Comparison: kardinal vs Kargo vs GitOps Promoter¶

This page compares kardinal-promoter with the two most similar tools in the GitOps promotion space.

Feature Matrix¶

Where kardinal Leads¶

Graph-native policy evaluation¶

kardinal PolicyGates are nodes in the kro DAG. They have access to the entire pipeline's state — not just the current stage:

# In a prod PolicyGate — reads upstream uat stage's soak time
expression: "bundle.upstreamSoakMinutes >= 60"

# Read upstream metrics
expression: "metrics.errorRate < 0.01"

# Combine schedule + soak + metadata
expression: '!schedule.isWeekend && bundle.upstreamSoakMinutes >= 30 && bundle.labels.hotfix != "true"'

Neither Kargo nor GitOps Promoter can express "do not promote to prod unless UAT has been healthy for 60 minutes." Kargo has per-stage approval with no expression engine. GitOps Promoter has webhook-based checks with no pipeline context.

Fan-out DAG topology¶

test ──► uat ──► staging ──┬──► prod-us ──┐
                           └──► prod-eu ──┴──► verified

kardinal models this natively. Kargo is sequential. GitOps Promoter has no DAG (roadmap item).

Structured PR evidence¶

Every promotion PR opened by kardinal includes:

## Promotion Evidence

**Pipeline**: my-app  **Bundle**: sha-9349a3f
**Image**: ghcr.io/pnz1990/kardinal-test-app@sha256:abc123...

### Gate Results
| Gate | Result | Expression |
|---|---|---|
| no-weekend-deploys | ✅ PASS | `!schedule.isWeekend` |
| require-uat-soak   | ✅ PASS | `bundle.upstreamSoakMinutes >= 30 (actual: 47)` |

### Upstream Environments
| Env | Status | Verified At |
|---|---|---|
| test | Verified | 2026-04-13 09:00 UTC |
| uat  | Verified | 2026-04-13 09:45 UTC |

Kargo tracks promotions in its own UI — PRs have no evidence body. GitOps Promoter PRs show the git diff only.

Auto-rollback and health-failure policy¶

kardinal opens a rollback PR automatically when a promotion fails health verification, using the previous Bundle. Each stage can independently configure onHealthFailure: rollback | abort | none. Combined with bake.policy: fail-on-alarm, this gives fine-grained control: critical stages abort and require human intervention; non-critical stages roll back automatically. Neither competitor has automated rollback or stage-level health failure policies.

Contiguous healthy soak¶

bake.minutes counts contiguous healthy minutes — if a health alarm fires during the soak window, the timer resets to zero. The deployment must survive a full bake.minutes window with no alarms. Kargo and GitOps Promoter both count elapsed time from deployment, regardless of whether the service was healthy during that window.

Change freeze management¶

A single ChangeWindow CRD in kardinal-system blocks all pipelines cluster-wide. Platform teams create one object during incidents, holidays, or maintenance windows — no per-pipeline configuration needed. Kargo has no equivalent. GitOps Promoter requires manually setting CommitStatus resources per-environment.

Wave topology for multi-region rollouts¶

The wave: field on Pipeline environments generates DAG dependency edges automatically: wave 2 cannot start until all wave 1 stages are verified. This makes the prod-wave-1 → prod-wave-2 → prod-wave-3 pattern idiomatic in three lines of YAML. Kargo has no wave concept. GitOps Promoter has no DAG support.

DORA metrics built-in¶

Bundle.status.metrics records commitToProductionMinutes, bakeResets, and operatorInterventions for every promotion. The kardinal metrics CLI surfaces these per pipeline. Neither Kargo nor GitOps Promoter tracks deployment efficiency metrics.

Where Kargo Leads¶

Artifact discovery: Kargo's Warehouse concept automatically monitors OCI registries and Git repos for new artifact versions, packages them as Freight, and feeds them into the pipeline. kardinal requires CI to create Bundles explicitly (via webhook or CLI). If you want passive artifact detection without modifying CI pipelines, Kargo wins.

Production maturity: Kargo is at v1.10.x with commercial support from Akuity, a larger community, and a longer track record in production.

ArgoCD integration depth: If you are all-in on ArgoCD, Kargo's argocd-update promotion step and ArgoCD-native health checks are deeply integrated. kardinal's ArgoCD adapter covers health verification but the update mechanism is GitOps-native (git commits).

Where GitOps Promoter Leads¶

Git-native purity: GitOps Promoter has no state outside Git. Every promotion is a branch, a commit, a PR. No additional CRDs tracking state. If your team wants "the only source of truth is git," GitOps Promoter's model is the cleanest.

CommitStatus extensibility: Any external system can participate in promotion gating by writing a CommitStatus CRD. The open interface is simpler to extend than kardinal's PolicyGate CEL (which requires running code in-cluster).

When to Choose kardinal-promoter¶

• You need parallel environment promotions — fan-out to prod-us and prod-eu simultaneously, gate on both completing
• You want expressive, cross-stage policy gates — soak time, upstream metrics, schedule, bundle metadata, PR approval state — without writing webhook servers
• You need contiguous healthy soak — deployments must survive bake windows with zero health alarms, not just elapsed time
• You want wave topology for multi-region production rollouts — promote to 1 region, bake, then expand to the next wave
• You want a centralized change freeze — one ChangeWindow object blocks all pipelines during incidents or holidays
• You use ArgoCD + Flux mixed, or neither — kardinal doesn't require a specific GitOps engine
• You want structured PR evidence so reviewers have full promotion context in the PR body
• You want auto-rollback triggered by health check failures, with per-stage abort vs. rollback vs. ignore policy
• You are a platform team that needs org-level policies automatically applied to all pipelines without teams being able to bypass them
• You want DORA metrics — time-to-production, rollback rate, operator interventions — surfaced per pipeline
• You need integration tests as promotion steps — run a Kubernetes Job as part of the promotion sequence
• You need emergency override with audit record — escape hatch that produces evidence, not a silent bypass

Further Reading¶

• Concepts — kardinal-promoter's core model
• Policy Gates — CEL expression reference
• Architecture — system design
• FAQ — common questions