Universal Tuya docsHomeSupported devicesRequest a deviceFAQChangelogArchitectureKnowledge baseCreditsSourcesContributing

P3065 — Memory audit (all branches)

Trigger: Bastien OOM on v1.0.161 / v1.0.209 (relays stop, app killed, JSON.parse ~62 MB in diag 097f1743). Method: static review + node harness scripts/perf/p3065-lifecycle-leak-harness.js (real driver/device classes, stubbed Homey/zclNode, N devices per driver, DP events, delete/re-pair cycles, --expose-gc heap after GC). Env knobs: DRV=a,b drivers, FLOOD=k DP flood blocks (out.flood), INTRO=1 per-device collection sizes (out.intro). Usage: node --expose-gc scripts/perf/p3065-lifecycle-leak-harness.js <root> <devicesPerDriver> <events> <cycles>.

Existing protections reused (not duplicated): MemwarnGuard, memwarn-cleaners, MemoryBudget (P3023), BoundedJsonCache, safeMemoryUsage, LogBuffer/log-buffer (bounded), LiveDataUpdater payload cap, TuyaEF00Manager _dpStateDedup (≤500).

Leaks / crashes found

#FindingEffectFixBranches
1homey.setInterval timers owned by a device (and its managers) are only cleared at app destroy, not on device delete/uninitevery delete/re-pair or re-init leaked intervals + the whole dead device graphlib/utils/OwnedTimerSweep.js called after _destroyDevice in TuyaZigbeeDevice onDeleted/onUninit (+ deferred 2nd sweep 30 s) and BaseUnifiedDevice onDeletedall 3
2this.warn() called on SimpleClass (no warn) — VirtualButtonMixinTypeError at initwarn() on TuyaZigbeeDevice + guardall 3
313 ReferenceError/TypeError at runtime (water leak DP1 alarm transform, contact_climate getManufacturer, radar divisor/unbounded lux Map, subDeviceId in 5 switch onDeleted, super.onNodeInit in onDeleted of rcbo/remote, climate flow temp, missing safeMultiply imports, TRV getter-only dpMappings)crash loops → re-init → leak amplificationtargeted fixes + ESLint no-undef ratchet testall 3 (TRV master only; stable air_purifier_motion extra)
4Late async init after delete (IntelligentDeviceAdapter maintenance, PeriodicAutoEnricher)re-armed timers on dead device_destroyed guardsall 3
5DeviceIOFacade startReportingPollFallback armed by the async interview after delete, never cleared1–5 forever intervals/run under CPU pressure (seen on CI bastien 40312247579)refuse when _destroyed, self-clear on tick, stopReportingPollFallback()all 3
6Per-device diag rings (receptionManager 200, diag loggers 100) and _dpDedup (keys embed raw JSON, purged only every 100 DPs, unguarded _protocolStats)bounded but large on big meshes; possible TypeErrormemwarn cleaner p3065-diag-rings (keep 20) + hard cap 256 keys + guarded purgeall 3

Measurements (heap after GC, MB)

Delete/re-pair cycles:

RunBeforeAfter
master, all 471 drivers — intervals left per cycle289 → 581 → 8730–3 (now 0 after #5)
master, all 471 drivers — heap afterDelete88.8 → 107.7 → 128.460 / 60
master — dead devices still reachable144 → 289 → 432~26–28 constant
stable-v5, 12 drivers × 120 devices, 60 ev, 3 cycles49.5 → 61.4 → 70.2 (timers 240/480/720)45.3 / 46.2 / 45.2 (timers 0)
bastien-home, same47.9 → 59.6 → 68.7 (> 64 MB Homey ceiling)41.4 / 42.8 / 43.4 (timers 0)
bastien-home, 7 drivers × 40, 6 runs on 1 CPU (CI-like) — intervals left1–5 (DeviceIOFacade)0 in 6/6

DP flood without deletes (bounded): 40 dev × 8 drivers × 25 blocks plateaus at 54.6 MB; 5 drivers × 60 dev × 4 blocks 57.4–58.6 MB.

Not fixed (noted, behavior changes or low value)

Tests

test/critical/p3065-owned-timer-sweep.test.js, p3065-undef-crash-fixes.test.js, p3065-memwarn-diag-rings.test.js, p3065-io-report-poll-leak.test.js (node:test, run by node22-tests).

Commits: master af5e17c403, d1eea228de, 0f624c0f6f, 391e6191f2 · stable-v5 24ae0c5c9c+1aad09f8eb, d3069acd74, 84f67aeafc, c06671ad17 · bastien-home 8ee643d6fe+e88235ac65, 13be6bf719, 945252c5b8, 1452bf3d56.

← Knowledge base index · source