> > By the way this patch is not really solves the problem, > > only partially. The second problem is that the "pending" > > list is referenced only in the sync queue, not copied. > > After calling peer_delete() the pending_put() > > immediately updates the values of the same > > structure. The data is read later in another thread > > to send to the peers. > > Hi > > Are you still working on it? > I sent the patch in the attachment, yahoogroups show that in fact there is an attachment as web link, but it is inaccessible. May be yahoogroups is broken? I advise to apply the patch but it does not solve the original problem. Please advise how can I provide the patch, If you want it. I see two options to go. 1. to change the access policy for the in memory structure to become append only, no update in place allowed. So a greylist entry cannot be promoted to a whitelist entry. Instead, an insertion of a new whitelist entry is needed, and the old greylist entry will expire at the appropriate time and subject of garbage collection. 2. Keep track the time values of the greylist entry outside of the memory structure and use them at the synchronization time. Independently all of that, if I may suggest a newer version of sync protocol which I believe conceptually is better. The only command to be used is the "encounter". It contains only the timestamp, sender, recipient, IP address quadruplet as seen by the milter-greylist instant on its own input. Than the sender milter-greylist replays this to all of his peers. The peers independently applies the quadruplet onto the in-memory structure based on the actual policy in force, similarly as the SMTP connection would have been made to the receiver peer with the same parameters. By the way that would be very hard to implement on a backward compatible way, so practically isn't a good choice.
Message
RE: [milter-greylist] Why autowhite isn't distributed over all MX?
2018-07-12 by Bruncsak, Attila
Attachments
- No local attachments were found for this message.