Limit the D&D feature to playbook preparation

This commit is contained in:
2026-09-21 05:14:06 +00:00
parent f9901f17d1
commit 1d1f34fc8b
3 changed files with 198 additions and 653 deletions

View File

@@ -1,88 +1,41 @@
# Feature 0016: Deploy the D&D Pipeline to `pipeline-dnd`
# Feature 0016: Prepare the D&D Deployment Playbook for `pipeline-dnd`
Status: in_progress
## Summary
Add a minimal playbook and host inventory to `rakestraw-ansible` that apply
`rakestraw.dnd.pipeline` to the existing `pipeline-dnd` host. Configure the
host through `host_vars`, enroll `eric` as the initial human operator, keep
credential values in Ansible Vault, and prove the complete pipeline through a
real repeatable deployment and representative D&D session.
[Feature 0015](0015-dnd-s3-storage-and-publishing.md) has implemented and passed
independent review of the storage and publishing behavior needed for an
operational end-to-end pipeline. Feature 0016 consumes that reviewed interface
and its suite defaults.
## Problem or Opportunity
The collection and its application roles are implemented and installable, but
the operational Ansible repository does not yet have a playbook for deploying
the new collection architecture. Its existing D&D playbook targets a retired
inventory name and invokes the proof-of-concept local `pipeline_dnd` role.
The current `pipeline-dnd` host inventory already records the connection
identity and useful legacy Narratio storage values. A production-shaped first
deployment now needs to express the new role's public interface, deliver
secrets safely, reconcile the controller's dependency manifest, and validate
the actual host rather than extending the role repositories with another
synthetic scenario.
Write and commit the consumer playbook, necessary host variables, dependency
manifest, and concise usage documentation in `rakestraw-ansible`. This feature
ends with repository preparation. Deployment to a controller or target host
and execution of the pipeline are separate operator work.
## Scope
- Add `rakestraw-ansible/playbooks/pipeline/dnd.yml`, targeting the existing
`pipeline-dnd` inventory host and applying only
`rakestraw.dnd.pipeline`.
- Keep the playbook limited to its host selection and role list. Put all
deployment choices in `inventory/host_vars/pipeline-dnd/`.
- Configure the external WhisperX transcription endpoint as
`https://transcription.ai.rakestrawhome.com/transcribe`.
- Enroll the existing `eric` account through
`dnd_pipeline_operator_users`.
- Map the role's `dnd_pipeline_secrets` interface to host-specific Vault
variables. Store credential values only in an Ansible Vault-encrypted host
variables file.
- Inherit Feature 0015's storage and publishing defaults, which already match
the intended Backblaze B2 deployment. Host inventory contains only actual
host choices and deliberate overrides, without copying suite defaults.
- Preserve the existing default campaign choice, `dilfs`, through the
standalone role's `narratio_campaigns_default_campaign_id` variable.
- Reconcile existing legacy Narratio host variables so they cannot silently
override the suite policy selected by the new pipeline role.
- Update `rakestraw-ansible/requirements.yml` so a clean controller can install
the reviewed collection snapshot, all five standalone role dependencies,
and required native collections from one manifest.
- Validate controller dependency resolution, playbook syntax, initial
convergence, immediate clean convergence, operator access, native Narratio
configuration, and one representative end-to-end pipeline run on the real
host.
- Document the operational invocation, Vault prerequisite, initial operator,
and expected validation path in `rakestraw-ansible`.
- `playbooks/pipeline/dnd.yml` targets the existing `pipeline-dnd` alias and
applies `rakestraw.dnd.pipeline`, using only hosts and roles.
- Host choices belong in `inventory/host_vars/pipeline-dnd/`: the WhisperX
endpoint, initial operator `eric`, default campaign `dilfs`, and credential
references. Unknown required values are explained in adjacent comments.
- Inherit the reviewed suite's storage, publication, caching, cleanup, and
application defaults. Retire obsolete legacy host variables.
- Root `requirements.yml` selects the reviewed public collection snapshot and
its role/native collection dependencies, preserving unrelated entries.
- Document local validation and eventual operator use. Commit the completed
repository changes without requiring remote access or real credentials.
## Non-goals
- Implementing storage or publishing behavior. Feature 0015 owns that
prerequisite.
- Changing application or role behavior merely to accommodate this
deployment.
- Replacing the existing inventory alias with `pipeline.dnd`, changing DNS,
or moving connection variables out of the current host inventory.
- Adding a wrapper role, pre-task layer, service, timer, scheduler, web UI, or
unattended session-ingestion mechanism.
- Creating or managing the `eric` account, its login policy, SSH access, or
sudo policy.
- Adding another production host, high availability, backup policy, or
disaster-recovery automation.
- Removing the proof-of-concept local `pipeline_dnd` role or its historical
playbook before the new deployment has demonstrated operational parity.
- Publishing a new coordinated suite release solely as part of this
deployment feature.
- Installing or updating anything on the operational controller or target.
- Connecting to `pipeline-dnd`, validating its platform or sudo access,
converging the host, checking runtime identities, or executing a real session.
- Obtaining API keys, choosing a Vault password, populating real credentials,
or selecting acceptance audio/session data.
- Changing application/role behavior, publishing a new suite release, adding
orchestration layers, or retiring the proof-of-concept role/playbook.
## Target End State
`rakestraw-ansible/playbooks/pipeline/dnd.yml` is an intentionally small
consumer playbook equivalent to:
The committed playbook is:
```yaml
---
@@ -93,12 +46,8 @@ consumer playbook equivalent to:
...
```
The existing inventory selects the host's connection identity. A dedicated
host-variable file selects the WhisperX endpoint, the initial operator list,
and the default campaign. Storage, caching, publication, and cleanup inherit
the reviewed suite defaults. The file maps runtime credential names to
Vault-backed variables without embedding credential values in the playbook or
ordinary plaintext inventory, for example:
Existing connection variables remain in the host's `main.yml`. Its
`narratio.yml` contains the following public configuration and comments:
```yaml
dnd_pipeline_operator_users:
@@ -109,6 +58,10 @@ dnd_pipeline_whisperx_transcribe_url: >-
narratio_campaigns_default_campaign_id: dilfs
# Before deployment, define the four vault_dnd_* variables below in an
# Ansible Vault-encrypted host_vars/pipeline-dnd/vault.yml file.
# Values are not known here; leave these references unresolved until supplied.
# Do not put API keys or object-storage credentials in this plaintext file.
dnd_pipeline_secrets:
AUDITA_LLM_API_KEY: "{{ vault_dnd_audita_llm_api_key }}"
OPENROUTER_API_KEY: "{{ vault_dnd_openrouter_api_key }}"
@@ -116,214 +69,77 @@ dnd_pipeline_secrets:
OBJECT_STORAGE_KEY: "{{ vault_dnd_object_storage_key }}"
```
The corresponding credential variables reside in a tracked file encrypted in
full with Ansible Vault. The Vault password or identity mechanism remains
outside Git and outside this feature's managed content.
The references and comments are a complete deliverable for unknown secrets.
No encrypted file, placeholder key, or real credential is required for feature
completion. Future operator-supplied values should be Vault-encrypted in Git;
the suite's existing restricted runtime secrets files remain the delivery path.
The controller dependency manifest selects public collection commit
`46913446765e86ecfb29ebad9955d579ef6a4c7e`, containing the reviewed Features
0014 and 0015 and pipeline commit
`51566afcf69083cdbb5b5718f2544c86b079001c`. Its compatibility manifest selects
the five standalone roles at `v0.1.0`, `ansible.posix` at `2.2.0`, and
`community.general` at `13.3.0`. A clean controller therefore does not depend
on sibling development checkouts or previously installed roles.
The manifest pins the collection to
`46913446765e86ecfb29ebad9955d579ef6a4c7e`, containing reviewed pipeline commit
`51566afcf69083cdbb5b5718f2544c86b079001c`. Its five standalone role dependencies
use their reviewed `v0.1.0` tags, with `ansible.posix` `2.2.0` and
`community.general` `13.3.0`. Suite Git dependencies use public Gitea URLs;
native collections come from Galaxy. Local sibling checkouts are not dependencies.
The published collection `v0.1.0` predates these features. The selected
development snapshot still declares collection version `0.1.0`, so the
deployment instructions explicitly refresh an already-installed collection
from the pinned commit instead of allowing an equal-version installation to
be skipped. Unrelated controller dependencies remain preserved.
On `pipeline-dnd`, the role converges the full application suite and its D&D
configuration. The `narratio` service identity can run the pipeline, and
`eric` can run it after starting a login session that observes the granted
supplementary group membership. A representative session can be transcribed,
processed, stored, and published through the configured production path. An
immediate identical Ansible run reports no changes.
The runbook explains that the selected collection snapshot still reports
`0.1.0`, so eventual operator installation must explicitly refresh it from
the pinned commit if an older `0.1.0` is installed. This is an instruction
for future use, not an operation required by this feature.
## Repository Ownership
- `rakestraw-ansible` owns the consumer playbook, host inventory, encrypted
host credentials, controller dependency manifest, and deployment runbook.
- `ansible-role-dnd-pipeline` owns the suite's public variables and the
opinionated cross-application behavior consumed by the deployment.
- `ansible-collection-dnd` owns the assembled `rakestraw.dnd.pipeline` role
snapshot and consumer-facing compatibility manifest.
- The standalone application roles continue to own installation and native
application configuration. Feature 0016 consumes their reviewed releases or
commits without reimplementing their behavior.
- `ansible-workspace-dnd` owns this roadmap, the later implementation plan,
progress tracking, and independent review record.
- Application and asset repositories are read-only sources of runtime
contracts and deployed content for this feature.
- `rakestraw-ansible`: consumer files, comments, dependency manifest, and runbook.
- Workspace: roadmap, staged plan, progress, and independent review record.
- Roles, collection, applications, and assets: read-only contract references.
## Constraints and Settled Decisions
- Feature 0015's implementation and independent review satisfy the storage
and publishing prerequisite. Its workspace completion record should be
closed out before the Feature 0016 sprint.
- The target is the inventory alias `pipeline-dnd` as currently defined in
`rakestraw-ansible`, not a new `pipeline.dnd` alias.
- The playbook follows the repository's established style: hosts and roles
only. Facts remain at Ansible's normal default, privilege escalation remains
task-owned by the roles, and deployment variables belong in host inventory.
- `eric` is the initial human operator and must already exist on the managed
host. The pipeline role grants only additive runtime-group membership.
- API credentials use the existing typed `dnd_pipeline_secrets` interface and
Ansible Vault. Plaintext secret values must not enter Git, command lines,
logs, examples, or ordinary variable files.
- Public Gitea repositories supply the suite's Git dependencies; native
collections come from Ansible Galaxy. The deployment must not reference
this workspace's sibling checkout paths.
- The collection is pinned to the reviewed public commit identified above.
Feature 0015 did not create a release; this deployment does not track an
unbounded branch or require a new coordinated release.
- The collection artifact does not itself install arbitrary Git-hosted role
dependencies. The operational manifest must therefore include the five
standalone roles as well as the collection and its native collection
dependencies.
- Remove the legacy `narratio_s3_*` host variables because their intended
values are supplied by the suite defaults. Replace
`narratio_default_campaign_id` with
`narratio_campaigns_default_campaign_id: dilfs`. The standalone role already
exposes this host-specific setting; no new pipeline variable is required.
- The initial real-host run is an operational acceptance exercise. It should
test observable pipeline behavior and convergence, not module task traces or
the current text of defaults.
The workspace [Ansible](../../ansible.md), [testing](../../testing.md), and
[planning](../../planning.md) policies apply. Feature 0015 has passed review
and supplies the required suite interface.
## Operational Prerequisites
Local repository preparation and syntax/dependency checks can proceed with the
workspace's existing Ansible virtual environment. Before live deployment,
establish the controller, target access, and Vault details below. Select the
session/audio before the end-to-end exercise, not before local implementation.
Record non-secret choices in the implementation plan or deployment runbook:
- Identify the operational Ansible controller and its deployment checkout.
The repository configuration currently selects `/etc/ansible/inventory`,
`/etc/ansible/roles`, and `/etc/ansible/collections`. Confirm the effective
paths so dependency installation and playbook execution use the same
inventory and installed content. Distinguish those paths from this
development workspace.
- The target must satisfy the roles' Ubuntu 26.04 and access requirements.
Normal connection, role guards, and privileged tasks can establish these;
separate platform or sudo probes are not required acceptance tests.
- Establish the controller's existing Vault password or identity mechanism
and operator availability to populate the four credential values in the
encrypted host file. Record how Vault is selected, never its password or
decrypted contents.
- Have the operator select the acceptance campaign, session identity, and
audio input, with the corresponding session file and accessible input
location. This is a real processing and publication exercise using the
selected providers; session selection remains an operator decision.
These are deployment inputs and operational checks, not additional production
tasks or a new automated preflight framework.
- Keep playbooks minimal and settings in inventory. Preserve `pipeline-dnd`
and its connection identity; do not rename it to `pipeline.dnd`.
- Preserve `dilfs` using the existing standalone Narratio variable. Remove
legacy `narratio_s3_*` and `narratio_default_campaign_id` host settings.
- Missing operational inputs are documented, not treated as implementation
blockers. No remote controller, SSH identity, Vault password, or session is
required for repository preparation.
- Local dependency installation and syntax/host-selection checks are permitted.
Existing passing evidence may be reused when relevant inputs are unchanged.
- `ansible-doc` rendering is not an acceptance gate. Missing role argument
descriptions are a separate documentation improvement.
## Simplest Viable Implementation
Use the reviewed collection revision and associated compatibility manifest to
update the controller's single root requirements file. Add the minimal playbook and one ordinary
host-variable file for the endpoint, operator, campaign, and credential
references, inheriting the suite's storage and publication defaults. Add one
fully encrypted host Vault file for the four credential values, populated by
the operator without exposing its plaintext during implementation or review.
Install or update dependencies from the root manifest at the confirmed
controller paths, explicitly refreshing the collection from its pinned commit
even when `0.1.0` is already installed. Confirm that Ansible
resolves `rakestraw.dnd.pipeline`, syntax-check the playbook, and converge only
`pipeline-dnd`. Start a fresh operator login after enrollment, validate the
native Narratio configuration under the intended runtime identity, rerun the
same play immediately to confirm clean convergence, and process one selected
session through the complete Feature 0015 publishing path. Record concise
operational instructions and results; do not add a second test harness around
the real deployment.
## Rationale and Alternatives
Keeping the playbook nearly declarative makes inventory the durable record of
what differs about this host and allows ordinary Ansible precedence to remain
the customization mechanism. Embedding endpoint, operator, storage, or secret
mapping variables in the playbook would make a one-host play harder to reuse
and would diverge from the established `rakestraw-ansible` style.
A local wrapper role could hide the collection interface, but it would create
another layer with no independent behavior. Applying the collection role by
FQCN proves the same consumer path that future hosts should use.
Tracking collection `main` would be convenient during development, but it
would make a later controller bootstrap select an unreviewed compatibility
set. An exact public commit retains development flexibility until the next
release while still making this deployment reproducible.
Feature 0015 now supplies the reviewed storage and publishing path needed for
meaningful end-to-end acceptance. Inheriting that policy keeps host inventory
focused on deployment choices and avoids maintaining a second copy of the
suite defaults.
Retain the already-committed minimal playbook, manifest, and public inventory.
Add adjacent comments for unresolved credentials, align the runbook with the
repository-only scope, perform appropriate local checks, and commit. No new
Ansible task, runtime test harness, or deployment automation is needed.
## Acceptance Criteria
- Feature 0015 is complete and independently reviewed before Feature 0016
implementation begins.
- `playbooks/pipeline/dnd.yml` targets only `pipeline-dnd`, applies only
`rakestraw.dnd.pipeline`, and contains no deployment variables or procedural
setup.
- Host inventory sets the confirmed WhisperX endpoint, enrolls only `eric`,
preserves `dilfs` as the default campaign through the existing standalone
role variable, and inherits Feature 0015's storage and publishing defaults.
- `dnd_pipeline_secrets` maps the required typed runtime keys to host Vault
variables. Credential values in the consumer repository are Vault-encrypted;
runtime delivery uses the suite's existing restricted secrets files.
- This feature introduces no plaintext credential values into commits, diffs,
command examples, validation output, or documentation. A repository-wide
historical secret audit is not an acceptance requirement.
- Legacy Narratio storage values are either represented through the current
pipeline interface or deliberately retired; no stale high-precedence host
variable can override the selected suite configuration accidentally.
- The root requirements manifest preserves unrelated existing dependencies
and installs the exact reviewed public collection snapshot, its five
standalone role dependencies, and its required native collections without
using local repository paths.
- The controller uses the selected collection commit even if collection
version `0.1.0` was already installed. The documented inventory, role, and
collection paths correspond to the operational controller.
- Public dependency installation succeeds, and the new playbook passes syntax
checking with the installed collection and roles. Documentation rendering
with `ansible-doc` is not required for deployment acceptance.
- The initial limited playbook run against `pipeline-dnd` completes
successfully using the encrypted inventory and ordinary sudo-capable remote
account.
- The immediate identical second run reports `changed=0` and `failed=0`.
- Narratio's native configuration validation succeeds under the intended
runtime identity, and a fresh session for `eric` can access and invoke the
shared pipeline through runtime-group membership.
- One operator-selected D&D session completes transcription, processing,
storage, and publishing, and its expected published output is observable at
the configured destination. Either supported runtime identity may execute
it. Production is the default profile; the operator may select the testing
profile while retaining the same processing and publication acceptance scope.
- The operational documentation explains dependency installation, Vault use,
deployment invocation, the login-session membership caveat, convergence
verification, and representative pipeline invocation without disclosing
credentials.
- The historical proof-of-concept role and playbook remain available during
the initial rollout; their later retirement requires a separate explicit
decision.
- The playbook targets only `pipeline-dnd` and applies only the collection role.
- Host inventory expresses the known choices above and documents unknown
credentials next to their references without real or fake secret values.
- The dependency manifest contains the reviewed public pins and preserves
unrelated dependencies. Public installation and local syntax/host-selection
evidence are available; no documentation-rendering check is required.
- The runbook distinguishes repository readiness from future deployment and
explains how the operator supplies missing values and eventually invokes it.
- Changes are committed in the owning repositories, with clean working trees
for the feature's changes. No controller deployment, target convergence,
clean-repeat run, or real pipeline execution is required or claimed.
- Historical role/playbook files remain available.
## Dependencies and Follow-up Work
[Feature 0015](0015-dnd-s3-storage-and-publishing.md) supplies the implemented
and independently reviewed pipeline interface, credentials, and publishing
boundary consumed here. Its administrative closeout remains before the
Feature 0016 sprint; its runtime implementation needs no further remediation.
Features 0013–0015 supply public dependencies, operator enrollment, and storage
policy. Feature 0016's consumer preparation already exists in `12b2462`, with
initial documentation in `716a80e`; remaining work aligns comments and docs
with the user's clarified scope.
Features 0013 and 0014 provide the public collection installation path and
existing-operator enrollment required here. Features 0009 and 0011 provide
the shared filesystem and managed credential-delivery policies.
After the real deployment demonstrates parity, a later cleanup feature may
retire the proof-of-concept `pipeline_dnd` role, its old playbook, and any
obsolete inventory values. Service scheduling, unattended ingestion,
additional human operators, backup policy, and additional hosts also remain
separate work.
The operator may later provision Vault values, install dependencies on their
controller, deploy to the host, and validate real processing/publication.
Those actions require a separate request and are not unfinished Feature 0016
acceptance requirements.