add address book pruning

[acud]

in the recent changes done in the addressbook, a wrapping data structure was added to indicate whether a bzz address was previously verified.

this type should be expanded to add a last seen timestamp. this timestamp should be updated every time we see the overlay over hive or in kademlia.

once an address is not seen for over a month - we can remove it from the address book (check on startup). this probably needs a migration that would set all existing addresses to seen now()

[gacevicljubisa]

Ass addition, maybe it would be more optimal, that before updating the LastSeen, we use Get to check the time diff is more then 24h, or something like this, before we update.

[acud]

Ass addition, maybe it would be more optimal, that before updating the LastSeen, we use Get to check the time diff is more then 24h, or something like this, before we update.

Why? what is the benefit?

[gacevicljubisa]

Ass addition, maybe it would be more optimal, that before updating the LastSeen, we use Get to check the time diff is more then 24h, or something like this, before we update.

Why? what is the benefit?

If we have to often updates of LastSeen, in case if we want to update when peer is detected in the hive, which is preaty agressive, this wlll create big amount of IO. We could GET to check the time diff is more then 24h, where db could use its cache, and write only if more then 24 hours.

[janos]

Ass addition, maybe it would be more optimal, that before updating the LastSeen, we use Get to check the time diff is more then 24h, or something like this, before we update.

Why? what is the benefit?

If we have to often updates of LastSeen, in case if we want to update when peer is detected in the hive, which is preaty agressive, this wlll create big amount of IO. We could GET to check the time diff is more then 24h, where db could use its cache, and write only if more then 24 hours.

I think that the IO concern is a valid one. The UpdateLastSeen method https://github.com/ethersphere/bee/pull/5513/changes#diff-b817056e32073e46ecb398adec8fcde59e8a72d424d90ea9e213039dbeb7b3a6R110, already calls both Get and Put on state storer. The check that @gacevicljubisa suggested can avoid Put calls with a just a few lines there.

But, I think that UpdateLastSeen should be called in hive as well to mark the node as seen even if not connected (as potential connection), which would create a much more IO. To protect from this, my suggestion is to have a batching internally in UpdateLastSeen function. That would be an isolated in memory storage of last seen times per overlay addresses that would be persisted on a period. There may not even be a worker at all, just to check what is the last save time, and to save them after some period is passed, like 10 minutes (it can be even longer). Loosing 10 minutes of seen times should not be a problem, and the IO protection is very benefitial. That gives the freedom to call UpdateLastSeen very frequently.