15 KiB
nenjim-hub-configuration Specification
Purpose
Defines the mandatory deployment configuration that selects which modern Nenjim processes the Hub constructs and starts, while preserving explicit order and fail-fast lifecycle semantics.
Requirements
Requirement: Hub startup requires one fixed runtime configuration
The reference Hub SHALL read only conf/NenjimHub.conf and SHALL complete all configuration parsing and validation before starting Registry or any other process. The path SHALL not be configurable through Main, command-line arguments, or the public NenjimHub API. A missing, unreadable, or non-regular runtime file SHALL fail Hub startup clearly and identify the configured path. The Hub SHALL not read, copy, generate, or fall back to conf/NenjimHub.conf.example, and SHALL not fall back to a hardcoded startup list.
Scenario: Runtime configuration is valid
- WHEN
conf/NenjimHub.confis a readable regular file containing a valid format 1 configuration - THEN the Hub derives its complete requested process order from that file before starting Registry
Scenario: Runtime configuration cannot be read
- WHEN
conf/NenjimHub.confis missing, unreadable, or not a regular file - THEN startup fails with a diagnostic identifying
conf/NenjimHub.confbefore any process starts
Scenario: Only the example exists
- WHEN
conf/NenjimHub.confis absent butconf/NenjimHub.conf.exampleexists - THEN startup fails without reading, copying, or using the example and without starting a hardcoded fallback set
Requirement: Format 1 has one exact ordered AUTO_START section
The first actual configuration entry SHALL be exactly FORMAT_VERSION=1, after optional leading blank lines and full-line comments. Version 1 SHALL be the only supported version; a missing, malformed, duplicate, version 2, or otherwise unsupported declaration SHALL fail. The next actual entry SHALL be exactly one [AUTO_START] header, optionally followed only by a trailing # comment. Unknown, malformed, repeated, valued, or out-of-order sections and entries before [AUTO_START] SHALL fail.
Outside [AUTO_START], leading and trailing whitespace SHALL be ignored, a line whose trimmed content begins with # SHALL be a full-line comment, and # after actual content SHALL begin a trailing comment. Inside [AUTO_START], surrounding whitespace around a process ID SHALL be ignored; full-line and trailing comments SHALL be ignored; comment-only lines SHALL not terminate the section; and the first whitespace-only line or end of file SHALL terminate it. After termination, only blank lines and comments SHALL be accepted. Format 1 SHALL provide no quoting or escaping for #.
Scenario: Leading comments and trailing comments are accepted
- WHEN blank lines or full-line comments precede
FORMAT_VERSION=1, and the version, section header, or process entries contain trailing#comments - THEN comment text and permitted surrounding whitespace are ignored while the actual entries are parsed normally
Scenario: Comment inside AUTO_START does not terminate it
- WHEN a comment-only line appears between two process IDs in
[AUTO_START] - THEN both process IDs remain part of the ordered section
Scenario: Empty line terminates AUTO_START
- WHEN a whitespace-only line appears after at least one
[AUTO_START]entry - THEN the section terminates at that line and later actual content fails as unexpected content
Scenario: End of file terminates AUTO_START
- WHEN the final process ID is immediately followed by end of file
- THEN the section terminates successfully at end of file
Scenario: Version declaration is invalid
- WHEN the version is absent, malformed, duplicated, not the first actual entry, or has a value other than
1 - THEN configuration validation fails before any process starts and identifies the path, relevant line and reason
Scenario: Section structure is invalid
- WHEN
[AUTO_START]is absent, duplicated, misspelled, has a value or non-comment trailing token, follows another actual entry, or is replaced by an unknown section - THEN configuration validation fails before any process starts and identifies the path, relevant line, offending content and reason
Requirement: The complete requested process plan is validated before startup
Every active [AUTO_START] line SHALL be parsed through the canonical NenjimComponentId rules and retained in source order with its line location. Before Registry starts, validation SHALL reject invalid ID syntax, duplicate IDs, unknown IDs, passive-only component IDs, and IDs the reference Hub cannot construct as canonical NenjimProcess objects. The section SHALL contain Registry exactly once at position one and Journal exactly once at position two. Applicable failures SHALL identify conf/NenjimHub.conf, the relevant line, offending value, and reason.
Scenario: Process ID is malformed
- WHEN an active section line is not a valid
NenjimComponentId - THEN validation fails before Registry startup and identifies the path, line, offending value and component-ID reason
Scenario: Process ID is duplicated
- WHEN the same canonical process ID appears more than once
- THEN validation fails before Registry startup and identifies the duplicate value and line
Scenario: ID is not constructible as a process
- WHEN an entry is unknown, identifies only a passive component, or is not in the Hub's modern process-construction catalogue
- THEN validation fails before Registry startup rather than constructing or starting any process
Scenario: Bootstrap entries are not first and second
- WHEN Registry is absent or not first, or Journal is absent or not second
- THEN validation fails before Registry startup and identifies the required placement
Requirement: Registry bootstraps eager passive composition
After validating the entire plan, the Hub SHALL construct the configured Registry subsystem, start Registry, register its service and process object as nenjim.registry.service, and record that ID only after successful startup. The Hub SHALL then eagerly construct and register the existing shared passive component graph. Dedicated liquidity deployment configuration, wallet signer, Jupiter swap service and direct Raydium service SHALL instead be constructed and registered only when the configured evelyn.liquidity.prod entry is reached. This passive phase SHALL not construct any canonical NenjimProcess, start background workers, initialize JavaFX, add startup IDs, infer process dependencies, or implicitly create an omitted process. Plain internal dependencies that are not NenjimComponent objects MAY be constructed as part of passive composition.
Scenario: Registry surrounds passive composition
- WHEN a valid Hub startup begins
- THEN Registry starts, is registered, and is recorded before the existing shared passive component graph is constructed and registered
Scenario: Passive composition completes
- WHEN Registry bootstrap succeeds
- THEN all existing shared non-process building blocks, adapters, and services are constructed and registered without starting a process
Scenario: Passive construction fails
- WHEN any passive component cannot be constructed or registered
- THEN startup stops immediately with Registry remaining recorded and no configured process after Registry constructed or started
Scenario: Liquidity is omitted
- WHEN the plan omits
evelyn.liquidity.prod - THEN its deployment configuration, dedicated signer/wallet, Jupiter service and direct Raydium service are neither loaded nor constructed nor registered
Requirement: Selected processes are constructed and started lazily in exact order
The Hub SHALL support the modern process IDs nenjim.registry.service, nenjim.journal.service, assetaz.ticker.default, jupiter-perps-alarm.default, evelyn.status.prod, evelyn.status.test, evelyn.iou-burner.prod, evelyn.liquidity.prod, nenjim.javafx-runtime.service, evelyn.mission-control.default, and nenjim.process-manager.default through implementation-only construction capability that creates no process instance during validation. After passive composition, the Hub SHALL visit every configured entry after Registry in declared order, beginning with Journal. At each entry it SHALL construct only that process and any explicitly composed process-specific passive dependencies or administration view, register the process under the exact configured ID, resolve it through the canonical NenjimProcess contract where applicable, call start(), and record the ID only after successful return.
The production IOU burner SHALL receive the already constructed Production Evelyn status service, and SHALL require both Ticker and evelyn.status.prod to precede evelyn.iou-burner.prod. The configuration SHALL be the complete manually ordered process dependency closure. The Hub SHALL not discover, insert, recursively construct, reorder, or automatically start dependencies. A process omitted from [AUTO_START] SHALL not be constructed, registered, started, or recorded, even when its implementation is available. Legacy process prototypes outside the modern Hub graph SHALL not be selectable.
Scenario: Configured process starts successfully
- WHEN a configured process entry is reached and its dependencies are already available
- THEN that process is constructed, registered, started, and recorded before the next entry is reached
Scenario: Process is omitted
- WHEN a known process ID does not appear in
[AUTO_START] - THEN no instance of that process is constructed or registered and its lifecycle is never started or recorded
Scenario: Jupiter Perps Alarm is selected
- WHEN
jupiter-perps-alarm.defaultis explicitly listed after its required predecessors - THEN the otherwise inactive modern alarm process is constructed, registered, started and recorded at that exact position
Scenario: Dependency order is invalid
- WHEN a selected process is reached before a process dependency required by its current constructor or startup behavior
- THEN construction or startup fails clearly without implicit dependency construction, repair, or reordering
Scenario: JavaFX processes are omitted
- WHEN a valid headless plan omits
nenjim.javafx-runtime.service,evelyn.mission-control.default, andnenjim.process-manager.default - THEN none of those objects is constructed or registered and the JavaFX toolkit remains uninitialized by the Hub
Scenario: Renamed status services retain dependency wiring
- WHEN
evelyn.status.prodandevelyn.status.testare listed after Ticker and before Mission Control - THEN the Hub constructs separate Production/Test status services, registers the renamed IDs, and injects the corresponding typed service views into Mission Control in the configured order
Scenario: Deployment still contains an old status ID
- WHEN a private deployment configuration contains
evelyn.service.prodorevelyn.service.test - THEN validation rejects the unknown ID before startup; deployment migration replaces only that ID with
evelyn.status.prodorevelyn.status.test, preserving selection and order
Scenario: Liquidity is explicitly selected
- WHEN
evelyn.liquidity.prodis reached in a valid startup plan - THEN the Hub loads its runtime configuration, binds the dedicated signer/wallet and Jupiter service to each other, constructs direct Raydium access and registers/starts the liquidity process without inserting a Ticker or status dependency or reordering the plan
Scenario: Production burner has its reference-price dependency
- WHEN
evelyn.iou-burner.prodis reached after Ticker andevelyn.status.prod - THEN the Hub injects the corresponding Production status service for buyback price decisions
Scenario: Production burner precedes its status dependency
- WHEN the burner is reached without a previously constructed Production status service
- THEN construction fails clearly without creating or starting the omitted dependency or the burner
Requirement: Configured startup is fail-fast and reports actual milestones
Registry and every selected process SHALL be registered before its start() call and appended to the Hub's immutable ordered startup snapshot only after start() returns successfully. Construction failure SHALL leave the selected process unregistered. Registration or startup failure SHALL stop the plan before any later entry is constructed or started; a startup failure MAY leave that process registered but SHALL not record it. Earlier successful processes SHALL remain in snapshots, no rollback or retry SHALL occur, and the Hub SHALL announce its existing online point and enter its blocking runtime wait only after every configured process succeeds.
Scenario: Selected construction fails
- WHEN construction of the current selected process throws
- THEN the process is not registered or recorded, no later process is constructed, and the Hub does not announce online or enter its final wait
Scenario: Selected registration fails
- WHEN registration of the constructed current process throws
- THEN the process is not started or recorded, no later process is constructed, and no rollback is attempted
Scenario: Selected startup fails
- WHEN the current process's
start()throws - THEN it remains unrecorded, no later process is constructed or started, earlier successful IDs remain visible, and no rollback is attempted
Scenario: Configured startup completes
- WHEN all configured process starts return successfully
- THEN the startup snapshot has the exact configured order, the online point is announced, and the Hub enters its existing blocking wait
Requirement: The version-controlled example is a copy-only template
The repository SHALL contain conf/NenjimHub.conf.example as a documented format 1 copy-only template with only the mandatory Registry and Journal entries active. It SHALL contain no optional deployment-specific process IDs. Its comments SHALL state that the file must be copied to conf/NenjimHub.conf before use, Registry and Journal are mandatory, order is preserved, and process dependencies must be listed manually before their consumers. Documentation SHALL show the same minimal bootstrap as a headless example and SHALL describe recursive dependency construction as future work rather than implemented behavior.
Scenario: Example provides a generic minimal bootstrap
- WHEN the example's uncommented process IDs are read in order
- THEN they are only Registry and Journal, with no optional deployment-specific process ID present
Scenario: Operator uses the minimal headless example
- WHEN the documented headless configuration is copied to the runtime path
- THEN it selects only Registry and Journal and leaves every optional process for the operator to add explicitly