Read 2026-11-03
seldonio/seldon-core: technical due diligence, read from its repository
What Borehole read in seldonio/seldon-core's code and history at 402619e: its licence, security advisories, releases, review practice, tests, dependencies and operations. Every finding cites the file and line it rests on. The full report also covers the team and asks the questions the code cannot answer.
| Dimension | Rating |
|---|---|
| Architecture & Codebase — including AI provenance | strong |
| Engineering Process & SDLC Maturity | adequate |
| Product & Engineering Maturity | strong |
| Security & Reliability Posture | adequate |
| Observability & Operations | adequate |
| Fit with the Acquirer | not assessable |
What the repository says
- Releases are cut from another branch than the one surveyed
LICENSE:1: The licence is BUSL-1.1: source available, not open sourceLICENSE: Relicensed: Apache-2.0 → BUSL-1.1 (2024-01-16)
Architecture & Codebase — including AI provenance
strongapis/mlops/v2_dataplane/schema_registry/kotlin/io/seldon/mlops/inference_schema/InferRequest.java is 9,064 lines, 75× this codebase's median file
adequateThe median source file here is 120 lines. 5 files exceed 1,200 lines, the largest being 9,064. Size is not a defect on its own — some problems genuinely live in one place — but these are where merge conflicts, review fatigue and single-owner knowledge concentrate, and they are the first thing that slows an inheriting engineer down.
apis/mlops/v2_dataplane/schema_registry/kotlin/io/seldon/mlops/inference_schema/InferRequest.java:1 — 9,064 lines (75× median) |
apis/mlops/chainer/kotlin/io/seldon/mlops/chainer/ChainerOuterClass.java:1 — 9,003 lines (75× median) |
apis/mlops/v2_dataplane/schema_registry/kotlin/io/seldon/mlops/inference_schema/InferResponse.java:1 — 7,541 lines (62× median) |
apis/mlops/v2_dataplane/schema_registry/kotlin/io/seldon/mlops/inference_schema/ModelInferRequestKt.kt:1 — 2,099 lines (17× median) |
apis/mlops/v2_dataplane/schema_registry/kotlin/io/seldon/mlops/inference_schema/ModelInferResponseKt.kt:1 — 1,847 lines (15× median) |
Engineering Process & SDLC Maturity
adequate7% of recent commits change more than 2,000 lines
adequate24 of 348 commits in the last 24 months changed over 2,000 lines of code, not counting reformatting, renames, vendored or generated code. A change that size is rarely reviewed line by line. Ask how they were reviewed, and what made them that large.
| 24 commits over 2,000 lines |
Product & Engineering Maturity
strongGo 1.24 is past end of life
adequateThe code is pinned to Go 1.24 (end of life 2026-02-10). None of these receives security fixes any more; the first stopped on 2026-02-10. Moving off an unsupported runtime is rarely a one-line change, so a buyer should ask whether the upgrade is planned and what it touches.
operator/Dockerfile — Go 1.24, end of life 2026-02-10 |
6 files still carry code deprecated more than two years ago
adequate6 files were marked deprecated over two years ago and still are, the oldest since Jan 2022. A deprecation is a promise to remove. Ask what still depends on these, and whether removing them is scheduled.
scheduler/pkg/agent/repository/triton/config/model_config.pb.go — deprecated since Jan 2022 |
apis/mlops/v2_dataplane/kotlin/io/seldon/mlops/inference/v2/V2Dataplane.java — deprecated since Mar 2022 |
apis/mlops/chainer/kotlin/io/seldon/mlops/chainer/ChainerOuterClass.java — deprecated since Mar 2022 |
apis/go/mlops/agent/agent.pb.go — deprecated since Sep 2022 |
apis/go/mlops/scheduler/scheduler.pb.go — deprecated since Jan 2023 |
Releases are cut from another branch than the one surveyed
adequateThe newest release, v2.10.2, is not on the branch this survey read. The shipped product may differ from what this report describes. Survey the release branch too.
| tag v2.10.2 |
Security & Reliability Posture
adequate9 dependencies float, and no lockfile is committed
adequateDependency versions are declared as ranges and no lockfile is committed, so a build today does not necessarily produce what a build last month produced. That matters twice over in a transaction: the thing being bought cannot be reproduced exactly, and a compromised upstream release arrives without anyone choosing to take it.
ansible/requirements.txt:1 — >=6.7.0,<8.0.0 |
ansible/requirements.txt:2 — >=5.0.0,<7.0.0 |
ansible/requirements.txt:3 — >=26.1.0,<27.0.0 |
samples/examples/huggingface/requirements.txt:1 — any version |
samples/examples/huggingface/requirements.txt:2 — any version |
The build runs third-party code it does not pin
adequate9 third-party CI actions are referenced by a tag or branch rather than a commit, so they run whatever that tag points to on the day; 1 step pipes a download straight into a shell. Tags have been moved to malicious commits in real supply-chain attacks, and a piped script is whatever the server returns that minute. Pinning to a commit is the fix, and it is small.
.github/workflows/changelog.yml:18 — action pinned by tag |
.github/workflows/changelog.yml:26 — action pinned by tag |
.github/workflows/draft-release.yml:84 — download piped into a shell |
2 outbound calls can wait forever
adequate2 HTTP calls use a client with no timeout (Python `requests` without `timeout=`, or Go's default client). When a dependency slows down, each one holds a worker until the connection gives up, and enough of them take the service down with the dependency.
scheduler/cmd/pipelinegateway/main.go:358 — outbound call without a timeout |
scheduler/pkg/agent/internal/testing_utils/rest_client.go:168 — outbound call without a timeout |
7 locked dependencies have published vulnerabilities, 4 high or critical in what ships
adequate7 of 170 locked dependencies match 9 published advisories in OSV, 4 of them rated high or critical. A lockfile says which version is installed, not whether the flawed code is reachable from this product, so each needs a look. Most are fixed by an upgrade, and the advisory names the version.
apis/go/go.mod — google.golang.org/grpc v1.73.0: GHSA-p77j-4mvh-x3m3 (critical), fixed in 1.79.3 |
scheduler/go.mod — go.opentelemetry.io/otel v1.37.0: GHSA-mh2q-q3fh-2475 (high), fixed in 1.41.0 |
scheduler/go.mod — go.opentelemetry.io/otel/sdk v1.37.0: GHSA-9h8m-3fm2-qjrq (high), fixed in 1.40.0 |
15 workload settings give a container the host
adequate15 places in the deployment configuration run a container privileged or in the host's network, process or IPC namespace. A compromise of that container is a compromise of the machine, and most acquirers' security reviews block it. Node agents (network, storage, monitoring) are excluded.
scheduler/all-host-network.yaml:6 — privileged or host namespace |
scheduler/all-host-network.yaml:30 — privileged or host namespace |
scheduler/all-host-network.yaml:56 — privileged or host namespace |
The licence is BUSL-1.1: source available, not open source
adequateLICENSE grants BUSL-1.1: source available, not open source; production use restricted until the change date. The reading comes from the licence's own title and clauses. Whether the terms fit the buyer's intended use, and whether any paid licence transfers with the code, is a question for counsel and the licensor.
LICENSE:1 — 51 lines, BUSL-1.1 |
Relicensed: Apache-2.0 → BUSL-1.1 (2024-01-16)
adequateThe root licence changed 1 time(s). The latest terms restrict what users, hosts or a buyer may do in ways the earlier ones did not. Code taken under the old terms stays under them, so both matter in diligence.
Observability & Operations
adequateNo error reporter is wired in
adequateNothing in the service initialises an error reporter (Sentry, Rollbar, Bugsnag, Honeybadger or an APM agent). An exception in production reaches a person only if someone is reading the logs, which usually means a customer reports it first.
| no error reporter initialised |
17 of 21 deployed images carry no version, or `latest`
adequate17 images in the deployment configuration are named without a version or as `latest`. A redeploy can change the software without a commit, a rollback has nothing to roll back to, and nobody can say from the repository what is running.
hodometer/k8s/deployment.yaml:23 — image without a version |
k8s/yaml/components.yaml:310 — image without a version |
k8s/yaml/components.yaml:466 — image without a version |
Readiness and liveness probes are identical in 4 of 13 workloads
adequateBoth probes call the same endpoint. When a dependency slows down, the liveness probe fails with the readiness probe, so the platform restarts pods that only needed taking out of rotation, and a partial outage becomes a full one. Liveness should ask whether the process is alive; readiness whether it can serve.
operator/config/serverconfigs-experimental/mlserver.yaml — same probe endpoint twice |
operator/config/serverconfigs-experimental/triton.yaml — same probe endpoint twice |
operator/config/serverconfigs/mlserver.yaml — same probe endpoint twice |
9 of 13 deployments run a single replica
adequate9 deployments set `replicas: 1`, and nothing (an autoscaler, a disruption budget) adds more. Every deploy and every node drain is downtime, and the availability the buyer inherits has a ceiling.
hodometer/k8s/deployment.yaml — replicas: 1 |
k8s/yaml/components.yaml — replicas: 1 |
operator/config/manager/manager.yaml — replicas: 1 |
Fit with the Acquirer
not assessableNot rated: structurally not assessable from a repository.
Read more, or read your own
The full report on seldonio/seldon-core: every dimension, the questions for the room, and the audit log.
Survey a repository: free on public repositories. Private code is read on your own machine by a collector whose source you can read first.