Kevin L'HereecandContinue
aad93911e4
- TR-069 CWMP state machine hardening (14 phases): shared event utility, multi-event routing, GPN circuit breaker, IdleState GPN guard, expanded GPN fault codes, CriticalReadsState, ConfigEval GPN guard, task queue race condition fix, explicit error handling, terminal states (OperationalState/PartialOperationalState), consolidated reclassify, crash pattern detection, removed enqueueImmediate, state flow wiring - Fix SOAP Fault not processed by state machine (soap-env:Fault detection) - Fix GPN circuit breaker not persisting after session drops - Fix CriticalReadsState GPV response parsing (xsi:type normalization) - Fix CriticalReadsState not processing GPV response values - Skip SPV when device is already compliant (ProvisioningCode check) - Add Device.DeviceInfo.ModelName to critical reads GPV - Fix TypeError on DeviceDetail parameters view (null guards) - Log empty 204 responses in exchange trace UI - TR-181 catalog: type corrections and new entries (+6640 lines) - SequenceViewer: replace session-nav buttons with compact dropdown select - SequenceViewer: fix CPE/ACS lifeline headers wrapping to 2 lines - Add 20 unit tests for state machine fixes (SOAP Fault, GPN circuit breaker, CriticalReads, stale session, ConfigEval/ComplianceUpdate skips) Generated with [Continue](https://continue.dev) Co-Authored-By: Continue <noreply@continue.dev>
70 lines
4.2 KiB
Markdown
70 lines
4.2 KiB
Markdown
# TR-069 CWMP State Machine — Problem Catalog
|
|
|
|
## CRITIQUE-1: Race Condition task queue / state machine
|
|
|
|
- File: `statemachinemanager.ts` (continue() + setupTaskQueueListeners())
|
|
- Detail: `addTask()` in DeviceActionExecutor emits 'taskSent' synchronously, setting `lastTaskId` BEFORE the XML is returned to the CPE. When both the state machine and task queue serve tasks in the same session, `lastTaskId` can point to the wrong task, corrupting CPE response routing.
|
|
- Impact: Responses routed to wrong callback, silent data corruption, session hangs.
|
|
|
|
## CRITIQUE-2: Multiple superfluous sessions on BOOTSTRAP
|
|
|
|
- File: `cwmpcontroller.ts` (handleInform()) + `statemachinemanager.ts` (startStateMachine())
|
|
- Detail: `enqueueImmediate(GPV)` in the controller + implicit ScheduleInform via SPV systematically generate 3 sessions instead of 1.
|
|
- Impact: 3x ACS load per bootstrap, unnecessary CPE wake-ups, slower onboarding.
|
|
|
|
## CRITIQUE-3: IdleState does not check gpnStatus before sending GPN
|
|
|
|
- File: `mainstatemachine.ts` (IdleState.handle())
|
|
- Detail: A device with gpnStatus='unsupported' (Livebox, buggy CPEs) still receives a GPN every session, causing systematic Fault or CPE crash/reboot.
|
|
- Impact: CPE crash loops, Fault flooding, device never reaches operational state.
|
|
|
|
## CRITIQUE-4: Silent safetyGuard in executeCurrentState()
|
|
|
|
- File: `mainstatemachine.ts` (executeCurrentState())
|
|
- Detail: The `while (safetyGuard++ < 15)` loop stops silently if the machine loops. No log of blocking state, no transition to an explicit error state.
|
|
- Impact: Sessions terminate with empty response, device not marked in error, monitoring blind spot.
|
|
|
|
## MAJEUR-5: Multi-event routing based on LAST event only
|
|
|
|
- File: `statemachinemanager.ts` (startStateMachine())
|
|
- Code: `const lastEvent = this.normalizeEvent(evts[evts.length - 1])`
|
|
- Detail: An Inform with ['0 BOOTSTRAP', '1 BOOT'] processes only '1 BOOT'. An Inform with ['6 CONNECTION REQUEST', '2 PERIODIC'] forces ImmediateCommandStateMachine when onboarding is required.
|
|
- Impact: Wrong state machine selected, missed onboarding, incorrect provisioning flow.
|
|
|
|
## MAJEUR-6: Incomplete GPN fault codes in DeviceActionExecutor
|
|
|
|
- File: `deviceactionexecutor.ts` (getParameterNames() wrappedCallback)
|
|
- Code: `if (code === 9000 || code === 9001) { device.setGpnStatus('unsupported') }`
|
|
- Detail: Codes 9003 (Invalid Arguments), 9004 (Resources Exceeded), 9005 (Invalid Parameter Name) should also trigger fallback. Network/timeout codes should NOT mark unsupported.
|
|
- Impact: Missing fallback on legitimate GPN failures, false unsupported marking on transient errors.
|
|
|
|
## MAJEUR-7: No GPN circuit breaker across sessions
|
|
|
|
- Files: `mainstatemachine.ts`, `configurationevaluationstate.ts`, `deviceactionexecutor.ts`
|
|
- Detail: No failure counter persisted between sessions. A CPE that crashes on GPN will be retried every Inform. No cooldown, no crash pattern detection.
|
|
- Impact: Infinite crash loops for vulnerable CPEs, ACS resource waste.
|
|
|
|
## MAJEUR-8: Critical GPV outside state machine (collision with ConfigEvalState)
|
|
|
|
- File: `cwmpcontroller.ts` (handleInform() enqueueImmediate block)
|
|
- Detail: GPV (HardwareVersion, SoftwareVersion, ConnectionRequestURL) is served by the task queue, not the state machine. It can interleave with GPN/GPV from ConfigurationEvaluationState, corrupting CWMP exchange order.
|
|
- Impact: Out-of-order CWMP messages, session corruption, device misidentification.
|
|
|
|
## MINEUR-9: Duplicated EventCode normalization
|
|
|
|
- Files: `cwmpcontroller.ts` (normalizeEventCode()), `statemachinemanager.ts` (normalizeEvent())
|
|
- Detail: Two identical functions, two maintenance points.
|
|
- Impact: Drift risk, inconsistent normalization.
|
|
|
|
## MINEUR-10: reclassifyDeviceFromCatalog() called at 4 different places
|
|
|
|
- Files: `cwmpcontroller.ts`, `deviceactionexecutor.ts` (x2), `mainstatemachine.ts` (CompletedState)
|
|
- Detail: Scattered fallback logic, hard to unit test.
|
|
- Impact: Inconsistent reclassification timing, test complexity.
|
|
|
|
## MINEUR-11: No explicit PartialOperational state
|
|
|
|
- File: `configurationevaluationstate.ts` (catch block -> { isComplete: true })
|
|
- Detail: On ConfigEval failure, the machine terminates without persisting a degraded status usable for monitoring/alerting.
|
|
- Impact: Monitoring blind to degraded devices, no alerting on partial provisioning.
|