java.lang.Object
net.i2p.router.peermanager.RouterInfoRefresher
class RouterInfoRefresherextends Object
Per-cycle state for the promotion deferral on a stale RouterInfo.

Promotion into the fast or high-capacity tier reads whatever RouterInfo the network database currently holds. A RouterInfo older than an hour is likely superseded — routers republish hourly — so promoting on one buys a tier slot that may not survive a build. ProfileOrganizer therefore holds such a promotion back and simply leaves the peer out of the tier; nothing is fetched and no network work is done on the peer's behalf.

What a cycle needs to remember is therefore only its own bounds and its two counters: the staleness threshold, the tier sizes the starvation guard compares against, how many promotions were held, and how many went ahead on a RouterInfo already known to be stale.

Locking: confined to the reorganisation write lock, which is also the only caller. Nothing here is written from a job or a selection thread.

See Also:

  • Field Summary

    Fields
    Modifier and Type
    Field
    Description
    (package private) static final long
    ADDRESS_REFRESH_MIN_INTERVAL_MS
    Minimum gap between drains, so repeated reorgs cannot stack requests.
    (package private) static final int
    MAX_ADDRESS_REFRESHES
    Most peers whose RouterInfo refresh may be requested in one drain.
    (package private) static final int
    MAX_MISSING_ROUTERINFO
    Ceiling on remembered peers missing a RouterInfo.
    (package private) static final int
    MISSING_ROUTERINFO_BATCH
    Peers looked up per drain when catching up on profiles loaded without a RouterInfo, giving roughly 64 lookups a minute.
    (package private) static final long
    MISSING_ROUTERINFO_MIN_INTERVAL_MS
    Minimum gap between missing-RouterInfo drains, in ms.

  • Constructor Summary

    Constructors
    Constructor
    Description
    RouterInfoRefresher()
    A fresh cycle record: no staleness threshold, no tier floors and no deferral until beginCycle(long, int, int) supplies them.

  • Method Summary

    All Methods
    Modifier and Type
    Method
    Description
    (package private) boolean
    beginCycle(long refreshAgeMs, int fastFloor, int highCapFloor)
    Open a cycle, resetting the per-cycle counters.
    (package private) int
    deferredCount()
    Get the number of promotions deferred this cycle.
    (package private) int
    fastFloor()
    Get the fast-tier size at which promotion may be deferred.
    (package private) int
    getAddressRefreshDropped()
    Peers dropped because the queue or the interval was full.
    (package private) int
    getAddressRefreshPending()
    Peers waiting for a lookup.
    (package private) int
    getMissingRouterInfoDropped()
    Peers dropped because the missing-RouterInfo set was already at its cap.
    (package private) int
    getMissingRouterInfoPending()
    Peers still waiting for their RouterInfo to be requested.
    (package private) int
    highCapFloor()
    Get the high-capacity size at which promotion may be deferred.
    (package private) boolean
    isDeferring()
    Test whether promotions are being held back this cycle.
    (package private) boolean
    mayDrainAddressRefreshes(long now)
    Whether enough time has passed since the last drain to issue another batch.
    (package private) boolean
    mayDrainMissingRouterInfo(long now)
    Whether enough time has passed to drain the missing-RouterInfo set.
    (package private) void
    noteDeferral()
    Record a promotion held back because of a stale RouterInfo.
    (package private) boolean
    noteMissingRouterInfo(Hash peer)
    Note that a stored profile was loaded for a peer the netdb cannot resolve.
    (package private) void
    noteStalePromotion()
    Record a promotion the starvation guard let through on a RouterInfo that is already known to be stale.
    (package private) int
    promotedStaleCount()
    The promotions the starvation guard let through on a RouterInfo that was already known to be stale.
    (package private) long
    refreshAgeMs()
    Get the staleness threshold applied this cycle.
    (package private) boolean
    requestAddressRefresh(Hash peer)
    Ask for a RouterInfo lookup for a peer held out of the tiers for want of a usable address.
    (package private) List<Hash>
    takeAddressRefreshBatch(long now)
    Hand over the queued peers, oldest first, clearing the queue.
    (package private) List<Hash>
    takeMissingRouterInfoBatch(long now)
    Hand over up to one batch of peers to look up, removing them from the set.

    Methods inherited from class Object

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

  • Field Details

    • MAX_ADDRESS_REFRESHES

      static final int MAX_ADDRESS_REFRESHES
      Most peers whose RouterInfo refresh may be requested in one drain.
      See Also:

    • MISSING_ROUTERINFO_BATCH

      static final int MISSING_ROUTERINFO_BATCH
      Peers looked up per drain when catching up on profiles loaded without a RouterInfo, giving roughly 64 lookups a minute.

      Faster than MAX_ADDRESS_REFRESHES on purpose. That cap is sized for held promotion candidates, a trickle of a few per reorganize; a router coming back from downtime may hold thousands of profiles whose RouterInfo has since expired, and at 16 a minute they would stay unusable for hours. Each lookup is a fire-and-forget lookupRouterInfo, and the netdb negatively caches known-dead keys, so peers that never come back cost little on repeat.

      See Also:

    • MISSING_ROUTERINFO_MIN_INTERVAL_MS

      static final long MISSING_ROUTERINFO_MIN_INTERVAL_MS
      Minimum gap between missing-RouterInfo drains, in ms.

      Sets the rate at MISSING_ROUTERINFO_BATCH per interval, about 64 a minute, so a couple of thousand profiles take roughly half an hour to recover.

      See Also:

    • MAX_MISSING_ROUTERINFO

      static final int MAX_MISSING_ROUTERINFO
      Ceiling on remembered peers missing a RouterInfo.

      The set is a recovery list, not a backlog to be worked off exhaustively. Past this size the extra peers are not worth the memory and the lookups: they are peers the netdb has already stopped telling us about.

      See Also:

    • ADDRESS_REFRESH_MIN_INTERVAL_MS

      static final long ADDRESS_REFRESH_MIN_INTERVAL_MS
      Minimum gap between drains, so repeated reorgs cannot stack requests.
      See Also:

  • Constructor Details

    • RouterInfoRefresher

      RouterInfoRefresher()
      A fresh cycle record: no staleness threshold, no tier floors and no deferral until beginCycle(long, int, int) supplies them.

  • Method Details

    • beginCycle

      boolean beginCycle(long refreshAgeMs, int fastFloor, int highCapFloor)
      Open a cycle, resetting the per-cycle counters.

      There is nothing to stagger here. The earlier startup grace period, quota and queue rotation only ever throttled how many lookups a cycle could issue; with no lookup to issue, each cycle now judges every candidate, and a fresh cycle resets the counters so they report one cycle rather than the life of the router.

      Parameters:
      refreshAgeMs - staleness threshold for this cycle, or <= 0 to disable
      fastFloor - fast-tier size at or above which a promotion may be held back
      highCapFloor - high-capacity tier size at or above which a promotion may be held back
      Returns:
      whether promotions are being held back this cycle

    • noteDeferral

      void noteDeferral()
      Record a promotion held back because of a stale RouterInfo. One integer increment: this runs on the promotion path, once per held tier insert, and has to stay cheap enough for the loop that calls it.

    • noteStalePromotion

      void noteStalePromotion()
      Record a promotion the starvation guard let through on a RouterInfo that is already known to be stale. This is the population the peer.promotedStaleRouterInfo stat reports.

    • refreshAgeMs

      long refreshAgeMs()
      Get the staleness threshold applied this cycle.
      Returns:
      the staleness threshold in force this cycle, in ms

    • isDeferring

      boolean isDeferring()
      Test whether promotions are being held back this cycle.
      Returns:
      whether promotions are being held back this cycle

    • fastFloor

      int fastFloor()
      Get the fast-tier size at which promotion may be deferred.
      Returns:
      fast-tier size at or above which a promotion may be held back

    • highCapFloor

      int highCapFloor()
      Get the high-capacity size at which promotion may be deferred.
      Returns:
      high-capacity size at or above which a promotion may be held back

    • deferredCount

      int deferredCount()
      Get the number of promotions deferred this cycle.
      Returns:
      promotions held back this cycle, counted across both tiers

    • promotedStaleCount

      int promotedStaleCount()
      The promotions the starvation guard let through on a RouterInfo that was already known to be stale.
      Returns:
      stale promotions counted since the last beginCycle()

    • requestAddressRefresh

      boolean requestAddressRefresh(Hash peer)
      Ask for a RouterInfo lookup for a peer held out of the tiers for want of a usable address.

      A held peer cannot be refreshed directly - that is precisely why it is held, since it has no address to send to - so the only way to re-qualify it is an iterative lookup via other routers. This only records the request; takeAddressRefreshBatch(long) issues it, because the scan that fills this queue runs under the reorganize write lock and must not do network work there.

      Bounded by MAX_ADDRESS_REFRESHES, so a scan over thousands of profiles cannot enqueue thousands of lookups.

      Parameters:
      peer - the held peer
      Returns:
      true if the request was recorded

    • mayDrainAddressRefreshes

      boolean mayDrainAddressRefreshes(long now)
      Whether enough time has passed since the last drain to issue another batch.
      Parameters:
      now - current time in milliseconds
      Returns:
      true if a drain may run

    • takeAddressRefreshBatch

      List<Hash> takeAddressRefreshBatch(long now)
      Hand over the queued peers, oldest first, clearing the queue.
      Parameters:
      now - current time in milliseconds
      Returns:
      the peers to look up, possibly empty

    • getAddressRefreshDropped

      int getAddressRefreshDropped()
      Peers dropped because the queue or the interval was full.
      Returns:
      the count of peers dropped because the queue or the interval was full

    • getAddressRefreshPending

      int getAddressRefreshPending()
      Peers waiting for a lookup.
      Returns:
      the count of peers waiting for a lookup

    • noteMissingRouterInfo

      boolean noteMissingRouterInfo(Hash peer)
      Note that a stored profile was loaded for a peer the netdb cannot resolve.

      Until the RouterInfo comes back the profile is inert: peer selection needs an address to send to, and ProfileOrganizer.isExcludedFromProfiling(Hash) declines to profile a peer with no RouterInfo. The capacity, latency and acceptance history in the file is then retained but unused, which only pays off if something goes and fetches the RouterInfo again.

      Parameters:
      peer - the peer whose profile was loaded
      Returns:
      true if the peer was recorded, false if the set was already full

    • mayDrainMissingRouterInfo

      boolean mayDrainMissingRouterInfo(long now)
      Whether enough time has passed to drain the missing-RouterInfo set.
      Parameters:
      now - current time in milliseconds
      Returns:
      true if the drain interval has elapsed

    • takeMissingRouterInfoBatch

      List<Hash> takeMissingRouterInfoBatch(long now)
      Hand over up to one batch of peers to look up, removing them from the set.
      Parameters:
      now - current time in milliseconds
      Returns:
      the peers to look up, possibly empty

    • getMissingRouterInfoDropped

      int getMissingRouterInfoDropped()
      Peers dropped because the missing-RouterInfo set was already at its cap.
      Returns:
      the count of peers dropped because the missing-RouterInfo set was full

    • getMissingRouterInfoPending

      int getMissingRouterInfoPending()
      Peers still waiting for their RouterInfo to be requested.
      Returns:
      the count of peers still waiting for their RouterInfo to be requested