GPU Availability Cache Methods SetGPUAvailability caches GPU availability data

SetHealthChecker wires the GPU health monitor so the handler can reject tasks while the provider is offline.

SetMetricsProvider wires a real GPU-metrics source (the daemon's detector) into the billing monitor.

SetProviderRates forwards a provider's rate update to the billing service.

SetReporter sets the reporter for the handler (used to break init cycle with NATS client).

SettleRentalJob finalizes the escrow to the provider after the challenge window (the provider payout, on-chain).

SettleRipeJobs is the provider-payout crank. It finalizes every delivered, undisputed Covenant job (taker == platform in Phase 1) whose challenge wind

SpecHash computes the 32-byte spec hash exactly as Covenant does: sha256(JSON.stringify(spec)) with no HTML escaping. The same hash must be sent to th

Start starts the GPU monitoring service

Start initiates the HTTP server listening process.

Start starts the device plugin server

Start runs the health loop until the context is cancelled. It blocks, so run it in a goroutine.

StartBilling informs the billing service that a task's billable period has begun.

StartConsuming subscribes to the NATS subject for job submissions and starts processing messages. This uses a JetStream PullSubscription for more cont

StartListening subscribes to the task subject and starts processing messages.

StartSession starts a new rental session

Stop gracefully shuts down the JobConsumer.

Stop gracefully shuts down the HTTP server.

StopBilling ends the rental session in the billing service. The session is keyed by the id the scheduler placed in the task; the daemon only needs to

ToInternalJobRepresentation converts a JobRecord from the database back to an InternalJobRepresentation.

TotalBalance returns the total balance including locked and pending

TransferTokens transfers USDC between wallets

UnlockFunds unlocks the specified amount

UpdateJobState updates specific fields of a job: its state, assigned provider ID, last error, and increments attempts. It also updates the UpdatedAt t

UpdateStatus updates the provider's status and last seen time.

Upload uploads a file to the specified bucket with the given key. reader is the content of the file, size is the total size, and contentType is the MI

ValidateJobRequirements validates that a job can be billed properly

ValidatePricingRequest validates a pricing request

ValidateRefreshToken validates a refresh token and returns the claims

VerifyEmail handles email verification

WithCorrelationID returns a new context with the correlation ID set

WithRetry executes a database operation with retries for transient errors

WithdrawTokens processes a dGPU token withdrawal

appleEstimateVRAMByName is a coarse per-chip fallback used only when sysctl is unavailable. It mirrors typical base unified-memory configurations and

appleUnifiedMemoryMB returns the machine's total physical memory in MB via sysctl hw.memsize. On Apple Silicon the GPU shares this unified memory, so

applyDefaultsIfNotSet applies default values to cfg fields if they are zero-valued. This is useful if the config file exists but is missing some keys.

applyDefaultsIfNotSet applies default values from `defaults` to `cfg` for zero-valued fields.

applyDefaultsIfNotSet applies default values to cfg fields if they are zero-valued.

applyDefaultsIfNotSet applies default values to cfg fields if they are zero-valued.

authenticate with API Gateway

bestProviderPowerW returns the highest rated power envelope (in Watts) across the provider's GPUs. Falls back to a model-name heuristic when the regis

bestProviderVRAM returns the largest single-GPU VRAM (in MB) the provider exposes. Heterogeneous rigs may pack different cards; we match against the s

broadcast sends message to all WebSocket clients

broadcastData broadcasts GPU data to WebSocket clients

buildEmailMessage builds the email message with headers

calculateSystemStats calculates overall system statistics

Helper functions

canonical renders the receipt deterministically (sorted keys) for hashing.

cleanupContainer removes a Docker container

collectContainerLogs collects logs from a Docker container

collectGPUData continuously collects GPU data

collectGPUMetrics collects current GPU metrics

collectMetrics collects system and GPU metrics during task execution

collectNVIDIAMetrics collects NVIDIA GPU metrics

collectSystemMetrics collects system and GPU metrics

convertAMR converts authentication methods reference

create2FASession creates a temporary session for 2FA verification

createSession creates a new user session

createWallet creates a new wallet for the user

defaultPricingConfig mirrors services/billing-payment-service/configs/config.yaml so the engine computes the same figures the live service would.

deleteObjectHandler handles object deletions.

deregisterServiceFromConsul deregisters the service from Consul.

detectAMDGPUs detects AMD GPUs (Linux only)

detectAMDGPUs detects AMD GPUs using rocm-smi

detectAppleGPUs detects Apple Silicon GPUs (macOS only)

detectAppleGPUs detects Apple Silicon GPUs

detectGPUs detects available GPUs on the system

detectIntelGPUs detects Intel GPUs

detectIntelGPUs detects Intel GPUs via sysfs (vendor id 0x8086). This works without any Intel tooling and reports real devices instead of a placeholde

detectNVIDIAGPUs detects NVIDIA GPUs using nvidia-smi

detectNVIDIAGPUs detects NVIDIA GPUs using nvidia-smi

downloadFile downloads a single file

downloadInputFiles downloads input files for the task

downloadObjectHandler handles object downloads.

endBillingSession ends the rental session that the scheduler created for this task (threaded in via JobParams["session_id"]). It runs for every termin

estimateJobCost handles cost estimation

executeDockerTask executes a task using Docker

executeScriptTask executes a script-based task