TimescaleDB

Author: Ludovic HENRY ludovic.henry@qti.qualcomm.com
Date: 2026-08-29
Readiness: yellow
Scope: RISC-V (riscv64/linux) support status for TimescaleDB
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

TimescaleDB is a PostgreSQL extension that adds time-series storage and query capabilities to PostgreSQL. It implements automatic partitioning (hypertables), columnar compression, continuous aggregates, and a vectorized aggregation engine on top of standard PostgreSQL infrastructure. It is written in C and requires PostgreSQL 13 or later.

Governance: TimescaleDB has no independent foundation. It is a corporate open-source project wholly owned by Timescale, Inc., which rebranded as TigerData on 2025-06-17 (legal entity remains “Timescale, Inc., d/b/a Tiger Data”). There is no Apache Software Foundation, CNCF, or other neutral foundation oversight.

License: Dual-licensed. The core engine is Apache 2.0. Enterprise features ship under the Timescale License Agreement (TLS), which is source-available but not OSI-approved. Debian and Ubuntu package only the Apache 2.0 core (+dfsg repack removes TLS-only code).

Active maintainers (derived from recent commit history; no MAINTAINERS or CODEOWNERS file exists in the repository): akuzm (Alexander Kuzmenkov), svenklemm (Sven Klemm), dbeck, Poroma-Banerjee, gayyappan – all Timescale/TigerData employees. No external corporate co-maintainers are identified.

Community stance on new ports: No public discussion of RISC-V porting exists. The project is tightly controlled by TigerData engineers with no history of community-driven architecture expansions. New platform support would require either a TigerData business case or a well-maintained community PR. Because TimescaleDB is a PostgreSQL extension in C, a RISC-V port is technically straightforward – it primarily depends on PostgreSQL supporting RISC-V on the target OS, which PostgreSQL does – but no such work is currently prioritized or planned.

RISE membership: Timescale/TigerData is not a RISE project member. The RISE member list (8 Premier + 11 General members as of 2026-08) does not include any Timescale entity.

2. Port History and Upstreaming Timeline

Date	Event	Source
2025-08-27	Debian Trixie ships `postgresql-17-timescaledb` 2.19.3+dfsg-1+deb13u1 for riscv64	Debian package tracker
2026-08-19	Debian sid ships `postgresql-18-timescaledb` 2.29.2+dfsg-1 for riscv64	Debian pool
(ongoing)	Ubuntu 26.04 (resolute) ships `postgresql-18-timescaledb` 2.25.1+dfsg-1 for riscv64	packages.ubuntu.com resolute

No upstream port exists. There are zero issues, PRs, commits, or code references to RISC-V in the timescale/timescaledb repository across all search vectors (GitHub Issues API, PRs API, Commits API, Code Search, and web search). The riscv64 packages available in Ubuntu and Debian are built entirely by Canonical and Debian maintainers from dfsg-repacked source, with no upstream involvement.

Key contributors: None identified – all riscv64 packaging work is done by Debian/Ubuntu maintainers, not by TimescaleDB upstream contributors.

3. Upstream Support Tier

TimescaleDB has no formal platform tier policy document (no PLATFORMS.md, no SUPPORT.md, no docs/platforms/ directory). Platform support is implicitly determined by the CI matrix. RISC-V is neither blocked nor planned – it is absent from all upstream considerations.

Tier aspect	amd64	arm64	riscv64
CI builds	Yes	Yes (custom runner)	No
CI tests pass	Yes	Yes	No
Upstream binary releases	Yes (Windows ZIP)	No	No
Docker image manifest	Yes	Yes	No
Official support statement	Implicit (CI)	Implicit (CI)	None
Distribution package	Ubuntu/Debian	Ubuntu/Debian	Ubuntu/Debian (distro-built only)

The arm64 CI runner (timescaledb-runner-arm64) uses CFLAGS=-march=armv8.2-a+crypto to enable ARM crypto extensions for vectorized grouping. The riscv64 distribution package is a downstream initiative with no upstream backing.

4. Technical Architecture and RISC-V-Specific Subsystems

TimescaleDB’s performance-sensitive code is concentrated in four areas:

UMASH hashing (vectorized text/multi-column GROUP BY)

UMASH is a vendored non-cryptographic hash library (tsl/src/import/umash.c) used for vectorized aggregation on text columns, UUID columns, and multi-column grouping keys. It requires either x86-64 PCLMUL (__PCLMUL__, -mpclmul) or ARM crypto (__ARM_FEATURE_CRYPTO, -march=...+crypto). The source contains a hard compile-time #error "Unsupported platform: umash requires CLMUL (-mpclmul) on x86-64, or crypto (-march=...+crypto) extensions on aarch64." for all other architectures. On riscv64, the build system (tsl/src/CMakeLists.txt) auto-detects this at configure time and sets USE_UMASH=OFF, disabling the text and serialized hashing strategies entirely. This is a silent functional gap: vectorized text-column and multi-column GROUP BY do not exist on riscv64.

RISC-V extensions that could close this gap: Zvbc or Zbc (carry-less multiply) or Zvkned (AES). No upstream code to wire these in exists.

CRC32 fast hash (fixed-width single-column grouping)

tsl/src/nodes/vector_agg/hashing/hash64.h uses SSE4.2 _mm_crc32_u64 under the USE_SSE42_CRC32C flag for fast hashing of 2/4/8-byte integer keys. On riscv64, this flag is not set and the code falls back to the SplitMix64 scalar finalizer. Fixed-width integer-key grouping works but without hardware acceleration. RISC-V Zbc/Zbkc extensions would enable hardware CRC, but no such code exists.

FastLanes compression (integer column compression)

tsl/src/compression/algorithms/fastlanes/ implements bit-packing and frame-of-reference (FFOR) compression for integer columns. The README states: “only portable C” with compiler auto-vectorization. No SIMD intrinsics, no #ifdef __riscv, no .S files. Compiles and runs on all platforms. The compiler may auto-vectorize this on riscv64 with -march=rv64gcv but no explicit RVV handling exists. This is a performance gap (auto-vectorization quality vs. hand-tuned SIMD) but not a functional gap.

Vector aggregation functions (SUM/MIN/MAX/AVG)

The columnar accumulation code is portable C with no ISA-specific dispatch. No architecture-specific implementations for riscv64 exist. The compiler may auto-vectorize.

Component	amd64 implementation	arm64 implementation	riscv64 implementation
UMASH text/multi-col hashing	PCLMUL intrinsics	NEON+crypto (`vmull_p64`)	Missing (compile-time disabled)
CRC32 fast hash (int keys)	SSE4.2 `_mm_crc32_u64`	Scalar SplitMix64	Scalar SplitMix64
FastLanes compression	Portable C (compiler auto-vec)	Portable C (compiler auto-vec)	Portable C (compiler auto-vec)
Vector aggregation (sum/min/max)	Portable C (compiler auto-vec)	Portable C (compiler auto-vec)	Portable C (compiler auto-vec)
JIT query compilation	PostgreSQL LLVM JIT	PostgreSQL LLVM JIT	Disabled (JIT package unavailable on riscv64)

5. Build System, Cross-Compilation, and Toolchain

Standard build commands (from docs/BuildSource.md):

git clone https://github.com/timescale/timescaledb.git
cd timescaledb
git checkout <release-tag>
./bootstrap [-DCMAKE_BUILD_TYPE=Release] [-DPG_CONFIG=/path/to/pg_config]
cd build && make
make install

The bootstrap script is a wrapper around cmake. CMake >= 3.15 is required (cmake_minimum_required(VERSION 3.15) in CMakeLists.txt). No minimum GCC or Clang version is stated; any compiler with C11 support (-std=c11 for PG <= 17, -std=c23 for PG >= 18) is accepted.

Native riscv64 build invocation:

./bootstrap \
  -DCMAKE_BUILD_TYPE=Release \
  -DPG_CONFIG=/usr/lib/postgresql/18/bin/pg_config \
  -DUSE_OPENSSL=ON \
  -DUSE_UMASH=OFF

USE_UMASH=OFF is not strictly necessary to specify – the configure step will auto-detect and set it OFF – but explicitly passing it avoids a confusing warning if the UMASH probe fails.

Key CMake flags for riscv64:

Flag	Default	riscv64 note
`USE_UMASH`	auto-detected	Auto-sets OFF on riscv64; do not force ON
`USE_OPENSSL`	ON	Keep ON if PostgreSQL was built with OpenSSL
`APACHE_ONLY`	OFF	Set ON to exclude TSL/proprietary code
`WARNINGS_AS_ERRORS`	ON (Debug), OFF (Release)	Can disable during porting
`EXPERIMENTAL`	OFF	Set ON to skip PG version check with snapshot PG builds

Cross-compilation: No riscv64 CMake toolchain file exists in the repository (cmake/ contains only GenerateScripts.cmake, GenerateTestSchedule.cmake, GitCommands.cmake, ScriptFiles.cmake). A custom toolchain file is required for cross-compilation:

set(CMAKE_SYSTEM_NAME Linux)
set(CMAKE_SYSTEM_PROCESSOR riscv64)
set(CMAKE_C_COMPILER riscv64-linux-gnu-gcc)
set(CMAKE_CXX_COMPILER riscv64-linux-gnu-g++)

When CMAKE_CROSSCOMPILING=TRUE, the UMASH -mpclmul probe is additionally guarded away (if(NOT CMAKE_CROSSCOMPILING) at tsl/src/CMakeLists.txt line 53), so USE_UMASH correctly remains OFF.

QEMU: No QEMU usage in the repository. The 32-bit CI uses a native i386/debian:bookworm-slim container. For riscv64 testing, QEMU user-mode emulation or a native board would need to be set up externally.

Known build failures: None documented, but the build has not been tested upstream. The Ubuntu 26.04 package build succeeding confirms the core build is functional on riscv64. The one structural issue is that USE_UMASH=OFF causes hash_strategy_single_text.c and hash_strategy_serialized.c to be excluded from compilation, resulting in a functional gap rather than a build failure.

6. Feature Coverage and Gap Analysis vs arm64 and amd64

Feature	amd64	arm64	riscv64
Hypertable storage	Full	Full	Full
Columnar compression (FastLanes)	Full	Full	Full
Continuous aggregates	Full	Full	Full
Vectorized aggregation (int keys)	Full	Full	Full (scalar fallback)
Vectorized aggregation (text keys)	Full (UMASH)	Full (UMASH+crypto)	Absent (UMASH disabled)
Vectorized aggregation (multi-col)	Full (UMASH)	Full (UMASH+crypto)	Absent (UMASH disabled)
CRC32-accelerated hashing	Hardware (SSE4.2)	Scalar fallback	Scalar fallback
PostgreSQL JIT (LLVM)	Available	Available	Unavailable (distro)
SSL/TLS (OpenSSL)	Full	Full	Full (software AES; no Zknd/Zkne)

Functional gaps on riscv64:

Vectorized text-column and multi-column GROUP BY are not compiled in. Queries that exercise these paths fall back to PostgreSQL’s standard (non-vectorized) hash aggregation. There is no crash, no error, and no data corruption – only a performance regression.
PostgreSQL JIT is unavailable on riscv64 in Ubuntu 26.04 (the postgresql-18-jit package is not built for riscv64). TimescaleDB’s own columnar execution engine operates independently of JIT, so this does not affect columnar scans, but complex query plans that would benefit from JIT run uncompiled.

Performance gaps on riscv64 (no benchmark data available):

Vectorized integer aggregation runs without hardware CRC32 (SplitMix64 fallback).
FastLanes compression runs on portable C without explicit RVV auto-vectorization tuning.
No benchmark data for TimescaleDB on riscv64 exists in any public source (GitHub, RISE blog, or web). Data not available: no riscv64 vs amd64/arm64 performance comparison.

Security hardening gaps: OpenSSL on riscv64 uses software AES (T-table implementation) unless the CPU implements Zknd/Zkne or Zvkned. The T-table AES is not constant-time. See the OpenSSL report for full details.

NaN/floating-point semantics: No riscv64-specific floating-point issues have been identified in TimescaleDB. The code does not contain architecture-specific floating-point dispatch.

7. CI/CD Infrastructure

No riscv64 CI exists in the timescale/timescaledb repository. All 47 workflow files in .github/workflows/ were read and searched; none contains the strings riscv, riscv64, risc-v, or RISCV.

CI aspect	amd64	arm64	riscv64
Build	Yes (ubuntu-22.04)	Yes (timescaledb-runner-arm64)	No
Unit/regression tests	Yes	Yes	No
Vectorized aggregation tests	Yes	Yes (with `-march=armv8.2-a+crypto`)	No
Sanitizer (ASAN/UBSAN)	Yes	No	No
Memory tests (Valgrind)	Yes	No	No
Package build (apt)	Yes (amd64 + ARM)	Yes	No
RISE CI runners	No	No	No

The arm64 build requires a self-hosted runner (timescaledb-runner-arm64) with -march=armv8.2-a+crypto to enable ARM crypto for UMASH. No RISE runners are used. No QEMU-based riscv64 emulation test job exists.

8. Distribution and Release Status

Upstream releases: GitHub releases provide only Windows/amd64 ZIP archives (timescaledb-postgresql-{16,17,18}-windows-amd64.zip). No Linux binaries, no riscv64 binaries, and no arm64 binaries are published upstream. The three most recent releases (2.29.2, 2.29.1, 2.29.0) each contain exactly three assets, all Windows/amd64.

Docker Hub: The official timescale/timescaledb Docker image supports linux/amd64, linux/arm64, linux/arm/v7, linux/arm/v6, and linux/386. No linux/riscv64 manifest entry exists.

Distribution packages:

Distribution	Package	Version	riscv64
Ubuntu 26.04 (resolute)	postgresql-18-timescaledb	2.25.1+dfsg-1	Yes
Debian sid	postgresql-18-timescaledb	2.29.2+dfsg-1	Yes
Debian Trixie	postgresql-17-timescaledb	2.19.3+dfsg-1+deb13u1	Yes
Docker Hub	timescale/timescaledb	latest	No
PyPI (`timescaledb`)	Django ORM wrapper	0.2.1	N/A (pure Python)

All distribution packages are built by Canonical/Debian maintainers from dfsg-repacked source. The +dfsg suffix indicates that non-DFSG-compliant files (TLS enterprise code) are removed. These builds are not tested by upstream CI and represent a downstream initiative.

To get a working riscv64 binary: Install postgresql-18-timescaledb from Ubuntu 26.04 or Debian via apt install postgresql-18-timescaledb. No upstream-provided binary exists.

9. Dependencies

Dependency	Role	riscv64 build	riscv64 test	riscv64 release	Blocking issues
PostgreSQL	Database engine hosting the extension	Green	Green (regression tests pass)	Green (ubuntu/debian packages)	LLVM JIT unavailable on riscv64; affects query compilation not columnar engine
OpenSSL (libssl3)	TLS for remote connections and telemetry	Green	Yellow (QEMU CI; known flaky test)	Green (libssl3t64 in Ubuntu 26.04 riscv64)	Software AES only unless Zknd/Zkne present; T-table not constant-time
LZ4 (liblz4)	Column and WAL compression	Green	Yellow (QEMU CI only)	Green (liblz4-1 in Ubuntu 26.04 riscv64)	LZ4_FAST_DEC_LOOP not enabled (PRs #1678, #1739 unmerged); performance gap
zstd (libzstd)	Higher-ratio column compression	Green	Yellow (QEMU CI only)	Green (libzstd1 in Ubuntu 26.04 riscv64)	Fast sequence decoding path not enabled (PR #4557 stalled); performance gap
UMASH (vendored)	Vectorized text/multi-col GROUP BY hashing	Red (disabled)	N/A	N/A	Hard #error at compile time; USE_UMASH=OFF on riscv64; entire text and multi-col vectorized grouping absent
FastLanes (vendored)	Integer column bitpacking compression	Green	Green	Green	None; pure portable C
LLVM JIT (via PostgreSQL)	Query plan JIT compilation	Red (unavailable)	N/A	N/A	postgresql-18-jit not built for riscv64 in Ubuntu 26.04
libpq5	PostgreSQL client library	Green	Green	Green	None

PostgreSQL (critical dependency): PostgreSQL builds and tests pass on riscv64. LLVM JIT is disabled in all riscv64 distribution packages due to segfaults in the LLVM JIT backend. TimescaleDB’s vectorized columnar engine operates independently of JIT and is not affected.

UMASH (critical gap): The vendored UMASH library has a hard #error for riscv64. The build system auto-detects and disables it. The RISC-V extensions Zvbc (carry-less multiply on vectors) or Zbkc (carry-less multiply on scalars) could provide the underlying operation needed, but no upstream code exists to wire them in.

LZ4 / zstd: Both dependencies have performance gaps on riscv64 (no fast decode paths, no unaligned-access optimization). These are performance gaps, not functional blockers.

11. Known Bugs and Active Issues

No riscv64-specific bugs have been filed against TimescaleDB. The following bugs on related architectures document the portability surface:

ID	Title	Status	Severity	Notes
#8984	Tests consistently fail on ARM (vector_agg_text, vector_agg_uuid, etc.)	Closed 2025-11-28	High	Vectorized aggregation nodes not enabled on ARM for certain query shapes; closest analog for riscv64 risk
#9039	TimescaleDB 2.24.0 Debian arm64 crashes with SIGILL on Raspberry Pi 4	Closed 2026-01-23	Critical	SIGILL from CPU features not supported by ARMv8.0 (armv8.2 binary run on armv8.0 CPU); regression in 2.24.0
#9126	SIGILL on Raspberry Pi 3B+ (ARMv8.0-A Cortex-A53) with 2.24.0	Closed 2026-01-23	Critical	Same root cause as #9039: arm64 package built with features not supported on older ARMv8.0 cores
#9900	Illegal instruction on Intel Atom (SIGILL in _mm_clmulepi64_si128 / umash.c)	Closed 2026-05-26	Critical	PCLMUL compiled in but CPU lacks it; crash path: `umash_fprint -> single_text_key_hashing_get_key -> vector_agg_exec`; confirms runtime SIGILL risk if UMASH is forced on an unsupported ISA
#9996	Domain constraint not respected in vectorized filter	Open 2026-06-08	Medium	Architecture-agnostic correctness bug in vectorized filter layer
#9925	Wrong null handling for degenerate scalar array operations in vectorized filters	Open 2026-05-29	Medium	Architecture-agnostic null handling bug in vectorized filter

The SIGILL pattern in #9039, #9126, and #9900 is the exact failure mode that would occur if UMASH were compiled in on riscv64 (it is not, so this is not an active risk – but it illustrates why the hard #error and auto-disable exist).

12. Objections and Upstream Blockers

No stated objections: No issues, PRs, or forum posts document any decision to block or defer RISC-V support. The absence is one of disinterest, not active rejection.

Technical blockers:

UMASH: Requires a RISC-V carry-less multiply implementation (Zvbc/Zbkc) or an alternative hash algorithm. This is the most significant engineering task for feature parity.
LLVM JIT: Blocked by the PostgreSQL/distro LLVM segfault issue on riscv64. Not a TimescaleDB issue to fix.
No riscv64 CI runner: Self-hosted riscv64 runners are required for testing (the ARM64 path uses a self-hosted runner); GitHub Actions does not provide hosted riscv64 runners as of 2026-08.

Organizational blockers:

TigerData has no disclosed commercial interest in RISC-V. There is no evidence of RISC-V hardware in their customer or internal deployment base.
All maintainers are TigerData employees. External contributions require a CLA (Contributor License Agreement). Community-driven RISC-V port PRs are possible but would need to be maintained by the submitter; TigerData is unlikely to own ongoing riscv64 CI costs without a business case.

Acceptance probability: Data not available – no upstream discussion exists to evaluate receptiveness to a RISC-V port PR.

13. Readiness Assessment

Color: yellow (Distribution floor – no upstream riscv64 CI, but Ubuntu 26.04 and Debian ship unpatched riscv64 packages built from source)
Release provider: distro

TimescaleDB has no upstream riscv64 CI, no upstream riscv64 binary releases, and zero riscv64-related activity in the repository. Ubuntu 26.04 and Debian ship postgresql-18-timescaledb for riscv64 as a distro-built package (packages.ubuntu.com), confirming that the build succeeds and the package is installable. The distribution floor rule applies: distro ships an unpatched riscv64 build, giving a floor of yellow. No upstream testing is performed on riscv64.

The build is functionally correct except for the UMASH vectorized hashing gap: text-column and multi-column GROUP BY fall back to non-vectorized PostgreSQL execution (no crash, no data error). FastLanes compression, continuous aggregates, and hypertable storage are fully functional.

Pending work that could change the grade:

An upstream riscv64 CI job with passing tests would raise the grade to blue.
A UMASH replacement for riscv64 (Zvbc/Zbkc-based or algorithmic fallback) would close the primary performance gap but would not by itself affect the color.
No RISE involvement in TimescaleDB exists at this time.

14. Investment Analysis

14.1 Functional Enablement

TimescaleDB installs and runs on riscv64 via the Ubuntu/Debian package. Basic time-series workloads (insert, query, compression, continuous aggregates) are functional. The only functional gap is vectorized text/multi- column GROUP BY via UMASH.

Work required to close the functional gap:

Implement a riscv64 carry-less multiply path in tsl/src/import/umash.c using Zvbc or Zbkc extensions (or a scalar software fallback), and update tsl/src/CMakeLists.txt to detect riscv64 CLMUL capability. [NEEDS VERIFICATION: whether TigerData would accept a Zbkc-scalar fallback vs. requiring a hardware extension path]
Alternatively, replace UMASH with a platform-agnostic hash (e.g., a wyhash or xxHash3 variant) for riscv64 builds.

14.2 Performance Optimization

Performance gaps relative to amd64 (no benchmark data – gap is structural):

CRC32 hash path: Add a Zbc/Zbkc-based CRC32 implementation in tsl/src/nodes/vector_agg/hashing/hash64.h (equivalent to the SSE4.2 path). Effort: 1-2 weeks.
FastLanes compression: Add explicit RVV (RISC-V Vector) intrinsics for the bit-packing inner loops in tsl/src/compression/algorithms/fastlanes/. This mirrors the auto-vectorization improvement pattern. Effort: 4-6 weeks, requires profiling on riscv64 hardware to justify.
JIT: Blocked upstream (PostgreSQL LLVM JIT unavailable on riscv64 in distributions). Not actionable within TimescaleDB.

14.3 CI/CD Infrastructure

Add a riscv64 build job to .github/workflows/linux-build-and-test.yaml using either a self-hosted riscv64 runner or QEMU-based emulation.
Add a riscv64 package build job to apt-packages.yaml and apt-installcheck.yaml.
Estimated effort: 2-3 weeks (CI setup, QEMU or runner provisioning, test exclusion list for UMASH-dependent tests analogous to the 32-bit job’s exclusions).

14.4 Ecosystem Enablement

No dependent package ecosystem requires separate enablement for riscv64. The PyPI timescaledb package (v0.2.1) is a pure-Python Django ORM wrapper (py3-none-any); it works on riscv64 without any changes. No npm, Maven, or OCI ecosystem components require separate riscv64 enablement.

14.5 Summary Table

Area	Work Item	Effort (person-weeks)	Owner	Priority
Functional	UMASH riscv64 carry-less multiply path or hash algorithm replacement	3-4	Timescale/TigerData or community (CLA required)	High
Functional	cmake detection for riscv64 UMASH support	1	Timescale/TigerData or community	High
Performance	CRC32 Zbc/Zbkc hash path in hash64.h	1-2	Community	Medium
Performance	RVV intrinsics for FastLanes compression	4-6	Community	Low
CI/CD	riscv64 build + test job (QEMU or self-hosted runner)	2-3	Timescale/TigerData or community	High
CI/CD	riscv64 package build verification	1	Timescale/TigerData or community	Medium

Total estimated effort: 12-17 person-weeks for full functional parity and basic CI. Performance optimization (FastLanes RVV) is optional and lower priority.

15. Updates

No updates yet – initial report dated 2026-08-29.