[Spread-users] problem re-using private names with 3.17.0

[Jeremy Hylton]

>>>>> "GS" == George Schlossnagle <george at omniti.com> writes:

  GS> I've witnessed this behavior in the past.  I believe the problem
  GS> is that when you call SP_disconnect(), the actual disconnect and
  GS> deregistration in the ring is asynchronous.  So you call
  GS> SP_disconnect, exit your process (or thread or what have you),
  GS> and then immediately reconnect with the same name, the
  GS> disconnect hasn't been processed by the ring yet.

Yes.  We've definitely seen this too.  For example, our regression
tests generate many very short spread sessions.  We have re-tries in
place to avoid failures when the disconnect isn't processed quickly
enough.  (This even happens with a single client and a single daemon.)

  GS> Especially when I was running the high-latency rings at CCI this
  GS> was constantly happening.  Sometimes it would take 30-60 seconds
  GS> for a disconnect to be fully propogated so that a new join on
  GS> the same private name could occur.

I don't have infinite patience <wink>, but I did wait for 5 minutes.

Jeremy