[tor-bugs] #13699 [Tor]: Remove rend cache object if not usable anymore

[Tor Bug Tracker & Wiki]

#13699: Remove rend cache object if not usable anymore
-------------------------+-----------------
     Reporter:  dgoulet  |      Owner:
         Type:  defect   |     Status:  new
     Priority:  normal   |  Milestone:
    Component:  Tor      |    Version:
   Resolution:           |   Keywords:
Actual Points:           |  Parent ID:
       Points:           |
-------------------------+-----------------

Comment (by special):

 Right now, by accident, I think this behavior is actually not so harmful:

 RendPostPeriod defaults to 1 hour, and the service will use a new IP (and
 publish a new descriptor) if any IP fails. If an IP is failing and the
 service is up, a new descriptor will quickly be available. Even if it
 doesn't, there will be a new descriptor within an hour, and clients will
 try all IPs again. Effectively, this is being timed out within an hour
 when the service is online, unless the service changes RendPostPeriod.

 That changes if we change the default RendPostPeriod, or if we implement
 #8239.

 I don't like using the descriptor retry behavior as the basis for IP retry
 behavior. We fetch descriptors _very_ often; for an offline service, it's
 6 times every REND_HID_SERV_DIR_REQUERY_PERIOD (15 minutes). If we tried
 every IP after every directory fetch, which I think is what you've
 proposed, we're looking at 18 IP circuits (assuming 3 IP) before deciding
 that a service is offline.

 I would rather see a more deliberate decision about when to retry an IP,
 preferably using knowledge about how it failed before (e.g. if the IP was
 reachable or not). What do you think?

 > Discussion on tor-assistant about this bug made by rransom, this might
 actually makes sense.

 Shame it was happening where we can't read it.

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/13699#comment:3>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online