Hi Jun, Thanks for the reply. Comments inline.
On Thu, Feb 16, 2017 at 2:29 PM, Jun Rao <j...@confluent.io> wrote: > Hi, Apurva, > > Thanks for the reply. A couple of comment below. > > On Wed, Feb 15, 2017 at 9:45 PM, Apurva Mehta <apu...@confluent.io> wrote: > > > Hi Jun, > > > > Answers inline: > > > > 210. Pid snapshots: Is the number of pid snapshot configurable or > hardcoded > > > with 2? When do we decide to roll a new snapshot? Based on time, byte, > or > > > offset? Is that configurable too? > > > > > When a replica becomes a follower, we do a bit log truncation. Having an > older snapshot allows us to recover the PID->sequence mapping much quicker > than rescanning the whole log. This is a good point. I have updated the doc with a more detailed proposal. Essentially, snapshots will be created on a periodic basis. A reasonable period would be every 30 or 60 seconds. We will keep at most 2 copies of the snapshot file. With this setup, we would have to replay at most 60 or 120 seconds of the log in the event of log truncation during leader failover. If we need to make any of this configurable, we can expose a config in the future. It would be easier to add a config we need than remove one with marginal utility. > > > > > > > 211. I am wondering if we should store ExpirationTime in the producer > > > transactionalId mapping message as we do in the producer transaction > > status > > > message. If a producer only calls initTransactions(), but never > publishes > > > any data, we still want to be able to expire and remove the producer > > > transactionalId mapping message. > > > > > > > > Actually, the document was inaccurate. The transactionalId will be > expired > > only if there is no active transaction, and the age of the last > transaction > > with that transactionalId is older than the transactioanlId expiration > > time. With these semantics, storing the expiration time in the > > transactionalId mapping message won't be useful, since the expiration > time > > is a moving target based on transaction activity. > > > > I have updated the doc with a clarification. > > > > > > > Currently, the producer transactionalId mapping message doesn't carry > ExpirationTime, but the producer transaction status message does. It would > be useful if they are consistent. > > You are right. The document has been updated to remove the ExpirationTime from the transaction status messages as well. Any utility for this field can be achieved by using the timestamp of the message itself along with another expiration time (like transactionalId expiration time, transaction expiration time, etc.). Thanks, Apurva
