Raw dependency graph data (JSON)

Vitess

Author: Ludovic HENRY ludovic.henry@qti.qualcomm.com
Date: 2026-08-30
Readiness: Orange
Scope: RISC-V (riscv64/linux) support status for Vitess
Audience: Technical leadership, resource allocation strategy
Verification policy: Every claim is cross-referenced to a primary upstream source. Items that could not be verified against a second source are marked [NEEDS VERIFICATION].

1. Project Overview

Vitess is a horizontal sharding and connection-pooling layer for MySQL, originally developed at YouTube (Google) around 2010 to handle MySQL at massive scale, open-sourced around 2012. Its core components are vtgate (query router), vttablet (per-shard tablet server), vtctld (cluster management), and VTAdmin (web UI). It is written almost entirely in Go with CGO_ENABLED=0 as the default build mode.

Governance: Vitess is a CNCF Graduated project (accepted February 5, 2018; graduated November 5, 2019), licensed Apache-2.0. Governance is meritocratic with a Steering Committee; roles progress from contributor to maintainer to Steering Committee member. Decision-making is by consensus with simple majority fallback. Governance documents are at GOVERNANCE.md and STEERING.md.

Corporate sponsors: PlanetScale is the dominant corporate sponsor. Current maintainers with confirmed PlanetScale affiliation include Florent Poinsard (@planetscale.com). Past maintainers with confirmed PlanetScale emails include Andres Taylor, Dirkjan Bussink, Rohit Nayak, Shlomi Noach, Nick Van Wiggeren, Noble Mittal, and Manan Gupta. Derek Perkins (Nozzle) and Tim Vaillancourt (independent) are also active maintainers. The project states it is “not controlled by any single company,” and CNCF graduation requires demonstrated independence.

Community culture on new ports: No formal platform tier policy exists. Platform support is implicitly defined by CI coverage. Maintainer statements on arm64 (the closest proxy for riscv64) are: “contribution welcome” (mattlord, 2023, issue #10501); “I don’t think there are any plans to add it to the testing suite” (derekperkins, 2021, issue #7600); “nobody is doing it, bugs will surely be found” (vmg, 2022, issue #9797). The path to acceptance for a new architecture is: open an issue, submit a PR with CI coverage, get maintainer review. No maintainer has publicly opposed new architectures.

2. Port History and Upstreaming Timeline

There is no riscv64 port of Vitess. The following table covers all confirmed riscv64-relevant events.

Date Event Source
Never First riscv64 commit None - zero commits matching riscv or riscv64 in the repository
Never riscv64 tracking issue opened None - zero issues or PRs mentioning riscv in vitessio/vitess
2022-03-29 PR #9830 merged: arm64 Docker images for vitess/lite (ubi8/mysql80) PR #9830
2022-03-29 docker buildx output in PR #9830 review shows linux/riscv64 in platform list PR #9830 comment by mattlord
2025-05-01 Arm engineer (ranimandepudi) commits to contributing arm64 CI and Docker work Issue #9797

Key contributors: None for riscv64. The only concrete multi-arch work is PR #9830 (arm64 Docker images) by aquarapid.

Upstream status: Not applicable - no port exists.

3. Upstream Support Tier

Vitess has no formal platform tier policy. There is no PLATFORMS.md or SUPPORT.md. Support tiers are implicitly defined by CI coverage.

Platform CI Release binaries Official Docker Status
linux/amd64 Full (all 49 workflows, oracle-vm-16cpu-64gb-x86-64) Yes (tar.gz, .rpm, .deb) Yes (amd64 only) Fully supported
linux/arm64 Partial (unit_test.yml only, mysql80, oracle-vm-16cpu-64gb-arm64) No No Nascent - multiple open requests since 2021
darwin/amd64 No No N/A Build guide exists [NEEDS VERIFICATION]
linux/riscv64 None None None Not supported, not mentioned

Evidence for arm64 CI: unit_test.yml contains arch: arm64 entries with oracle-vm-16cpu-64gb-arm64 runner, mysql80 platform only. This was confirmed by reading all 49 .github/workflows/ files.

Evidence for riscv64 absence: GitHub code search for riscv64, riscv, RISCV, and linux/riscv64 in repo:vitessio/vitess returns 0 results each. All 49 workflow files, the Makefile, and build.env contain zero riscv mentions.

4. Technical Architecture and RISC-V-Specific Subsystems

Vitess is pure Go with CGO_ENABLED=0 as the default. It has exactly two subsystems with architecture-specific assembly, both of which have complete pure-Go fallbacks that activate on riscv64.

4.1 128-bit Atomic CAS (go/atomic2/)

File Architecture Implementation ISA extensions
atomic128_amd64.s amd64 Full - native LOCK CMPXCHG16B CMPXCHG16B
atomic128_arm64.s arm64 Full - hand-tuned CASPAL (acquire/release) LDXP/STXP
atomic128_spinlock.go riscv64 (and all others) Scalar - spinlock via sync/atomic.Uint64 + runtime.Gosched() None

Build constraint for the spinlock fallback: //go:build !amd64 && !arm64. The fallback is correct but approximately 10-50x slower than a native CAS [NEEDS VERIFICATION on exact ratio].

4.2 HighwayHash (go/vt/vthash/highway/)

Used for internal row fingerprinting and checksumming.

File Architecture Implementation ISA extensions
highwayhash_amd64.s amd64 Full - SSE4.1 SSE4.1
highwayhashAVX2_amd64.s amd64 Full - AVX2 (runtime dispatch) AVX2
highwayhash_arm64.s arm64 Full - NEON (300 lines) NEON
highwayhash_ppc64le.s ppc64le Partial - VMX/AltiVec update-only VMX/AltiVec
highwayhash_ref.go riscv64 (and all others) Scalar - pure-Go generic None

Build constraint for the generic fallback: //go:build noasm || (!amd64 && !arm64 && !ppc64le). The fallback calls initializeGeneric, updateGeneric, finalizeGeneric from highwayhash_generic.go. Correct but significantly slower than the SIMD paths.

4.3 Other

  • go/hack/runtime.s: 3-line empty stub to force external linking. Architecture-neutral.
  • go/stats/prometheusbackend/: reads runtime.GOARCH as a Prometheus label string. Architecture-neutral.
  • No JIT backends anywhere in the repository.
  • No arch/riscv/ directories.
  • No CGo, no C files, no #ifdef __riscv guards.

Runtime behavior on riscv64: Vitess compiles and runs entirely via pure-Go fallbacks. HighwayHash uses highwayhash_ref.go; 128-bit atomics use atomic128_spinlock.go. No RVV, no Zba/Zbb/Zbc, no assembly of any kind executes on riscv64.

5. Build System, Cross-Compilation, and Toolchain

Vitess uses a Makefile over standard go build. There is no CMake, no Autoconf, no C build system.

5.1 Required toolchain versions

Tool Version Source Notes
Go >= 1.27.0 go.mod line 3; enforced by build.env via goversion_min() Go has supported riscv64 since 1.14
GCC/Clang Not required CGO_ENABLED=0 default No C toolchain needed for main binaries
protoc 21.3 build.env:PROTOC_VER=21.3 Hard-fails on riscv64 in bootstrap.sh - see below
etcd v3.6.7 build.env:ETCD_VER=v3.6.7 Hard-fails on riscv64 in bootstrap.sh - see below
consul 2.0.1 build.env:CONSUL_VER=2.0.1 Hard-fails on riscv64 in bootstrap.sh - see below
ZooKeeper 3.9.5 build.env:ZK_VER=3.9.5 Java-based, architecture-independent
Node.js >= 16.13.0 For vtadmin web UI only Skip with NOVTADMINBUILD=1

5.2 Cross-compilation command

The cross-build Makefile target:

cross-build:
    bash ./build.env
    mkdir -p ${VTROOTBIN}/${GOOS}_${GOARCH}
    CGO_ENABLED=0 GOOS=${GOOS} GOARCH=${GOARCH} go build \
        -trimpath $(EXTRA_BUILD_FLAGS) $(VT_GO_PARALLEL) \
        -ldflags "$(EXTRA_BUILD_LDFLAGS) $(shell tools/build_version_flags.sh)" \
        -tags "$(EXTRA_BUILD_TAGS)" \
        -o ${VTROOTBIN}/${GOOS}_${GOARCH} ./go/...

To cross-compile for riscv64:

source ./build.env
NOVTADMINBUILD=1 GOOS=linux GOARCH=riscv64 CGO_ENABLED=0 make cross-build
# Binaries land in: ${VTROOT}/bin/linux_riscv64/

No Go build tags need to be disabled for riscv64. All arch-specific code has automatic pure-Go fallbacks via Go build constraints.

5.3 bootstrap.sh blockers

bootstrap.sh (invoked by make tools) hard-fails on riscv64 for three dependencies. The install_protoc(), install_etcd(), and install_consul() functions each contain a case $(get_arch) in aarch64|x86_64|arm64) branch that exits with an error for any other architecture. Workaround:

BUILD_PROTOC=0 BUILD_CONSUL=0 BUILD_JAVA=0 ./bootstrap.sh

Then install protoc for riscv64 manually (build from source or use distro package). The dev.env file sets ETCD_UNSUPPORTED_ARCH=arm64 for aarch64; an equivalent ETCD_UNSUPPORTED_ARCH=riscv64 would be needed for riscv64.

5.4 Docker / QEMU

No Dockerfile.riscv64 exists. No riscv64 Docker support anywhere in the repository. The docker/bootstrap/build.sh hardcodes --platform=linux/amd64. The build_docker_image Makefile macro has explicit handling for arm64 only.

A riscv64 Docker image build would require modifications to docker/utils/install_dependencies.sh, which only handles amd64 and arm64 for MySQL packages. Oracle MySQL and Percona XtraBackup have no riscv64 packages; MySQL would need to be sourced from distro packages or built from source.

Notably, in the review of PR #9830 (2022), mattlord’s docker buildx ls output showed linux/riscv64 in the platform list alongside linux/arm64 - the cross-build infrastructure already supports riscv64 at the Docker layer, but no Vitess Dockerfile or CI workflow uses it.

5.5 Known build failures

No riscv64 build has been attempted or documented upstream. The Go cross-compilation path (GOOS=linux GOARCH=riscv64 CGO_ENABLED=0) is expected to succeed for the pure-Go binary layer. The bootstrap.sh failures for protoc, etcd, and consul are confirmed blockers for the development environment setup.

6. Feature Coverage and Gap Analysis vs arm64 and amd64

6.1 Functional gaps

Feature amd64 arm64 riscv64
Core vtgate/vttablet/vtctld binaries Full Full (cross-compile) Full (cross-compile, no CI)
MySQL wire protocol Full Full Full (pure Go)
gRPC inter-component RPC Full Full Full (pure Go)
Backup/restore (snappy, lz4, zstd) Full Full Full (pure-Go fallbacks)
HighwayHash row fingerprinting Full (SSE4.1 + AVX2) Full (NEON) Scalar fallback only
128-bit atomic CAS Full (CMPXCHG16B) Full (CASPAL) Spinlock fallback only
VTAdmin web UI Full Full Full (Node.js, arch-independent)
Docker images Official (amd64) Unofficial (PR #9830, ubi8 only) None
Release binaries Official None None
CI coverage Full (all workflows) Partial (unit tests, mysql80 only) None
FIPS compliance Possible (with BoringCrypto) Possible Not viable (Go FIPS+PIE broken, issue #74683 open)

6.2 Performance gaps

No riscv64 benchmark data exists for Vitess. The following gaps are inferred from the architecture-specific code analysis:

  • HighwayHash: riscv64 uses highwayhash_ref.go (pure-Go generic). amd64 uses SSE4.1 or AVX2 with runtime dispatch; arm64 uses NEON. The performance delta is significant for workloads with high row-fingerprinting throughput (e.g., online schema changes, VReplication). No measured numbers are available.
  • 128-bit atomics: riscv64 uses a spinlock. amd64 uses LOCK CMPXCHG16B; arm64 uses CASPAL. Under contention, the spinlock path will degrade. The impact depends on how frequently 128-bit CAS is exercised in production workloads. No measured numbers are available.

Data not available: riscv64 vs amd64 or arm64 throughput numbers for any Vitess workload. The upstream benchmark system (arewefastyet) runs on Equinix Metal t1.small.x86 (x86-64) only.

6.3 Security hardening gaps

FIPS compliance is not viable on riscv64 with the current Go toolchain. Go FIPS+PIE is broken (Go issue #74683, open). BoringCrypto has no riscv64 .syso. Deployments requiring FIPS compliance cannot use riscv64.

6.4 Floating-point and correctness

The open evalengine correctness bugs (section 11) are platform-agnostic - they affect all architectures equally. No riscv64-specific floating-point or NaN issues have been identified, because Vitess has never been tested on riscv64.

7. CI/CD Infrastructure

Architecture Runner Scope Status
amd64 oracle-vm-16cpu-64gb-x86-64, ubuntu-24.04 All 49 workflows Full
arm64 oracle-vm-16cpu-64gb-arm64, ubuntu-24.04-arm unit_test.yml only, mysql80 only Partial
riscv64 None None Does not exist

This was confirmed by reading all 49 .github/workflows/ files. The only non-x86 arch is arm64, and only for mysql80 unit tests. No QEMU emulation, no linux/riscv64 Docker platform target, no riscv64 runner anywhere in the repository.

RISE runners: RISE announced free native RISC-V CI runners on GitHub in March 2026 (announcement). Vitess has not adopted them. RISE has zero involvement with Vitess - no blog posts, no funded projects, no GitHub repos, not listed in the RISE wheel builder.

Other CI systems: No .gitlab-ci.yml, Jenkinsfile, .cirrus.yml, .travis.yml, .appveyor.yml, azure-pipelines.yml, or .buildkite/pipeline.yml exist in the repository.

8. Distribution and Release Status

Channel riscv64 available Notes
GitHub Releases (upstream) No All releases (v21 through v24) ship exactly 3 assets: _amd64.deb, .x86_64.rpm, .tar.gz (amd64 binary). Zero riscv64 assets across all releases.
Docker Hub (vitess/lite, vitess/vttablet) No All tags list ['amd64'] only. Confirmed across 25+ tags.
Ubuntu (any suite) No Vitess is not packaged in Ubuntu at all - not in Resolute (26.04), Jammy, Noble, or any other suite.
Debian No Not in Debian (tracker.debian.org/pkg/vitess returns 404).
Arch Linux / AUR No Not in official Arch repos or AUR.
Homebrew No Bottles exist for arm64_tahoe, arm64_sequoia, arm64_sonoma, sonoma, arm64_linux, x86_64_linux - no riscv64.
PyPI N/A Vitess is a Go project; no PyPI package exists.

What a user must do to get a working binary: Cross-compile from source using GOOS=linux GOARCH=riscv64 CGO_ENABLED=0 make cross-build with Go >= 1.27.0. The bootstrap.sh development environment setup will fail for protoc, etcd, and consul and requires manual workarounds. No pre-built binary is available from any source.

The release workflow (.github/workflows/create_release.yml + tools/make-release-packages.sh) runs on a single ubuntu-24.04 (amd64) runner with no matrix and no cross-compilation step. The Makefile supports GOARCH override for the cross-build target but this is never invoked in the release workflow.

9. Dependencies

9.1 Summary table

Dependency Role riscv64 build riscv64 test CI riscv64 release Notes
Go runtime (>= 1.27.0) Entire build toolchain Yes - secondary port Partial (secondary tier) Yes (golang-1.26-go in Ubuntu resolute/main) Miscompile bug #80127 fixed in Go 1.27; FIPS+PIE broken (#74683 open)
google.golang.org/grpc v1.83.0 All inter-component RPC Yes - pure Go No riscv64 CI Yes (Go module) No blocking issues
google.golang.org/protobuf v1.36.12 Wire format for all gRPC messages Yes - pure Go No riscv64 CI Yes (Go module) No protoc binary for riscv64 from upstream
github.com/go-sql-driver/mysql v1.10.0 MySQL wire protocol driver Yes - pure Go No riscv64 CI Yes (Go module) No runtime issues
MySQL 8.4 / 8.0 (runtime) Backend database No official Oracle support No No Oracle release Ubuntu 26.04 ships mysql-server for riscv64 from mysql-8.4 source (community build, not Oracle-supported)
go.etcd.io/etcd/client/v3 v3.7.1 Topology store Yes - pure Go No riscv64 CI No riscv64 release Maintainers stated “No plans” (PR #21510 closed); go-riscv/etcd community fork exists; Ubuntu 26.04 ships etcd-server in universe for riscv64
github.com/klauspost/compress v1.19.2 zstd/gzip/deflate/brotli compression Yes - pure Go with optional asm; riscv64 uses generic path 1 closed issue Yes (Go module) No blocking issues
modernc.org/sqlite v1.56.0 Pure-Go SQLite (VTAdmin, schema tracking) Yes - riscv64 transpile confirmed (GitLab issue #90 closed) Limited Yes (Go module) No blocking issues
github.com/prometheus/client_golang v1.24.1 Metrics exposition Yes - pure Go 4 riscv64 issues (all closed) Yes (Go module) No open blockers
github.com/shirou/gopsutil/v4 v4.26.7 System metrics Yes - Linux/riscv64 supported 16 riscv64 issues (2 open, FreeBSD-specific only) Yes (Go module) Linux riscv64 path unaffected by open issues
golang.org/x/sys v0.47.0 Syscall wrappers Yes - riscv64 fixes merged Included in Go CI Yes (Go module) 3 riscv64 issues (all closed)
github.com/cilium/ebpf v0.22.0 eBPF helpers (transitive) Yes - riscv64 Linux string added Limited Yes (Go module) 6 riscv64 issues (all closed)
github.com/golang/snappy v1.0.0 Snappy compression Yes - pure Go No riscv64 CI Yes (Go module) No issues
github.com/pierrec/lz4 v2.6.1 LZ4 compression Yes - pure Go No riscv64 CI Yes (Go module) amd64 asm bug (#20778) does not affect riscv64
github.com/cespare/xxhash/v2 v2.3.0 Fast hashing Yes - pure Go No riscv64 CI Yes (Go module) No issues
go.opentelemetry.io/otel v1.45.0 Distributed tracing Yes - pure Go No riscv64 CI (issue #8126 open) Yes (Go module) Compatibility matrix excludes riscv64
github.com/hashicorp/consul/api/v2 v2.0.0 Service discovery (optional) Yes - pure Go No riscv64 CI Yes (Go module) 6 riscv64 issues (all closed, all dependency bumps)
github.com/planetscale/vtprotobuf v0.6.1 Optimized protobuf serializer Yes - pure Go No riscv64 CI Yes (Go module) No issues
filippo.io/edwards25519 v1.2.0 Ed25519 crypto Yes - pure Go No riscv64 CI Yes (Go module) No issues
github.com/ebitengine/purego v0.10.2 CGo-free dynamic linking (via gopsutil) Yes - struct_riscv64.go + sys_unix_riscv64.s present Limited Yes (Go module) PR #397 open (trampoline generalization) - non-blocking

9.2 Critical dependency deep-dives

MySQL (runtime dependency - hard blocker): Vitess is a MySQL proxy and sharding layer; it requires a MySQL server to connect to. Oracle has no riscv64 MySQL release. The only riscv64 MySQL PR in the Oracle MySQL repository (#639) was auto-closed by the OCA bot. Ubuntu 26.04 ships mysql-server (MySQL 8.4) for riscv64 from the mysql-8.4 source package, but this is a community build, not Oracle-supported. For production deployments, this is the single most consequential dependency gap.

etcd (soft blocker for production topology): etcd is used for topology storage in production Vitess deployments. Upstream maintainers explicitly stated “No plans” for riscv64 (PR #21510 closed). A community fork (go-riscv/etcd) exists as a workaround. Ubuntu 26.04 ships etcd-server in universe for riscv64. For development and testing, the Ubuntu package is sufficient; for production, the lack of an official upstream release is a risk.

Go runtime: The struct-copy miscompile bug (Go issue #80127) that could corrupt []byte fields on riscv64 was fixed in Go 1.27 (backported to 1.25.13 and 1.26.6). Vitess requires go 1.27.0, so this is resolved. The remaining open issues are Go FIPS+PIE broken (#74683) and P256 unoptimized (#77069) - neither blocks non-FIPS deployments.

protoc: No official riscv64 protoc binary is released by Google. This blocks bootstrap.sh but does not block building Vitess binaries from a pre-generated source tree. Ubuntu 26.04 ships protobuf-compiler for riscv64 in universe.

11. Known Bugs and Active Issues

11.1 riscv64-specific bugs

None. Zero riscv64-specific issues or PRs exist in vitessio/vitess.

11.2 Architecture-specific bugs (other architectures)

ID Title Status Severity Architecture Notes
#20778 lz4 amd64 assembly decoder bug Open High - backup restore fails 4/20 runs amd64 only pierrec/lz4 v2 amd64 asm: MOVB spills only low byte across memmove call; upper 56 bits corrupted by Go >= 1.24 FSRM/ERMS-vectorized memmove. arm64 unaffected (uses pure-Go decoder). riscv64 unaffected (uses pure-Go decoder). Fix: upgrade to pierrec/lz4 v4.

11.3 Platform-agnostic correctness bugs

ID Title Status Severity Notes
#20718 evalengine/mysql/json: JSON and TIME equality wrong Open High - wrong results for IN, DISTINCT, hash joins on JSON columns Compiled vs interpreted evalengine paths disagree. Affects all platforms.
#20623 evalengine: wrong result for LIKE with NULL pattern in nested expressions Open Medium Affects all platforms.
#20750 evalengine: CAST to DECIMAL(M,D) does not round to target scale Open Medium Affects all platforms.

11.4 Recently fixed correctness bugs

ID Title Fixed Notes
#20724 mysql/json: reject nan as invalid JSON 2026-07-28 Go parser accepted nan; MySQL rejects it. Exposed compiled/interpreted path divergence.
#20722 mysql/json: read numbers the way MySQL does 2026-07-28 Vitess parsed 1e1025 as valid; MySQL rejects it.
#20475 ROUND() on DOUBLE rounds ties away from zero; MySQL rounds half-to-even 2026-07-04 ROUND(2.5e0) returns 3 in evalengine, 2 in MySQL. Root cause: Go math.Round (half-away-from-zero) vs C std::rint (IEEE round-half-to-even). Affects all platforms equally.

11.5 Open performance regressions

ID Title Status Severity Notes
#20845 OR FALSE causes 7.27x performance regression Open High Vitess does not recognize expr OR FALSE = expr; generates worse execution plan. x86-64 only (RIFT mutation testing).
#17467 vtgate planning regression for large UNION since v15 Open Medium 2x-30x planning regression (v15 to v19). linux/amd64, RHEL 7.9.

12. Objections and Upstream Blockers

12.1 Technical blockers

Blocker Severity Notes
No official Oracle MySQL riscv64 binary Hard blocker for production Ubuntu 26.04 ships a community MySQL 8.4 build for riscv64; sufficient for