java.lang.Object
net.i2p.router.JobImpl
net.i2p.router.tunnel.pool.TunnelPoolManager.FastTierProbeJob
All Implemented Interfaces:
Job
Enclosing class:
TunnelPoolManager
static class TunnelPoolManager.FastTierProbeJobextends JobImpl
Background reachability probe over the peers a client pool would provisionally use as a first hop.

Selection samples a bounded slice of the high-capacity tier (ProfileOrganizer.maxCandidateSample(int, int)), so an unconnectable peer can sit in that tier indefinitely and be drawn into builds that fail at the first hop. Nothing else records reachability cheaply: send-failure strikes only accumulate on tunnels already in use, and a peer that never answers produces no reply to blame. This job supplies that missing signal by pre-connecting to a bounded slice and letting TunnelPeerSelector.PreConnectFailJob demote whatever fails.

Sweep, pause, sweep: each pass probes at most PROBE_BATCH peers and then waits PROBE_INTERVAL, so the whole tier is covered over several passes without a burst of outbound traffic. Only peers we are not already talking to are probed — a peer with recent successful traffic has fresh evidence already, and probing it would spend the budget re-confirming what we know.

  • Field Summary

    Fields
    Modifier and Type
    Field
    Description
    (package private) static final long
    FAST_PASS_TARGET_MS
    Wall clock wanted for one pass over the whole tier, while young.
    (package private) static final long
    FAST_WINDOW_MS
    Uptime before which the sweep runs at its faster cadence.
    (package private) static final long
    MAX_PROBE_INTERVAL_MS
    Ceiling on the gap between passes, so coverage never stops refreshing.
    (package private) static final long
    MIN_PROBE_INTERVAL_MS
    Floor on the gap between passes regardless of band size.
    (package private) static final int
    PROBE_BATCH
    Peers probed per pass.
    (package private) static final int
    PROBE_BATCH_FAST
    Peers per pass while young, when nothing has been measured yet.
    (package private) static final int
    PROBE_BATCH_STEADY
    Peers per pass once the fast window has passed.
    (package private) static final long
    PROBE_INTERVAL
    Delay between passes, giving the probed slice time to be used or fail.
    (package private) static final long
    RECENTLY_ACTIVE_MS
    A peer we have built through this recently needs no probe.
    (package private) static final long
    STEADY_PASS_TARGET_MS
    Wall clock wanted for one pass once steady.

    Fields inherited from class JobImpl

    _id
    Modifier and Type
    Field
    Description
    final long
    _id
    Unique job identifier.

  • Constructor Summary

    Constructors
    Constructor
    Description
    FastTierProbeJob(RouterContext ctx, TunnelPoolManager mgr)
    Probe whether fast-tier peers are reachable, shortly after startup.

  • Method Summary

    Modifier and Type
    Method
    Description
    String
    getName()
    Descriptive name of the task.
    (package private) static boolean
    isRecentlyActive(RouterContext ctx, Hash peer, long now)
    Whether this peer has traffic recent enough that probing adds nothing.
    (package private) static int
    probeBatchFor(long uptimeMs)
    Peers to probe per pass, based on how much of the profile is still unmeasured.
    (package private) static long
    probeIntervalFor(long uptimeMs, int bandSize, int batch)
    Gap between passes, derived from the tier size so a full sweep takes a predictable wall-clock time whatever the population.
    void
    runJob()
    Actually perform the task.

    Methods inherited from class JobImpl

    dropped, getContext, getJobId, getMadeReadyOn, getTiming, madeReady, requeue, toString
    Modifier and Type
    Method
    Description
    void
    dropped()
    Default implementation does nothing.
    final RouterContext
    getContext()
    The router context that this job belongs to.
    long
    getJobId()
    The unique job ID for this job.
    long
    getMadeReadyOn()
    The timestamp when this job was last made ready for execution.
    JobTiming
    getTiming()
    The timing object for this job.
    void
    madeReady(long now)
    For JobQueue only, not for external use.
    protected void
    requeue(long delayMs)
    Requeue this job to run after the specified delay.
    String
    toString()
     

    Methods inherited from class Object

    clone, equals, getClass, hashCode, notify, notifyAll, wait, wait, wait

  • Field Details

    • PROBE_BATCH

      static final int PROBE_BATCH
      Peers probed per pass. Bounds the burst so a pass is not a flood.
      See Also:

    • PROBE_INTERVAL

      static final long PROBE_INTERVAL
      Delay between passes, giving the probed slice time to be used or fail.
      See Also:

    • RECENTLY_ACTIVE_MS

      static final long RECENTLY_ACTIVE_MS
      A peer we have built through this recently needs no probe.

      Deliberately generous, and comfortably wider than one build cycle: a peer we are already successfully sending through is confirmed for as long as that keeps being true, so the only peers worth spending budget on are the idle ones the sweep exists to find. A narrow window would re-probe the peers working hardest, which is where the budget is least useful.

      Widening this constant reduces coverage. It widens the skip set, not the probe set — so it is only safe because the test below keys on our own send successes rather than on incidental inbound traffic.

      See Also:

    • FAST_WINDOW_MS

      static final long FAST_WINDOW_MS
      Uptime before which the sweep runs at its faster cadence.
      See Also:

    • PROBE_BATCH_FAST

      static final int PROBE_BATCH_FAST
      Peers per pass while young, when nothing has been measured yet.
      See Also:

    • PROBE_BATCH_STEADY

      static final int PROBE_BATCH_STEADY
      Peers per pass once the fast window has passed.
      See Also:

    • FAST_PASS_TARGET_MS

      static final long FAST_PASS_TARGET_MS
      Wall clock wanted for one pass over the whole tier, while young.

      A young router has measured nothing, so the tier is unpopulated with first-hop latency and nothing can be judged until it is covered. Twenty minutes reaches a full sweep quickly without turning the probe into a sustained flood: probes carry establishment cost and the failures that come with it, and those failures are what demote peers.

      See Also:

    • STEADY_PASS_TARGET_MS

      static final long STEADY_PASS_TARGET_MS
      Wall clock wanted for one pass once steady. Matches the original cadence.
      See Also:

    • MIN_PROBE_INTERVAL_MS

      static final long MIN_PROBE_INTERVAL_MS
      Floor on the gap between passes regardless of band size.
      See Also:

    • MAX_PROBE_INTERVAL_MS

      static final long MAX_PROBE_INTERVAL_MS
      Ceiling on the gap between passes, so coverage never stops refreshing.
      See Also:

  • Constructor Details

    • FastTierProbeJob

      FastTierProbeJob(RouterContext ctx, TunnelPoolManager mgr)
      Probe whether fast-tier peers are reachable, shortly after startup.
      Parameters:
      ctx - the router context
      mgr - the pool manager this job reports reachability to

  • Method Details

    • probeBatchFor

      static int probeBatchFor(long uptimeMs)
      Peers to probe per pass, based on how much of the profile is still unmeasured.
      Parameters:
      uptimeMs - router uptime in ms
      Returns:
      the batch size for the next pass

    • probeIntervalFor

      static long probeIntervalFor(long uptimeMs, int bandSize, int batch)
      Gap between passes, derived from the tier size so a full sweep takes a predictable wall-clock time whatever the population.

      Computing the interval from targetPass * batch / bandSize rather than fixing it means the same cadence holds at 200 peers or 2000: a small tier is not over-probed and a large one is not starved. Clamped so a very small or very large band cannot produce an absurd interval.

      Parameters:
      uptimeMs - router uptime in ms
      bandSize - how many peers are selectable, non-negative
      batch - peers per pass, must be positive
      Returns:
      the delay until the next pass, in ms

    • getName

      public String getName()
      Description copied from interface: Job
      Descriptive name of the task.
      Returns:
      the descriptive name of the task

    • runJob

      public void runJob()
      Description copied from interface: Job
      Actually perform the task. This call blocks until the Job is complete.

    • isRecentlyActive

      static boolean isRecentlyActive(RouterContext ctx, Hash peer, long now)
      Whether this peer has traffic recent enough that probing adds nothing.

      Only our own successful sends count. Inbound hearing is deliberately ignored: PeerProfile.getLastHeardFrom() advances on DHT replies, explore traffic and transit acknowledgments, so it marks most of the tier as recently active without ever showing the peer works as a first hop. Keying on it made a 2 minute window behave like a much longer one and left a pass probing 33 peers out of a 128 budget.

      Parameters:
      ctx - the router context
      peer - the peer to test
      now - current time in ms
      Returns:
      true if recent traffic means we should skip probing