Core resource observations

Core capabilities are mandatory; provider placement may be shared or application-isolated. Additional isolation can add provider instances and resource consumption. The optional Console and application workloads are outside the Core totals below.

Measured Docker reference topology

The rootless Docker bootstrap qualification passed for public Core b25c4e25f0244648cc6a7641828e238005925933 on 2026-10-06. Both rows measure PostgreSQL + OpenBao + Keycloak, including the running provider dependencies, gateways and administrative helpers.

Machine-role default	Stabilized idle total, median	Observed idle range	Sampled startup/convergence peak
Development	2.62 GiB	2.60–2.62 GiB	3.79 GiB
Deployment	2.57 GiB	2.56–2.58 GiB	4.73 GiB

These are installation-shared observations, not universal minimums or application-isolated measurements. The two roles use the same provider topology; their differing observations do not establish a role-specific memory budget.

This existing reference realization has one SQL member and one OpenBao member, plus three Keycloak members backed by three PostgreSQL members and three etcd members. The retained idle inventory contains 17 running containers. Three mandatory capabilities therefore do not imply three containers. These figures must not be presented as measurements of a single-Keycloak realization.

Images include PostgreSQL 18, OpenBao 2.7.0 and Keycloak 26.8.0. Original JSON logs retain exact image IDs, per-service and per-capability memory, simultaneous Core totals, host memory/swap/PSI and timestamps. Provider memory limits in this run were unset. Container memory uses native Docker/cache accounting; it is not host RSS or a guaranteed allocation requirement.

Sampling targets a three-second interval. Each idle window contains eleven complete observations spanning at least 30 seconds, with total variation within 10%. Startup peaks are sampled observations and can miss shorter peaks.

Measured Podman reference topology

The rootless Podman bootstrap qualification passed for Core 55f7b1c4c72e57c5858a3476c097ca6e244c8d5d on 2026-10-06. It uses the same installation-shared 17-container reference realization.

Machine-role default	Stabilized idle total, median	Observed idle range	Sampled startup/convergence peak
Development	2.51 GiB	2.51–2.65 GiB	4.65 GiB
Deployment	2.56 GiB	2.55–2.60 GiB	4.59 GiB

These figures come from eleven complete native observations per idle window. Runtime-specific memory/cache accounting prevents treating differences between Docker and Podman as a performance comparison. Original JSON and image identities remain in the qualification artifact; application isolation is not measured.

Qualification limits

Remote execution-node memory evidence is separate from installation Core totals. The Connector reports current available RAM and swap from the selected node; runtime-info total RAM is not treated as available capacity. Missing, stale or inconsistent evidence is refused, as is exhausted RAM even without a reliable workload estimate. Core-host RAM is not a substitute for node evidence. This projection does not measure memory PSI, VM or cgroup limits and does not qualify the complete remote Application lifecycle.

Isolated placement and further host/topology calibration remain pending. Source tests and a provider count cannot supply missing figures. Host preflight continues to distinguish planning estimates from unavailable measurements for the selected installation.

Core preflight uses the largest observed Identity startup sample (4,976,065,638 bytes, including its dependencies) as a reference planning estimate on other hosts. It adds the existing topology-aware SQL/Secrets startup budgets. This estimate is explicitly ESTIMATED; it establishes no reliable minimum and does not claim that the selected host or Podman was measured. Reused owned capabilities are not budgeted as additional instances.