash` OSV-Scanner ignore added for `astral-tokio-tar`
- `tools-bump` rejects invalid upstream release tags

---

Summaries are produced by AI, so they can be wrong, and they are provided without warranty.
