Hi Arvind, Thanks for your reply. You are right in the fact the global state check and update to the sink will require each sink to explicitly support it. We can, of course have this implementation be in an abstract class which is inherited, but yes, this would also mean that there needs to be code changes.
It makes sense to check state in the channels, pretty much the same way as in the sinks. What is a bit concerning is that we will need to do this check at every agent that the event passes through, and probably make some changes in the channel interface to get rid of race conditions (not sure if that is the case, but I think we will need to). Given that an event is likely to pass through 2-3 tiers, each event gets delayed by the time taken by that many ZK round-trips. I am open to this as well, especially considering that it is likely to be a better OOB experience for many users (the ones who have their own custom sinks). Would it suffice to check at the sinks at the terminal agent to make sure that an event gets written out only once? Thinking about this, having a once-only delivery at the channel level also opens up some possibilities with regards to being able to do some sort of processing on events. Having a guarantee of seeing an event exactly once allows us to do some event processing like counters etc. That seems like a good side effect to have. Either way, I am glad we agree on the aspect of checking a global state manager to verify that events are deduped. Thanks, Hari On Tuesday, August 27, 2013 at 2:12 PM, Arvind Prabhakar wrote: > Hi Hari, > > Thanks for bringing this up for discussion. I think it will be tremendously > beneficial to Flume users if we can extend once-only guarantee. Your > initial suggestion seems reasonable of having a Sink trap the events and > reference a global state to drop duplicates. Rather than pushing this > functionality to Sinks is there any other way by which we can make it more > generally available? The reason I raise this concern is because otherwise > this becomes a feature of a particular sink and not every sink will have > the necessary implementation opportunity to get this. > > Alternatively what do you think about this being done at the channel level? > Since we normally do not see custom implementations of channels, an > implementation that works with the channel will likely be more useful for > the broader community of Flume users. > > Regards, > Arvidn > > > On Sun, Aug 25, 2013 at 9:07 AM, Hari Shreedharan <[email protected] > (mailto:[email protected]) > > wrote: > > > > Hi Gabriel, > > > > Thanks for your input. The part where we use replicating channel selector > > to purposefully replicate - we can easily make it configurable whether to > > delete deplicate events or not. That should not be difficult to do. > > > > The 2nd point where multiple agents/sinks could write the same event can > > be solved by namespacing the events into different namespaces. So each sink > > checks one namespace for the event, and multiple sinks can belong to the > > same namespace - this way, if multiple events are going to write to the > > same HDFS cluster, then if a duplicate occurs we can easily drop it. > > Unfortunately, this also does not work around the who > > HDFS-writing-but-throwing issue. > > > > I agree updating ZK will hit latency, but that is the cost to build once > > only semantics on a highly flexible system. If you look at the algorithm, > > we actually go to ZK only once per event (to create, there are no updates) > > - this can even happen per batch if needed to reduce ZK round trips (though > > I am not sure if ZK provides a batch API). > > > > The two phase commit approach sounds good, but it might require interface > > changes which can now only be made in Flume 2.x. Alse, if we use a single > > UUID combined with several flags we might be able to work duplicates caused > > by this replication. > > > > > > Thanks, > > Hari > > > > > > On Sunday, August 25, 2013 at 7:24 AM, Gabriel Commeau wrote: > > > > > Hi Hari, > > > > > > > > > I deleted my comment (again). The mailing list is probably a better > > avenue > > > to discuss this sorry about that! :) > > > > > > I can find at least one other way duplicate events can occur, and so what > > > I provided helps to reduce duplicate events but is not sufficient to > > > guaranty exactly once semantics. However, I still think that using a > > > 2-phase commit when writing to multiple channels would benefit Flume. > > > > > > > This > > > should probably be a different ticket though. > > > > > > Concerning the algorithm you offered, the case of replicating channel > > > selector should probably be handled, by creating a new UUID for each > > > duplicate message. > > > I hope this helps. > > > > > > > > > Regards, > > > > > > Gabriel > > > > > > > > > On 8/25/13 7:27 AM, "Gabriel Commeau (JIRA)" <[email protected] > > > (mailto:[email protected]) (mailto: > > [email protected] (mailto:[email protected]))> wrote: > > > > > > > > > > > [ > > https://issues.apache.org/jira/browse/FLUME-2173?page=com.atlassian.jira.p > > > > lugin.system.issuetabpanels:all-tabpanel ] > > > > > > > > Gabriel Commeau updated FLUME-2173: > > > > ----------------------------------- > > > > > > > > Comment: was deleted > > > > > > > > (was: I would approach the problem from a different angle. The way I > > see > > > > it, there are two main places where duplicates can occur: when using > > > > multiple channels for one source (using a replication channel > > > > > > > > > > > selector), > > > > and when the "output" of a sink cannot guaranty whether the event has > > > > truly been committed or not (as you pointed out for example, HDFS > > > > > > > > > > > writing > > > > the event but throwing an exception). > > > > Actually, I don¹t think there is a general solution to the problem of > > > > output systems (e.g. HDFS) that do not guaranty whether the event is > > > > truly committed or not, because we¹d need to enforce this requirement > > > > > > > > > > > on > > > > 3rd party systems (relative to Flume). I see it as a problem to be > > > > > > > solved > > > > on a case-by-case basis for each sink. > > > > > > > > However, I would like to suggest a solution to the first problem. Here > > is > > > > an example to illustrate it: Pretend an agent has a source that writes > > > > > > > to > > > > two (required) channels. As part of a transaction, the channel > > > > > > > processor > > > > will commit to the first channel, which succeeds, and then to the > > > > > > > second > > > > channel, which fails. The whole transaction will fail, but the event > > > > > > > has > > > > already been committed once to the first channel. When the transaction > > > > > > > is > > > > retried, the event will be duplicated. > > > > The solution I discussed a few months back with Mike P. was to use a > > > > two-phase commit when writing to channels. This insures that the events > > > > are not actually committed to a channel if the following ones fail. > > > > > > > > > > > This > > > > however will require an API change on the Channel interface. I would > > > > suggest adding a preparePut method returning a boolean, which would be > > > > the ³voting² phase. The put method becomes the commit phase. To make it > > > > backward compatible, we'd implement preparePut to always return true in > > > > the AbstractChannel. > > > > > > > > I hope this helps. > > > > ) > > > > > > > > > Exactly once semantics for Flume > > > > > -------------------------------- > > > > > > > > > > Key: FLUME-2173 > > > > > URL: https://issues.apache.org/jira/browse/FLUME-2173 > > > > > Project: Flume > > > > > Issue Type: Bug > > > > > Reporter: Hari Shreedharan > > > > > Assignee: Hari Shreedharan > > > > > > > > > > Currently Flume guarantees only at least once semantics. This jira is > > > > > meant to track exactly once semantics for Flume. My initial idea is > > > > > > > > > > > > > > > > > > > > to > > > > > include uuid event ids on events at the original source (use a > > > > > > > > > > > config to > > > > > mark a source an original source) and identify destination sinks. At > > > > > > > > > > > the > > > > > destination sinks, use a unique ZK Znode to track the events. If once > > > > > seen (and configured), pull the duplicate out. > > > > > This might need some refactoring, but my belief is we can do this in > > > > > > > > > > > > > > > > a > > > > > backward compatible way. > > > > > > > > > > > > > > > > -- > > > > This message is automatically generated by JIRA. > > > > If you think it was sent incorrectly, please contact your JIRA > > > > administrators > > > > For more information on JIRA, see: > > > > > > > > > > > http://www.atlassian.com/software/jira > > > > > > > > > > > > > >
