zookeeper-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jung Young Seok <jung.youngs...@gmail.com>
Subject Re: [Zookeeper] Zookeeper Cluster broken due to snapshot corrupted error
Date Thu, 27 Mar 2014 00:33:26 GMT
Thank, Michi.


We store same amount of data per transaction for user lock only.

Each user has own node for lock only. (few bytes)

Now I can guess, when various users use our service, snapshot file tends to
be a little bigger.

When users that use our service frequently use more our service then
snapshot file seems smaller. (because many transaction happened on same
node?)

I hope I understand zookeeper system properly.


I appreciate your previous explanation.

Have a great day.


2014. 3. 27. 오전 8:40에 "Michi Mutsuzaki" <michi@cs.stanford.edu>님이 작성:

> > Does snapshotting means dumping snap data on memory into disk?
>
> Yes.
>
> > Small snapCount would take little time for snapshotting, it means shorter
> > stopping world. Do I understand properly?
>
> The amount of time it takes to take a snapshot is proportional to the
> amount of data stored in ZooKeeper. Taking snapshots more frequently
> doesn't reduce the snapshot time.
>
> ZooKeeper takes what's called 'fuzzy snapshots' and it doesn't block
> writes (i.e. stop the world) during the snapshot. ZooKeeper recovers
> to a consistent state by loading the fuzzy snapshot and replaying all
> the transactions since the snapshot process started. This works
> because all the ZooKeeper operations are idempotent.
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message