Method Details

• getScore

  double getScore()

  Computed score.

  Returns:

  the health score, where 1.0 is healthy and lower is worse

• compositeScore

  static double compositeScore(double[] factors, double[] weights)

  Weighted geometric mean of the available factors.

  NaN factors mean "insufficient data" and are skipped along with their weights, so the remaining weights are renormalized to 1.0. Every participating factor is floored at MIN_FACTOR_SCORE first — without that floor a single saturated factor zeroes the aggregate and the score stops discriminating between bad and catastrophic.

  Parameters:

  factors - per-factor scores in [0,1]; NaN for no data

  weights - matching per-factor weights, summing to 1.0

  Returns:

  the composite score in [0,1], or 1.0 if no factor has data

  See Also:

  • MIN_FACTOR_SCORE

• compute

  private void compute()

  Scores every subsystem and combines them into the health score.

• scoreJobLag

  private double scoreJobLag()

  Job queue lag: 0ms = 1.0, >50ms = degraded, >200ms = 0.0 Also considers jobRunSlow (jobs >1s) as sustained overload signal.

  Returns:

  the job-lag health factor in [0,1], or NaN if the period holds fewer than three job-lag events

• scoreReadyJobs

  private double scoreReadyJobs()

  Ready (queued) jobs as a fraction of pool capacity. >50% of pool capacity queued = overload indicator.

  Returns:

  the ready-jobs health factor in [0,1], or NaN if the period holds fewer than three samples

• scoreDroppedJobs

  private double scoreDroppedJobs()

  Dropped jobs per interval — any dropped job is a severe overload signal.

  Returns:

  the dropped-jobs health factor in [0,1], 1.0 when nothing was dropped, or NaN if the stat is unavailable

• scoreJobWait

  private double scoreJobWait()

  How long jobs wait in the queue before execution. 0ms→1.0, 500ms→0.5, 2000ms→0.0.

  Returns:

  the job-wait health factor in [0,1], or NaN if the period holds fewer than three samples

• scoreBuildSuccess

  private double scoreBuildSuccess()

  Build success rate from 60-second event counts. More responsive than SystemVersion's 10-minute window and avoids NaN when that window hasn't accumulated events yet. Requires 10+ total events (exploratory + client) for stability. When firewalled, lower thresholds since inbound builds are rejected.

  Returns:

  the build-success health factor in [0,1], or NaN if fewer than 10 build events landed in the period

• scorePoolAlive

  private double scorePoolAlive()

  Fraction of registered tunnel pools that are alive. 1.0 = all alive, 0.0 = half or more dead.

  Returns:

  the pool-alive health factor in [0,1], or NaN when no tunnel pools are configured

• scorePoolDeficit

  private double scorePoolDeficit()

  Fraction of alive pools meeting their target tunnel count. 1.0 = all pools meet target, 0.0 = none do. Linear ramp: 0%→0.0, 70%→1.0. Pools still building at startup won't penalize until sufficient time has passed to meet targets.

  Returns:

  the pool-target health factor in [0,1], or NaN when no tunnel pools are configured

• isFirewalled

  private boolean isFirewalled()

  True if any transport reports firewalled status.

  Returns:

  true if a transport is firewalled

• getEventCount

  private double getEventCount(String statName, long period)

  Fetch the event count for a stat using the given period (ms).

  Parameters:

  statName - the statManager rate name to read

  period - the rate period in ms, e.g. RateConstants.ONE_HOUR

  Returns:

  the event count in that period, 0 if the stat or rate has no data

• getEventCount

  private double getEventCount(String statName)

  Fetch the event count for a stat using the 60s period.

  Parameters:

  statName - the statManager rate name to read

  Returns:

  the event count in the 60s period, 0 if the stat has no data

• getLifetimeEventCount

  private double getLifetimeEventCount(String statName)

  Return lifetime event count for a stat. Use for low-frequency stats where 60s window is too narrow.

  Parameters:

  statName - the statManager rate name to read

  Returns:

  the number of events recorded since the stat was created, 0 if the stat is unavailable

• scoreBuildStorms

  private double scoreBuildStorms()

  Build storms: 0 concurrent = 1.0, linear decay to 0 at 100. Startup bursts (30-50 concurrent) are normal; only penalize sustained high concurrency.

  Returns:

  the build-storm health factor in [0,1], or NaN if the period holds fewer than five samples

• scoreSendFailure

  private double scoreSendFailure()

  Transport send failure rate: failed sends vs total sends. Uses event counts (each send = 1 event, each failure = 1 event). Requires 500+ sends to skip noisy startup window. <1% = 1.0, >10% = 0.0.

  Returns:

  the send-failure health factor in [0,1], or NaN if fewer than 500 sends were recorded in the period

• scoreLatency

  private double scoreLatency()

  Message send latency: how long an outbound message waits in the transport send queue (transport.sendProcessingTime). ≤1s → 1.0, 4s → 0.5, ≥10s → 0.0.

  Anchored on what the router itself treats as normal versus congested: RouterThrottleImpl documents ~25ms for NTCP and ~1000ms for SSU as ordinary (SSU includes ack time and can be much more than RTT), and starts refusing participating tunnel requests once the smoothed average passes maxProcessingTime — 2000ms, 3000ms on slow systems. BuildExecutor likewise backs builds off above a 2000ms average. The old mapping reached 0.0 at 5000ms, only 2.5× the router's own congestion gate, so a plain SSU router sitting at its documented average scored 0.5 and one with slow acks scored 0.0.

  Returns:

  the send-latency health factor in [0,1], or NaN if the period holds fewer than three samples

• scoreGhostPeers

  private double scoreGhostPeers()

  Ghost exclusions as a fraction of the fast or high-capacity tier — the pool tunnel builds actually draw from. Up to GHOST_TIER_FREE of the tier excluded is free; the factor then ramps to 0.0 at GHOST_TIER_ZERO.

  Fraction, not absolute count: marks accrue routinely (a few per minute against multi-minute cooldowns), expire on their own, and clear the moment the peer answers, so any fixed count threshold fires on every healthy router — which is what pinned the composite to 0%. Scaling by the tier keeps a large router free while still catching the case that matters, where exclusions consume enough of the preferred pool to starve selection.

  Numerator and denominator must cover the same population: the count is tier-matched, because dividing an all-tier ghost count by a tier size would overstate whenever the tier is smaller than the ghost population.

  Returns:

  the ghost health factor in [0,1], or NaN if the tier is unknown

• scoreStreaming

  private double scoreStreaming()

  Streaming health: multi-signal score combining connection success rate, RTT at close, retransmission overhead, and inbound reset rate. Connection success rate dominates for client-heavy routers.

  Returns:

  the weighted streaming health factor in [0,1], or NaN when no signal has enough events to score

• clamp

  private static double clamp(double v)

  Limits a score to the valid range.

  Parameters:

  v - the raw score

  Returns:

  the score, clamped to 0 to 1

• computeSubsystemScores

  Map<String,double[]> computeSubsystemScores()

  Compute per-subsystem health scores for the ring chart dashboard. Each subsystem gets a score based on its most relevant stat(s).

  Returns:

  map of subsystem key → {score, ...detail}

• scoreTransit

  private double scoreTransit()

  Scores transit capacity against the tunnels offered.

  Returns:

  the score, 0 to 1

• scoreBuildQueue

  private double scoreBuildQueue()

  Scores the tunnel build handler queue depth.

  Returns:

  the score, 0 to 1

• scoreBanHit

  private double scoreBanHit()

  Scores how often peers are banned on contact.

  Returns:

  the score, 0 to 1

• scoreSendPool

  private double scoreSendPool()

  Scores the UDP send pool saturation.

  Returns:

  the score, 0 to 1

• scoreAvgRTO

  private double scoreAvgRTO()

  Scores the average UDP retransmit timeout.

  Returns:

  the score, 0 to 1

• scoreNetDbHealth

  private double scoreNetDbHealth()

  Scores network database lookup success and speed.

  Returns:

  the score, 0 to 1

• scoreI2cpHealth

  private double scoreI2cpHealth()

  Scores I2CP client session health.

  Returns:

  the score, 0 to 1

• scorePeerHealth

  private double scorePeerHealth()

  Scores peer connectivity and profile quality.

  Returns:

  the score, 0 to 1

• scoreCongestion

  private double scoreCongestion()

  Congestion: fraction of message send outcomes that triggered a window-reduction (congestion) event.

  The denominator is udp.sendConfirmTime, which fires once per ACKed message, so this is a true per-message congestion rate rather than a per-packet count. udp.congestionOccurred fires per RTO-timeout or fast-retransmit window reduction.

  Thresholds are calibrated for UDP/SSU over a lossy anonymizing overlay, where normal AIMD keeps a busy router in the 5-15% band. A wired-TCP-style 5% cap would flag healthy routers as congested. <=5% congestion → 1.0, 25% → 0.0.

  Requires at least 100 confirmed sends in the period for a stable ratio.

  Returns:

  the worst of the congestion, retransmit, expiry and inbound-drop health factors, each in [0,1]

• scoreCryptoPressure

  private double scoreCryptoPressure()

  Crypto: DH queue empty events vs used events. Frequent empty queue = key generation bottleneck. Only evaluates pools with actual activity; returns NaN when no crypto ops.

  Returns:

  the worst crypto-pool health factor in [0,1], or NaN when no pool did any work in the window

• getStatValue

  private double getStatValue(String statName, long period)

  Read a stat's average value using the given period (ms). Returns NaN if stat unavailable.

  Parameters:

  statName - the statManager rate name to read

  period - the rate period in ms, e.g. STAT_PERIOD_LONG

  Returns:

  the stat's average value over that period, NaN if unavailable

• getStatValue

  private double getStatValue(String statName)

  Read a stat's average value using the 60s period. Returns NaN if stat unavailable.

  Parameters:

  statName - the statManager rate name to read

  Returns:

  the stat's average value over 60s, NaN if unavailable