< Previous by Date | Date Index | Next by Date > |
< Previous in Thread | Thread Index | Next in Thread > |
Good point about the 503’s – I agree. I am a little worried
about only blacklisting a timeout (408) for 32 seconds, when there are
alternate servers that could handle requests. Does this mean that every 32
seconds we will try to send requests to the black listed server again - and
fail (assuming the server is down)? Or will the whitelisting logic kick in and
we will continue to use the good server? Scott From: Byron Campen
[mailto:docfaraday@xxxxxxx] We can't
unblacklist stuff that was blacklisted because of a 503. If all of the
available servers are keeling over from overload, and sending 503s, we are
supposed to just shut up and wait for the blacklists to expire. However, it is
reasonable to say that this behavior is not warranted in cases where we have
blacklisted due to a timeout. If we wish to treat these two scenarios
differently, maybe a greylisting concept is what we should use. Best regards, Byron Campen
Before
the recent changes, that way I understood it the blacklisting was supposed to
work as follows: 1. Only blacklist an
entry if there other DNS entries to try. 2. If we have ended up
blacklisting all entries from a particular lookup, then un-blacklist them all
and let the next request start the cycle again. Scott From:
resiprocate-devel-bounces@xxxxxxxxxxxxxxxxxxxx [mailto:resiprocate-devel-bounces@xxxxxxxxxxxxxxxxxxxx]
On Behalf Of Byron Campen This
is happening because the stack is blacklisting the tuple that your server was
running on, because it didn't respond. This blacklist (currently) lasts 32s. I
was thinking about making this duration configurable. Does anyone want to have
a discussion about whether blacklisting on a UDP timeout is something we want
to be doing? It seems to me that we might need to have a concept of
"greylisting", where greylisted tuples will only be used if they are
all that remains to be tried. (This is opposed to a blacklisted tuple, which we
should never try, since chances are we have been explicitly told, with a 503,
to leave the tuple alone for a while.) Any thoughts? (anyone?) Best
regards, Byron
Campen
_______________________________________________ resiprocate-devel mailing list |