Maybe the idea remove ledger from underreplicated when bookie
starts up has some problem, What important is, replicate legder puts so much
heavy loads on cluster. I want to reduce the affect.
suppose there are 5 copies and 4 bookies are lost, when one lost bookie is
online again, I don't simply mark the ledger replicated. When
the replicateWork gets the failed bookies, it will ignore the one which
online again. So, this will reduce replicating one copy of data.
------------------ ???????? ------------------
??????:
"dev"
<eolive...@gmail.com>;
????????: 2021??8??28??(??????) ????6:07
??????: "Bookkeeper-Dev"<dev@bookkeeper.apache.org>;
????: Re: Skip replicating ledger after bookkeeper server starts up again
Candy Rain,
Thanks for sharing your proposal
Il Ven 27 Ago 2021, 04:30 Candy Rain <gaozhangmin...@qq.com.invalid> ha
scritto:
> Describe
>
> When&nbsp; bookies in the cluster are down, the auto-recovery gets
> triggered, but as these come back online, the rereplication worker should
> Ideally skip the rereplication of the ledgers that are marked as
> underreplicated.
The replicator should be able to listen on zookeeper for changes I'm the
availability of a bookie.
Probably there is some space for improvements here.
I don't think it is a good idea to eagerly set the ledger as no more
underreplicated as soon as one of the bookie in the ensemble comes back
only.
The fact that a ledger is underreplicated is to be verified against the
requested number of replicas, as it is not enough that one bookie is running
Regards
Enrico
But the ledgers are rereplicated instead
>
>
>
> Expected behavior
>
> Ideally as and when the bookies come up the ledgers marked as
> underreplicated would be read by the rereplicaton worker and from the
> metadata the worked should skip these as the bookies are available.
