Hey Jun, Thanks much. I agree that we can not rely on committed offsets to be always deleted when we delete topic. So it is necessary to use a per-partition epoch that does not change unless this partition is deleted. I also agree that it is very nice to be able to uniquely identify a message with (offset, leader_epoch, partition_epoch) in face of potential topic deletion and unclean leader election.
I agree with all your comments. And I have updated the KIP based on our latest discussion. In addition, I added InvalidPartitionEpochException which will be thrown by consumer.poll() if the partition_epoch associated with the partition, which can be given to consumer using seek(...), is different from the partition_epoch in the FetchResponse. Can you take another look at the latest KIP? Thanks! Dong On Wed, Jan 10, 2018 at 2:24 PM, Jun Rao <j...@confluent.io> wrote: > Hi, Dong, > > My replies are the following. > > 60. What you described could also work. The drawback is that we will be > unnecessarily changing the partition epoch when a partition hasn't really > changed. I was imagining that the partition epoch will be stored in > /brokers/topics/[topic]/partitions/[partitionId], instead of at the topic > level. So, not sure if ZK size limit is an issue. > > 61, 62 and 65. To me, the offset + offset_epoch is a unique identifier for > a message. So, if a message hasn't changed, the offset and the associated > offset_epoch ideally should remain the same (it will be kind of weird if > two consumer apps save the offset on the same message, but the offset_epoch > are different). partition_epoch + leader_epoch give us that. global_epoch + > leader_epoch don't. If we use this approach, we can solve not only the > problem that you have identified, but also other problems when there is > data loss or topic re-creation more reliably. For example, in the future, > if we include the partition_epoch and leader_epoch in the fetch request, > the server can do a more reliable check of whether that offset is valid or > not. I am not sure that we can rely upon all external offsets to be removed > on topic deletion. For example, a topic may be deleted by an admin who may > not know all the applications. > > If we agree on the above, the second question is then how to reliably > propagate the partition_epoch and the leader_epoch to the consumer when > there are leader or partition changes. The leader_epoch comes from the > message, which is reliable. So, I was suggesting that when we store an > offset, we can just store the leader_epoch from the message set containing > that offset. Similarly, I was thinking that if the partition_epoch is in > the fetch response, we can propagate partition_epoch reliably where is > partition_epoch change. > > 63. My point is that once a leader is producing a message in the new > partition_epoch, ideally, we should associate the new offsets with the new > partition_epoch. Otherwise, the offset_epoch won't be the correct unique > identifier (useful for solving other problems mentioned above). I was > originally thinking that the leader will include the partition_epoch in the > metadata cache in the fetch response. It's just that right now, metadata > cache is updated on UpdateMetadataRequest, which typically happens after > the LeaderAndIsrRequest. Another approach is for the leader to cache the > partition_epoch in the Partition object and return that (instead of the one > in metadata cache) in the fetch response. > > 65. It seems to me that the global_epoch and the partition_epoch have > different purposes. A partition_epoch has the benefit that it (1) can be > used to form a unique identifier for a message and (2) can be used to > solve other > corner case problems in the future. I am not sure having just a > global_epoch can achieve these. global_epoch is useful to determine which > version of the metadata is newer, especially with topic deletion. > > Thanks, > > Jun > > On Tue, Jan 9, 2018 at 11:34 PM, Dong Lin <lindon...@gmail.com> wrote: > > > Regarding the use of the global epoch in 65), it is very similar to the > > proposal of the metadata_epoch we discussed earlier. The main difference > is > > that this epoch is incremented when we create/expand/delete topic and > does > > not change when controller re-send metadata. > > > > I looked at our previous discussion. It seems that we prefer > > partition_epoch over the metadata_epoch because 1) we prefer not to have > an > > ever growing metadata_epoch and 2) we can reset offset better when topic > is > > re-created. The use of global topic_epoch avoids the drawback of an ever > > quickly ever growing metadata_epoch. Though the global epoch does not > allow > > us to recognize the invalid offset committed before the topic > re-creation, > > we can probably just delete the offset when we delete a topic. Thus I am > > not very sure whether it is still worthwhile to have a per-partition > > partition_epoch if the metadata already has the global epoch. > > > > > > On Tue, Jan 9, 2018 at 6:58 PM, Dong Lin <lindon...@gmail.com> wrote: > > > > > Hey Jun, > > > > > > Thanks so much. These comments very useful. Please see below my > comments. > > > > > > On Mon, Jan 8, 2018 at 5:52 PM, Jun Rao <j...@confluent.io> wrote: > > > > > >> Hi, Dong, > > >> > > >> Thanks for the updated KIP. A few more comments. > > >> > > >> 60. Perhaps having a partition epoch is more flexible since in the > > future, > > >> we may support deleting a partition as well. > > >> > > > > > > Yeah I have considered this. I think we can probably still support > > > deleting a partition by using the topic_epoch -- when partition of a > > topic > > > is deleted or created, epoch of all partitions of this topic will be > > > incremented by 1. Therefore, if that partition is re-created later, the > > > epoch of that partition will still be larger than its epoch before the > > > deletion, which still allows the client to order the metadata for the > > > purpose of this KIP. Does this sound reasonable? > > > > > > The advantage of using topic_epoch instead of partition_epoch is that > the > > > size of the /brokers/topics/[topic] znode and request/response size can > > be > > > smaller. We have a limit on the maximum size of znode (typically 1MB). > > Use > > > partition epoch can effectively reduce the number of partitions that > can > > be > > > described by the /brokers/topics/[topic] znode. > > > > > > One use-case of partition_epoch for client to detect that the committed > > > offset, either from kafka offset topic or from the external store is > > > invalid after partition deletion and re-creation. However, it seems > that > > we > > > can also address this use-case with other approaches. For example, when > > > AdminClient deletes partitions, it can also delete the committed > offsets > > > for those partitions from the offset topic. If user stores offset > > > externally, it might make sense for user to similarly remove offsets of > > > related partitions after these partitions are deleted. So I am not sure > > > that we should use partition_epoch in this KIP. > > > > > > > > >> > > >> 61. It seems that the leader epoch returned in the position() call > > should > > >> the the leader epoch returned in the fetch response, not the one in > the > > >> metadata cache of the client. > > > > > > > > > I think this is a good idea. Just to double check, this change does not > > > affect the correctness or performance of this KIP. But it can be useful > > if > > > we want to use the leader_epoch to better handle the offset rest in > case > > of > > > unclean leader election, which is listed in the future work. Is this > > > understanding correct? > > > > > > I have updated the KIP to specify that the leader_epoch returned by > > > position() should be the largest leader_epoch of those already consumed > > > messages whose offset < position. If no message has been consumed since > > > consumer initialization, the leader_epoch from seek() or > > > OffsetFetchResponse should be used. The offset included in the > > > OffsetCommitRequest will also be determined in the similar manner. > > > > > > > > >> > > >> 62. I am wondering if we should return the partition epoch in the > fetch > > >> response as well. In the current proposal, if a topic is recreated and > > the > > >> new leader is on the same broker as the old one, there is nothing to > > force > > >> the metadata refresh in the client. So, the client may still associate > > the > > >> offset with the old partition epoch. > > >> > > > > > > Could you help me understand the problem if a client associates old > > > partition_epoch (or the topic_epoch as of the current KIP) with the > > offset? > > > The main purpose of the topic_epoch is to be able to drop leader_epoch > > to 0 > > > after a partition is deleted and re-created. I guess you may be > thinking > > > about using the partition_epoch to detect that the committed offset is > > > invalid? In that case, I am wondering if the alternative approach > > described > > > in 60) would be reasonable. > > > > > > > > >> > > >> 63. There is some subtle coordination between the LeaderAndIsrRequest > > and > > >> UpdateMetadataRequest. Currently, when a leader changes, the > controller > > >> first sends the LeaderAndIsrRequest to the assigned replicas and the > > >> UpdateMetadataRequest to every broker. So, there could be a small > window > > >> when the leader already receives the new partition epoch in the > > >> LeaderAndIsrRequest, but the metadata cache in the broker hasn't been > > >> updated with the latest partition epoch. Not sure what's the best way > to > > >> address this issue. Perhaps we can update the metadata cache on the > > broker > > >> with both LeaderAndIsrRequest and UpdateMetadataRequest. The challenge > > is > > >> that the two have slightly different data. For example, only the > latter > > >> has > > >> all endpoints. > > >> > > > > > > I am not sure whether this is a problem. Could you explain a bit more > > what > > > specific problem this small window can cause? > > > > > > Since client can fetch metadata from any broker in the cluster, and > given > > > that different brokers receive request (e.g. LeaderAndIsrRequest and > > > UpdateMetadataRequest) in arbitrary order, the metadata received by > > client > > > can be in arbitrary order (either newer or older) compared to the > > broker's > > > leadership state even if a given broker receives LeaderAndIsrRequest > and > > > UpdateMetadataRequest simultaneously. So I am not sure it is useful to > > > update broker's cache with LeaderAndIsrRequest. > > > > > > > > >> 64. The enforcement of leader epoch in Offset commit: We allow a > > consumer > > >> to set an arbitrary offset. So it's possible for offsets or leader > epoch > > >> to > > >> go backwards. I am not sure if we could always enforce that the leader > > >> epoch only goes up on the broker. > > >> > > > > > > Sure. I have removed this check from the KIP. > > > > > > BTW, we can probably still ensure that the leader_epoch always increase > > if > > > the leader_epoch used with offset commit is the max(leader_epoch of the > > > message with offset = the committed offset - 1, the largest known > > > leader_epoch from the metadata). But I don't have a good use-case for > > this > > > alternative definition. So I choose the keep the KIP simple by > requiring > > > leader_epoch to always increase. > > > > > > > > >> 65. Good point on handling missing partition epoch due to topic > > deletion. > > >> Another potential way to address this is to additionally propagate the > > >> global partition epoch to brokers and the clients. This way, when a > > >> partition epoch is missing, we can use the global partition epoch to > > >> reason > > >> about which metadata is more recent. > > >> > > > > > > This is a great idea. The global epoch can be used to order the > metadata > > > and help us recognize the more recent metadata if a topic (or > partition) > > is > > > deleted and re-created. > > > > > > Actually, it seems we only need to propagate the global epoch to > brokers > > > and clients without propagating this epoch on a per-topic or > > per-partition > > > basic. Doing so would simply interface changes made this KIP. Does this > > > approach sound reasonable? > > > > > > > > >> 66. A client may also get an offset by time using the offsetForTimes() > > >> api. > > >> So, we probably want to include offsetInternalMetadata in > > >> OffsetAndTimestamp > > >> as well. > > >> > > > > > > You are right. This probably also requires us to change the > > > ListOffsetRequest as well. I will update the KIP after we agree on the > > > solution for 65). > > > > > > > > >> > > >> 67. InteralMetadata can be a bit confusing with the metadata field > > already > > >> there. Perhaps we can just call it OffsetEpoch. It might be useful to > > make > > >> OffsetEpoch printable at least for debugging purpose. Once you do > that, > > we > > >> are already exposing the internal fields. So, not sure if it's worth > > >> hiding > > >> them. If we do want to hide them, perhaps we can have sth like the > > >> following. The binary encoding is probably more efficient than JSON > for > > >> external storage. > > >> > > >> OffsetEpoch { > > >> static OffsetEpoch decode(byte[]); > > >> > > >> public byte[] encode(); > > >> > > >> public String toString(); > > >> } > > >> > > > > > > Thanks much. I like this solution. I have updated the KIP accordingly. > > > > > > > > > > > >> > > >> Jun > > >> > > >> On Mon, Jan 8, 2018 at 4:22 PM, Dong Lin <lindon...@gmail.com> wrote: > > >> > > >> > Hey Jason, > > >> > > > >> > Certainly. This sounds good. I have updated the KIP to clarity that > > the > > >> > global epoch will be incremented by 1 each time a topic is deleted. > > >> > > > >> > Thanks, > > >> > Dong > > >> > > > >> > On Mon, Jan 8, 2018 at 4:09 PM, Jason Gustafson <ja...@confluent.io > > > > >> > wrote: > > >> > > > >> > > Hi Dong, > > >> > > > > >> > > > > >> > > I think your approach will allow user to distinguish between the > > >> metadata > > >> > > > before and after the topic deletion. I also agree that this can > be > > >> > > > potentially be useful to user. I am just not very sure whether > we > > >> > already > > >> > > > have a good use-case to make the additional complexity > worthwhile. > > >> It > > >> > > seems > > >> > > > that this feature is kind of independent of the main problem of > > this > > >> > KIP. > > >> > > > Could we add this as a future work? > > >> > > > > >> > > > > >> > > Do you think it's fair if we bump the topic epoch on deletion and > > >> leave > > >> > > propagation of the epoch for deleted topics for future work? I > don't > > >> > think > > >> > > this adds much complexity and it makes the behavior consistent: > > every > > >> > topic > > >> > > mutation results in an epoch bump. > > >> > > > > >> > > Thanks, > > >> > > Jason > > >> > > > > >> > > On Mon, Jan 8, 2018 at 3:14 PM, Dong Lin <lindon...@gmail.com> > > wrote: > > >> > > > > >> > > > Hey Ismael, > > >> > > > > > >> > > > I guess we actually need user to see this field so that user can > > >> store > > >> > > this > > >> > > > value in the external store together with the offset. We just > > prefer > > >> > the > > >> > > > value to be opaque to discourage most users from interpreting > this > > >> > value. > > >> > > > One more advantage of using such an opaque field is to be able > to > > >> > evolve > > >> > > > the information (or schema) of this value without changing > > consumer > > >> API > > >> > > in > > >> > > > the future. > > >> > > > > > >> > > > I also thinking it is probably OK for user to be able to > interpret > > >> this > > >> > > > value, particularly for those advanced users. > > >> > > > > > >> > > > Thanks, > > >> > > > Dong > > >> > > > > > >> > > > On Mon, Jan 8, 2018 at 2:34 PM, Ismael Juma <ism...@juma.me.uk> > > >> wrote: > > >> > > > > > >> > > > > On Fri, Jan 5, 2018 at 7:15 PM, Jason Gustafson < > > >> ja...@confluent.io> > > >> > > > > wrote: > > >> > > > > > > > >> > > > > > class OffsetAndMetadata { > > >> > > > > > long offset; > > >> > > > > > byte[] offsetMetadata; > > >> > > > > > String metadata; > > >> > > > > > } > > >> > > > > > > >> > > > > > > >> > > > > > Admittedly, the naming is a bit annoying, but we can > probably > > >> come > > >> > up > > >> > > > > with > > >> > > > > > something better. Internally the byte array would have a > > >> version. > > >> > If > > >> > > in > > >> > > > > the > > >> > > > > > future we have anything else we need to add, we can update > the > > >> > > version > > >> > > > > and > > >> > > > > > we wouldn't need any new APIs. > > >> > > > > > > > >> > > > > > > >> > > > > We can also add fields to a class in a compatible way. So, it > > >> seems > > >> > to > > >> > > me > > >> > > > > that the main advantage of the byte array is that it's opaque > to > > >> the > > >> > > > user. > > >> > > > > Is that correct? If so, we could also add any opaque metadata > > in a > > >> > > > subclass > > >> > > > > so that users don't even see it (unless they cast it, but then > > >> > they're > > >> > > on > > >> > > > > their own). > > >> > > > > > > >> > > > > Ismael > > >> > > > > > > >> > > > > The corresponding seek() and position() APIs might look > > something > > >> > like > > >> > > > > this: > > >> > > > > > > > >> > > > > > void seek(TopicPartition partition, long offset, byte[] > > >> > > > offsetMetadata); > > >> > > > > > byte[] positionMetadata(TopicPartition partition); > > >> > > > > > > > >> > > > > > What do you think? > > >> > > > > > > > >> > > > > > Thanks, > > >> > > > > > Jason > > >> > > > > > > > >> > > > > > On Thu, Jan 4, 2018 at 7:04 PM, Dong Lin < > lindon...@gmail.com > > > > > >> > > wrote: > > >> > > > > > > > >> > > > > > > Hey Jun, Jason, > > >> > > > > > > > > >> > > > > > > Thanks much for all the feedback. I have updated the KIP > > >> based on > > >> > > the > > >> > > > > > > latest discussion. Can you help check whether it looks > good? > > >> > > > > > > > > >> > > > > > > Thanks, > > >> > > > > > > Dong > > >> > > > > > > > > >> > > > > > > On Thu, Jan 4, 2018 at 5:36 PM, Dong Lin < > > lindon...@gmail.com > > >> > > > >> > > > wrote: > > >> > > > > > > > > >> > > > > > > > Hey Jun, > > >> > > > > > > > > > >> > > > > > > > Hmm... thinking about this more, I am not sure that the > > >> > proposed > > >> > > > API > > >> > > > > is > > >> > > > > > > > sufficient. For users that store offset externally, we > > >> probably > > >> > > > need > > >> > > > > > > extra > > >> > > > > > > > API to return the leader_epoch and partition_epoch for > all > > >> > > > partitions > > >> > > > > > > that > > >> > > > > > > > consumers are consuming. I suppose these users currently > > use > > >> > > > > position() > > >> > > > > > > to > > >> > > > > > > > get the offset. Thus we probably need a new method > > >> > > > > > positionWithEpoch(..) > > >> > > > > > > to > > >> > > > > > > > return <offset, partition_epoch, leader_epoch>. Does > this > > >> sound > > >> > > > > > > reasonable? > > >> > > > > > > > > > >> > > > > > > > Thanks, > > >> > > > > > > > Dong > > >> > > > > > > > > > >> > > > > > > > > > >> > > > > > > > On Thu, Jan 4, 2018 at 5:26 PM, Jun Rao < > j...@confluent.io > > > > > >> > > wrote: > > >> > > > > > > > > > >> > > > > > > >> Hi, Dong, > > >> > > > > > > >> > > >> > > > > > > >> Yes, that's what I am thinking. OffsetEpoch will be > > >> composed > > >> > of > > >> > > > > > > >> (partition_epoch, > > >> > > > > > > >> leader_epoch). > > >> > > > > > > >> > > >> > > > > > > >> Thanks, > > >> > > > > > > >> > > >> > > > > > > >> Jun > > >> > > > > > > >> > > >> > > > > > > >> > > >> > > > > > > >> On Thu, Jan 4, 2018 at 4:22 PM, Dong Lin < > > >> lindon...@gmail.com > > >> > > > > >> > > > > wrote: > > >> > > > > > > >> > > >> > > > > > > >> > Hey Jun, > > >> > > > > > > >> > > > >> > > > > > > >> > Thanks much. I like the the new API that you > proposed. > > I > > >> am > > >> > > not > > >> > > > > sure > > >> > > > > > > >> what > > >> > > > > > > >> > you exactly mean by offset_epoch. I suppose that we > can > > >> use > > >> > > the > > >> > > > > pair > > >> > > > > > > of > > >> > > > > > > >> > (partition_epoch, leader_epoch) as the offset_epoch, > > >> right? > > >> > > > > > > >> > > > >> > > > > > > >> > Thanks, > > >> > > > > > > >> > Dong > > >> > > > > > > >> > > > >> > > > > > > >> > On Thu, Jan 4, 2018 at 4:02 PM, Jun Rao < > > >> j...@confluent.io> > > >> > > > wrote: > > >> > > > > > > >> > > > >> > > > > > > >> > > Hi, Dong, > > >> > > > > > > >> > > > > >> > > > > > > >> > > Got it. The api that you proposed works. The > question > > >> is > > >> > > > whether > > >> > > > > > > >> that's > > >> > > > > > > >> > the > > >> > > > > > > >> > > api that we want to have in the long term. My > concern > > >> is > > >> > > that > > >> > > > > > while > > >> > > > > > > >> the > > >> > > > > > > >> > api > > >> > > > > > > >> > > change is simple, the new api seems harder to > explain > > >> and > > >> > > use. > > >> > > > > For > > >> > > > > > > >> > example, > > >> > > > > > > >> > > a consumer storing offsets externally now needs to > > call > > >> > > > > > > >> > > waitForMetadataUpdate() after calling seek(). > > >> > > > > > > >> > > > > >> > > > > > > >> > > An alternative approach is to make the following > > >> > compatible > > >> > > > api > > >> > > > > > > >> changes > > >> > > > > > > >> > in > > >> > > > > > > >> > > Consumer. > > >> > > > > > > >> > > * Add an additional OffsetEpoch field in > > >> > OffsetAndMetadata. > > >> > > > (no > > >> > > > > > need > > >> > > > > > > >> to > > >> > > > > > > >> > > change the CommitSync() api) > > >> > > > > > > >> > > * Add a new api seek(TopicPartition partition, long > > >> > offset, > > >> > > > > > > >> OffsetEpoch > > >> > > > > > > >> > > offsetEpoch). We can potentially deprecate the old > > api > > >> > > > > > > >> > seek(TopicPartition > > >> > > > > > > >> > > partition, long offset) in the future. > > >> > > > > > > >> > > > > >> > > > > > > >> > > The alternative approach has similar amount of api > > >> changes > > >> > > as > > >> > > > > > yours > > >> > > > > > > >> but > > >> > > > > > > >> > has > > >> > > > > > > >> > > the following benefits. > > >> > > > > > > >> > > 1. The api works in a similar way as how offset > > >> management > > >> > > > works > > >> > > > > > now > > >> > > > > > > >> and > > >> > > > > > > >> > is > > >> > > > > > > >> > > probably what we want in the long term. > > >> > > > > > > >> > > 2. It can reset offsets better when there is data > > loss > > >> due > > >> > > to > > >> > > > > > > unclean > > >> > > > > > > >> > > leader election or correlated replica failure. > > >> > > > > > > >> > > 3. It can reset offsets better when topic is > > recreated. > > >> > > > > > > >> > > > > >> > > > > > > >> > > Thanks, > > >> > > > > > > >> > > > > >> > > > > > > >> > > Jun > > >> > > > > > > >> > > > > >> > > > > > > >> > > > > >> > > > > > > >> > > On Thu, Jan 4, 2018 at 2:05 PM, Dong Lin < > > >> > > lindon...@gmail.com > > >> > > > > > > >> > > > > > > wrote: > > >> > > > > > > >> > > > > >> > > > > > > >> > > > Hey Jun, > > >> > > > > > > >> > > > > > >> > > > > > > >> > > > Yeah I agree that ideally we don't want an ever > > >> growing > > >> > > > global > > >> > > > > > > >> metadata > > >> > > > > > > >> > > > version. I just think it may be more desirable to > > >> keep > > >> > the > > >> > > > > > > consumer > > >> > > > > > > >> API > > >> > > > > > > >> > > > simple. > > >> > > > > > > >> > > > > > >> > > > > > > >> > > > In my current proposal, metadata version returned > > in > > >> the > > >> > > > fetch > > >> > > > > > > >> response > > >> > > > > > > >> > > > will be stored with the offset together. More > > >> > > specifically, > > >> > > > > the > > >> > > > > > > >> > > > metadata_epoch in the new offset topic schema > will > > be > > >> > the > > >> > > > > > largest > > >> > > > > > > >> > > > metadata_epoch from all the MetadataResponse and > > >> > > > FetchResponse > > >> > > > > > > ever > > >> > > > > > > >> > > > received by this consumer. > > >> > > > > > > >> > > > > > >> > > > > > > >> > > > We probably don't have to change the consumer API > > for > > >> > > > > > > >> > > > commitSync(Map<TopicPartition, > OffsetAndMetadata>). > > >> If > > >> > > user > > >> > > > > > calls > > >> > > > > > > >> > > > commitSync(...) to commit offset 10 for a given > > >> > partition, > > >> > > > for > > >> > > > > > > most > > >> > > > > > > >> > > > use-cases, this consumer instance should have > > >> consumed > > >> > > > message > > >> > > > > > > with > > >> > > > > > > >> > > offset > > >> > > > > > > >> > > > 9 from this partition, in which case the consumer > > can > > >> > > > remember > > >> > > > > > and > > >> > > > > > > >> use > > >> > > > > > > >> > > the > > >> > > > > > > >> > > > metadata_epoch from the corresponding > FetchResponse > > >> when > > >> > > > > > > committing > > >> > > > > > > >> > > offset. > > >> > > > > > > >> > > > If user calls commitSync(..) to commit offset 10 > > for > > >> a > > >> > > given > > >> > > > > > > >> partition > > >> > > > > > > >> > > > without having consumed the message with offset 9 > > >> using > > >> > > this > > >> > > > > > > >> consumer > > >> > > > > > > >> > > > instance, this is probably an advanced use-case. > In > > >> this > > >> > > > case > > >> > > > > > the > > >> > > > > > > >> > > advanced > > >> > > > > > > >> > > > user can retrieve the metadata_epoch using the > > newly > > >> > added > > >> > > > > > > >> > > metadataEpoch() > > >> > > > > > > >> > > > API after it fetches the message with offset 9 > > >> (probably > > >> > > > from > > >> > > > > > > >> another > > >> > > > > > > >> > > > consumer instance) and encode this metadata_epoch > > in > > >> the > > >> > > > > > > >> > > > string OffsetAndMetadata.metadata. Do you think > > this > > >> > > > solution > > >> > > > > > > would > > >> > > > > > > >> > work? > > >> > > > > > > >> > > > > > >> > > > > > > >> > > > By "not sure that I fully understand your latest > > >> > > > suggestion", > > >> > > > > > are > > >> > > > > > > >> you > > >> > > > > > > >> > > > referring to solution related to unclean leader > > >> election > > >> > > > using > > >> > > > > > > >> > > leader_epoch > > >> > > > > > > >> > > > in my previous email? > > >> > > > > > > >> > > > > > >> > > > > > > >> > > > Thanks, > > >> > > > > > > >> > > > Dong > > >> > > > > > > >> > > > > > >> > > > > > > >> > > > On Thu, Jan 4, 2018 at 1:33 PM, Jun Rao < > > >> > j...@confluent.io > > >> > > > > > >> > > > > > wrote: > > >> > > > > > > >> > > > > > >> > > > > > > >> > > > > Hi, Dong, > > >> > > > > > > >> > > > > > > >> > > > > > > >> > > > > Not sure that I fully understand your latest > > >> > suggestion. > > >> > > > > > > >> Returning an > > >> > > > > > > >> > > > ever > > >> > > > > > > >> > > > > growing global metadata version itself is no > > ideal, > > >> > but > > >> > > is > > >> > > > > > fine. > > >> > > > > > > >> My > > >> > > > > > > >> > > > > question is whether the metadata version > returned > > >> in > > >> > the > > >> > > > > fetch > > >> > > > > > > >> > response > > >> > > > > > > >> > > > > needs to be stored with the offset together if > > >> offsets > > >> > > are > > >> > > > > > > stored > > >> > > > > > > >> > > > > externally. If so, we also have to change the > > >> consumer > > >> > > API > > >> > > > > for > > >> > > > > > > >> > > > commitSync() > > >> > > > > > > >> > > > > and need to worry about compatibility. If we > > don't > > >> > store > > >> > > > the > > >> > > > > > > >> metadata > > >> > > > > > > >> > > > > version together with the offset, on a consumer > > >> > restart, > > >> > > > > it's > > >> > > > > > > not > > >> > > > > > > >> > clear > > >> > > > > > > >> > > > how > > >> > > > > > > >> > > > > we can ensure the metadata in the consumer is > > high > > >> > > enough > > >> > > > > > since > > >> > > > > > > >> there > > >> > > > > > > >> > > is > > >> > > > > > > >> > > > no > > >> > > > > > > >> > > > > metadata version to compare with. > > >> > > > > > > >> > > > > > > >> > > > > > > >> > > > > Thanks, > > >> > > > > > > >> > > > > > > >> > > > > > > >> > > > > Jun > > >> > > > > > > >> > > > > > > >> > > > > > > >> > > > > > > >> > > > > > > >> > > > > On Wed, Jan 3, 2018 at 6:43 PM, Dong Lin < > > >> > > > > lindon...@gmail.com > > >> > > > > > > > > >> > > > > > > >> > wrote: > > >> > > > > > > >> > > > > > > >> > > > > > > >> > > > > > Hey Jun, > > >> > > > > > > >> > > > > > > > >> > > > > > > >> > > > > > Thanks much for the explanation. > > >> > > > > > > >> > > > > > > > >> > > > > > > >> > > > > > I understand the advantage of partition_epoch > > >> over > > >> > > > > > > >> metadata_epoch. > > >> > > > > > > >> > My > > >> > > > > > > >> > > > > > current concern is that the use of > leader_epoch > > >> and > > >> > > the > > >> > > > > > > >> > > partition_epoch > > >> > > > > > > >> > > > > > requires us considerable change to consumer's > > >> public > > >> > > API > > >> > > > > to > > >> > > > > > > take > > >> > > > > > > >> > care > > >> > > > > > > >> > > > of > > >> > > > > > > >> > > > > > the case where user stores offset externally. > > For > > >> > > > example, > > >> > > > > > > >> > > *consumer*. > > >> > > > > > > >> > > > > > *commitSync*(..) would have to take a map > whose > > >> > value > > >> > > is > > >> > > > > > > >> <offset, > > >> > > > > > > >> > > > > metadata, > > >> > > > > > > >> > > > > > leader epoch, partition epoch>. > > >> > *consumer*.*seek*(...) > > >> > > > > would > > >> > > > > > > >> also > > >> > > > > > > >> > > need > > >> > > > > > > >> > > > > > leader_epoch and partition_epoch as > parameter. > > >> > > > Technically > > >> > > > > > we > > >> > > > > > > >> can > > >> > > > > > > >> > > > > probably > > >> > > > > > > >> > > > > > still make it work in a backward compatible > > >> manner > > >> > > after > > >> > > > > > > careful > > >> > > > > > > >> > > design > > >> > > > > > > >> > > > > and > > >> > > > > > > >> > > > > > discussion. But these changes can make the > > >> > consumer's > > >> > > > > > > interface > > >> > > > > > > >> > > > > > unnecessarily complex for more users who do > not > > >> > store > > >> > > > > offset > > >> > > > > > > >> > > > externally. > > >> > > > > > > >> > > > > > > > >> > > > > > > >> > > > > > After thinking more about it, we can address > > all > > >> > > > problems > > >> > > > > > > >> discussed > > >> > > > > > > >> > > by > > >> > > > > > > >> > > > > only > > >> > > > > > > >> > > > > > using the metadata_epoch without introducing > > >> > > > leader_epoch > > >> > > > > or > > >> > > > > > > the > > >> > > > > > > >> > > > > > partition_epoch. The current KIP describes > the > > >> > changes > > >> > > > to > > >> > > > > > the > > >> > > > > > > >> > > consumer > > >> > > > > > > >> > > > > API > > >> > > > > > > >> > > > > > and how the new API can be used if user > stores > > >> > offset > > >> > > > > > > >> externally. > > >> > > > > > > >> > In > > >> > > > > > > >> > > > > order > > >> > > > > > > >> > > > > > to address the scenario you described > earlier, > > we > > >> > can > > >> > > > > > include > > >> > > > > > > >> > > > > > metadata_epoch in the FetchResponse and the > > >> > > > > > > LeaderAndIsrRequest. > > >> > > > > > > >> > > > Consumer > > >> > > > > > > >> > > > > > remembers the largest metadata_epoch from all > > the > > >> > > > > > > FetchResponse > > >> > > > > > > >> it > > >> > > > > > > >> > > has > > >> > > > > > > >> > > > > > received. The metadata_epoch committed with > the > > >> > > offset, > > >> > > > > > either > > >> > > > > > > >> > within > > >> > > > > > > >> > > > or > > >> > > > > > > >> > > > > > outside Kafka, should be the largest > > >> metadata_epoch > > >> > > > across > > >> > > > > > all > > >> > > > > > > >> > > > > > FetchResponse and MetadataResponse ever > > received > > >> by > > >> > > this > > >> > > > > > > >> consumer. > > >> > > > > > > >> > > > > > > > >> > > > > > > >> > > > > > The drawback of using only the metadata_epoch > > is > > >> > that > > >> > > we > > >> > > > > can > > >> > > > > > > not > > >> > > > > > > >> > > always > > >> > > > > > > >> > > > > do > > >> > > > > > > >> > > > > > the smart offset reset in case of unclean > > leader > > >> > > > election > > >> > > > > > > which > > >> > > > > > > >> you > > >> > > > > > > >> > > > > > mentioned earlier. But in most case, unclean > > >> leader > > >> > > > > election > > >> > > > > > > >> > probably > > >> > > > > > > >> > > > > > happens when consumer is not > > >> rebalancing/restarting. > > >> > > In > > >> > > > > > these > > >> > > > > > > >> > cases, > > >> > > > > > > >> > > > > either > > >> > > > > > > >> > > > > > consumer is not directly affected by unclean > > >> leader > > >> > > > > election > > >> > > > > > > >> since > > >> > > > > > > >> > it > > >> > > > > > > >> > > > is > > >> > > > > > > >> > > > > > not consuming from the end of the log, or > > >> consumer > > >> > can > > >> > > > > > derive > > >> > > > > > > >> the > > >> > > > > > > >> > > > > > leader_epoch from the most recent message > > >> received > > >> > > > before > > >> > > > > it > > >> > > > > > > >> sees > > >> > > > > > > >> > > > > > OffsetOutOfRangeException. So I am not sure > it > > is > > >> > > worth > > >> > > > > > adding > > >> > > > > > > >> the > > >> > > > > > > >> > > > > > leader_epoch to consumer API to address the > > >> > remaining > > >> > > > > corner > > >> > > > > > > >> case. > > >> > > > > > > >> > > What > > >> > > > > > > >> > > > > do > > >> > > > > > > >> > > > > > you think? > > >> > > > > > > >> > > > > > > > >> > > > > > > >> > > > > > Thanks, > > >> > > > > > > >> > > > > > Dong > > >> > > > > > > >> > > > > > > > >> > > > > > > >> > > > > > > > >> > > > > > > >> > > > > > > > >> > > > > > > >> > > > > > On Tue, Jan 2, 2018 at 6:28 PM, Jun Rao < > > >> > > > j...@confluent.io > > >> > > > > > > > >> > > > > > > >> wrote: > > >> > > > > > > >> > > > > > > > >> > > > > > > >> > > > > > > Hi, Dong, > > >> > > > > > > >> > > > > > > > > >> > > > > > > >> > > > > > > Thanks for the reply. > > >> > > > > > > >> > > > > > > > > >> > > > > > > >> > > > > > > To solve the topic recreation issue, we > could > > >> use > > >> > > > > either a > > >> > > > > > > >> global > > >> > > > > > > >> > > > > > metadata > > >> > > > > > > >> > > > > > > version or a partition level epoch. But > > either > > >> one > > >> > > > will > > >> > > > > > be a > > >> > > > > > > >> new > > >> > > > > > > >> > > > > concept, > > >> > > > > > > >> > > > > > > right? To me, the latter seems more > natural. > > It > > >> > also > > >> > > > > makes > > >> > > > > > > it > > >> > > > > > > >> > > easier > > >> > > > > > > >> > > > to > > >> > > > > > > >> > > > > > > detect if a consumer's offset is still > valid > > >> > after a > > >> > > > > topic > > >> > > > > > > is > > >> > > > > > > >> > > > > recreated. > > >> > > > > > > >> > > > > > As > > >> > > > > > > >> > > > > > > you pointed out, we don't need to store the > > >> > > partition > > >> > > > > > epoch > > >> > > > > > > in > > >> > > > > > > >> > the > > >> > > > > > > >> > > > > > message. > > >> > > > > > > >> > > > > > > The following is what I am thinking. When a > > >> > > partition > > >> > > > is > > >> > > > > > > >> created, > > >> > > > > > > >> > > we > > >> > > > > > > >> > > > > can > > >> > > > > > > >> > > > > > > assign a partition epoch from an > > >> ever-increasing > > >> > > > global > > >> > > > > > > >> counter > > >> > > > > > > >> > and > > >> > > > > > > >> > > > > store > > >> > > > > > > >> > > > > > > it in /brokers/topics/[topic]/ > > >> > > > partitions/[partitionId] > > >> > > > > in > > >> > > > > > > ZK. > > >> > > > > > > >> > The > > >> > > > > > > >> > > > > > > partition > > >> > > > > > > >> > > > > > > epoch is propagated to every broker. The > > >> consumer > > >> > > will > > >> > > > > be > > >> > > > > > > >> > tracking > > >> > > > > > > >> > > a > > >> > > > > > > >> > > > > > tuple > > >> > > > > > > >> > > > > > > of <offset, leader epoch, partition epoch> > > for > > >> > > > offsets. > > >> > > > > > If a > > >> > > > > > > >> > topic > > >> > > > > > > >> > > is > > >> > > > > > > >> > > > > > > recreated, it's possible that a consumer's > > >> offset > > >> > > and > > >> > > > > > leader > > >> > > > > > > >> > epoch > > >> > > > > > > >> > > > > still > > >> > > > > > > >> > > > > > > match that in the broker, but partition > epoch > > >> > won't > > >> > > > be. > > >> > > > > In > > >> > > > > > > >> this > > >> > > > > > > >> > > case, > > >> > > > > > > >> > > > > we > > >> > > > > > > >> > > > > > > can potentially still treat the consumer's > > >> offset > > >> > as > > >> > > > out > > >> > > > > > of > > >> > > > > > > >> range > > >> > > > > > > >> > > and > > >> > > > > > > >> > > > > > reset > > >> > > > > > > >> > > > > > > the offset based on the offset reset policy > > in > > >> the > > >> > > > > > consumer. > > >> > > > > > > >> This > > >> > > > > > > >> > > > seems > > >> > > > > > > >> > > > > > > harder to do with a global metadata > version. > > >> > > > > > > >> > > > > > > > > >> > > > > > > >> > > > > > > Jun > > >> > > > > > > >> > > > > > > > > >> > > > > > > >> > > > > > > > > >> > > > > > > >> > > > > > > > > >> > > > > > > >> > > > > > > On Mon, Dec 25, 2017 at 6:56 AM, Dong Lin < > > >> > > > > > > >> lindon...@gmail.com> > > >> > > > > > > >> > > > wrote: > > >> > > > > > > >> > > > > > > > > >> > > > > > > >> > > > > > > > Hey Jun, > > >> > > > > > > >> > > > > > > > > > >> > > > > > > >> > > > > > > > This is a very good example. After > thinking > > >> > > through > > >> > > > > this > > >> > > > > > > in > > >> > > > > > > >> > > > detail, I > > >> > > > > > > >> > > > > > > agree > > >> > > > > > > >> > > > > > > > that we need to commit offset with leader > > >> epoch > > >> > in > > >> > > > > order > > >> > > > > > > to > > >> > > > > > > >> > > address > > >> > > > > > > >> > > > > > this > > >> > > > > > > >> > > > > > > > example. > > >> > > > > > > >> > > > > > > > > > >> > > > > > > >> > > > > > > > I think the remaining question is how to > > >> address > > >> > > the > > >> > > > > > > >> scenario > > >> > > > > > > >> > > that > > >> > > > > > > >> > > > > the > > >> > > > > > > >> > > > > > > > topic is deleted and re-created. One > > possible > > >> > > > solution > > >> > > > > > is > > >> > > > > > > to > > >> > > > > > > >> > > commit > > >> > > > > > > >> > > > > > > offset > > >> > > > > > > >> > > > > > > > with both the leader epoch and the > metadata > > >> > > version. > > >> > > > > The > > >> > > > > > > >> logic > > >> > > > > > > >> > > and > > >> > > > > > > >> > > > > the > > >> > > > > > > >> > > > > > > > implementation of this solution does not > > >> > require a > > >> > > > new > > >> > > > > > > >> concept > > >> > > > > > > >> > > > (e.g. > > >> > > > > > > >> > > > > > > > partition epoch) and it does not require > > any > > >> > > change > > >> > > > to > > >> > > > > > the > > >> > > > > > > >> > > message > > >> > > > > > > >> > > > > > format > > >> > > > > > > >> > > > > > > > or leader epoch. It also allows us to > order > > >> the > > >> > > > > metadata > > >> > > > > > > in > > >> > > > > > > >> a > > >> > > > > > > >> > > > > > > > straightforward manner which may be > useful > > in > > >> > the > > >> > > > > > future. > > >> > > > > > > >> So it > > >> > > > > > > >> > > may > > >> > > > > > > >> > > > > be > > >> > > > > > > >> > > > > > a > > >> > > > > > > >> > > > > > > > better solution than generating a random > > >> > partition > > >> > > > > epoch > > >> > > > > > > >> every > > >> > > > > > > >> > > time > > >> > > > > > > >> > > > > we > > >> > > > > > > >> > > > > > > > create a partition. Does this sound > > >> reasonable? > > >> > > > > > > >> > > > > > > > > > >> > > > > > > >> > > > > > > > Previously one concern with using the > > >> metadata > > >> > > > version > > >> > > > > > is > > >> > > > > > > >> that > > >> > > > > > > >> > > > > consumer > > >> > > > > > > >> > > > > > > > will be forced to refresh metadata even > if > > >> > > metadata > > >> > > > > > > version > > >> > > > > > > >> is > > >> > > > > > > >> > > > > > increased > > >> > > > > > > >> > > > > > > > due to topics that the consumer is not > > >> > interested > > >> > > > in. > > >> > > > > > Now > > >> > > > > > > I > > >> > > > > > > >> > > > realized > > >> > > > > > > >> > > > > > that > > >> > > > > > > >> > > > > > > > this is probably not a problem. Currently > > >> client > > >> > > > will > > >> > > > > > > >> refresh > > >> > > > > > > >> > > > > metadata > > >> > > > > > > >> > > > > > > > either due to InvalidMetadataException in > > the > > >> > > > response > > >> > > > > > > from > > >> > > > > > > >> > > broker > > >> > > > > > > >> > > > or > > >> > > > > > > >> > > > > > due > > >> > > > > > > >> > > > > > > > to metadata expiry. The addition of the > > >> metadata > > >> > > > > version > > >> > > > > > > >> should > > >> > > > > > > >> > > > > > increase > > >> > > > > > > >> > > > > > > > the overhead of metadata refresh caused > by > > >> > > > > > > >> > > > InvalidMetadataException. > > >> > > > > > > >> > > > > If > > >> > > > > > > >> > > > > > > > client refresh metadata due to expiry and > > it > > >> > > > receives > > >> > > > > a > > >> > > > > > > >> > metadata > > >> > > > > > > >> > > > > whose > > >> > > > > > > >> > > > > > > > version is lower than the current > metadata > > >> > > version, > > >> > > > we > > >> > > > > > can > > >> > > > > > > >> > reject > > >> > > > > > > >> > > > the > > >> > > > > > > >> > > > > > > > metadata but still reset the metadata > age, > > >> which > > >> > > > > > > essentially > > >> > > > > > > >> > keep > > >> > > > > > > >> > > > the > > >> > > > > > > >> > > > > > > > existing behavior in the client. > > >> > > > > > > >> > > > > > > > > > >> > > > > > > >> > > > > > > > Thanks much, > > >> > > > > > > >> > > > > > > > Dong > > >> > > > > > > >> > > > > > > > > > >> > > > > > > >> > > > > > > > > >> > > > > > > >> > > > > > > > >> > > > > > > >> > > > > > > >> > > > > > > >> > > > > > >> > > > > > > >> > > > > >> > > > > > > >> > > > >> > > > > > > >> > > >> > > > > > > > > > >> > > > > > > > > > >> > > > > > > > > >> > > > > > > > >> > > > > > > >> > > > > > >> > > > > >> > > > >> > > > > > > > > >