Hi Jason, There are a few problems we want to solve here: 1. The group metadata is too big to be appended to the log. 2. Reduce the memory footprint on the broker 3. Reduce the bytes transferred over the wire.
To solve (1), I like your idea of having separate messages per member. The proposal (Onur's option 8) is to break metadata into small records in the same uncompressed message set so each record is small. I agree it would be ideal if we are able to store the metadata separately for each member. I was also thinking about storing the metadata into multiple messages, too. What concerns me was that having multiple messages seems breaking the atomicity. I am not sure how we are going to deal with the potential issues. For example, What if group metadata is replicated but the member metadata is not? It might be fine depending on the implementation though, but I am not sure. For (2) we want to store the metadata onto the disk, which is what we have to do anyway. The only question is in what format should we store them. To address (3) we want to have the metadata to be compressed, which is contradict to the the above solution of (1). I think Jun's suggestion is probably still the simplest. To avoid changing the behavior for consumers, maybe we can do that only for offset_topic, i.e, if the max fetch bytes of the fetch request is smaller than the message size on the offset topic, we always return at least one full message. This should avoid the unexpected problem on the client side because supposedly only tools and brokers will fetch from the the internal topics, As a modification to what you suggested, one solution I was thinking was to have multiple messages in a single compressed message. That means for SyncGroupResponse we still need to read the entire compressed messages and extract the inner messages, which seems not quite different from having a single message containing everything. But let me just put it here and see if that makes sense. We can have a map of GroupMetadataKey -> GroupMetadataValueOffset. The GroupMetadataValue is stored in a compressed message. The inner messages are the following: Inner Message 0: Version GroupId Generation Inner Message 1: MemberId MemberMetadata_1 (we can compress the bytes here) Inner Message 2: MemberId MemberMetadata_2 .... Inner Message N: MemberId MemberMetadata_N The MemberMetadata format is the following: MemberMetadata => Version Generation ClientId Host Subscription Assignment So DescribeGroupResponse will just return the entire compressed GroupMetadataMessage. SyncGroupResponse will return the corresponding inner message. Thanks, Jiangjie (Becket) Qin On Tue, May 24, 2016 at 9:14 AM, Jason Gustafson <ja...@confluent.io> wrote: > Hey Becket, > > I like your idea to store only the offset for the group metadata in memory. > I think it would be safe to keep it in memory for a short time after the > rebalance completes, but after that, it's only real purpose is to answer > DescribeGroup requests, so your proposal makes a lot of sense to me. > > As for the specific problem with the size of the group metadata message for > the MM case, if we cannot succeed in reducing the size of the > subscription/assignment (which I think is still probably the best > alternative if it can work), then I think there are some options for > changing the message format (option #8 in Onur's initial e-mail). > Currently, the key used for storing the group metadata is this: > > GroupMetadataKey => Version GroupId > > And the value is something like this (some details elided): > > GroupMetadataValue => Version GroupId Generation [MemberMetadata] > MemberMetadata => ClientId Host Subscription Assignment > > I don't think we can change the key without a lot of pain, but it seems > like we can change the value format. Maybe we can take the > subscription/assignment payloads out of the value and introduce a new > "MemberMetadata" message for each member in the group. For example: > > MemberMetadataKey => Version GroupId MemberId > > MemberMetadataValue => Version Generation ClientId Host Subscription > Assignment > > When a new generation is created, we would first write the group metadata > message which includes the generation and all of the memberIds, and then > we'd write the member metadata messages. To answer the DescribeGroup > request, we'd read the group metadata at the cached offset and, depending > on the version, all of the following member metadata. This would be more > complex to maintain, but it seems doable if it comes to it. > > Thanks, > Jason > > On Mon, May 23, 2016 at 6:15 PM, Becket Qin <becket....@gmail.com> wrote: > > > It might worth thinking a little further. We have discussed this before > > that we want to avoid holding all the group metadata in memory. > > > > I am thinking about the following end state: > > > > 1. Enable compression on the offset topic. > > 2. Instead of holding the entire group metadata in memory on the brokers, > > each broker only keeps a [group -> Offset] map, the offset points to the > > message in the offset topic which holds the latest metadata of the group. > > 3. DescribeGroupResponse will read from the offset topic directly like a > > normal consumption, except that only exactly one message will be > returned. > > 4. SyncGroupResponse will read the message, extract the assignment part > and > > send back the partition assignment. We can compress the partition > > assignment before sends it out if we want. > > > > Jiangjie (Becket) Qin > > > > On Mon, May 23, 2016 at 5:08 PM, Jason Gustafson <ja...@confluent.io> > > wrote: > > > > > > > > > > Jason, doesn't gzip (or other compression) basically do this? If the > > > topic > > > > is a string and the topic is repeated throughout, won't compression > > > > basically replace all repeated instances of it with an index > reference > > to > > > > the full string? > > > > > > > > > Hey James, yeah, that's probably true, but keep in mind that the > > > compression happens on the broker side. It would be nice to have a more > > > compact representation so that get some benefit over the wire as well. > > This > > > seems to be less of a concern here, so the bigger gains are probably > from > > > reducing the number of partitions that need to be listed individually. > > > > > > -Jason > > > > > > On Mon, May 23, 2016 at 4:23 PM, Onur Karaman < > > > onurkaraman.apa...@gmail.com> > > > wrote: > > > > > > > When figuring out these optimizations, it's worth keeping in mind the > > > > improvements when the message is uncompressed vs when it's > compressed. > > > > > > > > When uncompressed: > > > > Fixing the Assignment serialization to instead be a topic index into > > the > > > > corresponding member's subscription list would usually be a good > thing. > > > > > > > > I think the proposal is only worse when the topic names are small. > The > > > > Type.STRING we use in our protocol for the assignment's > TOPIC_KEY_NAME > > is > > > > limited in length to Short.MAX_VALUE, so our strings are first > > prepended > > > > with 2 bytes to indicate the string size. > > > > > > > > The new proposal does worse when: > > > > 2 + utf_encoded_string_payload_size < index_type_size > > > > in other words when: > > > > utf_encoded_string_payload_size < index_type_size - 2 > > > > > > > > If the index type ends up being Type.INT32, then the proposal is > worse > > > when > > > > the topic is length 1. > > > > If the index type ends up being Type.INT64, then the proposal is > worse > > > when > > > > the topic is less than length 6. > > > > > > > > When compressed: > > > > As James Cheng brought up, I'm not sure how things change when > > > compression > > > > comes into the picture. This would be worth investigating. > > > > > > > > On Mon, May 23, 2016 at 4:05 PM, James Cheng <wushuja...@gmail.com> > > > wrote: > > > > > > > > > > > > > > > On May 23, 2016, at 10:59 AM, Jason Gustafson < > ja...@confluent.io> > > > > > wrote: > > > > > > > > > > > > 2. Maybe there's a better way to lay out the assignment without > > > needing > > > > > to > > > > > > explicitly repeat the topic? For example, the leader could sort > the > > > > > topics > > > > > > for each member and just use an integer to represent the index of > > > each > > > > > > topic within the sorted list (note this depends on the > subscription > > > > > > including the full topic list). > > > > > > > > > > > > Assignment -> [TopicIndex [Partition]] > > > > > > > > > > > > > > > > Jason, doesn't gzip (or other compression) basically do this? If > the > > > > topic > > > > > is a string and the topic is repeated throughout, won't compression > > > > > basically replace all repeated instances of it with an index > > reference > > > to > > > > > the full string? > > > > > > > > > > -James > > > > > > > > > > > You could even combine these two options so that you have only 3 > > > > integers > > > > > > for each topic assignment: > > > > > > > > > > > > Assignment -> [TopicIndex MinPartition MaxPartition] > > > > > > > > > > > > There may even be better options with a little more thought. All > of > > > > this > > > > > is > > > > > > just part of the client-side protocol, so it wouldn't require any > > > > version > > > > > > bumps on the broker. What do you think? > > > > > > > > > > > > Thanks, > > > > > > Jason > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > On Mon, May 23, 2016 at 9:17 AM, Guozhang Wang < > wangg...@gmail.com > > > > > > > > wrote: > > > > > > > > > > > >> The original concern is that regex may not be efficiently > > supported > > > > > >> across-languages, but if there is a neat workaround I would love > > to > > > > > learn. > > > > > >> > > > > > >> Guozhang > > > > > >> > > > > > >> On Mon, May 23, 2016 at 5:31 AM, Ismael Juma <ism...@juma.me.uk > > > > > > wrote: > > > > > >> > > > > > >>> +1 to Jun's suggestion. > > > > > >>> > > > > > >>> Having said that, as a general point, I think we should > consider > > > > > >> supporting > > > > > >>> topic patterns in the wire protocol. It requires some thinking > > for > > > > > >>> cross-language support, but it seems surmountable and it could > > make > > > > > >> certain > > > > > >>> operations a lot more efficient (the fact that a basic regex > > > > > subscription > > > > > >>> causes the consumer to request metadata for all topics is not > > > great). > > > > > >>> > > > > > >>> Ismael > > > > > >>> > > > > > >>> On Sun, May 22, 2016 at 11:49 PM, Guozhang Wang < > > > wangg...@gmail.com> > > > > > >>> wrote: > > > > > >>> > > > > > >>>> I like Jun's suggestion in changing the handling logics of > > single > > > > > large > > > > > >>>> message on the consumer side. > > > > > >>>> > > > > > >>>> As for the case of "a single group subscribing to 3000 > topics", > > > with > > > > > >> 100 > > > > > >>>> consumers the 2.5Mb Gzip size is reasonable to me (when > storing > > in > > > > ZK, > > > > > >> we > > > > > >>>> also have the znode limit which is set to 1Mb by default, > though > > > > > >>> admittedly > > > > > >>>> it is only for one consumer). And if we do the change as Jun > > > > > suggested, > > > > > >>>> 2.5Mb on follower's memory pressure is OK I think. > > > > > >>>> > > > > > >>>> > > > > > >>>> Guozhang > > > > > >>>> > > > > > >>>> > > > > > >>>> On Sat, May 21, 2016 at 12:51 PM, Onur Karaman < > > > > > >>>> onurkaraman.apa...@gmail.com > > > > > >>>>> wrote: > > > > > >>>> > > > > > >>>>> Results without compression: > > > > > >>>>> 1 consumer 292383 bytes > > > > > >>>>> 5 consumers 1079579 bytes * the tipping point > > > > > >>>>> 10 consumers 1855018 bytes > > > > > >>>>> 20 consumers 2780220 bytes > > > > > >>>>> 30 consumers 3705422 bytes > > > > > >>>>> 40 consumers 4630624 bytes > > > > > >>>>> 50 consumers 5555826 bytes > > > > > >>>>> 60 consumers 6480788 bytes > > > > > >>>>> 70 consumers 7405750 bytes > > > > > >>>>> 80 consumers 8330712 bytes > > > > > >>>>> 90 consumers 9255674 bytes > > > > > >>>>> 100 consumers 10180636 bytes > > > > > >>>>> > > > > > >>>>> So it looks like gzip compression shrinks the message size by > > 4x. > > > > > >>>>> > > > > > >>>>> On Sat, May 21, 2016 at 9:47 AM, Jun Rao <j...@confluent.io> > > > wrote: > > > > > >>>>> > > > > > >>>>>> Onur, > > > > > >>>>>> > > > > > >>>>>> Thanks for the investigation. > > > > > >>>>>> > > > > > >>>>>> Another option is to just fix how we deal with the case > when a > > > > > >>> message > > > > > >>>> is > > > > > >>>>>> larger than the fetch size. Today, if the fetch size is > > smaller > > > > > >> than > > > > > >>>> the > > > > > >>>>>> fetch size, the consumer will get stuck. Instead, we can > > simply > > > > > >>> return > > > > > >>>>> the > > > > > >>>>>> full message if it's larger than the fetch size w/o > requiring > > > the > > > > > >>>>> consumer > > > > > >>>>>> to manually adjust the fetch size. On the broker side, to > > serve > > > a > > > > > >>> fetch > > > > > >>>>>> request, we already do an index lookup and then scan the > log a > > > bit > > > > > >> to > > > > > >>>>> find > > > > > >>>>>> the message with the requested offset. We can just check the > > > size > > > > > >> of > > > > > >>>> that > > > > > >>>>>> message and return the full message if its size is larger > than > > > the > > > > > >>>> fetch > > > > > >>>>>> size. This way, fetch size is really for performance > > > optimization, > > > > > >>> i.e. > > > > > >>>>> in > > > > > >>>>>> the common case, we will not return more bytes than fetch > > size, > > > > but > > > > > >>> if > > > > > >>>>>> there is a large message, we will return more bytes than the > > > > > >>> specified > > > > > >>>>>> fetch size. In practice, large messages are rare. So, it > > > shouldn't > > > > > >>>>> increase > > > > > >>>>>> the memory consumption on the client too much. > > > > > >>>>>> > > > > > >>>>>> Jun > > > > > >>>>>> > > > > > >>>>>> On Sat, May 21, 2016 at 3:34 AM, Onur Karaman < > > > > > >>>>>> onurkaraman.apa...@gmail.com> > > > > > >>>>>> wrote: > > > > > >>>>>> > > > > > >>>>>>> Hey everyone. So I started doing some tests on the new > > > > > >>>>>> consumer/coordinator > > > > > >>>>>>> to see if it could handle more strenuous use cases like > > > mirroring > > > > > >>>>>> clusters > > > > > >>>>>>> with thousands of topics and thought I'd share whatever I > > have > > > so > > > > > >>>> far. > > > > > >>>>>>> > > > > > >>>>>>> The scalability limit: the amount of group metadata we can > > fit > > > > > >> into > > > > > >>>> one > > > > > >>>>>>> message > > > > > >>>>>>> > > > > > >>>>>>> Some background: > > > > > >>>>>>> Client-side assignment is implemented in two phases > > > > > >>>>>>> 1. a PreparingRebalance phase that identifies members of > the > > > > > >> group > > > > > >>>> and > > > > > >>>>>>> aggregates member subscriptions. > > > > > >>>>>>> 2. an AwaitingSync phase that waits for the group leader to > > > > > >> decide > > > > > >>>>> member > > > > > >>>>>>> assignments based on the member subscriptions across the > > group. > > > > > >>>>>>> - The leader announces this decision with a > > SyncGroupRequest. > > > > > >> The > > > > > >>>>>>> GroupCoordinator handles SyncGroupRequests by appending all > > > group > > > > > >>>> state > > > > > >>>>>>> into a single message under the __consumer_offsets topic. > > This > > > > > >>>> message > > > > > >>>>> is > > > > > >>>>>>> keyed on the group id and contains each member subscription > > as > > > > > >> well > > > > > >>>> as > > > > > >>>>>> the > > > > > >>>>>>> decided assignment for each member. > > > > > >>>>>>> > > > > > >>>>>>> The environment: > > > > > >>>>>>> - one broker > > > > > >>>>>>> - one __consumer_offsets partition > > > > > >>>>>>> - offsets.topic.compression.codec=1 // this is gzip > > > > > >>>>>>> - broker has my pending KAFKA-3718 patch that actually > makes > > > use > > > > > >> of > > > > > >>>>>>> offsets.topic.compression.codec: > > > > > >>>>>> https://github.com/apache/kafka/pull/1394 > > > > > >>>>>>> - around 3000 topics. This is an actual subset of topics > from > > > one > > > > > >>> of > > > > > >>>>> our > > > > > >>>>>>> clusters. > > > > > >>>>>>> - topics have 8 partitions > > > > > >>>>>>> - topics are 25 characters long on average > > > > > >>>>>>> - one group with a varying number of consumers each > hardcoded > > > > > >> with > > > > > >>>> all > > > > > >>>>>> the > > > > > >>>>>>> topics just to make the tests more consistent. wildcarding > > with > > > > > >> .* > > > > > >>>>> should > > > > > >>>>>>> have the same effect once the subscription hits the > > coordinator > > > > > >> as > > > > > >>>> the > > > > > >>>>>>> subscription has already been fully expanded out to the > list > > of > > > > > >>>> topics > > > > > >>>>> by > > > > > >>>>>>> the consumers. > > > > > >>>>>>> - I added some log messages to Log.scala to print out the > > > message > > > > > >>>> sizes > > > > > >>>>>>> after compression > > > > > >>>>>>> - there are no producers at all and auto commits are > > disabled. > > > > > >> The > > > > > >>>> only > > > > > >>>>>>> topic with messages getting added is the __consumer_offsets > > > topic > > > > > >>> and > > > > > >>>>>>> they're only from storing group metadata while processing > > > > > >>>>>>> SyncGroupRequests. > > > > > >>>>>>> > > > > > >>>>>>> Results: > > > > > >>>>>>> The results below show that we exceed the 1000012 byte > > > > > >>>>>>> KafkaConfig.messageMaxBytes limit relatively quickly > (between > > > > > >> 30-40 > > > > > >>>>>>> consumers): > > > > > >>>>>>> 1 consumer 54739 bytes > > > > > >>>>>>> 5 consumers 261524 bytes > > > > > >>>>>>> 10 consumers 459804 bytes > > > > > >>>>>>> 20 consumers 702499 bytes > > > > > >>>>>>> 30 consumers 930525 bytes > > > > > >>>>>>> 40 consumers 1115657 bytes * the tipping point > > > > > >>>>>>> 50 consumers 1363112 bytes > > > > > >>>>>>> 60 consumers 1598621 bytes > > > > > >>>>>>> 70 consumers 1837359 bytes > > > > > >>>>>>> 80 consumers 2066934 bytes > > > > > >>>>>>> 90 consumers 2310970 bytes > > > > > >>>>>>> 100 consumers 2542735 bytes > > > > > >>>>>>> > > > > > >>>>>>> Note that the growth itself is pretty gradual. Plotting the > > > > > >> points > > > > > >>>>> makes > > > > > >>>>>> it > > > > > >>>>>>> look roughly linear w.r.t the number of consumers: > > > > > >>>>>>> > > > > > >>>>>>> > > > > > >>>>>> > > > > > >>>>> > > > > > >>>> > > > > > >>> > > > > > >> > > > > > > > > > > > > > > > https://www.wolframalpha.com/input/?i=(1,+54739),+(5,+261524),+(10,+459804),+(20,+702499),+(30,+930525),+(40,+1115657),+(50,+1363112),+(60,+1598621),+(70,+1837359),+(80,+2066934),+(90,+2310970),+(100,+2542735) > > > > > >>>>>>> > > > > > >>>>>>> Also note that these numbers aren't averages or medians or > > > > > >> anything > > > > > >>>>> like > > > > > >>>>>>> that. It's just the byte size from a given run. I did run > > them > > > a > > > > > >>> few > > > > > >>>>>> times > > > > > >>>>>>> and saw similar results. > > > > > >>>>>>> > > > > > >>>>>>> Impact: > > > > > >>>>>>> Even after adding gzip to the __consumer_offsets topic with > > my > > > > > >>>> pending > > > > > >>>>>>> KAFKA-3718 patch, the AwaitingSync phase of the group fails > > > with > > > > > >>>>>>> RecordTooLargeException. This means the combined size of > each > > > > > >>>> member's > > > > > >>>>>>> subscriptions and assignments exceeded the > > > > > >>>> KafkaConfig.messageMaxBytes > > > > > >>>>> of > > > > > >>>>>>> 1000012 bytes. The group ends up dying. > > > > > >>>>>>> > > > > > >>>>>>> Options: > > > > > >>>>>>> 1. Config change: reduce the number of consumers in the > > group. > > > > > >> This > > > > > >>>>> isn't > > > > > >>>>>>> always a realistic answer in more strenuous use cases like > > > > > >>>> MirrorMaker > > > > > >>>>>>> clusters or for auditing. > > > > > >>>>>>> 2. Config change: split the group into smaller groups which > > > > > >>> together > > > > > >>>>> will > > > > > >>>>>>> get full coverage of the topics. This gives each group > > member a > > > > > >>>> smaller > > > > > >>>>>>> subscription.(ex: g1 has topics starting with a-m while g2 > > has > > > > > >>> topics > > > > > >>>>>>> starting ith n-z). This would be operationally painful to > > > manage. > > > > > >>>>>>> 3. Config change: split the topics among members of the > > group. > > > > > >>> Again > > > > > >>>>> this > > > > > >>>>>>> gives each group member a smaller subscription. This would > > also > > > > > >> be > > > > > >>>>>>> operationally painful to manage. > > > > > >>>>>>> 4. Config change: bump up KafkaConfig.messageMaxBytes (a > > > > > >>> topic-level > > > > > >>>>>>> config) and KafkaConfig.replicaFetchMaxBytes (a > broker-level > > > > > >>> config). > > > > > >>>>>>> Applying messageMaxBytes to just the __consumer_offsets > topic > > > > > >> seems > > > > > >>>>>>> relatively harmless, but bumping up the broker-level > > > > > >>>>> replicaFetchMaxBytes > > > > > >>>>>>> would probably need more attention. > > > > > >>>>>>> 5. Config change: try different compression codecs. Based > on > > 2 > > > > > >>>> minutes > > > > > >>>>> of > > > > > >>>>>>> googling, it seems like lz4 and snappy are faster than gzip > > but > > > > > >>> have > > > > > >>>>>> worse > > > > > >>>>>>> compression, so this probably won't help. > > > > > >>>>>>> 6. Implementation change: support sending the regex over > the > > > wire > > > > > >>>>> instead > > > > > >>>>>>> of the fully expanded topic subscriptions. I think people > > said > > > in > > > > > >>> the > > > > > >>>>>> past > > > > > >>>>>>> that different languages have subtle differences in regex, > so > > > > > >> this > > > > > >>>>>> doesn't > > > > > >>>>>>> play nicely with cross-language groups. > > > > > >>>>>>> 7. Implementation change: maybe we can reverse the mapping? > > > > > >> Instead > > > > > >>>> of > > > > > >>>>>>> mapping from member to subscriptions, we can map a > > subscription > > > > > >> to > > > > > >>> a > > > > > >>>>> list > > > > > >>>>>>> of members. > > > > > >>>>>>> 8. Implementation change: maybe we can try to break apart > the > > > > > >>>>>> subscription > > > > > >>>>>>> and assignments from the same SyncGroupRequest into > multiple > > > > > >>> records? > > > > > >>>>>> They > > > > > >>>>>>> can still go to the same message set and get appended > > together. > > > > > >>> This > > > > > >>>>> way > > > > > >>>>>>> the limit become the segment size, which shouldn't be a > > > problem. > > > > > >>> This > > > > > >>>>> can > > > > > >>>>>>> be tricky to get right because we're currently keying these > > > > > >>> messages > > > > > >>>> on > > > > > >>>>>> the > > > > > >>>>>>> group, so I think records from the same rebalance might > > > > > >>> accidentally > > > > > >>>>>>> compact one another, but my understanding of compaction > isn't > > > > > >> that > > > > > >>>>> great. > > > > > >>>>>>> > > > > > >>>>>>> Todo: > > > > > >>>>>>> It would be interesting to rerun the tests with no > > compression > > > > > >> just > > > > > >>>> to > > > > > >>>>>> see > > > > > >>>>>>> how much gzip is helping but it's getting late. Maybe > > tomorrow? > > > > > >>>>>>> > > > > > >>>>>>> - Onur > > > > > >>>>>>> > > > > > >>>>>> > > > > > >>>>> > > > > > >>>> > > > > > >>>> > > > > > >>>> > > > > > >>>> -- > > > > > >>>> -- Guozhang > > > > > >>>> > > > > > >>> > > > > > >> > > > > > >> > > > > > >> > > > > > >> -- > > > > > >> -- Guozhang > > > > > >> > > > > > > > > > > > > > > > > > > > >
